This policy page carries the ACM's standard for artifact review, version 1.1, dated 24 August 2020. It argues that an experimental result is not fully established until it can be independently reproduced, and it defines an artifact as a digital object that the authors created for the study or that the experiment generated, naming software systems, scripts used to run experiments, input datasets, raw data and scripts used to analyse results. Its useful contribution to a project is the vocabulary: the reproducibility package becomes a defined set of objects rather than a vague gesture at openness. The policy attaches three independent badges, artifacts available, artifacts evaluated, and results validated, and it distinguishes two levels of the middle badge, functional and reusable, defined by whether the artifacts are documented, consistent, complete and exercisable, and then whether they are well enough structured for reuse. The limits are that this review is optional for ACM venues, that it is a review standard rather than a study of reproducibility, and that it is written for computing broadly, so an HCI project must translate its categories into its own materials.

standard · 2020 · Open the source

This is the conference page that states what a CHI paper must contribute, and it is useful for a project because it shows how a contribution claim is assessed by reviewers rather than only by authors. The page says that every CHI paper must make an original research contribution, that papers are assessed on originality, correctness, novelty, importance and clarity of exposition, and that reviewers are directed to Wobbrock and Kientz for the types of contribution. For each type it lists the questions a paper should answer, including whether the paper motivates a real problem, whether it clearly reviews what is already known and the limits of that knowledge, and whether it describes the work in sufficient detail for others to reuse it. That set of questions is the most directly useful part for this module, because it turns positioning in the related work section into an explicit criterion. Two limits apply. The page governs one conference cycle and will be replaced, and it states expectations rather than evidence about how well they are met. It also carries no DOI, since it is a conference web page rather than a publication.

conference-page · 2027 · Open the source

Ancker and colleagues review experimental and focus-group research on graphics as a way of communicating quantitative health risks. Their abstract reports a finding that shapes the module's teaching: the graphic features that improve the accuracy of quantitative reasoning appear to differ from the features most likely to alter behaviour or intentions. Graphs that make the part-to-whole relationship visible, showing the numerator against the population at risk, appear to help readers attend to the denominator, while graphs that show only the numerator appear to inflate the perceived risk and can induce risk-averse behaviour. They also report that viewers often preferred design features, such as the use of colour, that were not the ones that improved reasoning. The review's limit is the state of the evidence when it was written: heterogeneous studies, small samples and few replications. Use it for the separation of comprehension from preference, and treat every further claim as provisional.

journal-article · 2006 · Open the source

Azuma wrote this as a survey and says so plainly: it presents no new results, and its contribution is to consolidate the field and give it a bibliography. Its lasting contribution is a definition that deliberately avoids naming hardware. A system is augmented reality when it combines real and virtual, is interactive in real time, and is registered in three dimensions. That third condition is doing the work, because it excludes a two dimensional overlay on live video and it excludes film, which is photorealistic but not interactive. The survey then describes six classes of application, including medical visualisation, maintenance and repair, annotation and military use, and it is honest that most were prototype demonstrations rather than evaluated products. For this material the most useful parts are the sections on registration error, the misalignment of virtual content with the real scene, and on sensing error, the uncertainty about where the head and camera actually are. Azuma calls these two of the biggest problems in the field, and the reason is that no amount of display quality compensates for them. The limits are its age and its method: it is a survey of work up to 1996, the hardware has changed completely, and the dated application claims should not be reused as current evidence.

journal-article · 1997 · Open the source

This short book argues that the work behind a qualitative study is rarely described, and that this is a gap rather than a sign that the work is trivial. Its central analogy is documentary film making: films were once presented as finished products, while the recent habit is to go behind the scenes and show how the final cut was made. The authors apply this to qualitative HCI, where there are many guides to a named method and few accounts of how to choose a method, adapt it to a setting, or handle the unexpected problems that arise once data collection begins. Their focus is semi-structured studies, which sit between ethnography and surveys and typically combine observation, interviews and systematic coding of the data. For a project, the book is the warrant for keeping a decision log and for treating a revised instrument as a version boundary rather than as a defect, which is exactly what this module asks for. The limits are that it addresses qualitative work rather than quantitative delivery, and that the full text sits behind a subscription, so only the publisher overview was read in this run.

book · 2016 · Open the source

This two-page editorial announces the registered report format at the journal Cortex and is the clearest short statement of what preregistration asks of an author. It argues that editorial decisions based on results reward questionable practices, cites a survey in which 67 per cent of researchers admitted selectively reporting experiments that produced desirable outcomes and 72 per cent admitted hunting for a p value, and proposes that a manuscript be reviewed before data collection. The mechanism matters for this module. A submission includes the hypotheses, procedure, analysis pipeline and an a priori power analysis, and acceptance in principle means the decision to publish is independent of the results. Chambers is explicit that this does not outlaw serendipity: post hoc analyses are welcome, and are simply distinguished from those planned in advance. That sentence is the module's licence to report an unexpected finding honestly. The limits are that it is an editorial rather than an empirical study, that it describes a policy at one journal, and that it names its own boundary, since not all modes of investigation are compatible with the format.

journal-article · 2013 · Open the source

Correll asks what obligations attach to the design and use of visualisations, beyond those that already bind scientists and engineers. Drawing on historical and contemporary examples, the paper addresses the moral components of visualisation, identifies areas of visualisation research that carry ethical dilemmas, and proposes a set of additional obligations for the people who design, build and research visualisations. Its argument is that a display has influence over how data are used to make decisions, and that this influence brings responsibilities that current methods rarely examine, from the framing of a result to the consequences of a display that readers treat as fact. The paper is a position piece, not an empirical study, and its contribution is to name the duties rather than to measure their neglect. It offers no checklist and no measurement, so it should be read as an argument about what designers owe, not as evidence about what they currently do. For the module it closes the evaluation unit by making the point that accuracy and honesty are different properties of a figure.

proceedings-article · 2019 · Open the source

This paper is useful because it replaces a slogan with a measurement at exactly the point where the slogan is most often used. The authors compared direct touch with a mouse on a tabletop display in two experiments, looking at both quantitative performance and subjective preference for unimanual and bimanual tasks. The result splits by task. For bimanual tasks, direct touch conferred an advantage, which is what the direct interaction argument predicts, since two hands can work at once without one device between them. For a single user doing single point interaction, the mouse was often the better instrument, because the hand does not have to reach across a large surface and pointing is more precise. Subjective preference did not always follow the performance data, which is itself a caution for design decisions taken on preference alone. The method is a controlled laboratory comparison, which is its strength and its limit: it isolates the effect of the input method, so it cannot say how the result survives a messy shared tabletop with several users. It remains the standard citation for the claim that direct input is task dependent rather than automatically superior.

conference-paper · 2007 · Open the source

Gigerenzer and colleagues review the evidence that doctors, patients, journalists and politicians routinely misread health statistics without noticing. Their abstract gives the worked example the module uses: many people are unaware that higher survival rates with cancer screening do not imply longer life, and that a 25 per cent reduction in the risk of dying from breast cancer can mean one fewer death in a thousand. They argue that this statistical illiteracy is common across audiences and is created by non-transparent framing, sometimes from a lack of understanding and sometimes by design, with serious consequences for health. The review's strength is that it collects evidence from several audiences rather than one, and its remedies are stated in terms of format rather than exhortation. It is a review and a position paper, and the module uses it as the clearest statement of why frequency formats and absolute risks change what readers understand.

journal-article · 2007 · Open the source

Mid-air interaction has a failure mode that a short demonstration hides, and this paper names and measures it. The authors call it the gorilla arm effect and observe that designers had been blaming fatigue for poor mid-air performance without a way to quantify it. Their contribution is a metric, Consumed Endurance, defined as the ratio of interaction time to the time the shoulder can sustain that contraction before it needs rest, derived from Rohmert's biomechanical formulation of endurance. They compute it from an off-the-shelf skeleton tracking system, which means the measure does not require electrodes or blood sampling, and they validate it against the Borg CR10 scale of perceived exertion. They then use it to compare designs. Holding the arm bent, on a plane midway between shoulder and waist, consumed the least endurance, and dwell selection consumed less than a press. They also derive an endurance efficient text entry layout. The practical value is that fatigue becomes predictable from posture, reach and duration before anyone touches the system. The limits are that the metric models the shoulder rather than the whole body, that it was validated against a self-report exertion scale, and that the interaction techniques it was tested on are now a decade old.

conference-paper · 2014 · Open the source

This is the paper to cite when a stereo display is justified by its specification alone. Its subject is the vergence and accommodation conflict. In natural viewing the eyes converge on an object and the lens accommodates to the same distance, but a conventional stereo display puts every object on one surface, so the lens must focus at the screen while the eyes converge at the depicted depth. The authors report four experiments using a display that presented focus cues correct or nearly correct for the depicted scene, and they describe the consequences of correcting the conflict: faster identification of a stereoscopic stimulus, better stereoacuity in a time limited task, less distortion in perceived depth, and less fatigue and discomfort. They also discuss the distortions that ordinary displays produce. The method is psychophysical, the manipulations are controlled, and the outcomes are perceptual, which makes this stronger evidence than most claims in the area. Its limit for a designer is that the corrective display was a laboratory instrument rather than a product, so the paper establishes the mechanism and the direction of the effect rather than a specification a product team can copy directly.

journal-article · 2008 · Open the source

Ishii and Ullmer argue that the graphical interface, for all its success, cut people off from the physical skills they already have. Their proposal is to couple digital information to graspable physical objects, to turn architectural surfaces into interfaces, and to use ambient media such as light, sound, airflow and water movement for background awareness. They illustrate the argument with three systems: the metaDESK, a table that reads the position of physical objects, the transBOARD, and the ambientROOM. The most teachable idea is the foreground and periphery distinction, that a person can be aware of information at the edge of attention and can pull it to the centre when it matters. The paper is candid about its own status. It says it is not a solution to any single problem, and that its aim is to propose a view of the interface and raise research questions. That is exactly how it should be read. It reports no controlled comparison, so it cannot support a claim that tangible interaction outperforms a graphical one, and treating it as such is the commonest misuse. Its influence is nonetheless real, and it is the reference against which later work on graspable and embodied interfaces positions itself.

conference-paper · 1997 · Open the source

Milgram and Kishino wrote this because researchers in the new field were using the same words for different systems. Their central device is the virtuality continuum, running from a purely real environment at one end to a purely virtual one at the other, with mixed reality as everything in between. Augmented reality is the case where an otherwise real environment is augmented with computer generated objects, and augmented virtuality is the converse case, where a virtual scene has real objects added to it. The paper then identifies six classes of hybrid display and argues that grouping them by whether they are video or graphics based produces inconsistent results. Its most useful contribution for design is the three dimensions it proposes instead: extent of world knowledge, reproduction fidelity, and extent of presence metaphor. Those three questions can be asked of any candidate system and they expose tradeoffs a single label hides. The limits are that the taxonomy is conceptual rather than empirical, that nothing in it was tested against user outcomes, and that the display technology it describes has changed almost entirely since 1994, so the categories have to be applied with judgement rather than copied.

journal-article · 1994 · Open the source

Munzner's textbook is the systematic framework the module uses to order its decisions. It teaches visualisation as three questions asked in a fixed sequence: what data the reader needs to see, why they need it, and only then how marks and channels can carry it. The book sets out the marks-and-channels model, the ranking of channels by the accuracy of the judgements they support, and the nested model that locates errors at the domain, task and encoding levels above the algorithm. That ordering is the reason this module begins with the commissioning conversation rather than with the chart. The book is a synthesis and a teaching text rather than a study, so it does not itself measure reader performance; its evidence comes from the perception experiments it cites. Read it for the framework and for the discipline of naming a task before drawing anything. Its limit is that it is written for analysts and students rather than for the practitioner who must answer to a client, which is the gap HCI3109 fills.

book · 2014 · Open the source

This is the book a student reaches for when the task is to make something physical behave, and its value in this material is the bridge it builds between a sensor and a behaviour. It works from electricity and circuits, through choosing and programming a microcontroller, to sensing movement, making movement and managing multiple inputs and outputs. That sequence is the skill set a tangible or physical prototype requires, and the book is written for a reader without a technical background, which is why it is the standard teaching text rather than a reference manual. Its stance is also useful. The authors treat the physical object and its behaviour as the design material, which is the same idea that the tangible interface literature argues for at a conceptual level, expressed here in terms of transducers and circuits. It is a technique text with a clear physical stance rather than a research monograph. Its limits are its age and its level. It predates the current generation of low cost boards, and it teaches technique rather than evaluation, so it can help a student build a prototype but cannot tell them whether the prototype supports a benefit claim.

book · 2004 · Open the source

Robinson's short paper is the origin of the term ecological fallacy. He showed that a correlation computed across areal units, such as states or districts, is not an estimate of the corresponding relationship between individuals, and that the two can differ markedly in size and even in sign. The argument matters to this module because a choropleth map invites exactly that error: a reader sees areas shaded by a rate and reads it as a statement about the people in those areas. Robinson's demonstration used census-type data and a correlation between two area-level variables, and the gap he described follows from how the units are constructed rather than from a mistake in the arithmetic. Use the paper as the module's authority for the claim that area-level visual patterns do not license individual-level conclusions, and pair it with the modifiable areal unit problem, which explains why the choice of boundary changes the correlation as well.

journal-article · 1950 · Open the source

This systematic review and meta-analysis asks a narrow question and answers it carefully. The authors searched six databases for studies that measured sickness after head-mounted display use with the Simulator Sickness Questionnaire, and fifty five studies met their criteria, covering just over three thousand participants. The pooled total score across all content was 28.00, which the authors note is high against recommended cut-off values, and gaming content scored highest at 34.26. Content type, the amount of visual stimulation, the style of locomotion and the length of exposure all influenced the score. Two findings can be turned into design rules: physically walking reduced sickness compared with controller based navigation, and longer exposure increased it, with one rollercoaster study in the set reporting that about two thirds of participants could not complete a fourteen minute session. The authors are honest about the weakness of their own evidence base. Almost all of it rests on self-report questionnaires, very few studies report objective physiological measures, and the terminology across the field is inconsistent, so the pooled figure should be read as a summary of what participants said rather than a physiological measurement.

journal-article · 2020 · Open the source

Sarikaya and colleagues examine how dashboards are actually used in practice rather than how they are defined in the literature. Their abstract reports that dashboards are one of the most common uses of data visualisation and that their design and contexts of use differ considerably from exploratory tools. Reviewing the surrounding literature and a body of dashboard examples and documentation, they construct a design space, identify major dashboard types, and characterise dashboards by their design goals, their levels of interaction and the practices around them. The paper's contribution is to show that a term used as if it named one thing covers a range of instruments with different purposes, owners and cadences, and it proposes research directions to support dashboard design and use. It is a literature review and design-space paper, not a study of dashboard outcomes. For the module it supplies the vocabulary for asking who owns a dashboard, what decision it serves and how it is governed.

journal-article · 2019 · Open the source

This is an opinion and survey article by the researcher most associated with presence research, and its value for teaching is that it separates the constructs the industry blurs. Immersion is defined as the technical capability of a system, the extent to which it lets a person perceive through natural sensorimotor contingencies, and it depends on tracking, field of view, resolution and latency. Presence is the subjective correlate, split into place illusion, the sense of being in the rendered place, and plausibility illusion, the sense that events there are really happening. The authors argue that presence is best evidenced by how people respond rather than by what they say, and they report that about ten per cent of participants ran out of a room when a virtual fire appeared, even though the fire looked unrealistic and they knew it was not real. The survey then works through applications in psychology, training, sport and cultural heritage, and it is candid that some areas rest on promising but inconclusive findings. It is a narrative review rather than a systematic one, so it is a guide to the field and to the authors' own position, not a meta-analysis.

journal-article · 2016 · Open the source

This comment is the first formal publication of the FAIR principles, a short and measured set of requirements for making research data reusable. It argues that the infrastructure supporting the reuse of scholarly data is inadequate, and that a diverse group of stakeholders, including academia, industry, funding agencies and publishers, has endorsed a concise set of principles: that data should be findable, accessible, interoperable and reusable. Its distinctive emphasis is machine-actionable reuse rather than human description alone, so metadata has to be structured for software as well as for people. For a project, the principles are most useful as a checklist that covers the materials and instruments as well as the data, and they answer the common objection that sharing is impossible by narrowing what sharing requires. The limit is that this is a principles document, not a study of their effect, and it does not say what to do when data cannot be shared at all, which is the situation many HCI projects face under a research ethics approval.

journal-article · 2016 · Open the source

This paper is a scoping review of a development that arrived after most user research teaching was written. It collects the recent proposals to replace human participants with large language models and other generative systems in research and in product development, a wave the authors count at more than a dozen reports and papers alongside several commercial products. The method is a thematic content analysis of those proposals, grouped by the argument each one makes for substitution. The motives it finds are lower cost, the protection of participants from potential harms and an apparent gain in the diversity of the collected data. Its argument is that the proposals conflict with three values that human participation serves: the representation of participants' interests, their inclusion and empowerment in the process, and the understanding a researcher builds through engagement with them. The paper is most useful for teaching because it states the case for simulated participants fairly before it answers it, so a student can weigh both the attraction and the objection. Its limits are that it is a conceptual and review paper rather than a measurement, that it argues from values as well as from evidence, and that only the abstract, the introduction and the opening of the background section were read here.

conference-paper · 2024 · Open the source

This is the field evidence for the module's point that an unheeded warning is evidence about the warning. Its argument is that the popular claim that security warnings are useless is too broad, and that measuring them shows both that some work and that their design decides how well. Its method is the study's strength: the authors used telemetry that Firefox and Chrome already collected, so they observed over twenty-five million warning impressions from real users browsing rather than from people recruited into a laboratory, which removes the usual demand characteristics of a warning study. The results cut against the easy story in one direction and support it in another. Users continued through only a tenth of Firefox's malware and phishing warnings and a quarter of Chrome's, so most people did stop for the phish, but behaviour varied sharply by warning, and users continued through 70.2 per cent of Chrome's secure-connection warnings. Two limits matter when citing it. The measure is a click-through rate, not whether a user who proceeded came to harm, so it is behaviour rather than safety consequence, and one browser's figure should not be read as all warnings. Only the abstract was read for this entry.

conference-paper · 2013 · Open the source

This is the best single book on security as a systems problem rather than a bag of algorithms, and the third edition is free to read chapter by chapter. Its argument is carried by its order. The third edition places a chapter on the opponent before any chapter on cryptography, and that ordering is the claim this module teaches: the choice of defence follows from an analysis of who the adversary is and what they want, so a design that defends everything equally defends nothing well. Its method is a wide synthesis rather than a systematic review, combining case studies of banking, payment, monitoring and command-and-control systems with research results and the author's own practice; the chapters on psychology and usability argue directly that security is something people must operate, and are the ones a designer should read first. For this module the most useful material is the classification of opponents in Chapter 2 and the treatment of unusable controls in Chapter 3. Its limits are those of any single-author survey of this size: it is stronger on some areas than others, the case studies are illustrative rather than controlled evidence, and the free edition circulates alongside a separate errata list. Note the page's two chapter sets; the earlier set is labelled with a different edition's titles.

book · 2020 · Open the source

This is a vendor's documentation rather than a study, and it is cited here for a platform rule rather than for a research finding, which is a distinction worth keeping. Its argument is operational: an application is expected to keep its main thread free, and the platform enforces that expectation with a deadline and a dialog. Its method is documentation of policy and diagnosis; the page states the trigger conditions, explains the trace files and debugging tools, and advises moving slow work and input and output off the main thread. For this module it turns an abstract scheduling point into a named and observable system behaviour. The single thread that draws and handles input is the thread the deadline applies to, so the freeze a person sees is the platform's response to that thread being busy, and the dialog offers to force quit the application. Its limits are the limits of vendor guidance. It describes one platform's policy rather than a measured user experience, the five-second figure is that vendor's design choice rather than a constant across systems, and the page is a dated snapshot of triggers that change as the platform changes. It also documents several triggers and tools that lie outside this module.

documentation · 2026 · Open the source

The paper argues that how developers debug in practice differs sharply from how they say they debug. Its method is mixed: the authors combined an online survey of 176 developers, an instrumented observation of 458 developers through a plug-in for the two popular IDEs Eclipse and IntelliJ, and follow-up interviews. The result is a gap between belief and behaviour. They report that developers did not run the debugger in 91 per cent of the observed sessions, and that only 132 of the 458 users, 28.8 per cent, started a debugging session at all. On the same data, 81.3 per cent of surveyed developers said they use the IDE debugger and 71.6 per cent reported using print statements. What is most useful for a course is that it is the honest counterweight to the instruction to use a debugger, and that it records what the tool is used for: line breakpoints were used by the vast majority, while the advanced kinds were unknown to most. The authors conclude that hands-on debugging experience should be strengthened in computer science teaching. Its limits are that it is observational, covers two IDEs over two months, draws a sample in which 84.1 per cent work in Java, and finds that self-report and instrumentation disagree.

conference-paper · 2018 · Open the source

This article is the standard reference for service blueprinting and the reason the technique appears in design courses that are otherwise about screens. The authors argue that services are fluid, dynamic, experiential and frequently co-produced in real time by customers, employees and technology, and therefore have few static physical properties. Methods developed for inventing and prototyping physical products, they say, do not work well for human and interactive services. Their answer is blueprinting, a technique grounded in the customer's experience that visualises a dynamic service process, including the parts the customer does not see, across five components: customer actions, on-stage employee actions, back-stage actions, support processes and physical evidence. Their working method is the authors' own consulting cases. Their own worked example is a one-night hotel stay, traced from the reservation through check-in and room service to check-out, with the support systems marked beneath each step; a second case, Yellow Transportation, describes a trucking firm that used blueprinting across sales, operations and customer service. For a first-year studio the most useful thing is the argument itself: a design's dependencies run past the screen into the work behind it. The limits are that it is a management article, the cases are the authors' own consulting work, and no outcomes are measured.

journal-article · 2008 · Open the source

Blandford, Furniss and Makri write for HCI researchers who need to plan, run and report a semi-structured qualitative study. The structure follows that work: purpose, planning, sampling and recruitment, data gathering, analysis, paradigms and strategies, reporting, and quality, with a checklist at the end of each part. The central claim is that qualitative research in HCI has its own standards, and that importing the criteria of a controlled experiment misjudges it in both directions. The authors are candid about the field's disputes, including the argument over what an ethnographic study is for: some hold that its only purpose is to inform system design, while others hold that describing the context is valuable on its own. That disagreement is useful to a student because it shows the criteria for good work are themselves contested, and this book is the clearest single place to see both positions set against each other. The limits are length and depth. It is a short synthesis rather than a methods textbook, it does not teach statistical analysis, and a student running a first study will still need a practical handbook beside it. Only the sample and the contents list were read here.

book · 2016 · Open the source

This is the definitive comparative study of attempts to replace the password, and the reason this module treats password failure as a design problem rather than as a user failing. Its argument is that the field has been asking the wrong question: the useful test is not whether a scheme is more secure than passwords but whether it is better across every property a real deployment needs. Its method is a structured expert evaluation rather than a controlled experiment, and the authors say so. They define twenty-five benefits grouped under usability, deployability and security, then rate thirty-five schemes against them, covering password managers, federated login, graphical and cognitive schemes, one-time passwords, hardware tokens and biometrics, with one representative scheme per category explained in detail and the remainder in a companion technical report. The headline finding is that no scheme comes close to providing all the benefits and that none even retains the full set that legacy passwords already provide, so the field sits at a point where almost every scheme is better on some criteria and worse on others. Its limit is that the ratings are judgements made against a defined framework, not measurements, and deployability is the benefit most likely to have changed since 2012.

conference-paper · 2012 · Open the source

This code is the module's source for what ethics approval is for and what it covers. It sets out four principles for research with human participants: respect for the autonomy, privacy and dignity of individuals and communities; scientific integrity; social responsibility; and maximising benefit while minimising harm. Around those principles it develops the practical apparatus a project must satisfy, covering valid consent, risk, confidentiality, debriefing and the principles of best practice in ethics review, and it addresses student research directly. Two features make it useful for teaching. The first is that it states the principles at a level of abstraction intended to catch any case, so a team must reason about its own project rather than look up a rule. The second is that it places responsibility for an adequate protocol with the researcher. Its limits matter too. It is a professional code rather than law, it applies specifically to psychological research, and a project run under a university's own policy will be judged against that policy. The four principles were confirmed in this run from the code's text.

guideline · 2021 · Open the source

Buxton argues that sketching is a distinct design activity rather than a poor version of prototyping, and the book is the standard statement of that position. The subtitle carries the argument: an early stage exists to get the right design, and later stages exist to get the design right, so a team that skips the first will refine the wrong concept very well. His stated contrast is that a sketch is quick, disposable and deliberately ambiguous, which invites change, while a polished rendering invites approval and is therefore harder to abandon. Its method is a case library rather than an argument from data, and its load-bearing case is a time-trial racing bicycle built for Lance Armstrong: the designer, Michael Sagan, generated thirty or more sketches a day early in the process, and later went to extra trouble to composite a photorealistic rendering with a hand sketch so that reviewers would still suggest changes. That case is the most useful thing in the book for a studio, because it shows the mechanism rather than the effect. Its limits are a trade book's: the argument rests on the author's own practice and illustrations, offering almost no experimental evidence, and this note rests on the publisher page, the contents list and one openly posted chapter, not a reading of the whole text.

book · 2007 · Open the source

Caine asks how a researcher should decide a sample size, then measures what the CHI community actually does. The method has two parts: a review of the available approaches, namely prospective power analysis, saturation, cost and return-on-investment reasoning, and guidelines such as local standards, followed by an analysis of every manuscript published at CHI in 2014, from which sample sizes and methods were extracted. Sample sizes that year ran from one to nine hundred and sixteen thousand, the most common single value was twelve, and the figure varied with study setting and method. Her recommendations are the most useful part for teaching: choose the method that matches the analysis, using a power analysis for statistical work and saturation for qualitative work; always report the number and how it was chosen; include demographic detail; and state cost and access constraints rather than conceal them. She names the sample-size fallacy, the habit of treating a small number as a reason to reject a study without qualification, and warns that relying on a local standard in isolation is not best practice. The limits are that the survey covers one venue in one year, that CHI is not the whole field, and that many papers reported too little for a reader to judge their reasoning.

conference-paper · 2016 · Open the source

Card, English and Burr compare four ways of selecting text on a screen: the mouse, a rate-controlled isometric joystick, step keys and text keys. The paper is the primary source for the use of Fitts's law in interface design, a use the module carried until now only from later analytical accounts. Its argument is that the choice of a pointing device should be settled by measurement, and that the law predicts how long a selection takes. Its method is a controlled comparison: participants selected targets with each device while positioning time and errors were recorded, and the results were fitted to the law. The mouse was fastest and most accurate, and the variation in positioning time for the mouse and the joystick was accounted for by the law. What is most useful is that the paper turns a psychological model into a design instrument, and reports the mouse's fitted slope close to that of other eye-hand tasks, the first HCI instance of the claim that a device can approach the minimal achievable time. Its limits are that it is one small laboratory study, that the key devices were modelled by keystroke count rather than by the law, and that the hardware dates from 1978. Only the abstract could be read, since the full text is paywalled.

journal-article · 1978 · Open the source

Pro Git is the standard account of the tool most software teams depend on, and its full text is free. Its argument is that version control is not a filing cabinet for finished work but a way of recording decisions, so that work can be branched, merged, compared and undone. Its method is demonstration on a toy repository, not measurement: commands, diagrams and a worked conflict. What is most useful is the account of branches and merges. Some people call Git's branching model its killer feature, and the book says Git encourages workflows which branch and merge often, even several times in a day. A branch is cheap because it is a file holding the 40-character checksum of the commit it points to, so creating one writes 41 bytes. Section 3.2 works a merge conflict through: two branches have changed the footer of index.html differently, Git stops the merge and marks the file with conflict markers, and the book resolves it by keeping a little of each side before staging the file and committing. Its limits are that it is a tool manual rather than an argument, that it dates from 2014, and that it never measures what a conflict costs, which is why the module no longer claims that a small merge costs less than a late one.

book · 2014 · Open the source

Chapman and Milham argue that the persona method has significant methodological and practical difficulties. The claims they examine are that personas can reflect empirical data and serve as an information source for product teams. The first difficulty is representativeness: it is difficult to determine how many, if any, users are represented by a persona, and therefore difficult to know whether the persona is relevant to the intended users at all. The second is testability: a persona cannot be adequately verified or falsified, so it has no demonstrable validity. The third is organisational, and it is the part students find most useful: personas are likely to produce political conflict inside teams and to weaken a researcher's ability to settle questions with data, because the attributes of a persona can be adjusted until they support the decision a team already wants to take. Their method is argument, not measurement. Its limits are those of an argument rather than a study: three pages, no data of its own, and no engagement with the specific evidence its proponents cite, so it is a critique rather than a refutation. What it establishes is the boundary of what a persona can support.

conference-paper · 2006 · Open the source

Chase and Simon asked how much of a chess position a player takes in at a glance, and whether skill changes the answer. The method is a small experiment: they tested three players, who ranged in strength from master to novice, on two tasks. In a perception task the player reproduced a position in plain view, and in a recall task from de Groot the player reproduced one after seeing it for five seconds. Pauses during reconstruction were used to segment it into perceptual structures, or chunks, and their size and number were analysed against chess skill. The well-known result, reported in the paper and repeated since, is that the strongest players rebuilt real game positions far more accurately than the weakest, at roughly four fifths of the pieces against about a third, while the advantage nearly disappeared when the same pieces were scattered at random. What is most useful is the mechanism: expertise works through a store of familiar configurations, so one glance loads whole patterns rather than single pieces, which explains why skill does not transfer to an unfamiliar domain. The limits are the very small sample of three players and the use of a static board rather than an interactive task. The abstract and metadata were read for this note; the full text is paywalled.

journal-article · 1973 · Open the source

This study is the module's current answer to a practical question: if the evidence still favours testing early and rough, why does so little studio work look like that? The authors interviewed ten UX/UI designers and design educators in North America, analysed the transcripts thematically, and report four pressures. Digital prototypes are easier to share and to comment on asynchronously, which matters in hybrid work. All-in-one tools such as Figma pull sketching, wireframing and polished mock-ups into one file. Design systems and reusable components make a finished-looking screen quicker to produce than a paper one, which erodes the speed advantage low-fidelity methods were said to have. Clients and internal stakeholders who are not designers ask for polish, and judge a sketch by its appearance rather than its behaviour. The paper also reads the comparison literature together and reports that the empirical studies generally have not found significant differences between low and high fidelity, while noting nuanced differences; the studies it cites include Walker, Takayama and Landay (2002), which this module also carries. The most useful point is that none of this is an evidential argument: the method lost its advocates' time, not its support. The limits are ten interviews, one region and fieldwork completed late in 2022, which the authors present as indicative rather than settled.

conference-paper · 2024 · Open the source

Cowan's target article is the source for the capacity figure this module uses, and for the range it reports alongside the figure. Its argument is that the older figure of seven was meant as a rough estimate and a rhetorical device rather than a measured limit, that others had suggested a limit of three to five chunks, and that the evidence he reviews supports a single central capacity limit averaging about four chunks. The most useful part for a designer is the conditionality in the second half of the argument: the limit is observable only under stated conditions, such as when information overload stops items being recoded into larger chunks. That reading stops the number becoming a slogan. The design consequence is that a screen requiring several items to be held and compared at once will fail for most people. The method is a review and synthesis of existing experiments, published as a target article with peer commentary rather than a single new study, which is why it reports a range and not one value. Its limits are that it contains nothing about interfaces, so the move to design is an inference, and that the argument is contested by the commentators published with it. A student can read the abstract and the opening argument.

journal-article · 2001 · Open the source

Dourish argues that HCI has adopted a habit of judging ethnographic and other field studies by the design implications they yield, and that this criterion is not the best way to measure such work. The concern is that confusion attends both ethnographic practice and the criteria by which it should be evaluated: when the demand for design implications becomes the test, it may fail to capture the value of an ethnographic investigation at all. The paper is a conceptual and methodological argument rather than an empirical study, and it works through the nature of ethnographic inquiry and the ways HCI has framed the relationship between fieldwork and design. It is most useful here as the primary statement of the position that describing a context well is a contribution in its own right, which this module sets against the opposing view that fieldwork must end in design consequences. Its limits are that it argues from reasoning and cases rather than new data, that only the abstract could be read here, and that it does not resolve how the two positions should be reconciled; the field has continued to debate the question since.

conference-paper · 2006 · Open the source

This is the strongest evidence in the module for producing several ideas before refining one. The authors report an experiment built on a problem they name: iteration helps, but it can also cause fixation, a designer improving one option without examining the others. Their method was a between-subjects study in which thirty-three participants designed a magazine banner advertisement. Everyone produced five prototypes and a final advertisement, and everyone received descriptive critique; only the order differed. The serial group was criticised after each prototype, while the parallel group produced three prototypes before any feedback, then two more, then the final design. The parallel condition performed better on every reported measure: click-through rates in a live advertising campaign, time on the client's site, and ratings by the client and advertising professionals. Independent raters found the parallel sets more diverse, and nearly half of the serial participants described negative reactions to critique where no parallel participant did. The most useful part is the explanation: the authors trace the gain to the comparison a designer can make between prototypes held at once, which one refining a single prototype cannot make. The limits matter: novices, an advertising task, short sessions, click-through as a proxy for attention, and few replications in the field.

journal-article · 2010 · Open the source

Think Python is written for people who have never programmed. Its argument is that programming is a way of thinking, and that a beginner should meet small complete programs rather than abstract rules. Its method is one runnable program per idea, with the reader expected to type it and see it work, moving from variables and statements through functions, conditionals, strings, lists, dictionaries, tuples, files and databases, and then to classes, methods, inheritance and objects. What is most useful here are two worked cases. Chapter 9 defines a reference, names aliasing where two names reach one list, and shows that passing a list to a function aliases it a second time. Chapter 10 sets two functions the same search task on a list of 113,783 words, and reports the list version at more than a minute against the dictionary version at under one hundredth of a second. Chapter 13 places files and databases mid-book, and testing arrives unevenly, with doctest from Chapter 7 and unittest only in Chapter 18. Its limits are that it teaches Python only, says little about interfaces or asynchronous code, and its timings are one machine's.

book · 2024 · Open the source

The study asks how reading speed and line length affect comprehension, reading rate and scrolling when reading from a screen. Participants read at a normal speed and a fast speed at several line lengths, and the authors measured comprehension, reading rate, and the pauses and movements of scrolling. Two findings appear in the abstract. Reading fast reduced overall comprehension, though it did not change the kind of information recalled. A medium line length of 55 characters per line supported effective reading at both speeds, producing the highest level of comprehension and being read faster than short lines. The authors conclude that effective reading can only be defined in relation to the aim of the task, which may favour either speed or accuracy. The study is used here because it disciplines a folklore rule and because it shows what a real finding looks like: a range and a trade off rather than a constant. Its limits are severe. It is a single study from 2001, in one language, on the screens of that period, and the abstract reports no effect sizes or sample details. The full text was not accessible to me, so this note describes the abstract rather than the method.

journal-article · 2001 · Open the source

Faulkner sets out to test, rather than repeat, the assumption that five participants are enough. Sixty users were tested, and random sets of five were drawn from the whole to see what a typical small test would have found. The result is a spread rather than a figure: some sets of five uncovered ninety-nine per cent of the problems and others only fifty-five per cent. At ten users the worst set rose to eighty per cent, and at twenty users to ninety-five per cent. The paper is most useful because it makes the variability itself the finding. The same sample size can look excellent or poor depending on which participants you recruit and what problems the interface contains, so the number should follow the problem set and the purpose of the test rather than a constant. Its limits are those of a single study on a single system, where the problem set was finite and known to the analysts. It also measures problem discovery rather than the depth of understanding a qualitative study seeks, so it is not a licence to substitute counting for interpretation.

journal-article · 2003 · Open the source

The Humble Dialog Box made an argument that is now standard in interface programming: do not put logic in the interface class. Feathers describes the problems that arise when interaction logic is written directly in a dialog box, including bugs that cannot be reproduced in a test harness and duplication that is hard to see because the code is busy with low-level component calls. His worked example is a program that chains audio filters, where he writes the smart object ChainComposer first, testing it against a mock view, and the dialog ComposeChainDialog last, so that its handlers only delegate. His method is a worked example and its counter-example, written in 2002 by a practitioner and illustrated with unit tests in CppUnitLite against a stand-in view, not an experiment. His counter-example is a tax calculator whose view can reach getters and setters, which lets logic accumulate in the dialog until, in his words, you would have to hire a lawyer to find it. The article adds one further point: once the logic sits in the tested object, it can be refactored however the developer needs. Its value is the design rule and a concrete method for reaching it. Its limits are that it is a 2002 practitioner essay with no empirical evaluation, and its examples use desktop toolkits.

article · 2002 · Open the source

Fitts's paper is where the index of difficulty and the notion of a human information-processing rate come from, and it remains the primary reference for every pointing model used in interface design. Its argument is that the average duration of a response is proportional to the minimum information per response, so the rate of performance is approximately constant over a wide range of movement amplitudes and tolerance limits; that is what licenses the reading of the result as a channel capacity. Its method is laboratory experiment: reciprocal tapping with a light and a heavy stylus and disc and pin transfer, with movement time measured against the ratio of amplitude to width. Two cautions attach to this note. In this run only the publisher's record and abstract could be read, so the conditions, means and regression quoted in the narrative still come from MacKenzie (1992), which reproduces the one-ounce tapping data in full. And the paper's units are inches and ounces, so any use of its coefficients in pixel-based design is an extrapolation, as the module states.

journal-article · 1954 · Open the source

The paper argues that the evidence for test-driven development is usually put the wrong way round: what improves code may be the size of the steps rather than the order of writing. Its method is an analysis of professional development sessions rather than a controlled trial. The authors analysed 82 data points from 39 professionals, each one capturing the process used while performing a specific development task, and built regression models relating four process characteristics to quality and productivity, separating sequencing, whether the test or the production code is written first, from granularity, uniformity and refactoring effort. What is most useful is the finding, uncomfortable for conventional advice: quality and productivity were associated with fine-grained, uniform steps, while sequencing had no important influence. Refactoring effort was negatively associated with both outcomes, which the authors explain cautiously as possibly mixed refactoring. Their conclusion is that the benefits claimed for test-driven development may come from small, steady steps rather than from the test-first rule, and they ask other researchers to attempt replication in varying contexts. Its limits are a small professional sample, industry settings, and quality reduced to functional correctness.

journal-article · 2017 · Open the source

Gibson's book is where the concept of affordance comes from, and the chapter this module uses is the theory of affordances. He argues that the environment offers possibilities for action, and that these possibilities are real and relational: a surface affords standing to a creature of a certain size and gait, not in the abstract. Perception, in his account, is the direct pickup of that information rather than an inference built from impoverished sensations. The method is theoretical synthesis. Gibson assembles findings from experimental psychology and from observation of vision in natural settings into a single ecological account, and argues for it rather than reporting one trial. That is what makes the book teachable, and also what dates it. The most useful part is the relational definition, because it stops a student equating affordance with a visual cue, and it explains why the design sense of the word, which Norman introduced later, departs from Gibson's. The limits are real. The direct-perception claim is disputed within psychology, long passages on natural vision have no bearing on interface design, and the 1979 edition is out of print, so the stable link is to a later reissue.

book · 1979 · Open the source

This is the most complete practical survey of user research techniques in one volume, and its structure follows the work: why research fits product development, then the techniques, then the reporting of results. The techniques chapters are the ones used in this module, covering research planning, recruiting and interviewing, focus groups, object-based methods, field visits and observation, diary studies, usability tests and surveys. Each is written as a process with its own preparation and pitfalls, and the material on recruiting and on note taking during a field visit is unusually concrete. The book's argument is that research should be balanced against the needs and the schedule of the organisation, and that a decision made without it is a decision made on assumption. Its limits are that it is a practitioner's guide rather than a research monograph, that it cites little primary evidence, and that its worked examples are commercial products rather than studies. The publisher page for this edition lists two authors while the standard citation names three; the Berkeley record resolves the discrepancy in favour of three, and only the shop page and contents list were read here.

book · 2012 · Open the source

This is a vendor engineering account of one browser's move to post-quantum cryptography, and the source for the module's claim that a security upgrade carries a latency cost. It is a primary description of a deployment rather than an independent study, and it is cited here for the mechanism and for one measured figure, not for a general finding about user experience. Its argument is that migrating key exchange is urgent while migrating authentication is not, because an adversary can record encrypted traffic today and decrypt it later, an attack the post names store-now-decrypt-later. Its method is the engineering report: it describes the hybrid key exchange that Chrome enabled by default for TLS 1.3 and QUIC in 2024, which combines a classical exchange with a post-quantum one. The two facts the module uses are size and cost. A post-quantum key share is around one kilobyte per peer against thirty-two bytes for the classical one, an increase of more than thirty times, and the larger first packet split in two, which the authors report added about four per cent to the median handshake time on desktop. Its limits are the limits of vendor reporting: one browser, one platform, a self-reported figure, and an account written before the final standards were published.

documentation · 2024 · Open the source

This is the primary definition of the metric set that the performance material is organised around. It defines the Core Web Vitals as three field-measurable metrics covering loading, interactivity and visual stability: Largest Contentful Paint, Interaction to Next Paint and Cumulative Layout Shift. It gives the thresholds to design against, being 2.5 seconds, 200 milliseconds and 0.1, judged at the 75th percentile of page loads on mobile and desktop separately, and it explains why the percentile rather than the average is the test. Its method is field measurement: the page names the Chrome User Experience Report, which collects anonymised real-user data from the browser, as the source of the numbers, and it documents a metric lifecycle of experimental, pending and stable, recording that Interaction to Next Paint was developed to address runtime performance more comprehensively than First Input Delay. The companion article on layout shift names the causes, including images and videos of unknown dimensions, fonts that render at a different size from their fallback, and third-party widgets that resize themselves. Its limits are real. This is one vendor's guidance, published by the company whose browser supplies the field data, and the deeper justification for each threshold sits in an article this run did not read. The metrics measure what a browser can instrument, not whether a page made sense.

practitioner-guidance · 2024 · Open the source

This is the free probability text the programme names, and it is well matched to a first-year interaction student, assuming some calculus but no measure theory. It opens with simulation and then develops discrete and continuous distributions, combinatorics, conditional probability, expectation and variance, sums of random variables, the law of large numbers, the central limit theorem, generating functions, Markov chains and random walks. The central limit theorem chapter is the one that licenses the standard-error arithmetic used to measure interaction, and the chapter on expectation is the one behind the reading of failure rates and latency. Its most useful feature is the habit of simulating before deriving: each chapter carries programs and exercises a student can run and check against the algebra. Its limits are that the examples are drawn from games, cards and classical probability problems rather than from interface data, and that the freely distributed text is a fixed edition, so its conventional topics are stable but it does not cover modern statistical practice.

book · 2003 · Open the source

This document argues that accessibility overlays, which are third party scripts added to a site in order to improve its accessibility, sit in the wrong place in the technology stack and cannot deliver conformance. Its case has three parts. Widgets that let a reader enlarge text or change contrast duplicate features the operating system already provides, and would be of no use on any other site. Automated repair applied at run time is unreliable for the things that matter most, including text alternatives, form labels and error handling, focus control and keyboard access; it also breaks down for component based interfaces that change state independently of the overlay, and it cannot repair content held in documents, canvas or media. And the claim of compliance fails on definition, because conformance means satisfying every success criterion rather than reducing a count of errors. It adds a privacy argument about detecting assistive technology on a reader's device. The document is a position statement maintained in a public repository with more than a thousand named signatories, and it states no publication date. It is advocacy within a commercial dispute rather than a controlled evaluation, so it should be read beside the peer reviewed evaluations that test overlay behaviour directly. Quoted alone, it is the practitioner side of an open argument.

position-statement · n.d. · Open the source

This is the paper to give a student who asks why HCI researchers disagree while studying the same technology. Harrison, Tatar and Sengers argue that the field contains three paradigms, each with its own account of what interaction is and its own test of truth. Human factors treats interaction as a coupling of person and machine and judges a repair by whether it worked. Classical cognitivism treats mind and computer as coupled information processors and aims at a general, predictive model, and GOMS sits in that tradition. The phenomenologically situated paradigm treats interaction as meaning making, with the setting carrying part of the explanation. The method is argument. The paper is a position piece, written for the alt.chi session at CHI 2007, that illustrates the three readings with one worked case, a 1960s United States Air Force cockpit warning system, and contrasts the paradigms in two tables. That case is the paper's own cartoon of the positions rather than a study, and the paper names no source for the performance claim it reports about familiar voices in the cockpit, so the module uses the case as an illustration and not as evidence. The value of the paper is taxonomic rather than empirical, and its main limit is that its claims rest on the authors' reading of the field.

conference-paper · 2007 · Open the source

This is the clearest published test of what an accessibility overlay actually changes, and it is the reason the automation question in this module is presented as contested rather than closed. The authors built two websites with different layouts and planted a range of known WCAG violations in them, then augmented the sites with three AI-powered overlays and checked the results by automated and manual methods across several devices and platforms. The finding that matters for teaching is two-sided. The overlays did correct some errors that have simple solutions, including missing alternative text and unlabelled form fields, so the claim that they change nothing is not supported. Serious errors remained, and the overlays reported pages as fully compliant when they were not, so the claim that they deliver conformance is not supported either. The authors conclude that automated overlays alone cannot remove the legal risk of inaccessibility, and that manual testing and repair are still needed for most errors. Its limits should be stated with it. Two constructed sites measure a mechanism rather than the frequency of failure in the wild, and the abstract reports no testing with disabled users. A student can use it as evidence about capability, not about harm.

conference-paper · 2026 · Open the source

Eloquent JavaScript is a book-length introduction for people who can already program a little. Its argument is that JavaScript repays proper study rather than grudging use: closures, objects and prototypes, the meaning of this, modules, and the asynchronous model are built from first principles instead of by analogy to another language. Its method is exposition with runnable code, carried by complete projects, so a robot, a platform game and a pixel editor are built in full before the reader moves on. This fourth edition, published in 2024 and updated to ECMAScript 2024, is the edition the module now cites; the third edition, cited in earlier loops, is still online at a separate address. What is most useful for interaction work is Chapter 11, Asynchronous Programming, which carries the callback, the promise that stands for a value still to come, and the await keyword. Its callback example reads one file and then reads a second inside the callback of the first. Its limits are those of any single-language book: it teaches JavaScript rather than programming, it covers neither testing nor version control, and a free online text can change under a citation.

book · 2024 · Open the source

Henrich, Heine and Norenzayan argue that a large share of published claims about human psychology rests on samples from Western, educated, industrialised, rich and democratic societies, and that researchers have treated those participants as a neutral standard. The method is a review of the comparative database: the authors assemble large cross-cultural experiments, smaller two-population comparisons and meta-analyses across visual perception, fairness, cooperation, spatial reasoning, moral reasoning, self-concepts and related motivations, and read them as a series of contrasts against small-scale, non-Western and non-student populations. Their finding is not that any single result is wrong, but that the usual basis for calling one universal is missing, because WEIRD samples are frequent outliers. For user research the argument matters twice over: it is the strongest available statement of what convenience sampling costs, and it supplies the vocabulary for asking whether a sample can carry the claim made from it. The limits are that the review concerns behavioural science experiments rather than product research, so the transfer is an analogy, and that its comparisons often set one project against another rather than holding conditions constant.

journal-article · 2010 · Open the source

This paper makes the argument that gives the module its contested question. Herley sets out to explain why users ignore security advice and rejects the usual explanation that they are lazy or ignorant. His claim is economic: the advice offers to shield a user from the direct costs of attack while loading that user with indirect costs, and because the direct costs are usually small the bargain is often a poor one for the individual. It is worth noting that this is also an argument about externalities, since the costs of an attack fall on others while the cost of compliance falls on the user, which is why the paper's title is a refusal rather than a complaint. The method is reasoning from economic models of cost and benefit, supported by worked examples rather than new measurement. He examines three areas, password rules, phishing-site identification and certificate warnings, and finds in each that the advice is complex and growing while its benefit is speculative or moot, and that where benefit can be estimated the burden of following the advice exceeds the losses it prevents. That method is the main limit and should be stated whenever the paper is cited: it models users as weighing costs and it follows no one as far as harm.

conference-paper · 2009 · Open the source

Hick's paper is where the claim that a person behaves like a channel with a roughly fixed information rate begins, and it is the companion to Fitts's law in any account of interaction as information processing. The abstract, which is what could be read in this run, reports two experiments: a conventional choice reaction task with up to ten alternatives and a negligible proportion of errors, and a ten-choice task in which participants deliberately traded accuracy for speed. The principal finding is that the rate of gain of information is on average constant within the duration of one perceptual-motor act, of the order of five bits per second, and that the distribution of reaction times is related to the objective uncertainty about which response will be required. The idea that matters for interface work is the coupling of uncertainty to time, which is what makes the depth of a menu a cost rather than a preference. Its limits are those of a 1950s information-processing experiment: a simple laboratory task, a short report, and a rate that later work shows is not the same across conditions. Only the abstract and the bibliographic record were read.

journal-article · 1952 · Open the source

This is the standard a medical device manufacturer works to when usability is a matter of safety rather than preference, and it is this library's evidence that interface quality is regulated where a failure can injure someone. Its summary states that it specifies a process for a manufacturer to analyse, specify, develop and evaluate the usability of a device as it relates to safety, and that this usability engineering, or human factors engineering, process lets the manufacturer assess and mitigate risks associated with correct use and with use errors, that is, normal use. Usability here is a risk-management activity with a documented file behind it, not a satisfaction score, which is what makes the standard useful in this module. The method is a required engineering process, defined by an international standard and issued as a consolidated edition of 246 pages in June 2020. Two limits matter for teaching. The standard states its own boundary, that it can be used to identify but does not assess or mitigate risks associated with abnormal use, which is why the module reports the scope clause rather than paraphrasing it. Only the summary and introduction are free to read, and a standard is not legislation: it does not itself set out the legal duties that reference it.

standard · 2020 · Open the source

The second edition of part 11 moved usability from a property of a product to an outcome of use, and the catalogue page confirms that wording. The standard defines usability as the extent to which a system, product or service can be used by specified users to achieve specified goals with effectiveness, efficiency and satisfaction in a specified context of use. Each qualifier does work. Specified users excludes the general public. Specified goals excludes open-ended use. Context of use imports the setting, which is why a laboratory result is not automatically a usability finding. Effectiveness concerns completion and the quality of the outcome, efficiency concerns the resources expended, and satisfaction concerns attitudes. The method is consensus: an international standard, agreed through committee, that explains a concept rather than testing one. Its most useful part is the boundary it states for itself, that it defines the concept and describes no methods, so a usability claim has to borrow its method from elsewhere. Its limits are practical: the full text is purchased, so the catalogue page is the only part of the record that could be fetched here.

standard · 2018 · Open the source

Part 210 is the process standard that sits beside the definition in part 11. Its catalogue abstract describes requirements and recommendations for human-centred design principles and activities across the life cycle of interactive systems, and names the range of systems covered, from off-the-shelf software to process control, banking systems, websites, vending machines and mobile phones. The method is the specification of a process. The standard sets out principles, including that the design is based on an explicit understanding of users, tasks and environments, that users are involved throughout, and that the design is driven and refined by user-centred evaluation, and it describes four activities: understanding and specifying the context of use, specifying requirements, producing design solutions, and evaluating against requirements. The most useful parts are that sequence and the discipline it imposes, naming the context of use before designing. The user-experience definition this module uses covers the use or anticipated use of a system, product or service and lists emotion, belief, preference, comfort and accomplishment among the responses that count, which is why experience is broader than usability. Its limits are that it states it does not cover methods in detail, and that the text is purchased, so only the catalogue page was checked.

standard · 2019 · Open the source

Jansson and Smith named design fixation and tested it experimentally, and the later literature on parallel prototyping cites them for the phenomenon. They argue that a blind adherence to a set of ideas or concepts limits what conceptual design produces, and that the block is measurable rather than anecdotal. Their method is a series of four experiments with engineering design students. Each student received a design problem and was asked to produce as many designs as possible within forty-five minutes to an hour. Half of each group received the problem alone. The other half received it with a sample design. The two groups produced about the same number of ideas, but the group given the sample copied its features, and in one experiment it copied the example's faults as well. In the bicycle rack problem twenty-five students took part, and suction cups appeared in 54 per cent of the fixation group's designs against 6 per cent of the control group's. The most useful thing in the paper is that it measures the pull of one worked example, which is the failure a critique on a single prototype produces. Its limits are the sample and the setting: senior engineering students, small groups, short written exercises with no interface task, and only the first two experiments read here.

journal-article · 1991 · Open the source

Kramer, Guillory and Hancock report an experiment run inside Facebook. For one week in January 2012 the researchers altered the emotional composition of the News Feed seen by 689,003 users: one condition reduced positive content, another reduced negative content, and each had a matched control. Users exposed to less positive content posted fewer positive words and more negative ones, and users exposed to less negative content showed the reverse. The authors read this as evidence that emotion spreads through a network without direct interaction and without nonverbal cues. The paper matters here for its ethics rather than its psychology. Participants were not told they were taking part; the authors offered Facebook's Data Use Policy, accepted at sign-up, as consent, and Cornell University judged the work to be internal company research outside its review process. PNAS published it and later issued a correction and an Editorial Expression of Concern about informed consent and the chance to opt out. The limits are that the effects are small, that emotion is measured by word counts, and that the study is now cited far more for its consent failure than for its findings.

journal-article · 2014 · Open the source

This NIST special publication is the practical reference for the step from a state space to a test plan, which is the calculation this module needs a source for. It defines t-way combinatorial testing and pairwise testing, in which every pair of parameter values is covered by at least one test, and it defines the covering array that delivers that coverage, with each column a parameter and each row a test. Its worked examples are the useful part, and two of them are used in the module: a word processor with ten binary settings whose 1,024 configurations are covered for all three-way combinations by thirteen tests, and a five-component platform with 72 configurations covered for all pairs by ten tests. It also states the interaction rule, that most failures are caused by a single factor or a pair and progressively fewer by three or more, and reports that pairwise testing may still miss between 10 and 40 per cent of faults. Two limits matter for this module. The subject is software test suites rather than interactive forms, so the translation to screens is the module's own work; and the estimates of fault detection come from the authors' own empirical studies rather than from replication.

report · 2010 · Open the source

This is the module's primary networking text, now in its ninth edition, and the reason the material is taught top-down, from the application layer towards the physical layer. It makes an argument by structure: the application layer is presented before the layers beneath it, so a reader meets HTTP, caching and DNS before the wires that carry them. Its method is explanatory and worked-example led, with review questions, problems and laboratory exercises, assuming little programming or mathematics. For this module it is most useful as the map and as the source of three specific claims: the layer model, the separation of packet delay into processing, queuing, transmission and propagation, and the statement that bulk transfer is limited by bandwidth where a chain of small requests is limited by round trips. It remains the account of HTTP as a stateless protocol that needs cookies to recognise a returning user, and of caching and DNS as devices that remove round trips. The ninth edition adds a section on HTTP/3 and QUIC, which the eighth edition the module used before did not carry. Its limits are the limits of a textbook: it reports settled material rather than the primary evidence for contested claims, and the record verified here exposes a table of contents and an overview rather than the printed text.

book · 2025 · Open the source

Leveson and Turner's investigation is the standard account of the Therac-25, a computer-controlled medical linear accelerator involved in six massive radiation overdoses between June 1985 and January 1987. It argues that the accidents were system accidents, produced by interactions between software, hardware, staff and management, not by one faulty part. Its method is a post hoc investigation from lawsuits, regulatory correspondence and public documents; the authors warn that the exact doses cannot now be determined and that the manufacturer's records were unavailable. Three pieces of mathematics are taken from the case. The first is the logic of the counterexample: after one overdose the manufacturer claimed the redesign improved safety "by at least five orders of magnitude", and when the first Yakima patient's striped burn was reported it replied that the damage "could not have been produced by any malfunction of the Therac-25 or by any operator error"; further overdoses followed and the machine was recalled. The second is state and timing: an edit made within the eight seconds the bending magnets took to set appeared on the console but was never re-read, so the screen did not represent the machine. The third is reachability: a corrective action made every malfunction code suspend treatment rather than pause it. Its limit: a case study, not a controlled comparison.

journal-article · 1993 · Open the source

Lusinchi re-examines the most famous failure in the history of polling. In 1936 the Literary Digest mailed more than ten million straw ballots, drawn from its own subscribers, telephone directories and automobile registration lists, and nearly 2.4 million were returned, a response rate of roughly twenty-four per cent. The magazine forecast an easy victory for the Republican challenger, Alf Landon, and Roosevelt won in a landslide. The received explanation is sampling frame bias: the lists over-represented well-to-do households, who leaned Republican. The method here is a reanalysis of the published poll data and of a 1937 Gallup telephone survey of n = 693, using techniques devised for quota-controlled polls of the 1930s, and it argues that telephone and automobile owners as a group favoured Roosevelt, at 53.9 per cent, so nonresponse bias rather than the frame carried the error. The paper is the best teaching case for sampling because it separates two faults that are usually fused and shows which one mattered. Its limits are that it rests on one retrospective poll, settles a historical dispute by reanalysis rather than new measurement, and spends much of its length on the Gallup survey's own shortcomings.

journal-article · 2012 · Open the source

MacKenzie sets out Fitts's law as an information-theoretic model of aimed movement: the index of difficulty is the logarithm to base two of twice the target amplitude divided by the target width, and movement time rises with it. Three parts are most useful for teaching. First, the analysis of failure, which records the upward curvature of observed movement times at low difficulty and the disproportionate cost of narrowing a target compared with extending the distance by the same amount. Second, effective target width, which replaces the nominal width with a value derived from the distribution of hits and brings the model back toward the theory it came from. Third, a review of six pointing studies which shows how inconsistent published performance measures are, and traces the inconsistency to device, task, selection technique, error handling and learning. This makes the source a caution as well as a tool: a throughput figure is only interpretable beside the conditions that produced it. The paper is a review and reanalysis rather than a new experiment, and it says so.

journal-article · 1992 · Open the source

McCarthy and Wright argue that any account of what is called user experience must take in the emotional, intellectual and sensual aspects of interaction, because people live with technology rather than merely use it. The method is conceptual rather than empirical: they build a framework from the pragmatist tradition, drawing on Dewey's claim that art is part of everyday lived experience rather than something isolated, and on Bakhtin, and then apply it to cases. What is most useful for a student is the framework of intertwined threads of experience and the accompanying sense-making processes, which gives an evaluation something concrete to reach for beyond liking. On their account a felt state is part of how a person makes sense of a situation, rather than a decoration on top of it, which is where this module's argument for measuring affect lands. The case list includes an online shop and an ambulance control room. The argument is the strongest available statement of why a task-only evaluation is incomplete, and its limits are real: the account is interpretive rather than measured, it supplies no standardised instruments, and the authors concede that the standards for good work in that tradition are themselves contested. The publisher description and metadata were read for this note; the full text was not read.

book · 2004 · Open the source

This paper is the origin of the response-time figures this module teaches, and reading it corrects how they are quoted. Its argument is that no single response time is right, because tolerable delay depends on what the person is doing: the literature 'abounds in controversy about the limits of system response time', and the first thing the paper does is refuse the premise, stating that a two-second response is not a universal requirement. Its method is a classification rather than an experiment. Miller lists seventeen transaction types a person conducts at a terminal and proposes a limit for each, drawing the time estimates from the psychological literature of the period and from informal observation, such as workers at a time clock, rather than from a controlled trial. What is most useful here are three of those limits, which the module now cites directly: no more than a tenth of a second for feedback to a control activation, a delay of more than one second seeming intrusive on the continuity of thought, and ten seconds as the point at which sustained problem solving breaks down. Its limits are those of 1968 and of its method: text terminals, light pens and keyboard entry, thresholds presented as design guidance rather than measured constants, and a taxonomy the paper itself calls not exhaustive.

conference-paper · 1968 · Open the source

MDN is included because it is the resource a working developer opens, and because it is kept current as no printed textbook can be. It is community-maintained reference documentation, and that is its method: each page states a rule or a mechanism, links to the specification that governs it, and carries its own last-modified date, so the collection as a whole has no single publication year and is dated n.d. here. The Learn area provides a structured course from beginner to comfortable. Its reference pages carry the mechanisms this module relies on: the cascade, specificity, flexbox and grid, responsive design, native constraint validation, and the same-origin policy, which defines two addresses as sharing an origin when the scheme, the host and the port match, whatever their paths are. It also states the definition of the accessibility tree, the first rule of ARIA use, that a native element be used where one exists, the heading rule that level is carried by the element rather than by styling, and the pointer and hover media features that test the primary input device. Its limits are those of community documentation: it is not a standard, has no single authorial voice, and describes what the platform does, not what helps users. Where evidence is at stake, read the standard or the research instead.

reference · n.d. · Open the source

MDN is the documentation the web is built against, maintained by Mozilla with community contributions, and it is the reference most developers use. Its argument is implicit but firm: an accurate, current, example-led reference beats a printed book. Its method is continuous public documentation: each page describes one feature and carries a last-modified date. Five groups of pages are useful here. The page on the JavaScript execution model lays out the heap, the stack and a queue of jobs, calls that queue the event loop, and states that a job which takes too long leaves the page unable to process clicks or scrolling. The pages on the input element, HTMLInputElement.value and getAttribute separate the current value property from the value attribute, which corresponds to the initial value. The guide on web workers states that a worker cannot directly affect the parent page, and that data is copied. The page on the accessibility tree says that browsers build it from the DOM tree for assistive technologies such as screen readers, and the page on the accessible name says that a form control takes that name from its associated label element. Its limits are that it is a reference rather than a course, and that pages change, so a reader should check its date rather than treat a URL as fixed.

documentation · 2026 · Open the source

Myers surveys the technical history of interaction and records where each idea was demonstrated and by whom. The module draws its dated instances from his entries: Sutherland's Sketchpad of 1963 at MIT, which let a light pen move and resize objects; the mouse, built at Stanford Research Laboratory in 1965 as part of Engelbart's project; the 1968 demonstration, with the mouse, tiled windows and participation by people at several sites; Xerox PARC's Bravo of 1974, the first editor to show the page as it prints; PROMIS at the University of Vermont in 1976, the first hypertext system released to its user community; VisiCalc for the Apple II in the late 1970s, with a solver based on dependency work by Sussman and Stallman at MIT; and the first commercial direct-manipulation systems, the Xerox Star of 1981, the Apple Lisa of 1982 and the Macintosh of 1984. The method is a short survey article for practitioners, and its value is that it dates each demonstration. Two limits matter. Each date summarises another source and should be followed back for assessed work, which is why this module now carries Sutherland's own thesis beside it. And the account is deliberately technology-centred, so it is sound for the sequence of events and weak for claims about how people behave.

journal-article · 1998 · Open the source

The paper argues the pro-test-first case with numbers, and it is the necessary counterweight to Fucci and colleagues in the module's contested question. Its method is a set of industrial case studies rather than a controlled trial. The authors report case studies with three development teams at Microsoft and one at IBM that had adopted test-first development. Across the four products the pre-release defect density fell by between 40 and 90 per cent relative to comparable projects that did not use the practice, while the teams judged their initial development time up by 15 to 35 per cent. What is most useful here is that it states the other side with concrete figures, so a reader can weigh two studies that disagree instead of hearing one side told as settled fact, and that its own method shows why it cannot settle the question. The teams chose the practice, there is no randomised control, and the time cost is described as subjective. Its limits are that only the abstract was reachable in this run, because the full text is behind a subscription, so the defect and time figures come from the abstract alone, and that the comparison rests on the authors' own choice of similar projects.

journal-article · 2008 · Open the source

This is the current edition of the United States federal authentication guidelines and the source for the module's account of how official password advice has moved. It is a normative standard rather than a study: its method is to state requirements, marked SHALL and SHOULD, in the language of compliance, and to justify them briefly from the research and from operational experience. What it requires is the substance of the module's section: a minimum length of fifteen characters where a password is the only factor and eight where it is one part of multi-factor authentication, with a maximum of at least sixty-four characters; a check of each new secret against a list of commonly used and compromised values; rate limiting of failed attempts; and allowance for pasting so that password managers work. It forbids arbitrary composition rules and forced periodic change, on the grounds that both push users towards weaker and more predictable secrets, and it states that a password is not phishing-resistant. It defines phishing resistance, and admits syncable authenticators, the class that includes synced passkeys, at the second assurance level but not the third. Two limits matter: the guideline dates from July 2025 and its mandatory force is limited to federal systems, and it is a specification rather than evidence about what users do.

standard · 2025 · Open the source

This is the paper behind the five-user rule, and it offers a model rather than a rule. Nielsen and Landauer fitted a Poisson process to eleven studies of usability testing and heuristic evaluation, in which each evaluator has an independent chance of finding each problem. On that assumption, detection rises steeply with the first few participants and then flattens, so each extra participant finds fewer new problems than the last. The model is most useful as a planning tool, because it turns "how many users" into arithmetic about the coverage and the cost you will accept, and because an early test can be used to estimate how many problems remain. Its limits are serious. Independence is an idealisation, since real participants share knowledge, follow the same habits and miss the same things, so the true curve can be flatter and wider than the model predicts. The eleven studies are small and predate modern interfaces, and they concern problem counting rather than the understanding a qualitative study seeks. The abstract was read here; its medium example estimates sixteen evaluations as worth their cost, with the best benefit-to-cost ratio at four.

conference-paper · 1993 · Open the source

This paper introduced heuristic evaluation and reported the first evidence about how well the method works, which is why it is worth teaching the finding before the ten heuristics that later accompanied it. The authors report that the method as described is informal: a number of evaluators are presented with an interface design and asked to comment on it. Across four experiments they found that individual evaluators were mostly quite bad at this, finding between 20 and 51 per cent of the usability problems in the interfaces they evaluated. Aggregating the evaluations from several evaluators to a single evaluation did rather well, even when the aggregate came from only three to five people. That pair of findings is the most useful thing in the paper for a peer evaluation: one list is a contribution rather than a verdict, and a handful of independent evaluators is the unit worth organising. The limits are real. The numbers come from four experiments in 1990 on a particular set of interfaces, the paper does not establish that aggregated expert findings match the problems users actually meet, and the abstract alone does not report the interfaces used, the evaluators' expertise or the tasks. The finding should be treated as evidence about detection, not about importance.

conference-paper · 1990 · Open the source

Norman argues that design failure is a property of the object rather than a fault in the person using it. The book's apparatus is the seven-stage action cycle, with the gulf of execution between an intention and the actions the interface allows, and the gulf of evaluation between what the system did and what the person can tell it did. The 2013 revision matters for teaching because Norman corrects a vocabulary problem of his own making: he had used affordance for a perceptible cue, and the revision separates affordance, the real possibility for action, from signifier, the signal that makes that possibility discoverable. The book also supplies this module's error vocabulary, separating slips, where a skilled action goes wrong, from mistakes, where the plan itself is wrong. Its method is argument from worked examples drawn from everyday objects, addressed to a general reader rather than to other researchers. That method sets its limits. It is a trade publication whose examples are chosen to persuade, and its claims carry no references in the way a journal article does, so it is not evidence for a measured effect. Its value is diagnostic: it is the fastest way to teach a student to see why an interface fails. The book is print, so only the publisher page could be checked.

book · 2013 · Open the source

Norman outlines a theory of action in which a sequence is represented by a parent schema and numerous child schemas, several of which can be active at once, each with triggering conditions and an activation value. The method is theoretical: he builds the schema model and organises slips reported in the literature rather than running new experiments, so the account is a model rather than a measured mechanism. The path from intention to action activates the parent schema and then the child schemas, and a slip occurs when the wrong schema is triggered, is lost, or is wrongly ordered. He sorts slips into three major categories: errors in forming the intention, such as mode and description errors; faulty activation of schemas, such as losing an intention or misordering components; and faulty triggering, such as blends and intrusions of a habitual action. What is most useful is that the taxonomy explains why a competent person does the wrong thing without being careless, and it points directly at design remedies such as distinct modes, confirmation of the object of action, and undo. The limits are that the schema account is a model, and that the categories overlap in practice, so one slip can sit in more than one. The abstract and metadata were read for this note; the full text is paywalled.

journal-article · 1981 · Open the source

This cheat sheet argues that cross-site scripting is prevented by controlling output rather than by policing input, and it is organised by output context for that reason. It is a community document in the OWASP Cheat Sheet Series, non-normative and maintained in a public repository, and it states no revision date on the page. Its central claim is perfect injection resistance: every variable must be validated and then encoded or sanitised for the context into which it is written, because browsers parse HTML, attributes, JavaScript, URLs and cascading style sheets differently, and the wrong encoding either creates a weakness or breaks the page. It names the cases a student meets in practice: HTML entity encoding for text, attribute encoding for attribute values, assigning text through the textContent property rather than parsing it as markup, sanitising with a maintained library when HTML really must be accepted, and the escape hatches where frameworks reopen the hole, including their direct document manipulation functions and their unsafe HTML helpers. It also argues against filtering interceptors as a primary defence, and notes that a content security policy is easy to misconfigure and should not be the root fix. Its limits are those of the genre: it sets out remedies rather than measuring how often each attack succeeds, and its framework-specific guidance is admittedly incomplete.

practitioner-guidance · n.d.-a · Open the source

This is the study behind the claim that meeting the guidelines is not the same as being usable, and it is quoted here in both directions. The authors ran task-based user evaluations with 32 blind participants across 16 websites, using screen readers, and logged 1,383 separate problems those users hit. Two results anchor the argument. Only 50.4 per cent of the problems fell under a WCAG 2.0 success criterion, so about half of what went wrong for a real user was outside the guidelines' coverage. And for problems that were covered, the abstract reports a figure of 16.7 per cent in a sentence that is not fully clear, describing websites where recommended techniques were implemented and still did not solve the problem. The paper argues for moving away from a problem based approach towards a design principle approach for web accessibility. It is used here because it disciplines the module's claim that semantic markup and conformance are the goal, and it is peer reviewed with real users, which is its strength. Its limits matter as much: one study from 2012 measured against WCAG 2.0, 16 sites, no reported effect sizes, and coding a user's problem onto a criterion involves judgement.

conference-paper · 2012 · Open the source

This is the standard library reference for opening URLs in Python, and it is the source for the Python half of the comparison the module draws between the two languages. Its argument is that the machinery for a network request belongs in the standard library rather than in a third-party package: the function takes a URL or a Request object, performs the HTTP or HTTPS exchange, and returns a response. Its method is reference documentation maintained by the language's stewards, written against the current source and published with the release it documents. What is most useful is the exact shape it gives the call. The page states that for HTTP and HTTPS URLs the call returns an http.client.HTTPResponse object, whose status, headers and body can then be read, and that a protocol error raises URLError rather than being returned as a value. That is what makes the JavaScript and Python versions of a request the same shape in the module: ask, receive a response, inspect it, and handle failure separately from success. Its limits are that it is a reference rather than a tutorial, that much professional Python uses third-party libraries it does not cover, and that the page tracks the current development version of the language.

documentation · 2026 · Open the source

Reason's book is a theoretical integration that treats error as a product of the same cognitive machinery that produces correct action, rather than as carelessness. Its method is synthesis: it draws together previously separate literatures, cognitive theory and accident analysis, rather than reporting controlled experiments. Its central distinction, used throughout this module, is between skill-based slips and lapses, in which the intention was adequate and the execution failed, and mistakes, in which the intention itself was inadequate because a rule was misapplied or knowledge was insufficient. The book develops performance levels and error forms, follows errors into systems where a local error meets a design or an organisation that lets it propagate into a disaster, and closes with how to assess and reduce human error risk. What is most useful for a designer is the insistence that the remedy depends on the error kind: constraints and forgiveness for slips, better information and feedback for mistakes. The limits are that the taxonomy is descriptive and its categories overlap, and that much of the evidence base is cognitive theory and accident analysis rather than controlled experiment. The publisher description and metadata were read for this note; the full text was not read.

book · 1990 · Open the source

This is the current standard for transport-layer encryption on the web and the primary source for what TLS promises. The paper's claim is narrow and worth stating exactly: the protocol is designed to prevent eavesdropping, tampering and message forgery, which is a statement about protection in transit and not about the honesty of the server at the other end, a distinction the module's threat-model section depends on. Its method is that of a specification: it defines message formats, states and required behaviours precisely enough for independent implementers to interoperate, and it is written for them rather than for designers. It was published in August 2018 and replaces the earlier protocol versions it obsoletes. For this module the two facts that matter are structural. Encryption is negotiated per connection, so a secure connection is established rather than assumed, and the handshake costs round trips, which is why resumption using pre-shared keys and session tickets exists, and why the 0-RTT mode saves a round trip at connection setup for some application data at the cost of certain security properties. The limit is readability: the document is dense and technical, and it is cited here for the protocol's guarantees and its round-trip structure, so read it with a friendlier explanation beside it.

standard · 2018 · Open the source

Rettig's column is the text that made low-fidelity prototyping ordinary practice, and it remains the clearest description. He argues that lo-fi prototyping is simple, effective and under-used, and that it lets a team demonstrate an interface's behaviour early, when changes are cheap. He prefers paper to a software prototype for three reasons: paper is fast, so more iterations fit into the time; its unfinished appearance turns feedback towards content and away from fonts and colours; and no one is attached to a sketch, so a failed idea can be thrown away without loss. His method is a practitioner's account rather than a study, and two cases carry it: an opening description of a team that codes a design before showing it to users and then cannot act on the comments it receives, and a workshop in which a visiting consultant put the developers into four teams that built paper ordering menus for a fast food restaurant and improved a design through repeated tests within about six hours. Most useful is the procedure he extracts from that work: assemble a kit, set a deadline, build models rather than illustrations, put one team member in the role of the computer, and write one observation per card. Its limits are a column's: no comparison group, and iteration counts from the author's own projects.

journal-article · 1994 · Open the source

This paper reports the field study of a persona initiative that failed for reasons that have nothing to do with the merits of the method. Ronkko and colleagues' method was fieldwork with the interaction design team at UIQ Technology, a Symbian subsidiary in Sweden that licensed a pen-based phone interface to competing handset makers. The team adopted personas to settle internal disputes with developers about what users wanted, built three personas, a teenager, a young woman and a middle-aged businessman, with use cases and wall charts, and expected marketing and sales to choose among them. No decision ever came, two clients joined the project late, the requirement set fragmented, and the team closed the persona work down. The authors later argue that the obstacles were branch-related: clients protected launch dates and new technology drove the requirements, so the reduction of scope never followed end-user needs. They also record an undocumented, inherited persona of the travelling business man that already governed decisions. It is the most useful counterweight to the method literature, because it shows a method defeated by its own organisation. Its limits are candid: one company, one failed attempt, and confidentiality prevents the authors from disclosing the full internal picture.

conference-paper · 2004 · Open the source

Rosen's book is the standard reference for the discrete half of this material, and its scope matches the module almost chapter for chapter: logic and proofs, sets, functions and relations, matrices, counting, discrete probability, graphs and trees. Its stance is conventional: mathematics is set out as a sequence of definitions, theorems and proofs, and the reader is expected to supply proofs rather than to trust examples. Its method is exposition with a very large graded exercise set, so that proof by induction, proof by contradiction and proof by counterexample become habits rather than slogans. What is most useful to a first-year reader is that the treatment is thorough and assumes no more than school algebra, and that the worked examples make each technique concrete. Two limits matter here. First, the applications are drawn from computing in general rather than from interaction design, so the translation into screens, gestures and test plans is the work of this module and not of the book. Second, the title is expensive and exists in many editions: the 2025 release is the current one on the publisher's page, earlier editions such as the eighth are widely used, and any page reference should be checked against the edition in hand.

book · 2025 · Open the source

Russell argues that affect is not a set of dimensions that vary independently, as factor-analytic work had suggested, but a systematic ordering of states that a spatial model represents as a circle, with two orthogonal axes beneath it: pleasure against displeasure and arousal against sleep. The method is multi-part and largely psychometric. He scaled 28 emotion-denoting adjectives in four different ways, among them a multidimensional scaling of perceived similarity among the terms, a unidimensional scaling on the hypothesised pleasure and arousal dimensions, and a principal-components analysis of 343 people's self-reports of their current affective states. What is most useful for this module is the pair of working dimensions that interface work now calls valence and arousal, together with the claim that these axes are the cognitive structure lay people use to organise affective experience. That is the primary statement behind the emotion section, which earlier drafts took from a review. The limits are that the evidence is self-report on English adjectives, that a spatial model describes rather than explains, and that later work has proposed more than two dimensions and questioned the circular structure. The abstract and metadata were read for this note; the full text is paywalled.

journal-article · 1980 · Open the source

This is the paper from which the vocabulary of secure design still derives. Its argument is that protection cannot be achieved by building a flawless system, since no one knows how to do that, so it should be designed so that the flaws that remain are few and mild; the principles are offered as the practical route to that property. Its method is a tutorial and a position paper rather than an empirical study. It was written when time-sharing made the sharing of information a live engineering problem, and its first section sets out eight design principles with the reasoning behind each. This module relies on two of them in particular. Fail-safe defaults establish that a protection mechanism should deny access in the absence of an explicit decision, which is the origin of the secure-default rule, and psychological acceptability asks that the human interface be designed for ease of use, so that people apply the protection correctly without effort, which is an early statement that usability is a security property rather than a concession. The paper is a tutorial, and the field has moved a long way since 1975, so use it for the principles and not for current mechanisms. The first section is free and needs little prior knowledge.

journal-article · 1975 · Open the source

Schon's book is the source of the term reflection-in-action and the argument behind studio teaching. Its claim is that competent practitioners know more than they can state: they act in situations that are uncertain, unique and conflicted, the situation answers back, and the next act is adjusted in the light of that answer. His method is close work with practitioners in five professions, engineering, architecture, management, psychotherapy and town planning, with architecture supplying the design-like material. The case the book returns to is a design review in an architecture studio, where the studio master, Quist, reframes a student's problem instead of correcting her drawing, and tells her to begin with a discipline, even an arbitrary one. Most useful for this module is the role he gives the coach: the critic does not hand over rules, but points at an adjustment the designer has already made, so the designer learns to see their own move. The limits are serious. The evidence is interpretive, drawn from cases and interviews, with no comparison group and no measured effect; the chapter reproduces a single protocol, and Schon himself notes that Quist's fluency may hide the reasoning from the student; and the vocabulary has since been flattened into a label for any reflective writing.

book · 1983 · Open the source

Shannon's paper is the origin of the ideas this module uses, and reading even its first part changes how the vocabulary is understood. He begins from the observation that the semantic content of a message is irrelevant to the engineering problem, and that what matters is that the message is selected from a set of possibilities; he argues for a logarithmic measure, defines the capacity of a discrete channel as the limit of the logarithm of the number of allowed signals of a given duration, and introduces entropy as the measure of uncertainty, with the properties this module uses: entropy is zero when the outcome is certain, maximal and equal to log2 n when the outcomes are equally likely, and no greater than the sum of its parts for a joint event. The paper is a technical report written for engineers, and much of it concerns coding for telegraphy and television, which a first-year reader can safely skip. The freely readable reprint used here is a scan of the 1948 text, so the typesetting is uneven and some formulae need care.

journal-article · 1948 · Open the source

Shneiderman's paper named a style of interaction that the graphical interface was already groping towards, and gave it an argument. Its three properties are continuous visibility of the object of interest, physical actions or labelled buttons in place of complex command syntax, and operations that are rapid, incremental and reversible, with the effect visible at once. The case for the style is about the cost of learning and the cost of error rather than about speed, which is why it outlived the arrival of faster machines. The method is a position statement illustrated with examples and supported by studies carried out by other people. The paper reports no trial of its own comparing the two styles, and its evidence is borrowed. What is most useful for teaching is the section on the problems of direct manipulation, because it concedes that graphic representations consume screen space and that experienced users may prefer a table of fifty document names to ten truncated icons, while the paper elsewhere claims that experts on these systems work extremely rapidly. The limit is that both sides of that expert case are argued rather than measured here. The version of record is paywalled, so the full text was read from the author's open copy.

journal-article · 1983 · Open the source

This paper is the practical companion to the theory, written to stop the Fitts's law literature from fragmenting further. The authors make seven recommendations for anyone building a model either to predict movement time or to compare experimental conditions, and they present them as a supplement to the ISO 9241-9 standard on pointing device evaluation, which they date to 2002 (the reference number is ISO 9241-9:2000(E)). The recommendations that matter most for teaching are to use the Shannon formulation of the index of difficulty, to spread amplitudes and widths over a wide range of difficulty, to record endpoint scatter or error rates rather than error-free times, to adjust for accuracy separately for each participant and condition, to fit with least-squares regression and report both intercept and slope, and to be explicit about how throughput is computed. The paper also shows why across-study comparison has been unreliable, and it sets the effective target width at 4.133 standard deviations of the observed endpoints. Its limits are that the recommendations are methodological rather than empirical, that they predate touchscreen and gesture pointing as the common case, and that the assessment of the earlier literature is the authors' own reading of it.

journal-article · 2004 · Open the source

This five-page reply states the situated-action side of the dispute this module names, and it is the version of that position a student can actually read. Suchman answers Vera and Simon, who had argued that the goals of situated action could be met inside a symbolic framework. Her claim is that plans are not determining of the actions they project. Plans, she writes, are conceptual and rhetorical devices that are deeply consequential, because people organise action in their terms, and the relation between a plan and the action it projects remains an open question. The part most useful for design is her claim about asymmetry: the limits of machine sense making lie in what a machine can access of the social and material resources people draw on, and she calls that asymmetry the central problem for human-machine communication. The method is argument in a numbered journal exchange. The reply answers specific criticisms rather than restating the book and reports no new study, so its weight is argumentative and not evidential. Two limits follow: it presupposes the reader knows her 1987 monograph, Plans and Situated Actions, which could not be read here, and its brevity compresses the argument.

journal-article · 1993 · Open the source

Sutherland's doctoral thesis is the primary source for the claim that Sketchpad was his 1963 work at MIT, a claim this module previously carried only from Myers's survey. The thesis argues that drawing can serve as a communication medium for a computer. A user sketches directly on a display with a light pen, which both positions parts of the drawing and points at them to change them, and a set of push buttons controls operations such as erase and move. The system stores explicit information about the topology of a drawing, so moving one vertex moves both adjacent sides and attached symbols follow it, and a separate mechanism satisfies constraints the user has set. The method is system building, and the most useful part is the description of how a drawing is stored. Sutherland built the system on the TX-2 computer at MIT Lincoln Laboratory and reports the ring structure that stores drawings, and the drawings in the thesis were made with Sketchpad itself. The limits are that there is no study of users and no comparison, and the Cambridge preface notes that executable versions ran only on a customized machine at MIT Lincoln Laboratory, so the influence came through the ideas. In this run only the abstract, preface and contents of the deposit could be read.

thesis · 1963 · Open the source

The JavaScript Signals standard proposal is a TC39 document, and it argues that the language should carry one reactive primitive rather than leaving every framework to invent its own. Its method is a design proposal written in the open: the README sets out the problem with hand-wired state, defines a small API of Signal.State for writable state and Signal.Computed for a derived value that is lazily calculated and cached, and records a plan to prototype the design inside several frameworks before asking to advance beyond the first stage. What is most useful for teaching is that it states the goal in one line, to decouple the reactive model from the rendering view, which is the same problem the humble dialog addressed in 2002, and that it names automatic dependency tracking as the mechanism, so a value notifies whatever reads it without manual subscriptions. Its limits are that it is a proposal at Stage 1, reached in April 2024, and not a standard: no browser ships a native Signal, the API may change, and the document is written for framework authors rather than for students.

proposal · 2024 · Open the source

This is the module's reference for the machine beneath the interface. It is not an argumentative book: it assembles the settled account of how an operating system works, and its structure is its contribution, since a designer can read one chapter and leave the rest. Its method is expository and example-led, with case studies of Unix, Linux, Android and Windows 11 that show the same mechanisms in different products. For this module the two chapters that matter are Processes and Threads, which supplies the process model, the scheduler and the distinction between threads that share memory, and Memory Management, which supplies the hierarchy from registers and caches to main memory and storage. The fifth edition adds coverage of solid-state storage and a security chapter, and the publisher's overview also claims coverage of interface design and performance tradeoffs, which is the nearest the book comes to this module's subject. Its limit for this module is that it is a textbook of operating systems rather than of human interaction. The link from an interface that stalls to a busy thread is an inference a designer draws from the scheduling and memory models, not something the book states for a design audience, and the printed text was not read.

book · 2023 · Open the source

This book is the standard cognitive account of what happens between a survey question and an answer, and its value here is that it converts item-writing rules into consequences. The model has four stages: the respondent interprets the question, retrieves relevant information from memory, forms a judgement from what is retrieved, and reports an answer, which may be edited for acceptability. Each stage is a place where the answer departs from the fact being asked about. A double-barrelled question fails at comprehension, a question about frequency depends on vague retrieval and estimation, and a sensitive question is distorted at the reporting stage by social desirability, the tendency to give the answer that looks good. The book is a research monograph rather than a how-to guide, and it draws mainly on laboratory studies and on an era of telephone and face-to-face surveys, so it says less about the behavioural traces a designer might prefer to self-report. Its three authors and its length were confirmed from the publisher record, but the text itself is purchased, so the four-stage model here rests on the book's established account rather than on a reading of the full text in this run.

book · 2000 · Open the source

Tversky and Kahneman name and describe three heuristics used in judgement under uncertainty. Representativeness is used when judging whether an object or event belongs to a class. Availability is used when assessing how frequent a class is or how plausible a development is. Adjustment from an anchor is used in numerical prediction when a starting value is supplied. The method is a short review that gathers their own laboratory demonstrations, so the paper is a position piece as much as an empirical report. Their own framing is deliberately moderate: the heuristics are economical and usually effective, and they produce systematic and predictable errors. What is most useful for interface work is the second half of that sentence, because a default, a suggested amount or the order of a list supplies an anchor, and the error is then predictable rather than random. The limits are that these are laboratory demonstrations, and the effect sizes found in field settings tend to be smaller and to depend on context. The paper's lasting influence has as much to do with the frame it created as with any single experiment. The abstract and metadata were read for this note; the full text is paywalled.

journal-article · 1974 · Open the source

This specification is the standard that defines how a two-dimensional transform is expressed and computed on the platform the programme teaches, and it is useful here because it is the point where matrix algebra stops being an abstraction. It defines the transform property, explains that a transformation matrix is a mathematical mapping from one coordinate system into another computed from the transform and the transform origin, states that a two-dimensional transform may be represented as a 3 by 2 matrix or as a 4 by 4 matrix with the unused entries fixed, and records that the 4 by 4 form is the one used in three dimensions. It also defines the current transformation matrix and the rules for composing a list of transform functions in order, which is the practical content of the claim that matrix multiplication does not commute. Its status is a Candidate Recommendation, so it is work in progress and may be revised. It is written for implementers, and a student needs a course alongside it for the underlying linear algebra.

standard · 2019 · Open the source

WCAG 2.2 is the specification this module teaches against. Its architecture matters as much as its individual rules: four principles, 13 guidelines, and testable success criteria at levels A, AA and AAA, with sufficient techniques and documented failures beneath them. The version read here is backwards compatible with 2.0 and 2.1, adds nine success criteria, and marks 4.1.1 Parsing obsolete. The criteria added in 2.2 are the ones a designer most often meets for the first time: focus not obscured (2.4.11), dragging movements (2.5.7), target size minimum (2.5.8), consistent help (3.2.6) and accessible authentication (3.3.8). The document is candid about its own boundary, and that candour is worth teaching: it says the guidelines give only some accommodation for cognitive and learning disabilities, records the difficulty of reaching consensus on testability, and states that testing requires automated checks combined with human evaluation. It is a normative standard, not a tutorial, so the reasoning behind each criterion lives in the supporting documents rather than here.

standard · 2023 · Open the source

The Authoring Practices Guide is where a designer discovers how expensive a custom widget really is. It is produced by the ARIA Authoring Practices Task Force, which the guide says relies on broad community representation and participation, and its patterns and examples are contributed and reviewed in a public repository. It provides design patterns with functional examples for common components, guidance on keyboard interaction, a practice on providing accessible names and descriptions, and instruction on using sectioning elements and ARIA landmark roles so that assistive technology users can understand the layout of a page. Two positions in it carry weight here. Accessible names and descriptions are described as one of the most important responsibilities an author has, which is why a control with no name is a defect rather than a refinement. And every pattern is presented as something to be implemented, which shows that a widget built from roles must reproduce the role, the states, the keyboard contract and the focus behaviour that a native element supplies for nothing. Its limits matter as much. It is guidance rather than a standard, so it cannot be cited as a requirement. Its examples demonstrate implementation rather than report evidence about users, and the guide records no testing with assistive technology users. No publication date is stated on the page.

guidance · n.d.-a · Open the source

This page is the standards body's own account of where WCAG 2 now stands, and it is cited here for the status of the standard rather than for any single success criterion. It argues, in effect, that WCAG 2 is a family of stable referenceable texts rather than one document: versions 2.0, 2.1 and 2.2 are all existing standards, 2.2 does not deprecate or supersede 2.1, and content that conforms to 2.2 also conforms to 2.1 and 2.0. Its method is institutional reporting rather than measurement: the page lists the publication and revision dates of each version, the ISO identity of the text, the standards that policy instruments use, and the state of the successor work. What is most useful is that currency. WCAG 2.2 is published as ISO/IEC 40500:2025, the 2026 version of EN 301 549 uses WCAG 2.2 and is the standard most organisations use in addressing the European Accessibility Act, and the successor guidelines remain a draft that the consortium says does not replace WCAG 2. Its limits follow from being a summary. It states status rather than evidence about users, it is revised in place without a publication date, and it says nothing about whether a criterion helps a particular person on a particular site.

guidance · n.d.-b · Open the source

The HTML Standard is the specification that says what each element means, so it is the primary source under this module's claim that choosing an element is a design decision. It states that the button element represents a button labelled by its contents, that the label element represents a caption in a user interface and takes a specific form control as its labelled control, and that the heading elements represent headings whose level is given by the number in their name. Read against those sentences, a styled div is not a button and a bold paragraph is not a heading, because the meaning is carried by the element rather than by its appearance. What is most useful is that exactness: the standard separates an element's meaning from its presentation and gives a designer a rule to check rather than a style to copy. Its method is the one the Web Hypertext Application Technology Working Group uses, maintaining a living standard revised continuously in public rather than issued in numbered editions, and that is also its limit. The text changes under the reader, it is written for implementers rather than designers, and it defines what a conforming user agent must do, not what helps a person. It is a definition, not evidence about users.

standard · n.d. · Open the source

The Fetch Standard defines requests, responses and the process that binds them, and it is the specification behind the fetch function a student uses to call an API. Its argument is unification. Before it, each way of fetching a resource, an image, a script, a stylesheet or a data file, behaved slightly differently on redirects, caching and cross-origin access, and the standard gives all of them one architecture. Its method is formal specification: an editor's draft maintained in the open by the WHATWG with a change log, a test suite and named implementers, written for people who build browsers. What is most useful for teaching is the error model, because it is the part a student will otherwise get wrong. The standard defines a network error as a response whose type is error and whose status is 0, which is distinct from a response that arrives carrying an error status. That distinction lets the module design an error state for a failed request and for a refusal, the shape of the API, a promise for a response object carrying a status, headers and a body, comes from the same text. Its limits are those of any specification. It is written for implementers, assumes a knowledge of HTTP, and is better read to confirm behaviour than to learn it.

standard · 2026a · Open the source

The HTML Standard is the primary specification for the browser platform, and this module now cites it for the execution model it previously took from MDN, a secondary reference. The standard argues by definition. Section 8.1.7 states the mechanism as precisely specified objects: an event loop has one or more task queues, a task queue is a set of tasks, task queues are sets rather than queues because the processing model takes the first runnable task rather than the first one, and the microtask queue is explicitly not a task queue. Tasks are raised by named task sources, among them events, parsing, callbacks and timers. Its method is formal specification agreed in the open by the WHATWG, with an editor's draft, a change log and a test suite. What is most useful for teaching is that it settles the vocabulary: MDN presents the heap, the stack and a queue of jobs and calls that queue the event loop, while the standard separates the event loop from the task queues it holds. Its limits are those of a specification: it is addressed to implementers, it states what a conforming user agent must do rather than how a person experiences it, and as a Living Standard it changes, so its last-updated date is part of the reference.

standard · 2026b · Open the source

This is the current extension of the fixation literature, and the reason the module now treats image generation as a source of fixation and not only of inspiration. The authors report a between-participants experiment with sixty participants, each asked to sketch as many ideas as possible for a chatbot avatar within twenty minutes, after reading a brief that contained an example avatar. They compared three sources of inspiration: none, an online image search, and an AI image generator. The generator group fixated more on the example, and produced fewer, less varied and less original ideas than the group given no inspiration at all. Independent raters scored the sketches on the four measures the older fixation studies use, so the result is comparable with Jansson and Smith's method. The most useful part is the mechanism the authors describe, fixation displacement, in which attention moves off the example and onto whatever the generator returns, with the participants' own prompts often recycling the brief's language. Their interviews suggest the outcome depends heavily on how a person prompts. The limits are one experiment with novices, one short task and one generator, measures of sketches rather than of design outcomes, and no replication.

conference-paper · 2024 · Open the source

This paper answers the question a studio always asks: is a sketch good enough to test with? The authors report that interfaces are tested at a mixture of fidelities and in a mixture of media, and that there was, at the time, insufficient research to say which combination produces the best feedback. Their method was a comparison of low-fidelity and high-fidelity prototypes of websites, in paper and on a computer, with task-based user tests, so that the testing medium was separated from the other attributes of the prototype. They report that low-fidelity and high-fidelity prototypes were equally good at uncovering usability issues, and that the results were independent of the medium despite differences in interaction style. The practical conclusion is the most useful part: choose the medium and the fidelity on grounds of cost, timing and design goal, not on evidential grounds. That is a genuinely useful permission for a first-year studio, and it also removes the excuse for delaying any user contact until a build exists. Its limits are that it is a short conference paper about websites rather than embedded or safety-critical interfaces, that only the abstract was available to me here, and that equal problem discovery does not mean equal severity judgements or equal transfer to a coded product.

conference-paper · 2002 · Open the source

This paper reports three experiments on what makes an unexpected object noticeable when attention is already busy. The method is online experimentation with 4,493 participants across the three studies. Study 1 repeated the gorilla video used in the original inattentional blindness demonstration and added faster versions of the gorilla, and those counting the black team's passes detected it far more often than those counting the white team's passes, which repeats the similarity effect. Studies 2 and 3 replaced the video with moving dots and one unexpected object whose speed and colour were varied, so speed could be separated from similarity. The consistent result is that speed decides detection: slow unexpected objects were often missed even when they matched the colour being tracked, while fast ones were noticed even when they did not, and the fastest object was seen despite the briefest spell on screen. What is most useful for interface work is the correction to a common design rule, because physical salience, size and contrast do not defeat a demanding primary task, whereas motion does. The limits are that studies 2 and 3 used simple moving dots rather than realistic scenes, the authors concede that speed was initially confounded with dissimilarity, and only the third study controlled that confound.

journal-article · 2023 · Open the source

Ware's book sets out what the science of perception tells designers about visualisation. Its method is reference synthesis: each chapter gathers primary findings on one perceptual mechanism and turns them into guidance, rather than reporting new research. The chapters move from foundations and optics through lightness, brightness, contrast and colour, then to visual salience and finding information, static and moving patterns, where the gestalt laws of grouping are set out as design principles, space perception, objects, images and the visual thinking processes. The fourth edition adds a chapter on designing for perception that tries to bridge specific guidelines and design decisions. What is most useful is that the treatment is organised by perceptual mechanism rather than by chart type, so a designer can reason from the mechanism to the case in front of them. Colour is treated as an opponent-channel construction with consequences for users whose colour vision differs, and contrast is handled separately from hue, which is the argument used in this module against encoding meaning in colour alone. The limits are that it is a reference text rather than a curriculum, that parts assume a strong background, and that its guidelines are heuristics rather than tested design rules. The publisher description, the table of contents and the Gestalt-laws chapter page were read for this note.

book · 2020 · Open the source

The WebAIM Million runs an automated evaluation of the home pages of the top million sites each year, using the WAVE engine on the rendered document object model, and reports error counts, page complexity, and the technologies correlated with failure. The 2026 edition reports detected WCAG 2 failures on 95.9 per cent of pages, up from 94.8 per cent the previous year, with an average of 56.1 detected errors per page, and the same six failure types dominating the total: low contrast text on 83.9 per cent of pages, missing alternative text on 53.1 per cent, missing form labels on 51 per cent, empty links on 46.3 per cent, empty buttons on 30.6 per cent, and missing document language on 13.5 per cent. Average page elements rose to 1,437, an increase of 14.3 per cent in a single year, and the authors link rising complexity to rising failure. Its limits are stated by its own authors and they matter. Detection is automated, so the absence of errors proves nothing, and because only automatically detectable failures are counted, true conformance is lower than the 4.1 per cent the report derives. The sample is home pages only, and the correlations reported are not causal.

technical-report · 2026 · Open the source

This is the module's evidence that the sequence of interaction technology it traces has a recent turn. The paper argues that systems which generate text, images, code or other artefacts from a prompt differ from earlier systems that produced a decision or a short output, because control shifts from the user naming each step toward the user stating the outcome wanted. Following Nielsen, the authors call this intent-based outcome specification and report it as a new interaction paradigm, the first in sixty years on that account. Their contribution is six design principles, three presented as new considerations for generative systems and three as reinterpretations of known problems in AI design, each paired with strategies for putting it into practice. The method is an iterative development and validation process: a literature review, feedback from design practitioners, validation against real applications, and use inside the design process of two generative AI applications. That is both the value and the limit. The principles are a considered proposal, not an experiment that measures whether following them improves an outcome, so the module cites the paper for the shift in interaction and for the state of the guidance rather than for a measured effect. It is open access.

conference-paper · 2024 · Open the source

This is the reference text that organises engineering psychology as the flow of information through a person, from the senses through the brain to action. Its method is textbook synthesis: it summarises primary studies rather than reporting new data, so the sample and the design behind any individual claim are usually not visible. Its chapters cover signal detection, attention in perception and display space, spatial cognition and manual control, language, memory and training, decision making, selection of action, multitasking, mental workload and stress, and human-automation interaction. The fifth edition adds a chapter on research methods and sections on interruption management and distracted driving. What is most useful for this module is the breadth of the attention and workload material under one consistent framework, which lets the attention, multitasking, skill-acquisition and reaction-time claims be traced to a single authoritative treatment rather than assembled from fragments. The limits are those of a textbook: it mediates primary work, so a student should follow its references to the original study before relying on a specific figure or effect size. The publisher description, the table of contents and the author details were read for this note; the full text was not read.

book · 2022 · Open the source

This review is the current map of where Fitts's law is settled and where it is not, which is the boundary a design decision runs into. Xiao and colleagues searched the Web of Science for work on the law in human-computer interaction and human motor behaviour, and set out the classical two-dimensional indices in one table: Fitts's own, Welford's, MacKenzie's Shannon form and Hoffmann's finger-pad correction. Their verdict is that MacKenzie's form is now the most widely used in two dimensions. They then turn to three dimensions and list the competing indices of Murata, Myung and Machuca, noting that each fits its own experiment and that no complete system exists for comparing them, so no general formula has been agreed. They also record that the law loses accuracy for very small touchscreen targets, and they review its use for children, older users, foot pedals and blind users. Its vulnerable-group section covers children, older users and visually impaired users, and states that the studies of those groups rest on insufficient sample sizes. Its method is a narrative review rather than a meta-analysis, so it reanalyses no data and its comparisons are qualitative. Read it as a statement of agreement and disagreement, not as evidence about any single device.

journal-article · 2024 · Open the source

The paper reports a study of modern code review at Microsoft whose headline finding is a gap between what review is expected to do and what it does. The authors observed 17 developers across 16 product teams, interviewed them, manually classified 570 review comments drawn from real discussions, and surveyed 165 managers and 873 programmers. Finding defects was the most common stated motivation, but defect-related comments were a small proportion of the total and mostly concerned small, local issues. The practice's other products were knowledge transfer, team awareness of what was changing, and alternative solutions to problems. The authors also found that understanding the change is the central requirement of review and that existing tools served it poorly. Its limits are that it covers one company, that comment classification was performed by the researchers, and that the survey measures perception rather than defect escape rates. Its teaching value is the argument that review should be judged on several outcomes, and that the intervention best supported is making a change easier to understand.

conference-paper · 2013 · Open the source

The System Usability Scale produces a single value from 0 to 100, and the score is convenient for comparison and useless on its own, because nothing in it says whether 62 is good. This paper addresses that gap by adding a seven-point adjective rating as an eleventh question to nearly a thousand surveys. The rating correlated strongly with the computed score, at 0.822, and the adjective means spread across almost the whole range, from awful at 20.3 through OK at 50.9 and good at 71.4 to excellent at 85.5. The paper also reports the practical guidance that carried into industry: scores in the seventies are acceptable, and anything below 70 is cause for concern. The companion study cited in the paper covered 2,324 surveys from 206 usability tests and reported a reliability coefficient of 0.91. Two limits matter. The data come from the authors' own accumulated usability practice rather than a random sample of products, so the adjectives are a reference point and not a population norm. And the scale is an ordinal aid attached after the fact to a score whose items are averaged rather than measured on a ratio scale.

journal-article · 2009 · Open the source

Bayer and McCreight set out to organise an index for a dynamic random access file kept on a backup store such as a disc or a drum, and the structure they describe is the B-tree. Their summary records that retrieval, insertion and deletion take time proportional to the base-k logarithm of the index size, where k is a device-dependent number chosen so that performance becomes near optimal, that storage utilisation is at least 50 per cent and generally much higher, and that they ran experiments with indexes up to 100,000 keys, reporting average transactions per second on an IBM 360/44 with a 2311 disc. The teaching value is the argument rather than the structure: the parameter that matters is set by the storage device, so a bound that mentions the logarithm still leaves the hardware to choose. Its limits are its age and its scope. The throughput figures belong to 1972 hardware and cannot be generalised, and only the published summary was read here.

journal-article · 1972 · Open the source

Beck's book proposes two rules: write new code only when a failing automated test demands it, and remove duplication. The cycle is red, green, refactor. The value for a student is not the worked example, which builds a multi-currency money class and then a test framework, but the reasoning. Beck frames the practice as a way of managing fear on problems whose end cannot be seen at the start, and he is explicit that the rules carry design consequences: you design organically with running code as feedback, you must write your own tests rather than wait for someone else, the development environment must respond quickly, and designs must consist of highly cohesive, loosely coupled components just to make testing easy. That last consequence is the most useful sentence in the book, because it shows that testability is a force that shapes design and a cost, not a neutral quality. It is a practitioner book with no data and no measured effect, written to persuade, so it should not be cited for a productivity claim.

book · 2002 · Open the source

This is the most useful single paper on why concurrency is a design decision rather than a checkbox. Its target is the way ANSI SQL-92 defines four isolation levels through three phenomena: dirty reads, non-repeatable reads and phantoms. The authors show that these phenomena and the standard's definitions fail to characterise several widely used isolation levels, including the ordinary locking implementations of the levels they describe. They investigate the ambiguity, arrive at more formal statements, and introduce further phenomena that distinguish the levels better. They then define snapshot isolation, a multi-version scheme that avoids the three standard anomalies and is still not serializable. The section list shows the shape of the argument: isolation definitions, drawbacks of the standard levels, snapshot isolation, and new anomalies. The teaching value lies in the last point. A system can pass the anomaly checklist a student has learned and still permit a result that no serial order of the transactions could produce, and an application designer needs to know that before relying on a level. The paper is technical and assumes familiarity with locking and serializability theory, so it is best read beside a textbook's transaction chapter.

conference-paper · 1995 · Open the source

Braun and Clarke argue that thematic analysis is a method in its own right rather than a step inside other approaches, and that the flexibility which makes it useful is also why so much of it is done badly. The paper sets out what thematic analysis is, locates it against methods that also look for patterns, and gives a six phase procedure: familiarising yourself with the data, generating initial codes, searching for themes, reviewing themes, defining and naming themes, and producing the report. It then names the failures: extracts strung together with commentary that paraphrases them, the interview schedule reported as themes, themes that overlap and do not cohere, and claims the extracts do not support. The most useful teaching material is the account of what a theme is, and the insistence that themes do not sit in the data waiting to be found. The limits are real. This is a methodological argument illustrated from the authors' own projects rather than a study, the six phases have been used as a recipe the authors did not intend, and the 2021 paper shifts the treatment of quality.

journal-article · 2006 · Open the source

This is the revision that a student should read beside the 2006 paper, because it exists to correct what the earlier one is blamed for. Braun and Clarke argue that thematic analysis now exists in several iterations that differ paradigmatically, philosophically and procedurally, and that this plurality is routinely unrecognised by editors, reviewers and authors who then promote coding reliability measures as universal requirements of quality. They identify ten common problems in published work that cites or claims to follow their reflexive approach, and many of those problems rest on the false assumption that thematic analysis is one homogeneous method. They close with twenty critical questions for reviewers and editors, which makes the paper unusually practical. For a student the most useful part is the reporting duty it implies: name the version of thematic analysis you used, say what it assumes about knowledge, and argue for quality in the terms that version supports. The limits are the paper's genre. It is a position statement and a set of appraisal questions, not an empirical comparison of quality across approaches, and it argues from the authors' experience as supervisors and reviewers.

journal-article · 2021 · Open the source

This is a single-volume introduction to the research methods used in human-computer interaction, written because students previously had to learn methods in the disciplines they came from and then adapt them to the field. Each chapter is written by active researchers who use examples from their own work, and the topics covered include experimental design, eyetracking, qualitative research methods, cognitive modelling, the development of new methodologies and writing up research. The book's argument, stated in the publisher description, is that the methods HCI borrows from computer science, psychology, cognitive science, and the organisational and social sciences need a single resource that presents them as they apply in HCI. For a project proposal the value is breadth: it is the reference to cite when the chosen method is not the obvious one, and it supplies the vocabulary for the alternatives a proposal must consider and reject. The limits are that it is an edited collection first published in 2008, so the tool-specific chapters have aged while the methodological ones have not, and only the publisher description and chapter titles were read in this run.

book · 2008 · Open the source

Campbell and Fiske argued that validity is not a property an instrument possesses on its own, but a claim about what its scores mean. Their method is a matrix of correlations among at least two traits, each measured by at least two methods. If two different methods that both target the same trait agree, the measure has convergent validity. If those same-trait correlations are larger than correlations between different traits measured by the same method, the measure has discriminant validity, and where they are not, what the instrument is capturing is largely the method. For HCI the argument matters because so much of the field's evidence rests on self-report scales and single behavioural proxies, and because a proxy that correlates with everything measures nothing in particular. The limits are real. Only the abstract was read here, and the approach is demanding: it requires a battery of measures that most interaction studies do not collect, and it treats reliability informally.

journal-article · 1959 · Open the source

Codd's paper is the origin of the model that most of this module assumes. Its argument is that users of large shared data banks should not have to know how the data is organised in the machine, and that a prompting service which supplies that information is not an adequate substitute. The abstract records the three constructions the paper introduces: a model based on n-ary relations, a normal form for database relations, and the concept of a universal data sublanguage. Section 2 applies operations on relations to the problems of redundancy and consistency in the user's own model. The most useful teaching point is the reason given for the model, which is data independence: application programs and terminal activity should survive changes in internal representation, and even some changes in external representation. That is a maintenance argument rather than a storage argument, and it is why the schema decisions taught here count as design decisions rather than plumbing. Two limits should be stated with it. This note rests on the abstract and the section summary quoted on the IBM page, because the full text is a scan and the ACM page blocked this server. The paper also predates SQL, and the normal form it defines is not the same as the later normal forms taught in textbooks.

journal-article · 1970 · Open the source

Cohen wrote this as a bare-bones treatment because researchers kept ignoring power analysis, and he opens with the evidence for that claim. His own review of a 1960 journal volume found mean power to detect a medium effect of .48, and a later review of 54 articles found that only two mentioned power and that seven studies asserting a null result had a median power of .25. He then gives conventional values for small, medium and large effects across eight tests, notably 0.20, 0.50 and 0.80 for the standardised difference between means and .10, .30 and .50 for a correlation, together with the sample sizes needed for .80 power. The most useful teaching point is his own candour: he says the conventions were set subjectively, describe a medium effect as one visible to the naked eye of a careful observer, and never claim they are facts about any particular research area. The limits are the brevity, the three levels of effect size and the age of the paper.

journal-article · 1992 · Open the source

Introduction to Algorithms is the reference the field teaches from, and its value here is the discipline of the analysis rather than any single structure. It separates worst-case, average-case and amortised cost, states the assumption each of those needs, and gives the recurrence methods that let a student derive a bound instead of quoting one. The publisher's own description of the fourth edition records new chapters on matchings in bipartite graphs, online algorithms and machine learning, and new material on solving recurrences, hash tables, potential functions and suffix arrays. Its method is definition, proof and pseudocode. That method is also its limit for this module: the book is silent about cache behaviour, disk seeks and measurement, so it can tell you how the work grows and not how long an operation takes on your hardware. This is a textbook that has not been read in full here, and the claims drawn from it are standard definitions rather than findings about practice.

book · 2022 · Open the source

Cumming makes two connected arguments. The first is about research integrity: studies should be prespecified where possible, analysed without selecting results after the fact, reported completely and replicated. The second is that null hypothesis significance testing is severely flawed and should be replaced by estimation. He then sets out an eight-step strategy that begins with a question phrased in estimation terms, moves through the choice of effect size, the declaration of the analysis plan and the calculation of intervals, and ends with figures, meta-analytic thinking and open data. The most useful part for a student is the disciplined link between the question and the quantity that answers it; the technique is ordinary, and the change is one of habit. The limits are that this is a position paper rather than a study, and that its strongest claim, that the test should be dropped altogether, is contested: many statisticians keep a decision rule and add estimation beside it. The paper is also dated in its examples, and it does not cover Bayesian alternatives.

journal-article · 2014 · Open the source

This report is the origin of the technical debt metaphor, and reading the original prevents most of the ways the phrase is now misused. Cunningham describes an object-oriented portfolio management system built by incremental growth from a working prototype, and argues that consolidation is what makes incremental development sustainable. The debt passage is explicit: shipping first-time code is like going into debt, a little debt speeds development so long as it is repaid promptly with a rewrite, and every minute spent on not-quite-right code counts as interest. He also warns that an organisation can be brought to a standstill under the load of an unconsolidated implementation, and that excess immature code leads to extreme specialisation of programmers and an inflexible product. It is evidence of practice rather than of outcomes: one team's account, with no measurements, and a metaphor that is generative rather than precise. Its teaching value is the discipline it demands, that debt is a loan whose repayment teams often omit.

conference-paper · 1992 · Open the source

Dynamo is the paper to give a student who assumes that a relational database is always the safe default. Its argument starts from the access pattern rather than from the technology. The authors state that many Amazon services need only primary-key access, and name best seller lists, shopping carts, customer preferences, session management, sales rank and the product catalogue as services where the common pattern of using a relational database would lead to inefficiencies and limit scale and availability. Dynamo therefore offers a simple primary-key-only interface, partitions and replicates data with consistent hashing, and resolves conflicts through object versioning and application-assisted reconciliation. To stay available it sacrifices consistency under some failure scenarios. That is the trade the paper is about, and the authors report that an eventually consistent store was used in production by demanding services. Two limits matter for teaching. The paper deliberately withholds detail: it says that intra- and inter-datacenter latencies, absolute request rates and outage lengths are given as aggregate measures to protect commercial interests. It also describes one company's services, so it supports an argument about access patterns rather than a general law.

conference-paper · 2007 · Open the source

This is the Council's own statement of the diagram that became the standard British description of a design process. It sets out the four phases, Discover, Define, Develop and Deliver, and explains what each is meant to achieve. Its value for teaching is the origin story it tells against itself. In 2003 the Council was promoting design management without a standard way to describe a process, so Richard Eisermann, then its Director of Design and Innovation, asked his team to describe one; the team codified, renamed and popularised an existing idea, and Eisermann says they were standing on the shoulders of giants. The page also records that the Council expanded the two diamonds into a framework for innovation because teams kept returning to the start, and it links to a piece by Eisermann asking whether the diagram is still fit for purpose. Its limits are plain. It is a promotional page rather than a research source, it cites no evidence that the process improves outcomes, and it states no method for any phase. Use it as the primary artefact and as the Council's own caveat, not as justification for a process.

web-resource · n.d. · Open the source

This paper is here for one practical question, and it answers it with a method that can be inspected. The abstract reports a closed card sort in which 191 participants sorted 45 cards into 7 predefined categories on an e-commerce website, so participants assigned items to categories that were supplied rather than inventing categories of their own. The analysis used a resampling procedure and two methods, computing the correlation between the data from all participants and the data from 100 random samples at every sample size from 1 to 191. The reported result is that 8 to 12 participants were sufficient for reliable closed card sort data, and that increasing the number beyond that was more costly than insightful. That is useful, because it turns a folklore number into a stated range produced by a stated method. The honest limitations belong with it. This is a single case study of one retail site, so the range is a starting point rather than a rule. The paper covers closed card sorting, where categories are given, and says nothing about open sorts, which are the method that generates categories. Nothing here was verified beyond the abstract, which was read through a secondary index rather than the publisher's page.

conference-paper · 2025 · Open the source

This is the book that made critical and speculative design legible outside the art school, and it is the primary source for the approach. Dunne and Raby propose design as a tool for creating ideas as well as things, using what-if questions to open debate about the kinds of future people want and do not want, rather than forecasting trends. The publisher's description records the method of physicalising speculation so that it can be argued with, and lists the examples the book uses, including a solar kitchen restaurant, a flypaper robotic clock, a menstruation machine, a cloud-seeding truck and devices for foraging that borrow from synthetic biology. It also records the authors' position that speculation should not be deployed for people to use. The most useful teaching point is the shift of the unit of design from product to future question, which is what distinguishes this work from product design with a social theme. The limits are serious and should be stated with the ambition. Only the publisher's description and reviews were read in this run, not the book itself, the argumentative weight is carried partly by images a text description cannot reproduce, and the critical response is carried by a separate source.

book · 2013 · Open the source

This is the reading list's modelling text and the source for the schema material. Its own publisher description states the aim plainly: to present the fundamentals of database modelling and design, the languages and models provided by database management systems, and implementation techniques, for a one or two semester course. The table of contents matters more here than the prose, because it shows how the topics divide. Entity relationship modelling and the enhanced model occupy two chapters; relational design and normalisation take two more; file structures, hashing and indexing take one; query processing and optimisation take two; transaction processing, concurrency control and recovery take three. The treatment is systematic and conservative, which is the point of it, and the normalisation chapters carry the functional dependency apparatus behind each normal form. Two things to keep in mind when citing it. It is a textbook, so its claims are consensus rather than primary findings, and any measured claim should be traced to the paper that measured it. The edition date is also genuinely muddled across sources: Pearson states a June 2015 publication date with a 2016 copyright year, and retailers list 2015, 2016 and 2022.

book · 2016 · Open the source

The Equality Act is the legal instrument that turns accessibility from good practice into an obligation in the United Kingdom. Section 29 says that a service provider must not discriminate against a person by not providing the service, and that a duty to make reasonable adjustments applies to service providers and to those exercising public functions. Section 20 then defines that duty as three requirements: where a provision, criterion or practice puts a disabled person at a substantial disadvantage, where a physical feature does so, and where an auxiliary aid would prevent such a disadvantage, the responsible person must take such steps as it is reasonable to take. It also provides that where information is involved, reasonable steps include ensuring the information is provided in an accessible format, and that the disabled person cannot be made to pay the cost. The most useful teaching point is that the duty is anticipatory and reasonable, not absolute and not a checklist. The limit is that the Act states duties in general terms, so what counts as reasonable is decided case by case rather than by a technical standard.

legislation · 2010 · Open the source

The European Accessibility Act matters because it extends accessibility duties beyond public bodies and the web to a defined list of consumer products and services, including e-commerce, banking, transport and electronic communications. The text read here sets out the scope in Article 2, applies the requirements to products placed on the market and services provided to consumers from 28 June 2025, and requires Member States to bring national measures into force by 28 June 2022 in Article 31. It also handles the cost question directly, listing the organisational and ongoing costs used to assess a disproportionate burden on an economic operator, which is the honest version of the exemption that students most often misapply. The most useful teaching content is the framing of accessibility as a market requirement with a stated exception, rather than a matter of goodwill. The limits are that a directive binds Member States rather than individual firms directly, so its practical effect depends on national transposition, and the requirements are expressed as essential and functional performance requirements rather than as WCAG success criteria.

legislation · 2019 · Open the source

The book's argument is that design and analysis are one decision made in a sequence, and that most bad studies go wrong before any data are collected. Its structure follows a study from finding a question, through variables and measurement, to design, then to choosing a test and reporting it. The most useful part for a student is the decision route for selecting a statistical test, which turns a confusing menu of procedures into a short set of questions about the design: how many conditions, whether the participants are the same or different, whether the outcome is continuous or categorical. The separate treatment of parametric and non-parametric tests supports this module's position that assumptions should be checked rather than assumed. The limits are that it is a psychology textbook, that its examples come from psychological experiments rather than interaction studies, and that it predates the current emphasis on effect sizes, intervals and pre-registration. The recommended test still needs a modern account of estimation alongside it.

book · 2003 · Open the source

Fielding derives representational state transfer by adding constraints to a null style, and presents each constraint with the properties it induces and the price it charges, which is why it teaches better than a list of rules. Client-server separates user interface concerns from data storage, improving portability, scalability and independent evolution. Statelessness requires every request to carry all the information needed to understand it, with session state kept on the client, and buys visibility, reliability and scalability. Caching trades reliability for latency, and a uniform interface trades efficiency for generality. Its limits are those of a design argument. Nothing here is measured, and the dissertation was written by the author of the HTTP and URI specifications, so parts of it read as retrospective justification of decisions already taken. It reports no developer outcomes and offers no comparison between architectures. It remains the sharpest available vocabulary for arguing about where state should live.

thesis · 2000 · Open the source

Finlay's article is the clearest short treatment of what reflexivity is for, and it is written against the version of it that appears as a confession in the opening pages of a report. She asks how researchers should reflexively evaluate the ways in which intersubjective elements transform their research, and describes the process as full of muddy ambiguity and multiple trails, a swamp of deconstructions, self analysis and self disclosure. Her contribution is a set of maps of five variants of reflexivity drawn from a range of research traditions and illustrated with the reflexive experience of other researchers: introspection, intersubjective reflection, mutual collaboration, social critique, and discursive deconstruction. Each offers different opportunities and challenges, and the diversity suggests competing though overlapping rationales rather than one settled practice. For a student the value is the choice it forces: decide which variant you are using, and what it is meant to do to the analysis. The limits are that the typology's variants overlap in practice, that the account is built from published reflexive writing rather than a study, and that only the abstract could be read here.

journal-article · 2002 · Open the source

The book's structure is its argument. It opens with a worked example, then principles, then a chapter on bad smells, then the catalogue: named transformations for composing methods, moving features between objects, organising data, simplifying conditional expressions, simplifying method calls and dealing with generalisation. The idea a team should take away is not any single refactoring but the practice of naming a small structural change that leaves behaviour intact. Naming makes a change reviewable and reversible, and it turns an argument about taste into a proposal that can be accepted or rejected on the record. The catalogue also supplies a shared vocabulary, so a reviewer can say which smell is being addressed and judge the fix against it. Its limits are those of a practitioner reference. There is no measurement of maintenance cost, the evidence that particular refactorings improve anything is experience and case report rather than controlled study, and the second edition rewrote the examples in JavaScript, a change from the first edition's Java. Use it for mechanics, not for an effect size.

book · 2018 · Open the source

This article is cited here for a single claim, and it is worth being precise about how little of it was read. The title states the result: a sparse table can be stored with constant worst-case access time. That is enough to establish the point the narrative needs, which is that the expected constant-time bound usually quoted for hash tables is not the only bound available, so a designer who says constant time has to say which kind. The bibliographic details were confirmed from Crossref and from the publisher record: Journal of the ACM, volume 31, number 3, pages 538 to 544, 1984. The record page carries no abstract and the article is paywalled, so the construction and its space cost were not read here and are not described in this module. Treated as a proof of possibility rather than as a method to implement, the citation is honest. Anyone repeating the result as a practical recommendation should read the paper first.

journal-article · 1984 · Open the source

This book is the fullest single statement of value sensitive design, written by the researchers who developed the approach. Friedman and Hendry argue that moral imagination and technical imagination can be brought to bear together on the design of technology, and that human values should be engaged at every stage rather than inspected at the end. The publisher's description records that the approach has been under development for more than two decades, that the book presents theoretical foundations and then explains seventeen methods including stakeholder analysis, value scenarios and multilifespan timelines, that experts from ten application domains report on practice, and that the authors close by naming open questions, among them the need for deeper investigation of indirect stakeholders and for further method development. The most useful teaching material is the vocabulary of direct and indirect stakeholders, which turns an ethical concern into something a project can list and act on. The limits are that the book is a normative framework rather than a source of measured effects, and that only the publisher's description was read in this run, so the methods are known by name and not by detail.

book · 2019 · Open the source

Gaver's essay is the corrective that keeps research through design honest, and it is why this module teaches the approach alongside its own critique. His argument has three parts. He concludes from two theories in the philosophy of science that research through design is likely to produce theory that is provisional, contingent and aspirational. He argues that the field's diversity of approaches is natural for a generative endeavour rather than a sign of low standards or a lack of progress. And he proposes that theory be treated as annotation of realised design examples, and especially of portfolios of related pieces, rather than as a body of general laws to be tested. His concluding warning is against impulses towards convergence and standardisation. The essay is most useful for setting the boundary of what such research can claim: statements about what is possible and what a design does, not statements about how a population will behave. The limits are that it is an essay rather than a study, and that only the abstract and institutional metadata were read in this run, so the specific arguments are known from the abstract's summary rather than from the full text.

conference-paper · 2012 · Open the source

Glaser and Strauss set out the programme that gave qualitative research its claim to produce theory rather than only to describe. Their stated enterprise is how the discovery of theory from data, systematically obtained and analysed in social research, can be furthered, in contrast to the literature's concern with how accurate facts can be obtained so that theory can be tested. Their strategy is the systematic choice and study of several comparison groups, which is the origin of constant comparison, and they divide the book into generating theory by comparative analysis, the flexible use of data including documentary and quantitative material, and the implications of grounded theory. The most useful part for teaching is the authors' own treatment of credibility, since the promise in the title is exactly what later writers dispute. The limits are that the book was written for sociology in the 1960s, that its rhetoric of discovery has been questioned on the ground that the analyst's concepts do so much of the work, and that this note rests on the publisher's abstract and the contents rather than on the text.

book · 1967 · Open the source

The authors begin by conceding that usability evaluation has a significant role when conditions warrant it, and then argue that it is harmful when done by rule rather than by thought. Three situations carry the argument. Evaluation of early designs can mute creative ideas or stop a team pursuing several competing ones. Evaluation of an immature invention measures the state of the technology rather than the merit of the idea, so problems that would disappear with development can kill a promising direction. And a confirmatory test that shows a prototype works is an existence proof rather than an attempt at refutation, which makes it weak science by the standard the scientific method sets. A fourth point, that traditional evaluation says little about how a product will be adopted culturally over time, is the one students find hardest and most useful. The value of the paper is the discipline it imposes on choosing a method to fit a question. Its limits are that it is a position essay supported by examples rather than a study, and that its claims about harm are argued rather than measured.

conference-paper · 2008 · Open the source

Guttman introduces the R-tree for the problem of retrieving data items by spatial location, and he opens by arguing that traditional indexing methods are unsuited to objects of non-zero size in multi-dimensional spaces. The applications he names are computer aided design and geo-data, which is exactly the pair this module needs: a map viewport is a geo-data query, and the broad phase of collision detection is a design query about objects that may overlap. The structure groups nearby objects under bounding rectangles so that a search descends only the branches whose rectangles overlap the query, and the paper gives algorithms for search and update and reports that a series of tests indicates good performance. Two limits should be recorded. The paper is a conference contribution from 1984, and only its abstract was read in this run, so the algorithms are described here at that level. And node overlap degrading search is a known property of the structure that the abstract does not discuss.

conference-paper · 1984 · Open the source

Hearst's article is the quickest route to the design trade-off behind faceted search. Clustering groups items by computed similarity. Its advantages are stated plainly: it is fully automatable, it applies to any text collection, it can reveal unexpected themes, and it is useful for clarifying a vague query or separating the senses of an acronym. Its disadvantages are equally plain: the groups are unpredictable, they conflate several dimensions at once, they are hard to label, and their sub-hierarchies are counterintuitive. The recipe example carries the argument. A clustering engine's groups for a chicken recipe query mixed levels of granularity, listing Salad, Crockpot, Chicken Breast and Italian together, and the members of a group did not correspond to its label. Hierarchical faceted categories avoid that by building one hierarchy per dimension, such as dish type, ingredient, cooking method and cuisine. The cost is that the hierarchies must usually be built by hand, so the categories of interest have to be known in advance. Hearst reports the usability studies she reviewed as finding that participants preferred categories, especially for browsing. The article is a short magazine piece and the studies it cites are not reported in detail, so it supports the direction of the trade-off rather than an effect size.

journal-article · 2006 · Open the source

This is the practical guide to using video in qualitative research, and it is written from within ethnomethodology and conversation analysis. The book's position is that audio-visual recording gives access to the fine detail of social action that other methods lose, and that the value of the material depends on how carefully it is gathered and analysed. The chapters follow the work: access, ethics and project planning; collecting audio-visual data; developing preliminary observations; objects, participation and institutional practice; and preparing presentations and publications. The most useful teaching material is the list of decisions the authors put early in the book, which asks how access and consent can be arranged for naturally occurring events, whether to collect in one intense period or successive phases, where the camera should be placed and what it should frame, and whether to make field notes at the same time. Its limit here is that the publisher page, the contents list and free Chapter One were read, while the later chapters were not.

book · 2010 · Open the source

This is a communications paper, closer to an argument than a study, and its argument is a boundary that students often blur. Henry, Abou-Zahra and Brewer contrast universal design, which they define as designing for the widest possible range of abilities and situations, with accessibility, which they say refers primarily to design for people with disabilities. Their claim is that the two overlap and that research and development in accessibility benefits everyone, particularly users with situational limitations such as device limits and environmental limits, but that addressing one set of needs does not guarantee the other. They therefore argue that accessibility should keep its focus on disability, because that focus is what drives work on those specific needs. The most useful part for teaching is the situational-limitation framing, which gives a concrete reason why a designer without a disabled user in the room still has a stake. The limit is that this is a short position piece with no empirical work behind it, so it should be used for the framing and not for evidence about outcomes.

conference-paper · 2014 · Open the source

Hertzum and Jacobsen name and document the evaluator effect: evaluators assessing the same interface with the same method detect markedly different sets of problems. Their review covers eleven studies of three widely used methods, cognitive walkthrough, heuristic evaluation and thinking-aloud studies, and finds that the average agreement between any two evaluators ranged from 5 per cent to 65 per cent. The effect held for novices and for experienced evaluators, for cosmetic and for severe problems, for problem detection and for severity assessment, and for simple and complex systems, and no method was consistently better than the others. The authors trace the variability to vague goal analyses, vague evaluation procedures and vague problem criteria, and they recommend involving multiple evaluators while accepting that the effect cannot be eliminated. For a student the most useful sentence is the one about authority: it is highly questionable to treat a thinking-aloud study run by one evaluator as an authoritative statement about an interface. The limits are that the review pools small studies with differing designs and that the field has no single agreed way to compute agreement.

journal-article · 2001 · Open the source

This paper asks whether the verbalisation that usability testing depends on distorts the behaviour it is used to observe. The authors distinguish classic thinking aloud, which follows the prescriptions of the classic verbal-report model, from relaxed thinking aloud, which follows what practitioners typically describe. Participants used web sites in a silent condition, in a classic thinking-aloud condition and in a relaxed condition, and the differences separate cleanly. Classic thinking aloud had little or no effect on behaviour apart from making tasks take longer. Relaxed thinking aloud changed behaviour in several ways: tasks took longer, participants spent more of the task on distributed visual behaviour, they issued more navigation commands within and between pages, and they reported higher mental workload as measured by the task load index. The teaching value is that the validity objection is specific rather than general, and that the remedy is a precise protocol. The limits are that the study used web sites in a laboratory task, that workload was self-reported, and that the authors say they did not test whether the two regimes find different problems.

journal-article · 2009 · Open the source

Hsieh and Shannon set out the three approaches to qualitative content analysis that most later work distinguishes. Conventional content analysis derives its coding categories directly from the text, which suits a question the literature has not yet addressed. Directed content analysis starts from a theory or from earlier findings that structure the initial codes, which suits a question where existing work is incomplete but usable. Summative content analysis counts, usually of keywords or content, and then interprets the context in which those words occur. The authors state that all three interpret meaning from text and sit within the naturalistic paradigm, and that the differences lie in the coding schemes, in the origin of the codes and in the threats to trustworthiness. Each approach is described procedurally and illustrated with hypothetical examples from end-of-life care, and those examples are what make the paper teachable. The limits are that the abstract is what could be read here, that the three-way division is an organising device rather than an empirically validated taxonomy, and that the threats to trustworthiness are described rather than measured.

journal-article · 2005 · Open the source

Jordan and Henderson's article is the reference that made interaction analysis a method rather than a habit, and it is the origin of the practice of reviewing recordings in a group rather than alone. As the method is described in the review literature, it treats talk, non-verbal interaction and the use of artefacts as parts of one organisation, and it asks what people actually do rather than what they later say they did. The seven foci most often taken from it are the structure of events, the temporal organisation of activity, turn taking, participation structures, trouble and repair, the spatial organisation of activity, and artefacts and documents. Those foci are ways of looking at a recording, not codes to be applied to it, which is the point a student is most likely to miss. The value of the article is the way it connects an analytic focus to the practical business of camera work, transcription and representation. The limits are that it is a long account grounded in the authors' own projects, and that this note describes it through a secondary summary.

journal-article · 1995 · Open the source

Kaplan and Meier solved a problem that arises whenever outcomes are observed over time: some items have not yet reached the event of interest when the observation window closes, so their outcome is unknown rather than negative. Their product-limit estimator uses such censored observations by treating them as informative up to the point at which they were last seen. The paper also states the assumption that makes this legitimate, namely that the lifetime is independent of the potential loss time, and adds that in practice the assumption deserves careful scrutiny. That caveat is the most useful sentence in the paper for an HCI reader, because interaction logs frequently violate it: engaged users generate more events and are observed for longer, so censoring is not independent of the outcome. The paper was written for lifetesting and medical follow-up, and the transfer to sessions and subscriptions is an analogy rather than a measurement. Only the abstract and the estimator's definition could be read here.

journal-article · 1958 · Open the source

This article is the field's own attempt to state what participatory design is committed to, written by two of its central researchers. Kensing and Blomberg characterise participatory design as a maturing research area and an evolving professional practice, and they organise the discussion around three issues that its researchers keep returning to: the politics of design, the nature of participation, and the methods, tools and techniques for participation. They also examine the conditions under which participatory results transfer to workers, user groups and design professionals, and they ask whether participation is sustainable inside an organisation once a project ends. That last question is the most useful thing in the paper for a student, because it names the failure that methods talk tends to hide. The account draws on a review of participatory design and computer-supported cooperative work literatures together with the authors' own practice, so it is a synthesis rather than a new study. It was published in 1998, before the participatory turn in commercial product development, and it says little about digital services. Only the abstract and metadata were read in this run.

journal-article · 1998 · Open the source

Kimbell's review is the strongest academic critique of design thinking as a management product. She identifies three accounts: design thinking as a cognitive style, as a general theory of design, and as a resource for organisations, and she traces the third into business schools, consultancies and the Davos and TED circuit. Her three objections are that the accounts lean on a dualism between thinking and doing, that a generalised design thinking ignores how situated and historically varied designers' practices are, and that the theory privileges the designer as the main agent while treating everyone else as a resource. Her alternative is to study design as a set of situated, contingent practices. It is most useful for teaching students to separate what a designer does from what a firm sells. The limits are that it is a conceptual review rather than an empirical test, that it is Part I of an argument continued in Part II, and that the link is the author's revised pre-publication PDF, which she asks readers not to quote from in place of the published version. Only that PDF, not the paywalled version of record, was read in this run.

journal-article · 2011 · Open the source

This is the book that gives the storage-engine and data-model material a treatment a textbook does not. Its scope is set out in the opening of Chapter 1: many applications need to store data so that it can be found again, remember an expensive result, search data by keyword or filter it, send messages to another process, and process accumulated data in bulk, and data systems are the successful abstraction that hides that work. The preface lists the forces behind the last decade of change, among them internet companies handling volumes of data and traffic that forced them to build new tools, businesses that need short development cycles and flexible data models, and the arrival of infrastructure as a service that lets a small team run a distributed system. Its value in a design course is that it treats the choice between a document model and a relational one as a question about access patterns and about which anomalies an application can tolerate, and then examines the storage engines underneath. The limitation to state is that it is a trade book rather than a research report. It synthesises primary sources and reports experience instead of presenting new measurements, so any claim needing an effect size should be traced to the study it draws on.

book · 2017 · Open the source

Kleppmann compares the designs behind storage and processing systems rather than teaching one product, and his claim is that the underlying principles and trade-offs change slowly even while the tooling changes quickly. Chapter 3, Storage and Retrieval, is the part this module leans on, and its sections run through hash indexes, SSTables and log-structured merge trees, B-trees, a comparison of B-trees with LSM-trees, and other indexing structures. That comparison is the clearest available statement of two points the module makes: an index is a structure the system maintains on every write, so the design that makes reads cheap makes writes expensive; and asymptotic similarity does not predict equal performance when one structure reads sequentially and the other seeks. Its method is synthesis from research and production experience, and it reports no original measurement, so cite it for trade-offs and judgement rather than for a measured effect. The full text is behind a subscription and was not read here.

book · 2017 · Open the source

Knuth, Morris and Pratt present an algorithm that finds every occurrence of a pattern in a text in running time proportional to the sum of the two lengths. The abstract makes three claims worth keeping separate. The bound is the guarantee, and it holds in the worst case rather than on average. The constant of proportionality is described as low enough to make the algorithm of practical use, which is the authors' assertion about their own implementation rather than a measurement anyone can check against modern hardware. And the paper also considers other algorithms that run faster on the average, which is the honest counterweight this module draws on: a worst-case guarantee and typical performance are different properties, and the paper does not pretend otherwise. The teaching use is the shift table, the mechanism that lets the text be scanned once. Only the abstract was read in this run, and the paper is from 1977, so its framing is historical even where the bound still holds.

journal-article · 1977 · Open the source

The book is written by experimentation leads at three large technology companies, and its value to a student is that it covers what a textbook on experimental design rarely does: the operating discipline of a platform that runs tens of thousands of controlled experiments a year. Two ideas drive it. The first is that a test needs a primary metric and an overall evaluation criterion agreed in advance, because a change that lifts a short-term metric such as clicks may damage a longer-term one. The second is that results require trustworthiness checks, since violated assumptions, sample imbalance and carry-over effects produce numbers that look decisive and are wrong. The authors repeatedly invoke Twyman's law, that an interesting or unusual figure is usually wrong. The limits are that this is practitioner experience rather than peer-reviewed research, that the examples are commercial and often disguised, and that the sound advice for a company running millions of users does not transfer directly to a student study with thirty participants. Only the publisher description was read here.

book · 2020 · Open the source

Krasner and Pope describe the model-view-controller paradigm as it was implemented in the Smalltalk-80 environment at Xerox PARC. Their argument is that an interactive application should be factored three ways: the model simulates the application domain, the view displays aspects of the model, and the controller turns input into messages to the model. Two claims carry the teaching value. The first is directional: views and controllers know their model explicitly, but models must not know about their views, which is what allows one model to be presented in several ways. The second is the change protocol: views and controllers register as dependents, and a model broadcasts when it changes, so every dependent updates rather than only the one that received the input. It is an experience report from a single system, with no measurements, and its examples are Smalltalk code that few readers now run. Read it for the design rationale, not as evidence that the pattern improves outcomes.

journal-article · 1988 · Open the source

This is the most directly useful recent evidence on tree testing, and it exists because the method is used far more often than it is tested. The abstract identifies a methodological gap: tree testing is an established way for software professionals to check that an information architecture is logically navigable, but the variants of the method and the software used to run it are not uniform, and the differences between them had not been examined. The study ran two between-subject experiments and collected a dataset of 180 participants and 1800 task completions across 6 experimental conditions, three tree testing variants and three high-fidelity prototype interfaces, analysed quantitatively and qualitatively. The reported result is that the variants produced significantly different results, and that they differed in how closely they approximated navigation in the same information architecture inside a real prototype. The variant with the greatest visibility of previous navigation choices, and direct controls to reverse them, was the most accurate. That is a practical finding for anyone choosing a testing tool. It is also one study of one comparison, the effect sizes are not visible in the abstract, and the work has been checked here only at abstract level.

journal-article · 2025 · Open the source

Landauer and Nachbar had people select goal items through a series of touch-menu choices among subdivided ranges of integers or alphabetised words, varying the number of alternatives at each step. They found mean response time per screen was well described by a function of the logarithm of the number of alternatives, consistent with a decision component in series with a movement component. On that function, breadth is favoured over depth, which is the claim the field usually carries from this paper. The useful part for teaching is the sentence after it: the authors state that this function favours breadth where others might not, and they speculate about when other functions would apply. That turns a design rule into a prediction whose truth depends on the assumed cost of a decision and the assumed cost of a movement. Two limits should be stated plainly. Only the abstract was read here, so the experimental detail is unavailable. And the manipulation was inversely confounded, since varying the number of alternatives also varied the size of the touch target, so breadth and target size move together.

conference-paper · 1985 · Open the source

The paper asks whether user stories, the most common textual notation in agile requirements, do what their advocates claim, and it begins by conceding that no scientific evidence on effectiveness existed before it. The method is a survey with 182 practitioner responses followed by 21 semi-structured interviews, reported as 12 findings. The result worth quoting is that practitioners agree that using stories, a story template and quality guidelines such as the INVEST mnemonic improves their productivity and the quality of what they deliver. Its limits matter as much as its findings. The evidence is perceptual and comes from people who have chosen to work this way, so it cannot show that stories produce better software than a written specification would, and there is no comparison group. The authors treat perceived effectiveness as the object of study rather than a proxy they have validated. Use it for the vocabulary of story quality and for the honest statement that the format rests on professional belief; look elsewhere for outcome evidence. Only the abstract and references were readable in this run.

conference-paper · 2016 · Open the source

The chapter list shows how closely this book tracks the retrieval half of the material. Chapter 1 introduces Boolean retrieval; chapter 2 covers the term vocabulary and postings lists; chapter 6 covers scoring, term weighting and the vector space model; chapter 8 covers evaluation. Those four chapters carry what a design student needs. The definitions used in the narrative come from them: an inverted index is a dictionary of terms with a postings list per term; the document frequency of a term is the number of documents in the collection that contain it; the tf-idf weight combines term frequency with inverse document frequency; similarity between vectors is computed as cosine similarity to compensate for document length; precision is the fraction of retrieved documents that are relevant and recall is the fraction of relevant documents retrieved; and the F measure is their weighted harmonic mean. Chapter 8 is the most valuable for teaching honest reporting, because it explains pooling and then discusses critiques and justifications of the concept of relevance. It is a textbook, so the framework is settled rather than argued, and the worked numbers in its exercises are illustrations rather than study findings.

book · 2008 · Open the source

Miller's paper is the source for this module's refusal of a universal response-time threshold. The abstract records that the literature of the time abounded in controversy about the limits of system response time, and that two semantic problems prevented anyone from settling it. The first is the question of response time to what, and the implication the author draws is that different human purposes and actions will have different acceptable or useful response times. That is the useful part, and it is a stronger claim than a list of numbers, because it makes the human task the unit of analysis for a latency budget. Its limits matter. The paper is an argument and a review rather than an experiment, several of its sources are 1960s project reports, and it was written about terminals, so nothing in it transfers as a measurement of a modern interface. Only the abstract was available in this run, and the three thresholds often attributed to it are not in the abstract, so this module quotes no number from it.

conference-paper · 1968 · Open the source

Muller's chapter is the standard survey of participatory design techniques, and it supplies the framing this module uses. He borrows the idea of a third space from cultural theory to describe a hybrid region that belongs neither to the software professionals nor to the end users, and argues that this in-between space is where new insights and shared languages are produced. The techniques he surveys include choosing a site for the work, workshops, storytelling through text, photography and drama, games for analysis and design, and the co-creation of descriptive and functional prototypes. The chapter opens with the field's political origin in the Scandinavian workplace-democracy movement, which makes it a good single source for both the history and the methods. It is also unusually candid about the field's move into commercial practice, noting that some greeted this mainstreaming with enthusiasm and others with dismay. The limits are that it is a handbook chapter rather than a study, that it is now more than twenty years old, and that the link is a university course copy rather than the published version of record. The abstract and introduction were read in this run.

book-chapter · 2003 · Open the source

Myers shows that finding a longest common subsequence and finding a shortest edit script are dual problems, and that both are equivalent to finding a shortest or longest path in an edit graph. From that view he derives an algorithm whose time and space are proportional to N times D, where N is the sum of the lengths of the two sequences and D is the size of the minimum edit script, and he reports expected time of order N plus D squared under a basic stochastic model. A refinement needs only N space, and the use of suffix trees gives a variant in N log N plus D squared. For teaching, the important consequence is that the cost tracks the size of the difference rather than the size of the files, which is why a diff between two nearly identical files is cheap and why a claim that diffing must be quadratic is unsafe. The limits are that the stochastic model is an assumption about the input, and that only the abstract was read here.

journal-article · 1986 · Open the source

Newell and Gregor ask a methodological question rather than a technical one: what does it take to design for everyone, including people with disabilities? Their answer is that universal usability is best served by extending the existing user-centred design methodology, which they call user sensitive inclusive design, rather than by inventing a new paradigm. The paper is most valuable for three claims a student can argue with. First, that ability and environment interact, so a capable user in a hostile environment (a spacesuit, a cold-weather glove, a moving train) is effectively disabled, and a disabled user in a favourable environment is not. Second, that users should be involved throughout a project, but that trained researchers should keep control of research questions and methods, which places their position in tension with participatory action research. Third, that involving disabled users raises real ethical and practical problems, including consent, communication and conflicting requirements. The limit is that the paper is a 2000 position piece, the paradigm it proposes is not then tested, and its own worked examples are retrospective.

conference-paper · 2000 · Open the source

This page states the three thresholds that interaction design uses for feedback decisions: 0.1 second is the limit for the user to feel that the system is reacting instantaneously, 1 second is the limit for the flow of thought to stay uninterrupted even though the delay is noticed, and 10 seconds is the limit for attention to remain on the dialogue. The second of these is the one that matters for direct manipulation, because the text says that beyond a tenth of a second and below a second the user loses the feeling of operating directly on the data. The page also notes that a response can be too fast, and that interface animation should be timed by a real-time clock rather than by the speed of the machine. Read it as an authority for the numbers and as a model for how to hedge them. The limits are perceptual tendencies rather than rules, they were written before current hardware, and the page is an excerpt from a book chapter with no method section. The primary studies it cites were not read in this run, so the thresholds are attributed to Nielsen's summary rather than to them.

web-article · 1993 · Open the source

This is the article students cite and rarely read. Nielsen reports earlier work with Tom Landauer on a model in which the proportion of usability problems found by a test with n users depends on the proportion any one user finds, with a typical value of 31 per cent averaged across many projects. At that value five users find about 85 per cent of the problems, and the article argues the money is better spent on three small tests than one large one. It also gives the caveats, which are the reason to assign the reading: the formula holds only for comparable users, and a site with several distinct user groups needs more participants. What the page does not do is examine its own assumptions. The model treats the probability of encountering each problem as the same for every user and independent between users, which is false for real interfaces, because ease of discovery varies sharply across problems. That is why the article is best paired with the evidence on variability in evaluation findings rather than quoted as a guarantee.

web-article · 2000 · Open the source

This is the most useful single article on the contested question the module opens with, because it takes both sides seriously. It explains what an intercoder reliability figure is, reviews the arguments for it, including systematicity, communicability, the discipline it imposes on coders and the dialogue it provokes inside a team, and reviews the arguments against, including the objection that reliability is imported from a paradigm that assumes a single reality. The practical sections are the part a student cannot get elsewhere: how many coders are needed, how much of the data to double code, how to segment data and why that choice moves the figure, how the number of codes affects agreement, and how the process can constrain the analysis it is meant to serve. The authors also warn about false precision, reporting research showing that nonsensical mathematical information can inflate judgements of a report's quality, and note that beginners can fixate on agreement at the expense of the analysis. Its limits are that it is a guidance paper based on a literature review and the authors' own practice rather than a study, and that it works inside the thematic and content analysis traditions.

journal-article · 2020 · Open the source

The project replicated 100 experimental and correlational studies drawn from three psychology journals, using high-powered designs and the original materials where those were available. Ninety-seven per cent of the originals had significant results and 36 per cent of the replications did; replication effects were about half the size of the originals; 47 per cent of original effect sizes fell inside the 95 per cent interval of the replication; and 39 per cent were judged to have replicated. The authors also report that replication success was better predicted by the strength of the original evidence than by characteristics of the original or replication teams, which is a more encouraging finding than the headline rate. The limits are important for honest use. This is psychology, weighted towards social psychology, so extending it to HCI is an argument rather than a measurement. Effect sizes can decline for legitimate reasons, including publication bias in the originals and moderators the replication could not reproduce, so a failed replication is not proof that the original was wrong.

journal-article · 2015 · Open the source

This is the standard reference for the assistive half of the topic, and its value is the framework rather than a set of design rules. Polgar and colleagues organise the field through the Human Activity Assistive Technology model, which places any technology inside four elements: the human, the activity, the assistive technology and the context, so that a device is only meaningful in relation to what a person is trying to do and where. The table of contents shows the breadth the module needs, from the human-technology interface to seating, mobility, manipulation, and access to mainstream information and communication technologies, and the sixth edition adds global provision, ethics and sustainability. Its most useful contribution to a design course is the reminder that assistive technology is a service as well as a product, involving assessment, configuration, training and review. The limits are important to state. This is a textbook for occupational therapy students rather than a research monograph, and the edition cited here was verified from the publisher's page only, so no claim about its contents should be attributed to it beyond what the publisher describes.

book · 2024 · Open the source

This paper is the honest counterweight to advocacy for test-first development. The authors pooled 27 primary studies comparing test-driven development with a control process, and measured external quality, meaning defects visible to users, and productivity. Their headline results are modest: a small positive effect on external quality, and little to no discernible effect on productivity. Subgroup analysis found the quality gain and the productivity drop to be larger in industrial studies than in academic ones, and a larger productivity drop where the difference in test effort between the two processes was substantial, while a positive correlation was found between task size and the size of the quality improvement. The limits are the usual ones for a meta-analysis. Primary studies use different control processes, some comparing against waterfall and others against a test-last iterative process, industrial data are scarce, and small pooled effects are easily overwhelmed by context. Read it as a model of careful synthesis, and quote the subgroup results rather than the headline.

journal-article · 2013 · Open the source

This is the reference for the information architecture half, and its standing comes from staying in print across four editions. The preface records the lineage: the first edition appeared in 1998 as Information Architecture for the World Wide Web, before the terms now used for the field existed, and the authors describe their own working life as one of continuous change in the technology of delivery. The book treats information architecture as the structural design of shared information environments and gives most of its attention to three systems, organisation, labelling and navigation, with search systems alongside them. Its practical value in a design course is that it turns each system into decisions a team can make and defend: what a category is called, how a person knows where they are, and how a site is extended without breaking the structure that already exists. The limits follow from the genre. It is a practitioner's guide rather than a research monograph, so its recommendations rest on experience and worked examples rather than measured effects, and any number or claim about a method should be traced to the study that produced it.

book · 2015 · Open the source

This paper is where conversation analysis begins as an organised programme. It takes turn-taking to be fundamental to conversation and to other speech-exchange systems, proposes a model for the turn-taking organisation of conversation, and examines that model for its compatibility with a list of grossly observable facts about conversation. The conclusion recorded in the abstract is that such a model will be characterised as locally managed, party-administered, interactionally controlled and sensitive to recipient design, with general consequences drawn out and contrasts sketched against other speech-exchange systems. For anyone analysing an interface, the value is the shift of the analytic object: not what an utterance means in the abstract, but how the participants jointly produce the order in which it is heard. The limits should be stated plainly. The paper is a formalisation of one system in one language, its data are recordings of ordinary conversation made in a particular period, and the model is a description of order rather than a causal account. Only the abstract was read in this run, so the note claims no more than the abstract supports.

journal-article · 1974 · Open the source

Sanders and Stappers describe how design research moved from designing for users to designing with them, and they argue that this shift changes the roles of everyone involved. The person who will be served by the design becomes the expert of their own experience and takes part in knowledge development and idea generation, while the researcher becomes a facilitator rather than a translator between the user and the designer. Their most useful contribution is a framework of levels of participation: four levels of creativity in people's lives, doing, adapting, making and creating, which differ in the expertise and interest they require. That framework lets a researcher pitch a method to the participants rather than to an ideal, and it explains why a workshop that asks everyone to create will serve a narrow group. They illustrate the argument with projects in which people build physical models, including nurses designing a future patient room with a three-dimensional toolkit. The paper is a position piece rather than a study, and the case is presented to illustrate rather than to evaluate.

journal-article · 2008 · Open the source

Sedgewick and Wayne cover fundamentals, sorting, searching, graphs and strings, and they motivate each structure by what it makes possible in an application. The distinctive choice is that the code is real rather than pseudocode, and the authors publish a companion booksite with excerpts, source code, exercises, test data and lecture videos at no cost, which makes the book unusually usable for a student without a library budget. For this module the useful parts are the treatment of stacks, queues and deques, the symbol table chapter covering hash tables and balanced trees, the trie material, and the substring search material. The limits should be stated. The implementations are Java, so the constants and the memory figures belong to the authors rather than to the reader's machine, and the performance discussion is illustrative. The book reports no controlled studies, so it is a source for method and for code, not for evidence about how people use an interface.

book · 2011 · Open the source

The authors argue that flexibility in collecting, analysing and reporting data raises the real false positive rate far above the nominal five per cent, without any intent to deceive. They simulate four common degrees of freedom and show the individual effects, from about nine per cent for choosing among two dependent variables, to 60.7 per cent when all four are combined. They then report two real experiments showing that listening to a Beatles song made participants about a year and a half younger by their own birth dates, achieved only through legitimate analyses reported selectively. The constructive half of the paper is a short disclosure checklist: six requirements for authors and four guidelines for reviewers, covering the reporting of all variables, the rule for deciding sample size, and the full set of analyses run. The limits are that the simulations assume data drawn from a true null and that disclosure improves transparency rather than correcting the underlying inference, so a reported result can be fully disclosed and still weak.

journal-article · 2011 · Open the source

The paper takes two statistics that text entry research had begun to use, the minimum string distance error rate and keystrokes per character, and shows that each is inadequate alone. Keystrokes per character measures effort and sits near 1.0 for error-free typing on a physical keyboard, while multi-tap entry on a numeric keypad averaged 2.03. The important finding for a student is that the two statistics move in opposite directions. A participant who leaves an error uncorrected lowers keystrokes per character and raises the string distance error rate; a participant who repairs everything does the reverse. A single number therefore reports a decision the participant made as well as the quality of the input method. Their replacement framework separates corrected from not-corrected errors, combines them into a total error rate, and adds correction efficiency, conscientiousness, and utilised and wasted bandwidth. The most useful contribution is conceptual rather than statistical: it shows that any text entry comparison must state which error rate it reports. The limit is that the field still has no single agreed measure.

conference-paper · 2003 · Open the source

This is the book that gave interaction research its scepticism about plans, and it is still worth reading for the precision of the objection. The publisher's own summary states the argument: the planning model of interaction favoured by most artificial intelligence researchers does not take sufficient account of the situatedness of most human social behaviour, and the problems this creates are illustrated through the careful analysis of a recorded interaction between novice users and an intelligent machine whose design failed to accommodate essential resources of successful human communication. The book then proposes a fresh characterisation of human-computer interaction drawing on the social sciences. For a designer the value is the question it forces: what does this system assume about the person using it, and what ordinary resources has that assumption written out. The limits are those of a single study. One machine and a small number of users in one setting cannot establish a general claim, and the second edition adds substantial new material, so a reference to the 1987 text and a reference to the 2007 book are not interchangeable.

book · 1987 · Open the source

These regulations are the public sector instrument that most UK designers will actually be held to, and they are narrower and more concrete than the Equality Act. They apply to public sector bodies, require their websites and mobile applications to meet an accessibility requirement, and require a published accessibility statement that is kept under review. The provisions read here also show that the regulations are not absolute: regulation 4 sets out a list of content that may be exempt, including live time-based media and third-party content that the public sector body neither funds, develops nor controls. The most useful teaching point is that the exemption list is where honest accessibility work happens, because a team that merely points at the third-party carve-out has usually failed to procure an accessible third party. The limits are that the regulations cover public sector bodies rather than private firms, that a published statement can describe non-conformance without curing it, and that they were made in 2018 and should be read against the later enforcement guidance rather than in isolation.

legislation · 2018 · Open the source

This review essay is the fair account of the objections to speculative design, and it is placed here so that the approach is not taught without its critics. Tonkinwise accepts the book's case for speculating about possible futures rather than realising market-feasible ones, and then argues that a shopping framework and a taste regime sit underneath the work. He objects that the authors' first-person plural slips between meaning themselves and meaning humanity, a habit that makes a private aesthetic look like a general claim. He objects that class difference is absent even though the 2008 financial crisis is central to the book's argument, and he notes that other kinds of diversity are thin. He also objects that the projects rarely make an affirmative case for a future anyone would want, so an audience is left hoping the vision does not arrive rather than debating how to reach a better one. The essay is a review by one critic rather than a study or a consensus, and it is polemical in tone. The published PDF was read in this run.

journal-article · 2014 · Open the source

This note exists because WCAG covers cognitive and learning disabilities least well, and it says so. The guidance sets out objectives rather than testable criteria: help users understand what things are and how to use them, help them find what they need, use clear content, and design so that users avoid mistakes and know how to recover. It recommends familiar conventions, plain words, short blocks of text and consistent placement of help. It also includes user stories and personas, and asks that people with cognitive disabilities are included in design and testing. What makes it useful in teaching is that it is honest about its status. It is a working group note, not a Recommendation, the W3C states that publication does not imply endorsement and that its groups do not expect it to become a Recommendation, and it cannot therefore be used as a pass or fail test. Its value is that it names needs the criteria omit, and it gives a designer something concrete to design against where conformance would otherwise stop.

standard-guidance · 2021 · Open the source

The Understanding documents are where the reasoning behind WCAG lives, and they are what turn a list of criteria into something a designer can apply. Each page states the goal, what to do, why it matters, and the intent, so a reader can see what problem a criterion was written to solve. The pages read here show why the reasoning matters. Target Size Minimum defines a 24 by 24 CSS pixel floor together with five exceptions, and the spacing exception explains why a small inline link is excused. Non-text Contrast sets a 3:1 ratio for the cues that identify a control or a state, and notes that thin strokes can pass on the computed colour and fail in practice. Accessible Authentication names copy and paste and password managers as the helper mechanisms that satisfy the criterion. There is a limit to record: these documents are informative rather than normative, they carry no stated publication date, and a student citing a criterion should cite WCAG 2.2 itself for the requirement and the Understanding page only for the explanation.

standard-guidance · n.d. · Open the source

The authors set out ten advantages of the Bayesian approach and then answer the objections they expect from practising researchers. The two advantages most useful to a student are that a Bayesian analysis quantifies evidence for and against a hypothesis rather than only rejecting a null, and that the evidence can be monitored as data accumulate without the analyst having to declare the intention behind the data collection. That second point bears directly on the recurring problem of peeking at results and stopping when they look favourable, which inflates the error rate in a frequentist analysis. The paper is a position piece, and it is candid that priors must be chosen: a reader who rejects the prior will read the posterior differently, which is a genuine cost rather than a technicality. It is open access in the Psychonomic Bulletin and Review, and the companion paper covers the JASP software. Only the abstract was read in this run, so the ten advantages are summarised rather than examined one by one.

journal-article · 2018 · Open the source

Wagner and Fischer define the string-to-string correction problem as determining the distance between two strings, measured by the minimum cost sequence of edit operations needed to change one into the other. The operations they consider are changing one symbol into another, deleting a single symbol, and inserting a single symbol, and they give an algorithm whose running time is proportional to the product of the two lengths. The applications they name are automatic spelling correction and finding the longest sequence of characters common to two strings, which is why this module uses the paper for fuzzy matching and for the logic that sits behind a diff. The limits matter for how it is taught. The product-of-lengths cost is the full table, so the method ranks candidates well and searches a large corpus badly, and the module therefore pairs it with an indexed first pass. The abstract reports no evaluation on real misspellings, and only the abstract was read here.

journal-article · 1974 · Open the source

This book is the clearest public account of how a large engineering organisation organises teamwork, and its full text is free to read. It is a collection of practice chapters written by practitioners, curated by three Google engineers, rather than a research monograph. The chapters used here are Chapter 2 on working well on teams, Chapter 9 on code review and Chapter 16 on version control and branch management. Chapter 9 supplies the concrete flow the narrative draws on: every change is reviewed before commit, one reviewer's agreement recorded as looks good to me permits the commit, and review is explicitly not the occasion to reopen settled design decisions. Chapter 16 argues that formal version control is a baseline no organisation above a few people can do without. The value is the specificity of the practices and the honesty about trade-offs. The limit is context and method: these are experience reports from one very large company, written to persuade, with no comparison group. Nothing here should be cited as a measured effect.

book · 2020 · Open the source

This is the reframing the module teaches. Wobbrock and colleagues argue that accessible computing has tended to centralise disability, and that a dis-ability is defined by a lack rather than a capacity, which limits what the field builds. They shift the guiding question from what disability a person has to what a person can do, and they tie the argument to the observation that ability is not static but changes with context. They offer seven principles and describe the research projects behind them, including SUPPLE, which measures a user's pointing performance and then rearranges and resizes the interface to fit it. The most useful teaching passage contrasts this with assistive technology that inserts a component between the user and an unchanged system, placing the burden of change on the person. The limits are worth stating plainly: this is a conceptual and position article, the systems it describes are research prototypes, the paper reports the high abandonment of specialised access technology without the underlying study, and the difficulty of reliably measuring ability is acknowledged by the authors themselves.

journal-article · 2011 · Open the source

The authors argue that gesture sets invented by designers are not necessarily what users would do, and they test this by eliciting gestures rather than proposing them. Each participant was shown the effect of a command and asked to perform the cause, which removes the problem of a participant simply repeating a gesture they had been shown. Twenty participants produced 1080 gestures across 27 commands performed with one or two hands, and the results were paired with think-aloud data. Four findings carry the argument: users cared little how many fingers they used, one hand was preferred to two, desktop idioms shaped what people attempted, and agreement between participants varied greatly, with some commands producing almost no consensus. The last of these is the operational lesson, because the authors conclude that commands with little agreement need on-screen widgets rather than gesture alone. The method is also reusable: agreement per command is the closest thing the field has to a guessability score. The limits are the single tabletop prototype, twenty non-technical participants, and gesture conventions that are culturally local.

conference-paper · 2009 · Open the source

WCAG 2.2 is the normative specification for web content accessibility, written so that its requirements can be tested. It organises success criteria under four principles, perceivable, operable, understandable and robust, and its criteria are technology-independent. Its own framing states the limits that teaching should respect. It is designed to be testable through a combination of automated testing and human evaluation, so an automated conformance check is necessary and not sufficient. It will not address every user need for people with disabilities, and the introduction records the difficulty of agreeing testable criteria for cognitive, language and learning disabilities. Conformance is cumulative in one direction: content conforming to WCAG 2.2 also conforms to 2.1 and 2.0, and publication of 2.2 does not deprecate the earlier versions. For an engineering module the value is that accessibility arrives as testable statements that can be attached to a build and to acceptance criteria, with a stated human half. The specification is not a study and makes no measured claim about usability.

standard · 2024 · Open the source

This is the paper that made research through design legible to human-computer interaction research. Zimmerman, Forlizzi and Evenson argue that the field had struggled to integrate design into research even though design was strong in practice, and they propose a model in which the designer produces novel integrations of research in order to make the right thing, a product that moves the world from its current state to a preferred one. On their account this lets design researchers contribute through their strength in addressing under-constrained problems, which scientific method handles poorly. To make the contribution assessable, the paper offers four lenses for evaluating a design research contribution, and it illustrates the model with three worked examples. The most useful thing to teach from it is the claim that the artefact can carry knowledge, which is the assumption the whole approach rests on and the one that later criticism tests. The limits are that only the abstract and metadata were read in this run, so the four lenses are known by name and not by content, and that the paper is a proposal whose criteria were widely adopted without being validated empirically.

conference-paper · 2007 · Open the source

The ACM Code is the profession's most cited statement of obligation, and its value for teaching is its concreteness. It is organised into general ethical principles, professional responsibilities, leadership obligations, and compliance. Two principles do most of the work for a designer. Principle 1.2 requires a professional to avoid harm, to consider the impacts of a decision on everyone affected, to analyse the consequences of data aggregation and the emergent properties of systems, and to report signs of system risk, naming whistleblowing when leaders will not act. Principle 1.6 requires only the minimum personal information necessary, defined retention periods, no reuse of personal data without consent, and transparency about what is collected. The companion case studies, which apply the principles to concrete situations, are the most useful part to assign. The limits are substantial. The Code binds ACM members and award recipients rather than firms, so it constrains an individual inside an organisation that may not share it. Enforcement exists but is rarely used, and the document does not fix the sanctions. Treat it as a source of reasons and a standard to be judged against, not as evidence that the profession behaves well.

professional-code · 2018 · Open the source

Abowd and Mynatt survey a decade of ubiquitous computing research and organise it around three themes: natural interfaces, context-aware applications, and automated capture and access. Their most useful contribution is the argument about scale. Ubicomp raises questions of device number, physical space and number of people, and they add a fourth, time. Scaling interaction with respect to time produces everyday computing, a continuous relationship with a system that moves computing from a tool to a companion. That framing is what makes interruption and resumption core problems rather than edge cases, because a companion is present while other work happens. The paper is a review, so its individual claims summarise other work and should be followed back to their sources. It is also dated: it treats the personal digital assistant as the emerging form and does not anticipate the smartphone, so its technology examples have aged while its analytic frame has not. Use it for the map, not for the inventory.

journal-article · 2000 · Open the source

Acquisti, Brandimarte and Loewenstein review empirical work on privacy behaviour from the social and behavioural sciences, and they organise it under three themes that are immediately useful in design. First, people are uncertain about both the consequences of disclosure and their own preferences over those consequences. Second, concern about privacy depends on context, so the same person will disclose readily in one setting and refuse in another. Third, concern is malleable, and can be manipulated by commercial and governmental interests. The third theme explains why the privacy paradox, the gap between stated concern and observed disclosure, is less paradoxical than the phrase implies. The authors draw policy observations from the review. Two limits should be recorded. It is a review of a heterogeneous literature rather than a new experiment, so it offers mechanisms rather than an effect size. And it examines behaviour rather than law, so it cannot tell a practitioner what the UK GDPR requires; the regulator's guidance does that.

journal-article · 2015 · Open the source

Amershi and colleagues codified over 150 design recommendations from academic and industry sources into 18 guidelines organised by the phase at which they apply, and validated the set in a study with 49 design practitioners who applied it to 20 popular AI-infused products. The most useful contribution here is the vocabulary. Guidelines that ask a system to make clear what it can do, and how well it can do it, give a team a shared way to talk about the uncertainty problem, while the guidelines on learning from behaviour, granular feedback, global controls, scoping when in doubt, efficient correction and dismissal, cautious updating and notifying users about changes map onto the design problems that adaptation creates. The example application and violation lists are the teaching device, and the participant report of recommendations shifting weekly with no explanation of what changed is one of them. Two limits matter. Validation established that the guidelines are relevant to real products, not that following them improves any measured outcome, and the authors do not claim otherwise. They also concede that inconsistency is intrinsic to some AI components, so established advice on consistency and error prevention is hard to apply directly.

conference-paper · 2019 · Open the source

This paper established the phrase the field now uses for the problem the module teaches, the unintended consequences of clinical information systems. Ash, Berg and Coiera draw on qualitative studies of patient care information systems in the United States, the Netherlands and Australia, and report errors they call silent, meaning that they emerge only when a technical system meets the reality of clinical work. Their framework has two parts. Errors in entering and retrieving information include the loss of cognitive focus that follows from demanding elaborate structured data entry. Errors in communication and coordination follow from treating communication as information transfer, so that a physician believes that entering an order has told the team about it. The value of the paper is that it names the mechanisms and locates them in the mismatch between rigid software and contingent work. Its limits are equally clear. It is qualitative and argumentative, it establishes mechanisms rather than rates, and its categories come from the authors' own studies and judgement.

journal-article · 2004 · Open the source

The accreditation guidelines are the document that makes individual accountability a requirement rather than a preference, and they are the reason this module exists in the form it does. The relevant clauses sit in section 2.5 on major projects. They state that an Integrated Masters degree must include a major group project, commonly at Level 6 or 7; that its assessment is typically made up of a group mark, an individual mark and peer review; and that the marks each student receives must reflect their individual achievement and input. An earlier clause requires that, where a major activity is undertaken as a group enterprise, individual contribution is measured against all the learning outcomes. The list of things a project must let a student demonstrate also names critical self-evaluation of the process. The most useful thing about the document is its directness: it tells a team what the assessment must be able to show. Its limits are equally clear. It is a regulatory standard, not evidence, and it says nothing about which accountability mechanism is fair or effective.

standard · 2022 · Open the source

The BCS Code of Conduct is the standard a member signs up to on joining, and it is enforced through disciplinary action that can end in expulsion. This module uses it because it turns the vague idea of professionalism into clauses a student can act on. The four principles are public interest, professional competence and integrity, duty to the relevant authority, and duty to the profession. The competence clause bears most directly on practice: a member must work only within their competence, must not claim competence they lack, and must keep their knowledge current. The public interest clause names discrimination and equal access to the benefits of IT as professional matters rather than private ones, which is the basis for treating inclusion as a duty. The criticism clause asks members to respect alternative viewpoints and to give and accept honest criticism of work, which is the standard this module applies to reviewing a colleague's design. The code is short and general, so it states obligations without prescribing how to meet them. Version 8 is dated 8 June 2022.

professional-code · 2022 · Open the source

The CITP standard is the framework behind Chartered IT Professional registration, and it is useful to a student mainly because it shows what evidence a professional claim rests on. Applicants must demonstrate depth and breadth of technical knowledge against SFIA skills, with depth at level 5 for at least two skills and level 3 for two more, and breadth at level 3 across five skills outside their specialism. Beyond technical expertise, the standard asks for the professional behaviours of leadership, decision making and problem solving, evidenced from real projects, and for the values of a chartered practitioner: keeping skills current, acting as a role model, and considering the ethical implications of the work. Continuing professional development is the mechanism by which those values are evidenced, with a plan for future development. The requirement is deliberately evidenced rather than asserted, which is what makes it usable as a template for a development plan. The page carries no visible date, so it is cited without a year. A student should check which BCS specialism applies to their planned route, because the mapping is not automatic.

professional-standard · n.d. · Open the source

The BCS Code of Conduct is the instrument a UK member of the chartered institute accepts, and it is shorter and blunter than the ACM Code. It states four principles: you make IT for everyone, you share what you know, you respect the organisation or individual you work for, and you pass IT on. Each principle carries numbered duties. The duties most likely to bite on a designer sit in professional competence and integrity and in the duty to the relevant authority: to work only within your competence, to understand and comply with the legislation that applies, to respect alternative viewpoints, and not to misrepresent or withhold information about the performance of products, systems or services. BCS publishes a route for reporting a breach, which is what makes the code an instrument rather than an aspiration. The limits are the same as for any professional code. It binds members rather than employers, and no version date is printed on the page, so a student should cite the date it was read. Enforcement depends on a complaint, so the code is a backstop rather than a routine control.

professional-code · n.d. · Open the source

This page is a short practice report from BCS describing how one department runs assessed group work, and it is useful precisely because it records a design decision and its reasoning rather than a result. King's College London requires students to complete three group projects in sequence rather than one, on the argument that a single project makes a student's experience depend heavily on who else was in their team. The first two are smaller and differ in how teams are composed, so that every student works in a homogeneous and a heterogeneous team. The substantial third project is a ten-week engagement with a real external client. Throughout, a monitoring system records attendance at meetings, minutes, contribution statistics from GitHub and a weekly experience survey, and runs a peer assessment afterwards, with the stated purpose of detecting social loafing, communication failure, unresolvable conflict and non-participation early. The most useful element is that evidence of contribution is collected continuously as a by-product of the work. The limits are that this is an account of practice, with no outcome data and no comparison group.

web-page · n.d. · Open the source

Bainbridge's paper is four pages long and forty years old, and it is still the cleanest statement of a problem that modern AI features reproduce exactly. Her argument is that automating the routine parts of a task expands rather than removes the human's problems, because the designer who removes the operator still leaves the operator the tasks that could not be automated. What remains is the abnormal case, which is hardest to handle and the case for which the operator now has least practice; she notes that physical skills deteriorate when unused, so an experienced operator who has spent a long period monitoring may take over as an inexperienced one. She also argues that sustained monitoring is not something people sustain, citing vigilance research for the claim that effective visual attention to a source where little happens cannot be held for more than about half an hour. The teaching value is that these are properties of the task allocation rather than of any technology. The paper is a brief communication drawing on process-control and flight-deck experience rather than a programme of experiments, so it is cited for the mechanism, not for a measured size of effect.

journal-article · 1983 · Open the source

Balasubramanian and colleagues measured how much energy 3G, GSM and WiFi consume on real phones, and their central finding is that energy tracks the traffic pattern rather than the data volume. On 3G, nearly sixty per cent of the energy of a typical transfer was tail energy, spent lingering in a high-power state after the bytes had moved, and the tail time was longer on 3G than on GSM. The consequence is counter-intuitive and directly useful in design: a few hundred bytes sent intermittently can cost more energy than a megabyte sent at once, because each burst pays the tail again. The authors build a scheduling protocol, TailEnder, that batches and defers delay-tolerant traffic, and report lower energy for email, news feeds and web search. The most useful part for this module is the reframing of battery as a constraint on when an application may transmit rather than only on how much it sends. The limits are the era and the hardware: the measurements predate later radio standards and modern phones, so the direction of the effect is more durable than the specific figures.

conference-paper · 2009 · Open the source

Bao and Intille collected accelerometer data from twenty people wearing five biaxial sensors, without a researcher supervising the sessions, and trained classifiers to label twenty everyday activities. Their best decision trees reached about 84 per cent overall accuracy. The number is not the useful part. The study's contribution is the demonstration that recognition accuracy is conditional. Some activities were recognised well using training data drawn from other people, and others required training data from the person being recognised, which means a product cannot assume one figure transfers to a new user. The paper also reports that reducing to two sensors, on thigh and wrist, cost only a little accuracy, which is a practical result for anyone budgeting sensor count against battery and comfort. The limits are the activity set and the recording protocol: twenty scripted everyday activities are not open-ended behaviour, and the sessions were semi-naturalistic rather than unconstrained. Read the figure as an upper bound under those conditions, not as a product specification.

conference-paper · 2004 · Open the source

This paper is the modern statement of the socio-technical position for a systems audience, and it is useful here for the diagnosis it offers rather than for evidence. The authors note that a socio-technical approach to system development is widely acknowledged to produce systems that are more acceptable to end users and to deliver better value to stakeholders, and that it is nevertheless not widely practised. They analyse why, including problems with the better known design methods, and set out a framework for socio-technical systems engineering that joins organisational change and system development through two kinds of activity, sensitisation and awareness, and constructive engagement. They then list research problems, among them how to apply the approach cost-effectively and how to integrate it with existing software engineering practice. Its value in teaching is the explicit claim that the technical system and the work organisation are one problem, and its candour about the approach's own record. The limit is methodological: only the abstract could be read here, and the framework is argued from a review rather than validated as an intervention.

journal-article · 2011 · Open the source

Beresford and Stajano define location privacy as the ability to prevent other parties from learning one's current or past location, and their argument turns on scale. Knowing a single position is different from holding a second-by-second record at sub-metre accuracy, and they argue that this quantitative change becomes a qualitative one. Their remedy is the mix zone, an area in which identities are unlinked and pseudonyms are exchanged, borrowing techniques from anonymous communication, together with two metrics, one based on anonymity sets and one on entropy. They test the framework against a corpus of more than three million location samples from an indoor system. The most useful part for a designer is the framing of privacy as a property of the whole data flow. Controlling who may query a coordinate is not sufficient, because the coordinate can be joined to other data later, so retention and purpose limitation are design decisions rather than paperwork. The paper is a construction and an analysis rather than a field study, and it predates the smartphone, so its threat model is simpler than the present one.

journal-article · 2003 · Open the source

This paper is the honest account of what interdisciplinary work on interactive digital health interventions is actually like, and it is open access. Blandford and colleagues set out seven lessons organised around a research lifecycle, from establishing what is already known through development, implementation and evaluation to reporting results. The contrast that carries the paper is methodological. Human-computer interaction prefers formative evaluation, which asks how a design performs with users and feeds the next iteration, while health research prefers summative evaluation, which asks whether the intervention worked against a comparison. The authors also describe differences in what counts as acceptable evidence, in assumptions about development lifecycles, and in the culture of reporting. Their stated audience is researchers working across the two areas for the first time, which makes the paper unusually usable in teaching. Its limits are that it is a reflection on the authors' own experience rather than a systematic review, and that it presents the two cultures as poles while noting that real practice sits between them and that middle ways are developing.

journal-article · 2018 · Open the source

Bodker's keynote is the paper that framed the third-wave debate, and it remains the best short account of what actually changed. The second wave, in her description, was work-centred: groups working with a collection of applications in established settings, analysed with situated action, distributed cognition and activity theory, and designed through participatory workshops, prototyping and contextual inquiry. In the third wave the use context and application types are broadened and intermixed, technology spreads from the workplace into homes and everyday life and culture, and elements such as culture, emotion and experience enter the field. Her topics are multiplicity, context, boundaries, experience and participation. Two points are usually dropped from summaries and matter for teaching: third-wave research is defined partly in the negative, as non-work, non-purposeful and non-rational, and it has partly moved away from a commitment to users towards a more exploratory stance. Her own case material, including the study of tax officers' physical desktops, shows her working with the older theories rather than discarding them.

conference-paper · 2006 · Open the source

Bowles and Box wrote this short book for the designer who wants to do user experience work in an organisation that has not asked for it. The publisher describes it as a pragmatic guide from the front line, giving frank advice on making UX work in real companies with real problems, and on fitting research, ideation, prototyping and testing into a daily workflow under the constraints of time, budget and culture. That framing is the book's value here. Most professional texts assume a mandate and a budget; this one assumes neither, and it treats the political problem of getting work adopted as part of the craft. It is also a document of its moment, published in 2010 and no longer in print, and it predates the widespread adoption of research operations and design systems. Its advice on the tools of the trade is dated, and its claims are assertions from practice rather than findings. Read it for the stance, not for the evidence. Copyright is 2011 and the ISBN is 978-0-321-71990-4.

book · 2011 · Open the source

Brush and colleagues made semi-structured visits and household tours with fourteen households that already had home automation, split between people who installed it themselves and people who paid a professional, and asked why a technology available for three decades had not spread. Their answer is four barriers: high cost of ownership, inflexibility, poor manageability, and difficulty achieving security. The most useful part for a designer is the shift of attention from sensing to installation and upkeep. A sensor that works in a laboratory was installed, powered, named and maintained by the researchers. In a real home the same sensor demands structural changes, a compatible hub and a decision about who may control it, and each demand loses households. The recommendations follow: remove the need for structural change, provide security primitives people can configure with confidence, and let devices compose rather than each needing an app. The limits are the sample and the method: fourteen households that had already adopted automation cannot speak for the households that never did, and the findings are qualitative rather than statistical.

conference-paper · 2011 · Open the source

This is the study behind the planning fallacy, and it is the module's evidence that estimating error is systematic rather than a personal failing. The authors tested three hypotheses: that people underestimate their own completion times but not other people's, that they build predictions from an imagined plan rather than from relevant past experience, and that they explain away past overruns as external and temporary. Five studies with 465 undergraduates supported all three. Predictions were optimistic across academic and everyday tasks; think-aloud procedures showed participants focused on future scenarios; and the bias was eliminated when they were instructed to connect relevant past experience to the prediction. Observers judging other people used past experience more and were less optimistic. The practical reading is that an outside view, built from the record of comparable finished work, beats an inside view built from the plan. The limits are real: the participants were students, the tasks were short, and none of it involved a design engagement. The effect is well established; its size in professional work is not.

journal-article · 1994 · Open the source

This is the socio-technical model that health human factors now reaches for, and it is the source of the module's claim that the person who made the error is usually not the cause of it. Carayon and colleagues present the Systems Engineering Initiative for Patient Safety as a framework relating the structures, processes and outcomes of care, and they demonstrate it with an application in outpatient surgery. The argument's force is practical. If a work system produces the performance, then an investigation that stops at the individual has answered a question nobody needed answered, and a model of the work system tells the investigator where else to look. The paper sits naturally beside Reason's system approach, but it is specific to healthcare and it names the elements of the setting rather than treating them as background. The limits are that this is a conceptual paper with an illustrative application rather than a comparative test, and that only its abstract and published summary were read in this run, so the detail of its components should be taken from the paper itself.

journal-article · 2006 · Open the source

Consolvo and colleagues built UbiFit Garden, a system that infers physical activity from on-body sensors and presents it on a glanceable mobile display, and ran it for three weeks with twelve participants in their own lives. The paper is the module's model of a field study and of an honest one. It reports what the sensing got right and wrong, how participants reacted to being inferred, and how they manipulated or mistrusted the record. The most useful teaching point is the two-part display design. A glanceable garden of flowers and butterflies carries the current state at low attention cost, while the interactive application carries the detail, and the split is a deliberate trade of precision for peripheral readability. The limits matter as much as the findings. Twelve participants over three weeks can show how a system behaves in use, and it cannot show whether a behaviour change persists. The authors state that their aim at this stage is user experience rather than sustained change.

conference-paper · 2008 · Open the source

Dey's short paper is where most later work takes its definition of context, and the definition is the point. Context is any information that can be used to characterise the situation of an entity, where an entity is a person, place or object considered relevant to the interaction. The paper unpacks context into the familiar questions of identity, activity, location, time and reason, and separates context-awareness from the older term situation-awareness. The most useful element for a designer is the Context Toolkit, an architecture that decouples sensing from application logic so an application can ask for a type of context instead of a specific device. The proposed abstraction of a situation bundles context into a higher-level interpretation. The limit is that the definition is deliberately broad and cannot decide what a given product should sense, which is the selection problem the module returns to. The full text sits behind a paywall, and only the abstract could be checked in full in this run.

journal-article · 2001 · Open the source

Dourish sets out to give human-computer interaction a philosophical basis, and the position he develops is embodied interaction. His claim is that meaning arises from a person's engaged, skilled practice in a physical and social setting, rather than from a disembodied process of reasoning over representations. That stance is drawn from the phenomenological tradition, and the book works through it in a way that is unusually readable for a philosophical argument. For this material the useful move is that he treats tangible and social approaches to interaction as two views of one phenomenon, because both depend on a shared physical context and on practice rather than on internal models. That is a stronger and more general claim than the tangible interface literature usually makes, and it is why the book is the reference for the design stance rather than for any particular technique. The limits are real and should be stated. This is a work of argument and synthesis, not an empirical study, and it offers no measured comparison between embodiments. It also relies heavily on named philosophers, so a reader who wants a design rule will not find one. It is best read as the conceptual frame within which the tangible and embodied claims are made.

book · 2001 · Open the source

This study is the strongest single piece of evidence in the module on alarm design, and it is open access, so a student can read the whole of it. Drew and colleagues recorded every alarm from every monitor in five adult intensive care units for 31 days, covering 461 consecutive patients, and annotated 12,671 arrhythmia alarms using a written protocol. The scale is the finding. 2,558,760 unique alarms occurred in the month, 381,560 of them audible, which is 187 audible alarms per bed per day, and 88.8 per cent of the annotated arrhythmia alarms were judged false positives. The paper then identifies the causes, which are mostly inappropriate settings and non-actionable events rather than broken hardware, and derives design and policy recommendations from them. Its limits matter for how the numbers are used. The data come from one medical centre, one monitor model and one software version, and alarm fatigue itself is inferred from the alarm conditions rather than measured in the clinicians. The recommendations are sensible but untested by the study.

journal-article · 2014 · Open the source

This is the paper that fixed the three-level model on which almost all later work on situation awareness rests. Endsley defines the construct as the perception of the elements in the environment within a volume of time and space, the comprehension of their meaning, and the projection of their status in the near future, and she argues that it must be treated as separate from decision making and from performance. The model's mechanism is that attention and working memory limit what an operator can hold, while mental models and goal-directed behaviour relieve those limits. The most useful part for teaching is the taxonomy of errors at each level, worked through accident reports, including cases where the information the crew needed was available and the crew did not connect it to their goals. The paper is a theoretical synthesis rather than new empirical work, and its accident readings depend on the investigation reports it cites. It also defers measurement to its companion paper, which is listed next. Its value is that it gives a student a vocabulary for saying precisely what an operator did not know.

journal-article · 1995a · Open the source

Where the theory paper defines situation awareness, this companion piece asks how anyone could measure it without destroying the task being studied. Endsley reviews the candidate methods and their drawbacks, then reports two studies of a query-based technique in which a running simulation is interrupted and the operator is asked about the state of things. Her findings are that the interruption did not disturb normal behaviour during the trial, and that the results did not appear to be limited by memory decay, which is the evidence against the objection that such a probe measures recall rather than awareness. For a student the useful content is the logic rather than the numbers. A measure has to capture the construct, predict performance and respond to changes in the task, and a workload rating will not do, because situation awareness and workload vary independently. The limits are serious. The validity evidence comes from aviation simulation run by the technique's own author, the freeze is itself an intervention in the task, and the paper's detail sits behind a paywall, so this note rests on the publisher abstract.

journal-article · 1995b · Open the source

Engestrom's paper is the most compact statement of the third generation of activity theory, and it is where a designer finds the five principles in one place: the activity system as the unit of analysis, multi-voicedness, historicity, contradictions as the driving force of change, and expansive cycles as a possible form of transformation. It also sets out the three generations, from Vygotsky's mediated act through Leont'ev's distinction between an individual action and the collective activity that gives it point, and argues that the third generation takes two interacting activity systems as its minimal unit. The empirical material is an intervention study in the multi-organisational field of medical care for children in the Helsinki area, in which the object of the work moves from an unreflected case to a jointly constructed understanding of the child's situation and care plan. That is the analytic payoff and the reason to teach the paper. The limit is the usual one for such a framework: it interprets a setting rather than predicting it, and contradictions are identified in retrospect.

journal-article · 2001 · Open the source

The Artificial Intelligence Act is the first comprehensive statutory regime for artificial intelligence, and its organising device is a set of risk categories carrying different duties. Article 5 prohibits a defined list of practices that includes systems using subliminal, manipulative or deceptive techniques, systems that exploit the vulnerabilities of a person or a group, social scoring, and real-time remote biometric identification in publicly accessible spaces. High-risk systems must meet requirements for risk management, data governance, technical documentation, logging, human oversight, and accuracy and robustness. The recitals place education, credit scoring, judicial assistance and voting among the high-risk domains, and the annexes set documentation duties for general-purpose models with systemic risk. The application dates are staggered: the regulation applies from 2 August 2026, while the prohibitions and general provisions applied from 2 February 2025. The limits are that it is European Union law, binding in the union rather than in the United Kingdom, and that its centre of gravity is documentation, which is not the same as safety. Read the definitions before the detail.

legislation · 2024 · Open the source

This is the founding paper of the persuasive technology tradition, and it is worth reading because later work, both positive and critical, is answering it. Fogg argues that computers can be designed to change what people think and do, and that this capacity deserves study in its own right. He surveys the mechanisms then available, including interactivity, simulation, and the human tendency to treat computers as social actors, and he sets out a research agenda. The most useful content for this module is the framing of persuasion as a designed effect rather than an accident, because that makes the ethical question unavoidable rather than external to the work. The limits are the ones the field now teaches. The paper assumes the persuasive goal is benign and treats ethics as a matter for later research, which is the gap the critical literature fills. The studies it cites are small and often concern novelty. Only the entry metadata and reference list were readable in this run, so the account of the argument is from those rather than from the paper text.

conference-paper · 1998 · Open the source

Fowler's article is the standard practitioner statement of continuous integration, and its definition is the one the narrative teaches: each member merges changes into a shared mainline at least daily, with every integration verified by an automated build that includes tests. The article sets out the supporting practices in a list a team can audit itself against, including putting everything in version control, automating the build, making it self-testing, committing to the mainline daily, triggering a build on every push, fixing a broken build immediately, and keeping the build fast. It also does something rarer in practice writing: it names the conditions under which the practice is the wrong choice, and it distinguishes continuous integration from continuous delivery and from deployment. The opening anecdote, of a site where integration had run for months with no completion date in sight, frames the risk being managed. The article is candid that its account rests on experience across many projects rather than controlled comparison, which is its main limit as evidence.

web-article · 2024 · Open the source

Gebru and colleagues argue that the machine learning community has no standard way of documenting datasets, and that the omission has serious consequences in high-stakes work. They propose the datasheet, borrowed from electronics, where every component ships with a document describing its operating characteristics, test results and recommended uses. A dataset datasheet records the motivation for collecting the data, its composition, the collection process, any preprocessing, and the recommended and prohibited uses. The most useful parts for a designer are the composition and collection sections, because they force answers to the questions that cause harm later: who is in the data, who is missing, how were they recruited, and did they agree to this use. The paper is a proposal with a template, not an empirical study, so its value lies in the structure it imposes rather than in demonstrated outcomes. It explicitly notes that a datasheet records the creators' account, so it inherits whatever the creators were willing or able to write down. Publication is voluntary, and the template does not audit itself.

preprint · 2021 · Open the source

This is the paper that gave the field its working account of dark patterns, and the method is why it is worth teaching. Rather than define the term in advance, Gray and colleagues assembled a corpus of examples identified by practitioners and analysed them by content analysis. What emerged was a wide range of ethical concerns gathered under one umbrella term, and a shared worry among designers that they could easily become complicit in manipulative or unreasonably persuasive work. The paper closes by arguing for broader research on the ethics of user experience and for changes in design education, which is the passage a student can use directly. Two limits matter. It is a content analysis of posted examples, so it maps what practitioners said rather than what products do or what users experience. And the authors report that the umbrella term conceals real disagreement, so the taxonomy they propose organises a debate rather than settling it. Read alongside a measurement study before using it to claim prevalence.

conference-paper · 2018 · Open the source

Greshake and colleagues argue that a language model integrated into an application dissolves the boundary between data and instruction, and that this turns ordinary content into an attack surface. Their method is adversarial. They derive a taxonomy of impacts from a computer security perspective and then demonstrate working attacks against deployed systems, including Bing's GPT-4 powered chat and code-completion engines, and against synthetic applications they built. The reported consequence is the one a designer has to plan for, namely that processing retrieved text can amount to arbitrary code execution and can control how and whether other services are called. The paper is most useful here because it moves the problem out of the text-processing category and into the category of system boundaries, which is where the guardrail decisions in this material come from. Its limits are candid and should be repeated to a stakeholder. The work is a demonstration, so it establishes that attacks are possible and not how often they occur in use, and the authors state that effective mitigations were lacking at the time of writing, which leaves a designer without a validated defence. The specific attacks also date with the systems tested, and only the abstract was read in this run.

preprint · 2023 · Open the source

This paper gives the module its working account of what teamwork quality is and why it can be measured at all. The authors define teamwork quality through six facets: communication, coordination, balance of member contributions, mutual support, effort and cohesion. They then test the construct against project success using data from 575 team members, team leaders and managers drawn from 145 German software teams, estimating structural equation models. Teamwork quality as rated by team members was significantly associated with team performance, and it was also associated with members' personal success in the form of work satisfaction and learning. One result is especially useful for teaching: the strength of the association with performance varied with the perspective of the person rating the performance, whether manager, team leader or team member, which is a reminder that performance is partly in the eye of the rater. The inclusion of balance of member contributions as a defining facet is the direct link to individual accountability. The limits are that the data are self-reported, the sample is German software teams, and only the abstract was read here.

journal-article · 2001 · Open the source

This white paper is the primary statement of the argument that the Swiss cheese model, and the safety thinking that goes with it, asks the wrong question. Hollnagel, Wears and Braithwaite characterise Safety-I as treating safety as the absence of accidents, presuming that failures come from identifiable malfunctions of components, and casting people as the most variable and therefore most suspect component. Safety-II redefines safety as the system's ability to succeed under varying conditions, treats everyday performance variability as the reason work usually goes right, and changes the purpose of investigation from explaining failure to understanding success. The paper also states that improvement in healthcare safety has been slow since the 2000 report To Err is Human, and that harm rates appear stable while demand and complexity rise. Its status limits it. It is a position paper published by its own authors rather than a peer-reviewed study, and it reports no data. It argues for combining the two perspectives rather than replacing one with the other, which is easy to miss and matters for practice.

report · 2015 · Open the source

Horvitz's paper is where the vocabulary of shared control in this material comes from. Its argument is that interface agents and direct manipulation should be coupled rather than opposed, and it sets out ten principles for that coupling, illustrated by the Lookout system for scheduling and meeting management. The decision-theoretic core is what a designer should carry away. Whether an automated service should act is a comparison of expected costs and benefits under uncertainty about the user's goals and attention, not a report of the model's accuracy. The paper names the costs explicitly, including the frustration caused by an unnecessary interruption, and it sets a threshold above which action is warranted and a band within which a system should ask rather than act. It also asks a system to reduce the precision of its service as uncertainty rises, which is the earliest clear statement of the design problem of showing what a system does not know. The limits are important. This is a position statement built on one prototype, with no controlled evaluation of the principles, and the systems described are those of 1999. Cite it for the framing and the vocabulary, not for an effect size.

conference-paper · 1999 · Open the source

This is the study that made distributed cognition arguable, and it is short enough to teach from. Hutchins takes a memory task in the cockpit of a commercial airliner, the computation and retention of the airspeed and wing-configuration speeds needed for an approach, and analyses it with the cockpit as the unit of analysis rather than the pilot. The claim that carries the most weight for designers is that the speed card booklet and the moveable speed bugs are not aids to individual memory. They are parts of the process by which the cockpit system remembers, because the speeds are read, written and compared across several representations. The paper states plainly that the cognitive properties of such a system can differ from those of the individuals inside it. Its value is the worked demonstration, and the design consequence that follows: change the representation and the memory system changes with it. Its limits are the single detailed setting and the reasoning by analogy rather than comparison, since the paper illustrates a framework rather than testing two designs against each other.

journal-article · 1995 · Open the source

The Information Commissioner's Office is the UK regulator for data protection, and its guidance is what a practitioner actually reads when applying the UK GDPR and the Data Protection Act 2018. Three parts matter here. The principles guidance sets out the seven principles from Article 5: lawfulness, fairness and transparency, purpose limitation, data minimisation, accuracy, storage limitation, and integrity and confidentiality, with accountability requiring a controller to demonstrate compliance. The lawful basis guidance states that consent is one of seven bases, and treats valid consent as freely given, specific, informed and signalled by an unambiguous indication. The impact assessment guidance treats the assessment as a defined process to be completed before processing that is likely to result in high risk. The limits are that guidance is the regulator's interpretation rather than the statute, and each page read here carries a notice that it is under review following the Data (Use and Access) Act, so any summary is provisional.

guidance · n.d. · Open the source

This short guidance note from the Intellectual Property Office explains changes to UK designs law that came into force on 1 October 2014, under the Intellectual Property Act 2014. The change this module uses is the ownership rule: the owner of a commissioned design is now the designer and not the commissioner, unless a contract states otherwise. That reverses the earlier position, and it matters to any freelance or agency designer, because it means the default protects the designer and the client must negotiate by contract if they want ownership. The note also warns that a contract may be written, oral or implied, and that an employment relationship can exist even without documentation, so the surrounding facts matter as much as the paperwork. It covers other changes too, including protection for good-faith prior use and a criminal offence for intentionally copying a registered design. It is government guidance rather than legal advice and it says so. It addresses designs rather than software, and a designer should take advice before relying on it in a contract.

government-guidance · 2014 · Open the source

Iqbal and Bailey built a system that defers notifications to breakpoints, the transitions between units of a task, using statistical models to detect those moments during real work. Two studies give the paper its force. In the first, the models detected breakpoints reasonably well, around 55 per cent on unfamiliar tasks, but could not reliably tell the three types apart, so detection worked better than classification. In the second, scheduling notifications to breakpoints reduced frustration and reaction time compared with immediate delivery, at the cost of an average deferral of about a minute and a half, and the relevance of the content decided which breakpoint suited a message. Participants preferred the managed behaviour, which the authors read as evidence of likely adoption. The most useful result for a designer is counter-intuitive: immediacy is not a virtue for a notification. The limits are the task domains, diagram editing and programming, and the modest detection rate, which the authors report plainly rather than conceal.

conference-paper · 2008 · Open the source

Ji and colleagues survey the production of unintended text by language generation systems, the failure this module calls hallucination. The survey covers metrics, mitigation methods and open problems, and then reviews the problem in summarisation, dialogue, generative question answering, data-to-text generation, machine translation and visual-language generation. The most useful chapters for a designer are the ones on abstractive summarisation and dialogue, because they show that a fluent output can be unfaithful to its source in ways that no overlap-based metric detects. The survey's overall impression is that hallucination is measured more thoroughly than it is solved, which is the honest position to take into a design review. Two limits should be stated. It is a survey of a fast-moving area, so specific numbers and methods date quickly. And it is a taxonomy of the problem in generation systems rather than a study of how users respond to a hallucinated sentence, so the interface consequences are inferred rather than measured. A student citing it for a specific mitigation should follow its reference to the primary paper.

preprint · 2022 · Open the source

John and Kieras wrote the paper to read before trusting a GOMS prediction. It takes one small task, moving a marked-up phrase in a word processor, and analyses it four ways: the keystroke-level model, the original formulation from Card, Moran and Newell, a structured notation called NGOMSL, and a parallel-activity technique. It then compares the cognitive architecture each assumes, the operator times each uses and the predictions that follow. The same task comes out at 14.38 seconds in two variants and 16.38 seconds in a third, the highlighting sub-method alone is put at 2.21 seconds in the parallel-activity model, and NGOMSL also predicts 784 seconds of procedure learning. The most useful parts for teaching are the component definitions, the explicit statement that the boundary between a goal and an operator is the analyst's choice of detail, and the honest constraints: serial processing, no interruption, no overlapping activity, and empirical validation only at the keystroke level. Its limit is that it compares and reports others' validation rather than running new studies.

journal-article · 1996 · Open the source

Kaptelinin and Nardi wrote the book that lets a designer use activity theory without first reading the Soviet tradition from the 1920s onward. The publisher's description presents it as the first systematic entry-level presentation of the principles, together with the theory's history, its relation to other approaches and its application to the analysis and design of technology. The parts that matter for teaching are the treatment of the object of activity, including an empirical study of how the notion is used, the account of the debates inside activity theory, and the comparison with the theory's competitors in interaction design, named as distributed cognition, actor-network theory and phenomenological approaches. That comparison is the reason to put this book beside the primary sources here: it tells a student what activity theory claims against the alternatives, rather than only what it claims. The limits are that it is a synthesis written for designers rather than a report of new evidence, and that its case material comes from work settings, so its application to personal and leisure technology needs care.

book · 2006 · Open the source

Kleinberg, Mullainathan and Raghavan formalise three fairness conditions that are each defensible on their own terms: calibration within groups, balance of error rates for the negative class, and balance of error rates for the positive class. Their result is that no method can satisfy all three at once except in highly constrained special cases, and that even approximate satisfaction forces the data into an approximate version of one of those cases. The abstract states the conclusion plainly, and it is the conclusion rather than the proof that a designer needs. Fairness is not a property a system either has or lacks; it is a choice between definitions that trade against each other, and the trade changes who receives help and who is wrongly excluded. The paper is most useful in a design review as a reason that a fairness claim must be stated as a named criterion with its consequences admitted. Its limits for teaching are that it is a theoretical result about probabilistic risk scores and says nothing about how a team should choose, and reading only the abstract, as this module does, leaves the conditions and the proof unchecked.

conference-paper · 2017 · Open the source

This is the paper that named the free-rider effect in its psychological form. Latane, Williams and Harkins asked whether people exert less effort in a group than alone. Their two experiments used clapping and shouting tasks with 84 undergraduates, and they replicated an earlier unpublished rope-pulling finding that individual effort fell as group size increased. Their important refinement was methodological: the decline was not explained by losses of coordination alone, so it could not be dismissed as an artefact of trying to pull a rope together. They called the effect social loafing and argued that the mechanism involves the reduced identifiability of any one person's contribution. That mechanism is the design hook for the module, because it implies that a team which makes contribution visible addresses the cause rather than the symptom. The limits are those of a laboratory study from the late 1970s. The tasks are physical and trivial, the participants were students, and the setting is unlike a year-long software project. Later work has qualified the size and conditions of the effect.

journal-article · 1979 · Open the source

Latour's chapter is the accessible entry to actor-network theory and the reason the theory has any place in a design module. Its argument is that accounts of society fail to balance because sociologists look only at human ties, and the missing mass is in things. The two worked examples are a door-closer and a seat-belt system: engineers link the ignition to a sensor in the belt, so a rule that a person should buckle up becomes a property of the assembly of driver and car. His term for this is the programme of action, a rule written into the assembly rather than into a person, and his point is that a technical choice is organisational and moral at the same time. The most useful teaching move is to keep the method and drop the metaphysics: follow the artefact and ask what work it is doing, and on whose behalf. The limits used in teaching are not stated in the chapter: the theory was built to describe science and technology in the making, it offers no design criteria, and its symmetry between human and non-human agency can be used to justify almost anything.

book-chapter · 1992 · Open the source

The MRS Code of Conduct is a professional code for market, social and opinion research, and it is the module's source for what consent and participant handling actually require. The 2023 edition, updated in May 2023 and operative from 15 May 2023, opens with twelve principles, including transparency about the subject and purpose of data collection, respect for the rights and well-being of individuals, a duty not to harm or adversely affect them, and the exercise of independent professional judgement in design, conduct and reporting. Its rules then become concrete. Participants must give permission before a data collection exercise, informed consent requires a specified set of information, the right to withdraw must be respected at any stage of a project, re-contact requires permission obtained during the initial data collection, and participant details must not be passed to a third party without consent. The code is written for the research industry, so some vocabulary is commercial, and it is not a substitute for data protection law. It remains a precise account of the infrastructure a research practice needs.

professional-code · 2023 · Open the source

This is the measurement study that turned dark patterns from a matter of example into a matter of prevalence, and the numbers are why it is cited. Mathur and colleagues built automated techniques to detect dark patterns, then crawled shopping websites. From about 53,000 product pages across about 11,000 websites they identified 1,818 instances, organised into 15 types and 7 broader categories. They classified 183 websites as using deceptive practices, and found 22 third-party entities offering dark patterns as a turnkey service. That last finding is the one that matters most for regulation, because it shows a supply chain rather than isolated choices. The paper also develops a taxonomy of the underlying influence and the potential harm. The limits should be stated plainly. The crawl observed the page each site served, so it cannot see patterns that are personalised, shown to some visitors only, or varied in an experiment. The true figure is therefore a floor, not a total, and detection relied on the researchers' own rules, so the categories inherit their assumptions.

journal-article · 2019 · Open the source

This report is the best known attempt to put a number on the business value of design, and the module uses it as much for its limits as for its findings. McKinsey tracked the design practices of 300 publicly listed companies over five years, drawing on interviews, surveys, more than two million pieces of financial data and more than 100,000 recorded design actions. It built the McKinsey Design Index from twelve actions and reported that top-quartile companies increased revenue by 32 percentage points and shareholder returns by 56 percentage points more than industry peers over the period, across medical technology, consumer goods and retail banking. The report states that this is a correlation, and its method cannot establish causation, because design practice and financial performance were observed together rather than manipulated. Two findings are safer ground: over 40 per cent of surveyed companies were not speaking to end users during development, and just over 50 per cent had no objective way to set targets for design output. Use the report to describe practice, and be explicit that its headline numbers are associational.

industry-report · 2018 · Open the source

Miller reviews work on explanation from philosophy, cognitive science and social psychology, and argues that explainable AI has largely been built on the researchers' own intuition of what a good explanation is rather than on what the social sciences have established. The findings he draws out are the ones this module uses: people seek contrastive explanations, asking why this outcome rather than an expected alternative; they prefer a small number of selected causes to a complete account; and they prefer causes that are relevant and actionable, including counterfactuals, over a restatement of the computation. The paper is most useful as a corrective to the assumption that a feature attribution is an explanation. Its limits are important. Most of the evidence he reviews concerns how people explain to each other in conversation rather than how they read a machine's reasons in an interface, so the design conclusions drawn from it are inferences and not tested claims about interfaces. The review is long and surveys several literatures; the sections on contrastive and selected explanation carry most of the teaching weight.

journal-article · 2019 · Open the source

Mitchell and colleagues propose that a released machine learning model be accompanied by a short document recording what it is for and how well it works. The card reports the intended use cases, the evaluation conditions, benchmarked performance across cultural, demographic and phenotypic groups and intersectional groups such as age alongside race, and the known limitations. The authors demonstrate the format with two cards, for a model detecting smiling faces and a model detecting toxic comments, which is the most useful part for a student because it shows the level of detail involved. The paper's wider significance is that it converts undocumented assumptions into claims that people outside the team can review and contest. Two limits matter. Publication of a card is voluntary, so the practice is a norm rather than a guarantee. And a card can be accurate and still not prevent harm: it documents what the model does without deciding whether it should be used. The claimed benefits are transparency and communication, and the paper offers the framework rather than evidence that cards change outcomes.

conference-paper · 2019 · Open the source

Parasuraman and Riley integrate theoretical, empirical and analytical work on how people use automation, and give the field its standard vocabulary for the ways that use goes wrong. Use is the voluntary engagement or disengagement of automation. Misuse is overreliance, which produces monitoring failures and decision bias. Disuse is underuse or neglect, and the paper traces it to alarms that fire falsely, explaining that the designers set the false alarm rate without considering the base rate of the condition being detected, so for a rare condition almost every alarm is wrong. Abuse is a design and management failure, where a function is automated without regard for the consequences for human performance, which in turn promotes misuse and disuse. The most useful part for a designer is that the four terms separate failures usually lumped together as user error, and that each has a different remedy. The paper is a synthesis rather than a new study, so it is cited for the framework and the mechanisms, not for a measured effect size.

journal-article · 1997 · Open the source

Raji and colleagues address a specific failure in the accountability literature: most audits are performed from outside an organisation, after deployment, and they struggle to trace an emergent harm back to the decision that caused it. Their proposal is an internal audit framework that runs across the whole development life cycle, from conception to deployment and monitoring, and that produces a set of documents which together form an audit report. The framework draws on an organisation's stated values or principles to assess whether the decisions made at each stage fit them, and the authors argue that this is what closes the accountability gap and gives an audit integrity. The most useful teaching content is the claim that accountability is a process built into development rather than an inspection added at the end. The limits are those of any framework paper. It is a proposal rather than an evaluation, so it reports no evidence that internal audits prevent harm, and it depends on an organisation being willing to document its own choices honestly. The full text is paywalled; the abstract was read here.

conference-paper · 2020 · Open the source

Reason wrote this short BMJ piece to separate two ways of thinking about the same problem, and it is the clearest statement of the position in his own words. The person approach blames the sharp end for forgetfulness, inattention, carelessness, negligence and recklessness, and its countermeasures are poster campaigns, another procedure, retraining, disciplinary measures and blame. The system approach holds that humans are fallible and that errors are to be expected, and that errors are consequences rather than causes whose origins lie upstream. The paper supplies the vocabulary used in the narrative: active failures, listed as slips, lapses, fumbles, mistakes and procedural violations, and latent conditions, which arise from decisions by designers, builders, procedure writers and senior managers. It also sets out the Swiss cheese model and the reporting culture and just culture argument. Its limits are worth stating. It is a short discursive article rather than a study, and it reports no data of its own. The figure of about 90 per cent blameless lapses in aviation maintenance is cited to another source, and the Chernobyl judgement rests on Reason's reading of accident reports. The fuller treatment is in his book Human Error (Cambridge University Press, 1990), whose publisher page was checked in this run but whose text is not free to read.

journal-article · 2000 · Open the source

Rogers delivers the critique that reframes the whole area. She argues that three dominant UbiComp programmes, context awareness, ambient intelligence and monitoring and tracking, have been fruitful but have not produced anything like Weiser's world. Her diagnosis is about initiative. Research that tried to make computers proactive, acting on a person's behalf, aimed at the wrong target, because people are resourceful and can extend their own capabilities if technology is designed to engage them. Her alternative is a shift from proactive computing to proactive people, and from calm living to engaged living, with technology that provokes learning and reflection instead of quietly removing decisions. The most useful part is the redirection it gives a designer: ask what the person could do better with the technology, not what the system can decide for them. The paper is a position paper rather than an empirical study, so it argues from the field's record rather than reporting new data. That is its licence and also its limit.

conference-paper · 2006 · Open the source

Rogers's short book is the orientation text for this area, and its value is the shape it gives the field rather than any single claim. It charts how theory in HCI developed, divided into classical, modern and contemporary theory, and asks what the rhetoric promised against what the reality delivered, which theories were used and which had an impact, and why. It closes by asking whether theory will keep a role. Its most useful sentence for a student writing about the purpose of theory is the publisher's summary of that argument: the role has expanded from testing hypotheses about user performance to describing, explaining and critiquing, and to generating new designs. The limits are the format and the level. A synthesis of a little over a hundred pages cannot treat any tradition in depth, and it reports on primary work rather than evaluating it, so it should be used to place a source rather than to settle a question about it.

book · 2012 · Open the source

This is the empirical study behind the treatment of code review as a team obligation. It investigates modern, tool-based review at Google through three methods used together: twelve interviews, a survey with forty-four respondents, and analysis of review logs for nine million reviewed changes. It reports why the practice was introduced, how it operates now, and developers' satisfaction and the challenges they report. The most useful finding for a student team is that review's benefits are not confined to defect detection; the practice also spreads knowledge of the code and creates a second person who understands each change. The scale of the log analysis is what gives the paper its authority. Its limits are stated plainly by its own design. It is a case study of one company whose engineering culture, tooling and staffing a student team does not share, and the interview and survey material is self-report. The paper's own reference list includes work questioning how often review finds bugs. Only the abstract and references were read here.

conference-paper · 2018 · Open the source

This meta-analysis is the source for the claim that vigilance decrements are larger than intuition suggests. See, Howe, Warm and Dember pooled 42 vigilance studies to test and refine the then-dominant taxonomy, which attributed the decline in perceptual sensitivity to the type of discrimination a task required and to the rate at which events were presented. Their analysis confirmed that the sensitivity decrement is substantial, and that it is a function of those two task characteristics. Their further finding is the one to teach: the taxonomy needed a new dimension, which they call the sensory-cognitive distinction, because discrimination type and event rate did not account for all of the variation they found. What it does not give is a number a designer can use in a particular setting. The pooled studies are laboratory tasks, and their control is what makes the effect visible, so extending the result to a ward at night with interruption and sleep debt is an inference rather than a finding. The abstract and the bibliographic record were read in this run; the article body was not.

journal-article · 1995 · Open the source

This is the reference text for the management scaffolding a project needs, and the chapter list on the publisher's page shows why. Chapters 22 and 23 cover project management and project planning, and the surrounding chapters on requirements engineering, configuration management and quality management map closely onto the artefacts a team must produce. Sommerville's framing is comparative rather than prescriptive: plan-based and agile approaches appear as choices with different risk profiles, which suits a project that must both deliver an agreed baseline and adapt as it learns. Its value for a student is that it gives a name and a structure to each management artefact, from the plan to the risk register, so that the team can decide which to keep and which to drop. The limits are worth stating. It is a teaching textbook, so its examples illustrate rather than evidence, and it reports little primary research of its own. Chapter numbering differs between editions, so a reader holding an earlier printing will find the same topics under different numbers.

book · 2021 · Open the source

Suchman's book is the source of the situated-action critique, and it is also the clearest example in the field of a design argument built on a philosophical claim. Her target is the model of purposeful action in which a plan determines what is done, a model that was then being built into interactive machines. The contrast she opens with is the navigation case in her preface: the European navigator begins with a charted course and relates every move to it, while the Trukese navigator begins with an objective and steers by wind, waves, the stars and the feel of the boat, and can point to the objective at any moment but cannot describe a course. Her conclusion is that all activity, even the most analytic, is concrete and embodied, so plans are best viewed as a weak resource for primarily ad hoc activity. The preface is the honest place to begin, and the linked copy is the publisher's own. Its limit here is that the empirical chapters are not in that file, and the critique itself is contested by cognitive modellers whose models can still be validated on routine work.

book · 1987 · Open the source

Towsey's handbook is the module's account of research operations as a discipline rather than a set of chores. It is written for research leaders, researchers and ResearchOps specialists, and the publisher's description sets out its scope: research strategy, a research operating system, eight elements of research operations, a planning matrix, participant recruitment, research knowledge management, building trust with stakeholders, and making ethics and data privacy more than a checklist. It also covers funding and hiring. The book's contribution is the argument that scaling research is a systems problem, and that those systems have to be designed deliberately or they emerge as bottlenecks and inconsistency. It is a practitioner text, published in September 2024, and its claims rest on the author's experience and on practice across organisations rather than on controlled studies, so it should be read as expert guidance. Only the publisher's page and description were read in this run; the book itself was not. The research operations framing is the most useful part, because it names components a student can plan for.

book · 2024 · Open the source

Unger and Chandler's book is a practical guide to running a UX project from the first conversation to release, and this module uses it for the shape of a project rather than for any research method. The third edition opens with the project ecosystem, devotes a chapter to the operations functions, and later covers project objectives and approach, discovery, research and product definition. The framing matters because it places the designer as one role inside a system of roles, budgets and schedules rather than at the centre of it. The material on defining objectives and approach collaboratively is the most useful for scoping, and the operations chapter introduces the idea that a practice needs infrastructure. The book is a practitioner text from an imprint that publishes trade guidance, so its claims are advice drawn from experience and are not referenced to studies. It is a good orientation and a weak source for any measured effect. The third edition is dated November 2023 and carries a 2024 copyright, which is why the year used here is 2024.

book · 2024 · Open the source

Weiser's essay is the founding statement of ubiquitous computing, and it is worth reading in the original because summaries flatten it. His argument is that the most profound technologies disappear into daily life, and that computing had not, because a personal computer keeps attention fixed on a single box. His alternative, embodied virtuality, draws computing out of its case and into the world, at three scales he called tabs, pads and boards, with location and scale as the variables that matter. The most useful part for a designer is that the claim is about attention rather than hardware: the goal is technology that recedes to the periphery and is ready at a glance. The essay is a speculation and says so. It offers no measurements, and its social predictions are optimistic, expecting ubiquitous systems to increase privacy and community rather than to produce surveillance. Later work tests that optimism, and the module treats this text as a vision to be judged rather than as evidence.

journal-article · 1991 · Open the source

The article summarises the Menlo Report, a framework for ethical guidelines in research on information and communication technology, written by four people involved in producing it. The framework builds on the 1979 Belmont Report and keeps three of its principles, respect for persons, beneficence and justice, translating each into practice: identification of stakeholders and informed consent, systematic balancing of risks and benefits, and fair allocation of burdens. It adds a fourth principle, respect for law and public interest, which charges researchers to engage in legal due diligence, be transparent in methods and results, and be accountable for their actions. Its most useful move for a human-computer interaction project is the wider range of people who count, since the guidelines apply to research with the potential to harm humans whether or not those humans are the direct research subjects. The limits are its status and scope: it is a magazine summary rather than the report, it is aimed at computer and information security research, and it argues for a framework rather than evaluating one.

journal-article · 2012 · Open the source

Bender and colleagues ask how large a language model should be and what risks arrive with size, which makes this paper the critical counterweight to capability claims about generative interfaces. Its evidence is a synthesis of environmental, data and scientific arguments rather than a new experiment. Two points carry into interface design: these models are not performing natural language understanding and succeed on tasks that can be approached by manipulating linguistic form, and ingesting datasets larger than a team can document creates what the authors call documentation debt. Their framing of the human half of the problem is the most useful passage here, because people impute meaning where there is none, and that tendency misleads researchers and the public into treating synthetic text as meaningful. That is the mechanism behind hallucination as an interface problem rather than a model bug. The recommendations concern research and release practice, including weighing costs first, budgeting for curation and documentation, and considering directions beyond larger models. Its limits for teaching are that the argument is in part a values and policy position, its environmental figures are not re-derived here, and it predates the assistants students now use, so it is cited for the mechanism rather than for measurements of today's systems.

conference-paper · 2021 · Open the source

This short book is the HCI-specific account of qualitative research, written to fill a gap the authors identify: there are many qualitative methods texts in the social sciences and very few that address the concerns of human-computer interaction. Their position is that qualitative methods matter in HCI for gathering requirements, for understanding the situations in which technology is used, and for evaluating use in practice, and that HCI's interest is technology use for informing design rather than social phenomena in general. The book follows a study from planning through conduct to reporting, and the authors draw on their own work in healthcare technologies and in interacting with information, including the difficulty of engaging patients and handling sensitive material inside complex clinical processes. Its value for a student is the attention to the parts of the work that published papers omit. The limits are that it is a short practitioner-facing lecture rather than a methods treatise, and that only the open Introduction chapter was read in this run, with the remaining chapters known from the record and the contents.

book · 2016 · Open the source

Blomberg and Karasti look back over 25 years of ethnographic research in computer-supported cooperative work and ask what the tradition has contributed and what it is for. Their argument is that ethnography in this field has changed the understanding of work as both social and material, and that the idea of the field site has itself become a construct: not simply a bounded workplace, but a span of activity that reaches across settings and moves over time. They also address the persistent difficulty of drawing design implications from ethnographic findings, and describe the strategies researchers developed to couple the two more tightly. They record that the field still argues about what constitutes useful and proper ethnographic research. The article is a reflective review rather than a report of new data, and it is written from within CSCW, so its examples belong to workplace studies. Only the abstract was read in this run, which limits what can be attributed to it; the narrative therefore uses it for framing claims rather than for any specific finding.

journal-article · 2013 · Open the source

The paper traces how cultural probes were adopted and adapted across HCI, and found a proliferation of divergent uses and derivatives. The authors argue that the flexibility which made probes popular also exposed disagreements the field had not stated: about what counts as valid interpretation, and about how a method relates to the methodology that lies behind it. They frame the problem as epistemological rather than as a matter of technique, and they aim to clarify how methods are defined and evaluated in HCI, especially those grounded in unfamiliar conceptions of how research should be done. For a student reading deployment work, the value is a diagnostic question: what does this study assume that knowledge is, and do its claims sit inside that assumption? The paper is a conceptual analysis and not an empirical study, and it offers no measurement of its own. Only the abstract could be read in this run, so the account of particular derivatives would need the full text.

conference-paper · 2007 · Open the source

Braun and Clarke developed thematic analysis as a method, and this book is its fullest statement. It is built around reflexive thematic analysis, the version in which the analyst's interpretive work is treated as part of the method rather than a bias to be subtracted. Section One follows the work in sequence: getting ready, familiarisation with the data, coding, developing themes, and writing the report. Section Two turns to theory, to the role of interpretation, to the differences between reflexive thematic analysis and its methodological siblings and cousins, and to what makes the approach good. The practical value for a student is that the sequence is presented as a series of decisions with alternatives, which is the posture this module asks for in study design. Its limit here is evidential rather than conceptual. The publisher's description and the full table of contents were read in this run and the book was not, so a student citing an individual chapter should check the text itself.

book · 2021 · Open the source

Brown, Reeves and Sherwood treat a field trial as an object of study rather than only as a way to test a system. Their trial of trials examined a photo-sharing deployment and documented three effects that shape what any such study can find. Demand characteristics are the adjustments participants make to fit what they take to be the researchers' expectations; in their trial, users described trying to give the team the results it seemed to want. Lead participants are the few users who carry a deployment, and here one participant organised and encouraged use by others while also supplying much of the insight. Interdependence of methods and results is the way in which how a trial is run and what it asks narrow the behaviour that can appear. The paper also notes how rarely field trials are discussed as a method, with the debates around probes as the main exception, and it describes the rise of living labs as a way to keep some features of a field trial under more control. The evidence is a single deployment examined closely, so the three effects are offered as general characteristics rather than as measured frequencies.

conference-paper · 2011 · Open the source

Cairo's book is the journalistic and ethical counterpart to the perceptual material, and it is where the vocabulary of truthfulness comes from. Its argument is that truthfulness in a graphic is a continuum rather than a property, and that a reader should ask whether the variation on the page is variation in the data. The contents move through what visualisation is and what makes a visualisation good, then to the truth continuum and to conjecture and uncertainty, then through practical chapters on simple charts, distributions, change, relationships and maps, ending on confidence and significance. The examples come from news organisations and are part of the argument as much as its illustration. The book is a practitioner's guide written from craft experience and it is not a research monograph, so its claims are not referenced in the way a journal article's would be, and it should not be cited as evidence for a measured effect. Its value here is the framing and the worked examples.

book · 2016 · Open the source

This is a review of Registered Reports, the publication format in which a study proposal is peer reviewed and pre-accepted before the research is undertaken, so that the decision to publish rests on the question, the theory and the methods rather than on the results. The abstract states that the authors reflect on the history, progress and future prospects of the format and offer practical guidance for authors, reviewers and editors, that they review early evidence that Registered Reports are working as intended, and that they acknowledge these are not a universal solution for irreproducibility. They also consider how the policies around the format are changing or must change. For a project of this kind the useful content is what in-principle acceptance commits a researcher to, and the acknowledgement that the format suits confirmatory questions better than exploratory ones. The limit is access: the article is subscription content, so only the abstract and the reference list were read in this run, and this note rests on the abstract rather than on the body of the paper.

journal-article · 2022 · Open the source

This is the paper that gave visualisation its ranking of encodings. Cleveland and McGill set out to build a scientific foundation for graphical methods and identified a set of elementary perceptual tasks that people perform when they read quantitative information from a chart. They then ordered those tasks by how accurately people perform them, and tested the ordering in experiments in which participants judged values encoded on graphs. The ranking that follows, with position on a common scale at the most accurate end and colour and area nearer the least accurate, is the basis for the module's claim that the encoding decides which judgements are easy. The paper is honest about its own scope: it isolates perceptual tasks from one another, and its authors note that the set of tasks had to be expanded as the work proceeded. Display conditions and the interaction of channels are outside what it measures. Use it for the ranking and for its limits, not as a law that decides a design on its own.

journal-article · 1984 · Open the source

This CHI paper argues that HCI experiments should be preregistered, on the ground that preregistration lets reviewers and readers be confident that the evidence offered for a hypothesis is not the result of HARKing, hypothesising after the results are known. The authors review the motivation and the outcomes of preregistration in other disciplines and earlier commentary on evaluation in HCI, then discuss how the practice could be adapted to the distinctive features of HCI empirical research. For a project writer the useful contribution is the framing: the boundary between what was planned before data collection and what was added afterwards is a property the reader is entitled to see, and stating it is not an admission of weakness. Two limits should be stated. Only the abstract and metadata could be read in this run, because the full text is not open, so the authors' proposed adaptation is not summarised here in detail. The paper is also a position piece that argues for a policy; it is not a study measuring what preregistration achieves in HCI, and its case rests on evidence imported from other fields.

conference-paper · 2018 · Open the source

Crameri and colleagues argue that the accurate representation of data requires colour chosen with the same care as the encoding itself, and they document what goes wrong when it is not. Their abstract states that colour maps which visually distort data through uneven gradients, or which are unreadable to readers with colour-vision deficiency, remain prevalent in science, naming rainbow-like and red-green maps among them. They present a guide for the scientific use of colour, show that scientifically derived maps report true data variation and reduce complexity, and describe how the community, publishers and the press can recognise and prevent the misuse of colour. For the module this is the central source for the claim that perceptual uniformity and accessibility are reproduction requirements rather than finishes. Its evidence is a mixture of the measurable lightness profiles of published palettes and the settled physiology of colour vision, not a controlled trial of reader accuracy, and the note should say so when the claim is used.

journal-article · 2020 · Open the source

Dourish and Bellotti compare two ways a collaborative system can supply awareness. Active mechanisms ask people to report what they are doing, or generate information in a channel separate from the work. Passive mechanisms display activity inside the shared object. Their objection to the active approach is that it penalises whoever supplies the information, presumes it is relevant to whoever receives it, and is awkward to consult. Their evidence is a study of ShrEdit, used by four groups of three designers given ninety minutes to design an unstaffed twenty-four-hour post office, with sessions video-recorded and participants interviewed afterwards. The groups moved between close and loose collaboration, allocated authorship without assigned roles, and made little use of the explicit find and track controls, while complaining about the missing edit cursors. The value is the distinction and the case detail. The limits are a single system, four small groups, one task and a 1992 laboratory setting, reported qualitatively with no effect sizes, so the claim that structure should not be preordained rests on that small corpus.

conference-paper · 1992 · Open the source

This paper is the anchor for a decolonial reading of the field. Dourish and Mainwaring argue that ubiquitous computing has a grand, universalising vision, and that in this it follows a long tradition of projects that try to unite scattered elements into one conceptual whole. They link that tradition to colonial intellectual history, identify the problems that follow, and discuss the locatedness of innovation, meaning that where a technology is made shapes what it assumes about the world. From this they set out strategies for decolonising the research method. The value for a design student is that it changes the question from how a system should work to whose world the system assumes, and where that assumption came from. The limit is its kind. It is an argument built from history and theory rather than a study, and it reports no design outcome and no measurement. It should be cited for its critique, and a reader who wants evidence about what decolonising a project changes in practice must look elsewhere.

conference-paper · 2012 · Open the source

The framework sets out the Economic and Social Research Council's expectation of good practice for social science research, and it is addressed to researchers, research organisations and research ethics committees. Its most directly usable content is a list of six principles: research should maximise benefit and minimise harm, respect the rights and dignity of individuals and groups, secure voluntary and appropriately informed participation wherever possible, be conducted with integrity and transparency, define lines of responsibility and accountability clearly, and maintain independence with any conflicts of interest made explicit. It states that ethics issues should be considered across the whole lifecycle of a project, including design, dissemination and the archiving and future use of data, and it offers case studies alongside the principles while saying that it cannot replace self-critical ethical reflection. The limit a student must record is that this is funder guidance rather than an institution's policy: the reader's own university policy and its committee decide the application, and this page carries a last-updated date rather than an edition date.

report · 2025 · Open the source

This paper is a road map, and reading it as a road map is the point. Ens and colleagues report 17 key research challenges for immersive analytics, the area that combines visualisation, immersive environments and interaction to support the analysis of data. The challenges were developed over several sessions by 24 international experts, starting from a workshop at CHI 2020, and the paper presents them in order to coordinate future work. The value for a student is twofold. It names the open problems in spatial and immersive computing in the words of the people working on them, which is useful when positioning a project. And it is a clear example of a synthesis contribution whose evidence is agreement among experts rather than measured outcomes. The limit follows directly. A list of challenges shows what a community expects to be hard. It is not evidence that the technology will deliver on its promise, and it cannot tell you which challenge will be solved and which will stay open. The paper is candid that its purpose is to coordinate, not to predict.

conference-paper · 2021 · Open the source

Eysenbach argues that eHealth studies have a characteristic that drug trials do not: a substantial proportion of users stop using the application or drop out. He calls this the law of attrition, and he rejects the reflex that treats a high dropout rate as a failed study, because researchers then gloss over it or do not publish at all. His examples are stark. In an online panic-disorder programme only 12 of 1161 users completed the twelve-week course. In an open depression service 97 of 19607 users completed all five modules, against 41 of 182 participants in a trial setting, so the same programme behaved differently by setting. Attrition, he argues, should be measured, analysed and discussed, with usage metrics and determinants of dropout reported alongside treatment outcomes. He recommends methods from survival analysis, including Kaplan-Meier estimation and proportional hazards models, a usage half-life as a summary, and reporting the characteristics of the subgroup for whom the application works. He notes that intention-to-treat analysis over heavy dropout can understate the effect on continuing users, and proposes a run-in and withdrawal design. The paper is a viewpoint, so its figures are drawn from other studies.

journal-article · 2005 · Open the source

This article is the practical account of integration in mixed methods research, and it stands in for the Creswell and Plano Clark design typology on a student's shelf. Fetters, Curry and Creswell describe integration at three levels. At the study design level they distinguish exploratory sequential, explanatory sequential and convergent designs, and note four advanced frameworks including participatory and case study designs. At the methods level they set out four approaches: connecting, building, merging and embedding. At the interpretation and reporting level they name narrative, data transformation and joint display. They add the idea of the fit of integration, which is the extent to which the two sets of findings cohere. The most useful teaching point is that integration is a property to be designed and shown rather than a virtue claimed in a discussion. Its limits are that it is a conceptual and illustrative article rather than an empirical study, its examples come from health services research, and the version of record is subscription-only, so the free author manuscript is the copy read here.

journal-article · 2013 · Open the source

Flaxman, Goel and Rao test the filter bubble claim rather than asserting it, which is why the paper is the reference for this contested question. Their method is an observational analysis of web-browsing histories for 50,000 users located in the United States who regularly read online news, which lets them compare different routes to news, with social networks and search engines set against direct visits to publishers. The findings support both sides of the debate. Social networks and search engines were associated with an increase in the mean ideological distance between individuals, and the same channels were also associated with an increase in a person's exposure to material from the side of the spectrum they preferred less. Most news consumption was direct visits to the home pages of favourite mainstream outlets, and the authors describe the magnitudes as relatively modest. What is most useful for teaching is exactly that mixture, because it shows which claim the data can carry. The limits are structural. The design cannot separate what an interface offered from what a person chose to open, the outcome measures exposure rather than belief, and the panel covers one country at one time. The paper is evidence about the shape of news consumption, not a verdict on whether recommendation changes minds.

journal-article · 2016 · Open the source

Frauenberger argues that the field's intimate entanglement with digital technology is unsettling its foundations, and that the resulting uncertainties about what is real, what can be known and what is right call for a shift in the research paradigm. The proposed shift, Entanglement HCI, draws on post-humanism and relational ontologies, and the paper reviews Actor-Network Theory, post-phenomenology, object-oriented ontology and agential realism as the sources it builds on. It develops the position through four perspectives: the performative relationship between people and technology, the reframing of knowledge generation around phenomena rather than users, the tracing of accountability and responsibility, and design practices that move beyond user-centred design. The value is that it gives a student a serious vocabulary for the more-than-human turn rather than a slogan. The limit is the limit of all position papers. It tests nothing. It argues that the paradigm should change, and it offers no measurement of whether the change improves research or design. Read as theory it is generative; read as evidence it is thin.

journal-article · 2020 · Open the source

This is the source to reach for when a sustainability claim needs a number rather than a gesture. Freitag and colleagues review peer-reviewed estimates of the share of global greenhouse gas emissions attributable to information and communication technology and put the current figure at 1.8 to 2.8%. Their own contribution is a correction: adjusting for truncation of supply-chain pathways raises the estimated share to between 2.1 and 3.9%. They then examine the assumptions behind analysts' projections, find that the projections vary because of those assumptions, and give three reasons to expect emissions to rise without intervention. They also assess sector pledges and argue that not all are ambitious enough for climate targets, and that current policy mechanisms are too weak to enforce sector-wide compliance. For an interaction designer the useful move is to treat this as a design constraint with a measured magnitude. The limits are real: estimates depend on contested system boundaries, the projections are scenarios rather than predictions, and the figures will date.

journal-article · 2021 · Open the source

Ganguli and colleagues describe an early programme of red teaming, the practice of finding harmful outputs before users do. Their method is deliberately transparent. They red team four model types across three sizes, report how that effort scales, release the 38,961 attacks they collected, and set out their instructions, statistical methods and their own uncertainty about what the exercise can support. The most useful finding for a design review is a negative one: models trained with human feedback became increasingly difficult to red team as they grew, while the other model types showed a flat trend with scale. What is most useful about the paper is its candour, because it lets a reader see how much judgement sits inside a headline. Two limits should be carried into any red team report. The exercise samples attacks rather than estimating how often a harm would occur in use, so an absence of findings is not a clearance, and it finds what the team thought to look for. The findings describe the models of 2022, and only the abstract was read in this run, so the scaling figures were not verified here.

preprint · 2022 · Open the source

Glasgow, Lichtenstein and Marcus address the gap between research and practice, and they locate one cause in an assumption: that effectiveness research naturally and logically follows a successful efficacy study. They argue that the two traditions have evolved different methods and values, so the characteristics of a successful efficacy intervention differ from those of an effective one. Their thesis is that the trickle-down model, moving from efficacy studies to effectiveness trials to dissemination, is inherently flawed or at least incomplete, because an intervention that works under tight conditions is unlikely to carry its effects unchanged into a service. Moderating factors that limit robustness across settings, populations and staff need attention during planning, not afterwards. Their recommendations centre on documenting reach, adoption, implementation and maintenance, and on adapting trial reporting to include external validity. The paper is a position piece with recommendations rather than an experiment, so it argues a case from the literature. It is nonetheless the source of the reach and maintenance vocabulary that later deployment work uses.

journal-article · 2003 · Open the source

Grudin's article is the standard explanation of why collaboration software fails. Its argument is that most expensive failures are not technical: they arise from eight challenges created by the gap between software for one person and software for a group. The first is a disparity between who does the work and who gets the benefit, illustrated by digitised voice annotation that saves the speaker time and costs the listener it, and by group calendars that ask everyone to maintain a record serving a manager. Others include critical mass and free riding on discretionary databases, compatibility with social practices, exception handling where enforced procedure can halt work, the design of infrequently used features, the underestimated difficulty of evaluation, unreliable intuition about groups, and acceptance. The most useful part for a student is the worked contrast with single-user software and with organisational systems, and the remedy that follows: design the process of use alongside the technology, and give every member a direct benefit. The limit is that the article argues from the author's product-development experience and from published cases rather than from a measured study, and the eight challenges have not been separately quantified.

journal-article · 1994 · Open the source

Heer and Bostock reran Cleveland and McGill's comparisons with crowdworkers on the web, using the same error measure so that the new results could sit beside the old ones. The shape and order of the judgement types were preserved, which is why the module still teaches the ranking; two things changed, with the most accurate tasks moving closer together and length judgements improving. They also added a new experiment on rectangular area, of the kind treemaps produce, and found that accuracy matched circular area, and that aspect ratio mattered, with the squares a squarified treemap aims for producing the worst judgements. The paper states its own limits: display, lighting and viewing distance were uncontrolled, so only a subclass of perception experiments can run this way, and visual variables interact. Its value to the module is as the second test of a first result, and as a warning against reading a ranking as a specification.

proceedings-article · 2010 · Open the source

This is a systematic review of studies that have tried to test saturation, and it is the most useful single source on the number question. Hennink and Kaiser searched four databases and identified 23 articles that assessed saturation with empirical data or statistical modelling. The studies using empirical data converged on a narrow range of 9 to 17 interviews, or 4 to 8 focus group discussions, and that convergence is the headline. The conditions matter as much as the range: it held particularly for relatively homogeneous study populations with narrowly defined objectives, and the outliers, including multi-country research and work seeking saturation of meaning rather than of codes, needed larger samples. The authors are careful that the synthesis gives guidance rather than a rule, and they call for further work on how study parameters affect sample size. The method is a literature review, so the finding is about what other researchers observed rather than about any one study, and no single figure in it belongs in a methods section without its accompanying conditions.

journal-article · 2022 · Open the source

Hoffmann and colleagues built TIDieR because the description of interventions in publications is poor, with the result that clinicians cannot implement what works and researchers cannot replicate or build on it. An international group developed the checklist from a literature review, a Delphi survey of experts and a panel meeting, and produced twelve items: brief name, why, what materials, what procedure, who provided it, how, where, when and how much, tailoring, modifications, and how well it was delivered as planned and in fact. It is an extension of item 5 of the CONSORT 2010 statement for trials and of an item in the SPIRIT statement for protocols, and the authors intend it to apply across evaluative designs. The evidence for the problem is telling: one comparison found 67 per cent of drug interventions described adequately against 29 per cent of non-pharmacological ones, and another found only 39 per cent of 137 non-drug interventions adequately described, rising to 59 per cent after authors were contacted. The checklist is a reporting standard, not a method for running a study. Its value here is that its items transfer almost unchanged to a deployment report.

journal-article · 2014 · Open the source

The paper tests a claim the field repeats without evidence: that HCI contains too few replications. The authors examined four publication outlets, covering 891 papers, and found about 3 per cent attempting to replicate an earlier result. They also report that the replications they found usually treated replication as a confirm or do not confirm decision, rarely analysing effect sizes or comparing in detail with the paper being replicated, and that when asked, most authors agreed that studies they had run were replications, though they had rarely planned them as such. The last finding is the most useful for a project writer, because it shows that the label of replication is often applied after the fact rather than designed in. The limits are scope and age: the count covers four outlets in a period the authors defined, so it will not describe the field today, and the study is descriptive rather than a judgement that replication is always the right design for a question. The paper was open access and its abstract was read in this run; the full text was not read.

conference-paper · 2014 · Open the source

Hullman and colleagues address a specific difficulty: uncertainty visualisation is important, but evaluating it is hard because people read uncertainty poorly and because correct behaviour with uncertainty information is difficult to define. They present a taxonomy that differentiates six levels of decisions made in designing an evaluation, from the behavioural targets of the study and the expected effects of uncertainty through to the analysis. Applying the taxonomy to published studies, the survey reports that evaluation practice concentrates on performance and satisfaction measures, and that those measures assume more predictable judgement than the research on judgement and decision making supports. The paper is a survey and a framework, so its contribution is to structure a problem rather than to settle it. For the module it is the evidence that a display can score well on an abstract task and still mislead a reader, which is why the uncertainty unit is taught as a question about inference rather than about numbers.

journal-article · 2019 · Open the source

This is the origin of the technology probe, and the definition matters more than the technology. A probe is an instrument deployed to find out about the unknown, and the authors accept that it may fail or return something unexpected. Technology probes carry three goals at once: a social science goal of understanding needs and desires in a real-world setting, an engineering goal of field-testing the technology, and a design goal of inspiring users and researchers to think about new technology. The paper distinguishes probes from prototypes on functionality, flexibility, usability, logging and design phase. The honesty is instructive: a probe is not mainly about usability, is not changed during the deployment, and may deliberately leave out a function in order to provoke a reaction. The interLiving project deployed a messageProbe and a videoProbe with families in France, Sweden and the United States for a month or longer, and the authors record the administrative and hardware failures that required extra home visits. The evidence is a small number of families and a qualitative report, so it supports a method claim rather than an effect.

conference-paper · 2003 · Open the source

Jannach and colleagues wrote the technical introduction that most recommender courses still use, and for this material it is a source for mechanism rather than for claims about interfaces. The free first chapter sets out the requirement that every recommender maintains a user model, and separates acquiring preferences implicitly, by monitoring what a person views or buys, from asking for them explicitly. It states the collaborative rationale as a claim about resemblance, that users who agreed in the past will probably agree in future, and distinguishes collaborative from content-based and knowledge-based approaches. The most useful parts for a designer are the bookstore framing, which treats a recommendation as an interface object rather than an algorithm output, and the structure of the later chapters, with chapter 6 on explanations and chapter 7 on evaluation. Its limits are those of a textbook. The treatment turns mathematical where design material needs judgement, the retail examples date from 2010, and the evaluation chapter sits behind a paywall. Only the front matter and chapter 1 were read in this run, so the later chapters are cited as the publisher's contents listing rather than as text.

book · 2010 · Open the source

Klasnja, Consolvo and Pratt ask how technologies for health behaviour change should be evaluated within HCI, and they reject the obvious answer. Assessing whether a technology produced the intended change in behaviour is, they argue, too limited: demonstrating behaviour change is often infeasible and unnecessary for a meaningful contribution, particularly in the early stages of design or when the technology is novel. Their alternative has two parts. First, efficacy evaluations should be tailored to the specific behaviour-change strategies embodied in the system, such as self-monitoring or conditioning, rather than to a general outcome. Second, studies should aim at a deep understanding of people's experience with the technology. The paper is a position piece written against a trial-centred standard, and it is best read beside Glasgow and colleagues, who argue from the public health side that reach and maintenance have to be measured. The two positions are the live disagreement in this area. Only the abstract could be read in this run.

conference-paper · 2011 · Open the source

Kraut and Resnick set out to translate social science into design claims for online communities, stating with each claim the evidence behind it, drawn from theory, experiment or observational study. The book organises the area around five challenges: starting a new community, attracting and incorporating newcomers, encouraging commitment, encouraging contribution, and regulating misbehaviour and conflict. Contribution is analysed as a public goods problem, which explains why free riding is the default prediction and why design has to change the incentives, and commitment is analysed through interpersonal bonds. The treatment of newcomers is the most directly usable in teaching: because members leave, every surviving community must absorb new ones, who are sensitive to its public image and to their own first experience of it. The book is honest that most online communities never get off the ground. Its limit is structural: the design-claim format organises evidence without predicting which claim binds in a new platform, and the evidence behind different claims varies in kind, so each should be checked before it is transferred.

book · 2012 · Open the source

The article states the purpose of a pilot study and the limits of what its results can support. The authors argue that a pilot examines the feasibility of an approach intended for a larger study, and that the feasibility of recruitment, randomisation, retention, assessment procedures and intervention implementation can each be quantified. They are equally clear about what a pilot is not. It is not a hypothesis test, so safety, efficacy and effectiveness are not evaluated. Contrary to common practice, it does not provide a meaningful effect size estimate for planning the main study, because an estimate from a small sample is too imprecise. Feasibility results also do not necessarily generalise beyond the inclusion and exclusion criteria of the pilot itself. The paper is written for clinical trials, and it uses a randomised controlled trial as its running example, so the transfer to a human-computer interaction study is an analogy. The part most useful in a project proposal is the table of feasibility aspects and the insistence that the objectives be stated plainly, because those give the panel something measurable.

journal-article · 2011 · Open the source

Luger and Sellen report 14 semi-structured interviews with people who used conversational agents in everyday life, including Siri, Cortana and Alexa, at a point when HCI had little evidence about how such agents were actually used. Their finding is that user expectations were dramatically out of step with how the systems operated, particularly about the machine's intelligence, its capability and its goals, and they analyse the result with Norman's gulfs of execution and evaluation. The most useful material for design is their account of the deep gulf of evaluation: users could not judge how much the system had understood, they confirmed visually on all but the simplest tasks, and they avoided the agent for anything complex or sensitive. The paper also sets out where dialogue was expected to pay, including filtering large sets, building a query over several turns and referring to what cannot be pointed at, and then records the recognition errors, unintelligible responses and hard to navigate dialogues that spoil it. Two limits should be stated. The sample is small, mostly male and United Kingdom based, and the systems are those of 2016, so the mechanisms travel further than the figures. The analysis is thematic rather than a measurement of system capability.

conference-paper · 2016 · Open the source

Marwick and boyd interviewed Twitter users with different kinds and sizes of following to ask how they decide what to post. Their argument is that social media collapse multiple audiences into a single context, so the face-to-face technique of varying self-presentation by audience becomes unavailable, and users work instead from an imagined audience assembled from cues in the environment. They describe the strategies users reported: targeting different audiences, concealing subjects, self-censoring, and balancing the personal against the strategic, alongside practices resembling micro-celebrity and personal branding. The most useful teaching point is the finding that the imagined audience is often the most sensitive reader a user can picture, which makes public posting a continuous act of self-censorship rather than unguarded disclosure. The limits are those of interview research conducted in 2009: a small purposive sample, self-reported strategies, one platform, and a period before the site's audience and norms had settled, so the study describes a mechanism rather than establishing its prevalence.

journal-article · 2011 · Open the source

This is a widely cited programme for making science more reliable, written by ten researchers across psychology, neuroscience, statistics and meta-research. It argues for measures that optimise five elements of the process: methods, reporting and dissemination, reproducibility, evaluation, and the incentives that reward researchers and journals. Its honest self-assessment is what makes it useful for a project paper, and it is worth quoting precisely. The authors write that there is some evidence from simulations and empirical studies for the likely effectiveness of these measures, and that broad adoption by researchers, institutions, funders and journals will require iterative evaluation and improvement. That sentence is a model for how to present a reproducibility package: as a set of practices with plausible but not settled support, not as a guarantee of truth. The limit is that the manifesto concerns the scientific literature as a whole, largely written with the life and behavioural sciences in view, and it says little about design or qualitative HCI work, where the package has a different shape and different constraints.

journal-article · 2017 · Open the source

Munzner's book is the framework this module is organised around, and it is the source of the what, why and how sequence. Design is treated as a series of choices about what data the reader must see, why they must see it, and how marks and channels can carry it, with function settled before form. The book covers abstract data types and dataset types, the channel ranking and its perceptual basis, the trade-offs between one view and several linked views, and the ways of reducing what each view shows. Six case studies close the book, each analysed through the same framework. It is a textbook and a synthesis rather than a report of new experiments, so its perceptual claims summarise the work of others, and a student citing a specific result should follow the reference back rather than stop here. The publisher page describes the contents but the text is not free to read, and the companion site at the author's university carries figures, slides and further reading at no cost.

book · 2014 · Open the source

The paper argues that generating a hypothesis from existing observations and testing a hypothesis on new observations are different acts, and that ordinary biases such as hindsight bias make the difference difficult to maintain in practice. Its remedy is preregistration, defined as committing to research questions and an analysis plan before the outcomes are observed, usually by posting the plan to a registry, which preserves it and makes it discoverable. Preregistration then distinguishes analyses that were predictions from analyses that were postdictions. The authors are explicit about the strength of their own case, describing the theoretical argument as strong and the empirical support as correlational, and stating that experimental evidence establishing preregistration's superiority for reproducibility is not yet sufficient. They add that the benefit is lost if the plan is not followed. For a proposal the useful part is the vocabulary of prediction and postdiction and the honest caveat. The limit is that the paper is an advocacy article rather than a review, and the caveat should travel with any use of it.

journal-article · 2018 · Open the source

Olson and Olson review over ten years of laboratory and field studies of collocated and remote group work, against the claim that communications technology had abolished distance. Their synthesis uses four concepts: common ground, the coupling of the work, readiness to collaborate, and readiness for the technology. Remote work has a chance when common ground is high and the work is loosely coupled, and deviation from any of the conditions puts strain on the arrangement. The most useful material for teaching is the detail. In the laboratory, audio plus a shared editor produced lower quality work than meeting face to face, while video produced equivalent output at the cost of more clarification and management overhead. In the field, remote tools were used less and less until the work was reorganised, with design repartitioned into loosely coupled modules owned by single sites. The limits are that the review selects the authors' own studies, the technology described is that of 2000, and video's benefit is shown to be conditional rather than general.

journal-article · 2000 · Open the source

PRISMA 2020 is the reporting guideline for systematic reviews, and it is the document that turns a literature search into a method with a record. Its core is a 27-item checklist for the review, with a short separate checklist for the abstract and revised flow diagrams for new and updated reviews. Each item asks for a decision a reader would otherwise have to guess at: the review question, the search strategy, the eligibility criteria, the number of records at each stage of screening, the assessment of bias in the included studies, and the way results were synthesised. The paper is a reporting standard, not a method for conducting a review, and it does not tell you which question to ask or how to search. Two limits are worth stating. It grew out of health research, so some items map onto interaction research loosely. And the 27 items describe what must be reported, not what makes a review good, so a fully compliant review of a badly framed question is still a badly framed review.

journal-article · 2021 · Open the source

Palinkas and colleagues review purposeful sampling as it is used in mixed-methods implementation research, and the result is a practical taxonomy rather than a defence of one strategy. They separate strategies that emphasise similarity, such as criterion, typical case, homogeneous and snowball sampling, from those that emphasise variation, such as maximum variation, extreme case and intensity sampling, and they give each an objective, an example and a caution. Their own audit of 28 implementation studies found that 21 used some form of criterion sampling, and that most studies did not describe their sampling strategy at all. The most useful teaching material is the table of strategies, with the blunt note that convenience sampling is neither purposeful nor strategic. The paper also records that some methodologists reject systematic sampling of any kind as a positivist import, which is the honest counter-position. The limits are that every example comes from mental health implementation research, the percentages come from a small review rather than a representative sample, and the authors' recommendation that strategies be combined is argued rather than tested.

journal-article · 2015 · Open the source

Perski, Blandford, West and Michie asked how engagement with digital behaviour change interventions is conceptualised and measured. They searched four databases in November 2015, identified 117 articles, and synthesised them using principles from critical interpretive synthesis. Their conclusion is that engagement has two faces. The experiential face covers attention, interest and affect; the behavioural face covers what people do with the intervention. They propose a framework in which engagement is influenced by the intervention's content and delivery, by the context of use including the setting and the population, and by the behaviour being targeted, with a possible feedback loop between engagement and effectiveness. The measurement section is the most useful part for a deployment researcher: self-report questionnaires and interviews reach experience, automatic records of logins and page views reach behaviour, and physiological measures have shown weak associations with self-report. The authors describe the relationships in their framework as tentative, because they have not been much studied. The review is conceptual and does not test the framework.

journal-article · 2017 · Open the source

Rogers and Marshall argue that in-the-wild research differs from other applied approaches in where it starts and ends. Ethnographic and user-centred work typically begins by observing existing practice and then offers requirements or design implications. Research in the wild instead creates and evaluates a new technology in situ, and it often aims to augment a person, a place or a setting rather than to answer a stated user need. The authors also describe a shift in intent: rather than fitting a solution to existing practice, the researcher installs an intervention and observes how people react, change and integrate it into daily life. The book is short and structured as a framework, covering approaches, methods, case studies and outcomes, with a separate chapter on practical and ethical issues. For a student, the most useful parts are that chapter and the framing of the approach as a position with costs rather than as a single method. Its limit is that it is a synthesis and an argument, not a programme of experiments.

book · 2017 · Open the source

Rosenblat and Stark spent nine months, between December 2014 and September 2015, collecting and analysing Uber drivers' posts in driver forums, both retrospectively and in real time, alongside a critique of the application's design and of the company's public claims. Their argument is that the information and power asymmetries produced by the application are what structure control over drivers, and that the language of technology and algorithms legitimises an unequal relationship presented as independent partnership. The concrete case is the ride request: about fifteen seconds to accept, with destination and fare withheld, low minimum fares, and the risk of deactivation for cancelling unprofitable trips, which the authors read as soft control and gamified engagement rather than direct command. It is the clearest available demonstration that an interface decision about who knows what is a decision about power. The limits are the method: public forum posts and interviews with self-selected drivers, no access to company data, and an interpretive reading of corporate rhetoric that another analyst could frame differently.

journal-article · 2016 · Open the source

The article asks how researchers construct research questions from existing literature and answers with a typology built from close reading. The authors reviewed 52 articles published in leading United States and European management journals, and they report that the most common way across paradigmatic camps is to spot gaps in the literature and formulate a specific question from the gap. Against the variants of gap spotting they set problematization, which identifies and challenges the assumptions underlying existing work and formulates the question from that challenge. The paper is a conceptual contribution supported by textual analysis, and it is written for organisation studies rather than for human-computer interaction, so any transfer is an analogy rather than a finding. Its most useful material in a project proposal is the vocabulary and the distinction: a student can ask whether the proposed question rests on a gap that has been demonstrated or on an assumption nobody has examined. The limit is that the paper argues a position and does not show that problematization produces better research; the claim that it leads to more influential theory is the authors' assertion.

journal-article · 2011 · Open the source

Saraiya, North and Duca argue that the standard evaluation of a visualisation, a controlled task with predetermined answers, does not capture what an analysis tool is for. They define insight as an individual observation about the data and build a method that recognises and quantifies insight in open-ended user tests, then apply it to five microarray visualisation tools, reporting the amount and types of insight each supported and the time taken. The method is the module's example of an evaluation that measures something closer to real analysis rather than to a taught search. Its costs are stated in the paper: coding insight is labour intensive and needs a domain-expert coder and self-motivated participants, each session was short, and a person analysing their own data works for far longer. The module uses it to make the point that no single measure of a visualisation is neutral, and that the choice of measure is itself a design decision with ethical consequences.

journal-article · 2005 · Open the source

Schmidt and Bannon argue that CSCW had no settled object and would dissipate for want of one. Their proposal is that the field should be concerned with the support requirements of cooperative work arrangements, rather than with computer support for groups, and they contrast that with the small-group conception which dominated the early systems work. The concept they introduce for the missing object is articulation work, the coordinating work that holds a distributed arrangement together, and the abstract states that the problems it raises are marginalised in work on small group support while being critical to systems designed for work in the everyday world. The paper is a position statement, published in the first volume of the journal it helped define. It argues a definition and reports no empirical study, so it should be used for what the field takes as its object rather than for evidence about systems. Its vocabulary, articulation work above all, is still in use. Read with that limit in mind it is the clearest short statement of why CSCW is not simply groupware. The full text is paywalled.

journal-article · 1992 · Open the source

Segel and Heer systematically review the design space of narrative visualisation, drawing on case studies from news media and visualisation research. They identify distinct genres, among them the magazine style, the annotated chart, the slide show and the video, and characterise the differences along with interactivity and messaging in terms of the balance between the narrative flow intended by the author and story discovery on the part of the reader. That balance is the organising idea the module uses when it treats annotation and sequence as an authorial act rather than as decoration. The paper is a review and a taxonomy built from a curated set of published examples, so it names the genres and the design moves without measuring their effect on readers. Its value is that it gives a vocabulary for the choices a designer makes when a static figure has to carry an argument, and it makes visible how much of the author's reading is built into the artefact.

journal-article · 2010 · Open the source

Sharma and colleagues investigate why assistants agree with the person using them, and this is the source for the claim that sycophancy arrives through the training signal rather than from a single model flaw. Their method combines a demonstration that five state-of-the-art assistants exhibited sycophancy across four free-form text-generation tasks with an analysis of existing human preference data, and an experiment in which model outputs were optimised against preference models. The results are what a design review needs. Human raters and preference models both preferred a convincingly written sycophantic response over a correct one a non-negligible fraction of the time, and optimising against a preference model sometimes sacrificed truthfulness. The most useful consequence is that a preference model is not a neutral referee, which links this paper directly to the circularity problem in model-based evaluation. The limits should be stated plainly. The study reports the behaviour of five specific assistants at a particular time, so the pattern should be re-checked rather than assumed for a new system, and only the abstract was read in this run, so the task list, the exact rates and the optimisation procedure were not verified here. The finding concerns aggregate preference, not any individual user's experience.

preprint · 2023 · Open the source

This short paper is the origin of the phrase that structures interactive visualisation teaching, and of the task and data type taxonomy behind it. The abstract states the visual information seeking mantra directly: overview first, zoom and filter, then details on demand. Around it the paper proposes seven data types, from one, two and three dimensional data through temporal, multidimensional, tree and network data, and seven tasks: overview, zoom, filter, details on demand, relate, history and extract. The value to a student is that it turns interaction into a set of named tasks that can be designed for and checked, which is what brushing, linking, small multiples and coordinated views are built on. The paper is a position statement and an organising framework rather than a report of experiments, and it says so, noting that more tasks and refinements would be natural next steps. It contains no measured comparison of interaction designs.

conference-paper · 1996 · Open the source

Taylor's short book is an introduction to discourse analysis as an empirical research approach, and it is unusually clear about what the approach asks of data. Her working definition is that discourse analysis is the close study of language and language use as evidence of aspects of society and social life. She sets out the premises that travel with it: language is constitutive rather than a neutral channel, meanings are socially derived and also situated, negotiated and co-constructed, and language use is a functional social practice. She is explicit that there is no single discourse analysis, and that narrative analysis is one recognisable form of it, which is why the two can be taught together. The book also covers the practical work, including data collection, processing, transcription and selection, and it draws on published studies on topics such as gender, health and national identity. The limits: it is an introductory text rather than a manual, its examples come mainly from British social research, and only the repository record, abstract and opening pages were read here.

book · 2013 · Open the source

Tracy offers a deliberately broad answer to the question of what makes qualitative research good. She distinguishes the means of qualitative research, its methods and practices, from its ends, and proposes eight markers of quality: a worthy topic, rich rigour, sincerity, credibility, resonance, significant contribution, ethics, and meaningful coherence. The argument is for a big tent set of criteria that many traditions can recognise, and she frames the list as a pedagogical model and a shared language rather than a test. That is the useful part for teaching: it gives a student a vocabulary for a review conversation, and it separates rigour in the method from the significance of the contribution, which are routinely conflated. Its limits are that it is a conceptual synthesis rather than an empirical study, so it does not show that work meeting the criteria is better, and the author explicitly leaves space for dialogue and improvisation. Only the abstract was read in this run, because the full text is paywalled.

journal-article · 2010 · Open the source

This is the origin of the data-ink argument and of the small multiple as a design form. Tufte's position is that a graphic should be stripped of any ink that does not carry data, that the ratio of data-ink to total ink is a measure of graphical quality, and that graphical deception is detected by comparing design variation with data variation. Small multiples, the same encoding repeated across panels with fixed scales, are offered as a way to make comparison cheap by separating comparison between panels from reading within one. The book is a craft argument made by example: its before-and-after figures are chosen by the author to persuade, and it reports no experiments on reader accuracy. That is the ground for reading it critically rather than dismissing it. A purist reading of the ratio can remove gridlines and tick labels, which lowers the accuracy of the estimate a reader has to make, and the perceptual literature ranks that accuracy highly.

book · 1983 · Open the source

Wobbrock and Kientz argue that HCI produces a small, stable set of knowledge types and that a paper should be read and judged against the type it is making. They summarise seven contribution types, which are empirical, artifact, methodological, theoretical, benchmark or dataset, survey, and opinion, and give the criteria each attracts: empirical contributions on the importance of the findings and the soundness of the method, artifact contributions on what they make possible and how, methodological ones on utility, reproducibility, reliability and validity, opinion contributions on the credibility of their evidence and their fair treatment of alternatives. The column also reports how CHI 2016 papers were distributed across the types, which is the most useful part for a project writer deciding what to claim. Its limits are real. It is a magazine column rather than a study, its chief instrument is a taxonomy the first author had posted as an unpublished whitepaper in 2012, and the acceptance figures describe one conference in one year, so they should not be generalised to the field. The value is that it gives a project paper a vocabulary for its own contribution, and a reason to state that contribution rather than imply it.

journal-article · 2016 · Open the source

Yang and colleagues treat the pandemic shift to remote work as a natural experiment, separating the effect of firm-wide remote work from other pandemic-related changes by using employees who were already working remotely beforehand as a comparison. They analyse emails, calendars, instant messages, calls and workweek hours for 61,182 employees of one firm over the first six months of 2020. Their reported findings are that firm-wide remote work made the collaboration network more static and siloed, with fewer bridges between separate parts of the organisation, and that communication shifted from synchronous to asynchronous media, which the authors say may make it harder to acquire and share new information across the network. The teaching value is measured behaviour at a scale no interview study reaches. The limits matter: one firm, no random assignment, telemetry standing in for collaboration, no direct measure of output, and a correction published in October 2021. The authors state that the data are withheld because of employee privacy and legal restrictions, which is itself a teaching case in research ethics.

journal-article · 2022 · Open the source

This paper comes from an expert consensus process, with an international steering committee and a workshop of 42 experts, and it reflects the state of the art rather than reporting a single study. Its central move is to distinguish engagement from effective engagement, which the authors define empirically as sufficient engagement with the intervention to achieve the intended outcomes. That definition resists the assumption that more use is better use, and the authors state that sustained engagement is not always required. The paper reviews how engagement has been conceptualised and measured, argues that the relationship between engagement with the intervention and the target behaviour has to be established rather than assumed, and asks for valid combinations of measures rather than a single number. It also discusses tailoring to individual needs and combining digital delivery with human support, and notes that digital interventions can record engagement in detail over time. Its status as consensus is also its limit: it is selective by design and poses questions as much as it answers them.

journal-article · 2016 · Open the source

Zheng and colleagues address the evaluation problem that arrives with open-ended generation, where a task may have no single right answer and human judgement is expensive. Their contribution is two benchmarks, MT-bench, a multi-turn question set, and Chatbot Arena, a crowdsourced battle platform, together with a study of using a strong model as the judge. The reported result is that a strong judge such as GPT-4 matched controlled and crowdsourced human preferences with over 80 per cent agreement, which the authors describe as the same level of agreement found between humans. They also document the biases the method carries, including position, verbosity and self-enhancement, along with limited reasoning, and propose mitigations for some of them. That pair of findings is what makes the paper teachable. Model-based judging is usable, and it is not neutral. The limits follow from its own account. Agreement with aggregate human preference is not agreement with a domain expert, the biases are documented rather than removed, and a judge that shares a model family or training data with the system under test is not an independent rater. Only the abstract was read in this run, so the scoring procedure was not re-run.

preprint · 2023 · Open the source

This article is the introduction to a special section, and it is the source of the definition that most later work cites: web-based services that allow a person to construct a public or semi-public profile within a bounded system, articulate a list of other users with whom they share a connection, and view and traverse those lists. The authors choose network rather than networking because the defining practice is making an existing social network visible rather than meeting strangers. They trace the form to SixDegrees, launched in 1997, which combined the three features and still failed as a business, and they show that sites differ mainly in their visibility defaults, from profiles crawled by search engines to profiles a user chooses to publish and profiles visible to a network. Two limits matter. The article is a definitional and historical introduction rather than an empirical study, and the landscape it describes is that of 2007, before smartphone-dominant use and before feed-based distribution. Read as a definition it remains serviceable; read as a description of social media today it is a period document.

journal-article · 2007 · Open the source

The AoIR guidelines are the reference document that internet researchers in many disciplines work from when a study involves people's online traces. The third version was approved in October 2019 and published in 2020. Its approach is deliberately plural: it sets out several ethical frameworks, notes that they disagree, and concludes that the best it can offer is guidelines rather than recipes, because internet research raises problems that admit more than one defensible answer. The structure is useful in practice, covering stages of research, informed consent, a general structure for ethical analysis, involved subjects, and separate questions for data acquisition, analysis, storage and dissemination, with companion sections on algorithms and learning models, feminist research ethics and industry partnerships. On the question this module teaches, its starting point is who counts as a subject and the principle that greater vulnerability brings a greater duty of protection, with downstream harms, minors, politically sensitive research and named vulnerable groups. The limit is that it is guidance rather than regulation: it does not decide a case, and institutional and national requirements still govern.

report · 2020 · Open the source