Legibility: one enforced threshold and no settled measure

Legibility is two questions wearing one word. The first is whether a reader can see the text, and it has a magnitude: luminance contrast. A rule can therefore be written and tested for it. The second is whether a reader can read the text well, and it has no settled measure. The same text can be read fast and understood poorly, or read slowly and understood well, and the studies hunting for the best line length disagree about what to count. A designer therefore works with one enforced threshold and no settled measure. The enforced threshold is the contrast criterion the conformance unit takes in full, and the risk is mistaking the testable figure for the property that matters. The strongest counter to this unit's thesis is that the search for a figure is the wrong search. Dillon (1992) concluded that no single-variable explanation captures reading from screens, so a settled line length may be unattainable. The unit before it arranged the page. This one asks how the text is read, and where the evidence stops.

The one legibility variable with a direction is luminance contrast

Buchner and Baumgartner (2007) compared dark text on light with light text on dark across proofreading experiments under varied illumination and chromaticity. Positive polarity was consistently better. A final experiment found that colour contrast could not compensate for a lack of luminance contrast. What predicts the performance is a quantity, and a ratio can describe it.

Buchner, Mayr and Brandt (2009) tested the explanation rather than the effect, varying polarity and display luminance independently. No polarity advantage appeared once the two display types were equally bright. The advantage was therefore a luminance advantage, so the effect survives and the mechanism does not.

The room and the display are inputs the document does not carry

Yu and colleagues (2018) had participants adjust surface brightness for paper and for a tablet below 300 lx, to find the optimal and minimum acceptable readability. The character contrast ratio did not influence the surface luminance the tablet needed, and paper could sit below ambient luminance given at least 27 cd/m2 of its own. Only one side of that relation is in the document.

Gattullo and colleagues (2014) took the point onto a display where the author controls almost nothing. Twelve participants produced about 3,400 readability measurements on two head-worn display technologies, optical see-through and video see-through, at 1,000 lx and 4,000 lx. The two technologies wanted opposite polarities. Readability depended on the lighting, the display and the text style together.

Work one page in two rooms. A tablet held in a normally lit office renders a document at a fixed contrast ratio legibly. Carried outdoors on a bright day, the same document at the same ratio is not legible, because the display must now out-shine the ambient light. The page does not control how much light falls on the screen.

The line-length experiments do not count the same thing

Line length is the variable practice treats as settled, and the experiments do not agree. Dyson and Haselgrove (2001) tested two reading speeds against several line lengths, measuring comprehension, rate and scrolling. Reading fast reduced comprehension without changing the type of information recalled. A medium line of 55 characters gave the highest comprehension and was read faster than short lines.

Shaikh and Chaparro (2005) put the same question to online news. Twenty students read articles at 35, 55, 75 and 95 characters per line, and comprehension was assessed by six question types. The 95-character passages were read fastest, satisfaction was unaffected, and preference went to the extremes. Structure questions were hardest of the six types. The samples are students and volunteers, so the figures describe skilled readers rather than a reader with low literacy or a reader in later life.

Cerepinko and colleagues (2017) moved the reader to a tablet: 220 students read from an iPad, a computer screen and paper, and lines of 79 characters and above were judged readable enough there.

The Baymard Institute's review (2022) gives 50 to 75 characters for body text, with a ceiling of 80 or fewer and a recommended maximum width of about 70 characters. Dyson (2004) names characters per line the critical variable, and asks for work on its interactions. That is a different kind of number from an optimum.

The medium question that will not settle

Dillon (1992) found the paper-against-screens literature dominated by attempts to overcome speed deficits from poor image quality. Dillon concluded that single-variable explanations cannot capture reading from screens. Noyes and Garland (2008) repeated the exercise fifteen years later. They conclude that total equivalence is not achievable, though better technology, better measures and more positive attitudes have moved the media closer, and that any remaining difference is specific to the task and outcome. Cerepinko and colleagues (2017) judge tablet reading good enough to replace printed text. Both positions can hold, because equivalence is a claim about a task and not a medium.

Legibility depends on the reader, and the popular remedies do not survive contact

Rello and Baeza-Yates (2013) tracked the eye movements of 48 people with dyslexia, each reading 12 texts in 12 fonts. They reported that sans serif, monospaced and roman styles improved reading performance over serif, proportional and italic. It is an eye-movement experiment on one group of readers, so the effect is in how the text was read.

Galliussi and colleagues (2020) separated the parts of the dyslexia-friendly package. Two groups of 64 children, with and without dyslexia, read eight equivalent texts while letterform and spacing varied. Letterform made no difference, and increasing inter-letter spacing without also increasing inter-word spacing impaired reading speed. The remedy that is sold therefore does nothing measurable, while the variable that does move reading is one a reader can set.

Vecino and colleagues (2022) tested the everyday version with 246 volunteers on an e-commerce prototype. They compared serif and sans serif within one family on four measures of usability and preference. No effect on reading speed appeared, and only gender predicted preference, so the study found no support for the industry's sans serif default.

Contested question: is there an optimum line length, and what would settle it?

Two positions sit inside these sources. The first holds that an optimum exists, and that characters per line is the variable to find it by. On this view 55 characters gave the highest comprehension and a fast rate at both speeds (Dyson and Haselgrove, 2001), and the review names characters per line critical (Dyson, 2004). The second holds that the optimum belongs to the measure and the device. On this view 95 characters was fastest while satisfaction stayed flat and preference went to the extremes (Shaikh and Chaparro, 2005). On a tablet, 79 characters and above was judged adequate (Cerepinko et al., 2017). Practice works to 50 to 75 against a ceiling of 80 (Baymard Institute, 2022).

What would settle it is a study that fixes content, typeface, size and reader group. It would vary line length over the same values from 35 to 95 characters per line, run the same readers on two devices, and report rate, comprehension by question type and preference separately. A difference between measures could then not be read as a difference between samples. None of these sources does that. Dyson and Haselgrove report comprehension and rate but no preference, on one display. Shaikh and Chaparro add preference, on one monitor. Cerepinko and colleagues report speed and difficulty rather than comprehension.

The smaller dispute over polarity has the same shape, and it is partly resolved. Buchner and Baumgartner (2007) attributed the advantage to polarity, while Buchner, Mayr and Brandt (2009) equated display luminance and found a luminance effect. Resolution needs a magnitude as the variable. The line-length argument has no equivalent, because its confound is the measure itself.

While the question is open, a designer should set the contrast figure as a floor and choose a line length as a working default rather than a law. Meet the contrast criterion, then check the result in the light the reader's device will meet. Give the reader control of the spacing and the text size, because the evidence shows those to move reading.

Where the argument is heading

Three directions are visible in named work, and all move away from a single figure for a block of text. The first puts the variable under the reader's control: spacing rather than letterform moved reading in the dyslexia literature (Galliussi et al., 2020). The second moves the arithmetic into the environment. The low-illuminance models (Yu et al., 2018) and the work under industrial lighting (Gattullo et al., 2014) both ask for guidance conditioned on the situation rather than the page. The third is methodological. Rello and Baeza-Yates (2013) found their effect with eye tracking on fixation duration, while work measuring speed and preference found nothing (Vecino et al., 2022). The prediction is that per-reader settings and environment-aware luminance will carry legibility before any new consensus figure does. The study that would confirm it, a factorial comparison of length, device and measure, has not been run.

The sources this unit rests on

• Buchner, A. and Baumgartner, N. (2007). Text-background polarity affects performance irrespective of ambient illumination and colour contrast. Ergonomics. doi:10.1080/00140130701306413 Proofreading experiments compared dark text on light with light text on dark under varied ambient illumination and chromaticity. Positive polarity was consistently better, and colour contrast could not compensate for a lack of luminance contrast. It is the unit's luminance-contrast evidence.
• Buchner, A., Mayr, S. and Brandt, M. (2009). The advantage of positive text-background polarity is due to high display luminance. Ergonomics. doi:10.1080/00140130802641635 The experiment varied polarity and display luminance independently. No polarity advantage appeared once the two display types were equally bright, so the 2007 effect was an effect of luminance.
• Yu, H., Akita, T., Koga, T. and Sano, N. (2018). Effect of character contrast ratio of tablet PC and ambient device luminance ratio on readability in low ambient illuminance. Displays, 52, 46-54. doi:10.1016/j.displa.2018.03.002 Participants adjusted surface brightness on paper and on a tablet below 300 lx. Contrast ratio did not influence the surface luminance the tablet needed, so the contrast a reader must have is set by the room.
• Gattullo, M., Uva, A. E., Fiorentino, M. and Monno, G. (2014). Effect of text outline and contrast polarity on AR text readability in industrial lighting. IEEE Transactions on Visualization and Computer Graphics. doi:10.1109/tvcg.2014.2385056 Twelve participants produced about 3,400 readability measurements on optical see-through and video see-through displays at 1,000 lx and 4,000 lx. Readability depended on the lighting, the display and the text style together, so the page alone cannot fix the ratio.
• Dyson, M. C. and Haselgrove, M. (2001). The influence of reading speed and line length on the effectiveness of reading from screen. International Journal of Human-Computer Studies, 54(4). doi:10.1006/ijhc.2001.0458 Two reading speeds were tested against several line lengths with comprehension, reading rate and scrolling measured. A medium line of 55 characters gave the highest comprehension and was read faster than short lines. It is the figure practice repeats.
• Shaikh, A. D. and Chaparro, B. S. (2005). The effects of line length on reading performance of online news articles. Proceedings of the Human Factors and Ergonomics Society Annual Meeting, 49(5). doi:10.1177/154193120504900514 Twenty college-age students read online news at 35, 55, 75 and 95 characters per line, with comprehension measured by six question types. The 95-character passages were read fastest while satisfaction did not differ, so the best length depends on the measure chosen.
• Cerepinko, D., Perisa, M. and Kecek, D. (2017). Text readability and legibility on iPad with comparison to paper and computer screen. Tehnicki vjesnik - Technical Gazette, 24(4). doi:10.17559/tv-20160225140202 Two hundred and twenty university students read from an iPad, a computer screen and paper. Lines of 79 characters and above were judged readable enough on the tablet, so the number moves with the device.
• Dyson, M. C. (2004). How physical text layout affects reading from screen. Behaviour and Information Technology. doi:10.1080/01449290410001715714 A critical review of empirical work on screen text layout covers line length, columns, window size and interlinear spacing. It names characters per line the critical variable and asks for work on its interactions.
• Baymard Institute (2022). Readability: the optimal line length. Published 10 May 2022. Baymard Institute research article. https://baymard.com/research-articles/line-length-readability A practitioner review gives 50 to 75 characters for body text and notes the standard's guidance of 80 characters or fewer, translating it into a maximum width of about 70 characters. It reports no new experiment.
• Dillon, A. (1992). Reading from paper versus screens: a critical review of the empirical literature. Ergonomics. doi:10.1080/00140139208967394 The review found the literature comparing paper with screens dominated by attempts to overcome speed deficits caused by poor image quality. It concludes that single-variable explanations cannot capture reading from screens.
• Noyes, J. M. and Garland, K. J. (2008). Computer- versus paper-based tasks: are they equivalent? Ergonomics. doi:10.1080/00140130802170387 The review covers the fifteen years of comparative work since Dillon. It concludes that total equivalence is not achievable and that any remaining difference is specific to the task and the outcome required.
• Rello, L. and Baeza-Yates, R. (2013). Good fonts for dyslexia. Proceedings of the 15th International ACM SIGACCESS Conference on Computers and Accessibility. doi:10.1145/2513383.2513447 Forty-eight people with dyslexia read 12 texts in 12 fonts while their eye movements were tracked. Font type had a significant effect on reading time, with sans serif, monospaced and roman styles beating serif, proportional and italic, so legibility depends on the reader.
• Galliussi, J., Perondi, L., Chia, G., Gerbino, W. and Bernardis, P. (2020). Inter-letter spacing, inter-word spacing, and font with dyslexia-friendly features: testing text readability in people with and without dyslexia. Annals of Dyslexia. doi:10.1007/s11881-020-00194-x Two groups of 64 children, with and without dyslexia, read eight equivalent texts while letterform and spacing varied. Letterform made no difference, and increasing inter-letter spacing without increasing inter-word spacing impaired reading speed, so the remedy that sells loses to spacing.
• Vecino, S., Mehtali, J., de Andres, J., Gonzalez-Rodriguez, M. and Fernandez-Lanvin, D. (2022). How does serif versus sans serif typeface impact the usability of e-commerce websites? PeerJ Computer Science. doi:10.7717/peerj-cs.1139 Two hundred and forty-six volunteers worked on an e-commerce prototype with serif and sans serif versions of one family. No effect on reading speed appeared and only gender predicted preference, so the industry default has no support here.