Individual accountability inside a shared mark: social loafing, peer assessment and the proxy problem

A shared mark records that a team finished; it cannot record who did the work. This is the seventh of ten units, and it turns from the team's practice to the mark the practice produces. The evidence that contribution varies inside a group is strong. Individual effort falls as a group grows, and the balance of member contributions is one of six facets of teamwork quality that predict project success. The evidence that a better number can close the gap is much weaker. Every instrument a team reaches for counts a proxy: commits, attendance, lines changed, peer ratings. A commit count rewards volume and says nothing about the member who writes the documentation, reviews a change or runs the client conversation. Peer rating is the more promising proxy, though it is weakest on the multi-dimensional judgement a contribution mark needs. The defensible design is not a single fairness score. It uses evidence from more than one source, read against criteria set in advance, with a route by which a member can challenge the reading.

Social loafing is real, and a small group is where it hides

The founding demonstration is small and physical. Latane, Williams and Harkins (1979) had participants clap and shout alone and then in groups, and separated two possible reasons for any drop in output. Individual effort fell as the group grew, and fell by more than faulty coordination of the group alone could explain, which left a loss of individual motivation. They named the effect social loafing: people work less hard when their effort is pooled with other people's.

Karau and Williams (1993) meta-analysed 78 studies and found loafing robust across tasks and populations. Evaluation potential, expectations of co-workers' performance, task meaningfulness and culture moderated it. Identifiability is the part of that mechanism a team can act on: a member whose contribution is invisible has been permitted to slacken. The effects are moderate and the laboratory tasks far simpler than a term-long build.

Balance of contribution is a facet of teamwork quality, not an add-on

Loafing matters because it is not marginal to performance. Hoegl and Gemuenden (2001) built a six-facet model of teamwork quality: communication, coordination, balance of member contributions, mutual support, effort and cohesion. They tested it on data from 575 members, leaders and managers of 145 German software teams. Teamwork quality was associated with performance and with members' satisfaction and learning, and balance of member contributions is one facet among six, not an afterthought.

Their result carries a warning: the strength of the association varied with who rated the performance, team members, team leaders or team-external managers. A contribution mark assembled from those ratings inherits that disagreement rather than resolving it.

The instruments were built to measure behaviour, not volume

The peer-rating tradition worked around the proxy problem rather than falling into it. Loughry, Ohland and Moore (2007) built the Comprehensive Assessment of Team Member Effectiveness from the teamwork literature. They tested candidate items on two student surveys of 2,777 and 1,157 respondents, and refined them by factor analysis. The final instrument has 87 items measuring 29 kinds of contribution in five categories. They are contributing to the team's work, interacting with teammates, keeping the team on track, expecting quality, and having relevant knowledge, skills and abilities. Each item is behaviourally anchored, so a rater judges described behaviour rather than a tally.

Kaufman, Felder and Fuller (2000) show the mechanism in a course. There, confidential self- and peer ratings in two sophomore classes became weighting factors on the team grade. The factor moves a mark, but the process rewards responsibility, not output.

The proxy problem: what a commit count actually rewards

If the validated instruments judge behaviour, the instruments a team reaches for first judge something else. A commit count, an attendance register or a line-change total is easy to collect, and each rewards what is visible in the repository or the room. The work that keeps a project alive, writing the documentation, reviewing a teammate's code, talking to the client, is exactly the work that leaves few commits. Visibility is not effort: a member whose hours are cut by illness or caring responsibilities shows fewer commits.

Three findings explain why the resulting numbers mislead. Peer marks are valid only in part: Falchikov and Goldfinch (2000) meta-analysed 48 studies and found peer and teacher marks align more closely for global judgements on well-understood criteria than when several dimensions are marked at once. Topping (1998) found among 31 reliability studies that 18 reported peer assessment adequate and seven unacceptably low. Second, raters cannot always see what they rate. Freeman and Greenacre (2010) found that actively contributing students had difficulty distinguishing teammates who were socially loafing from those genuinely struggling. Third, the instrument may measure a different thing: Britton and colleagues (2015) built a tool to assess an individual's teamwork skill rather than a share of contribution.

The proxy measures something real but not the thing the mark prices, and it measures it through a rater who may not be able to see the difference.

The proxy problem now has a machine version. Usher (2025) had 76 undergraduates' group projects graded by an AI chatbot, by peers and by the instructor. The chatbot marked higher than either human while peer and instructor marks aligned closely, and its written feedback was rated higher in quality but was occasionally irrelevant or contradictory. The fourth rater tracks the expert's mark least closely.

What actually reduces loafing on a project

Aggarwal and O'Brien (2008) found that loafing rose with the project's scope and the group's size, while multiple peer evaluations reduced its incidence. Bacon, Stewart and Silver (1999) found that clarity of instructions, team longevity and self-selection of teammates most influenced whether a student had a good or a miserable team experience. Team size and the grade share attached to team performance did not separate the best from the worst. Strong and Anderson (1990) reviewed the causes of free-riding and set out fifteen recommendations for reducing it, with preliminary data on how students judged six of the techniques.

The through-line is early structure and agreement rather than surveillance. The controls students rated effective were largely ones they had a hand in setting, which seeds the challenge route this unit defends.

Contested question: is fairness a mark that agrees with the tutor or a process a member can contest?

Whether a shared mark can be fair to an individual is the contested question, and two positions answer it differently. The calibration position holds that fairness is a measurement problem. A well-built instrument approximates the judgement an expert would have made. The goal is a single adjusted number that agrees with a tutor, with less measurement error (Falchikov and Goldfinch, 2000; Loughry et al., 2007).

The recognition position holds that fairness is recognisability. What students want is that their particular contribution be seen for what it was, which a number need not deliver even when it agrees with the tutor. Rasooli and colleagues (2025) studied fairness in group-work assessment across three large UK universities. They found the majority of students concerned with individual recognition, and the solutions they proposed were mostly instructor-focused.

the majority of students were concerned with individual recognition in their evaluation of fairness in group work assessments (Rasooli et al., 2025)

What would settle it is a study that holds the team, the task and the deadline constant and varies only how the mark is produced. One arm would use a single weighted number, the other a portfolio of evidence with a challenge route. Both would face two criteria fixed in advance: the agreement of the mark with an expert's independent judgement, and the share of students who felt recognised. Nobody has run it at student-team scale. The calibration case holds a meta-analysis of 48 studies and a factor-analysed peer instrument (Falchikov and Goldfinch, 2000; Loughry et al., 2007); the recognition case holds one multi-university study of student views together with the finding that seven of 31 reliability studies rated peer assessment unacceptable (Rasooli et al., 2025; Topping, 1998). While the dispute is open, publish the criteria and the evidence to be collected before the work begins, and treat a challenge as information about the instrument. The strongest objection is that the preferred process has no evidence behind it either, and a challenge route can raise the cost of marking without moving the mark.

Where the argument is heading

The direction of travel is away from the single score and towards evidence plus agency. Rasooli and colleagues (2025) show the debate moving to recognition and to the values group assessment is built on. Britton and colleagues (2015) point the same way from the measurement side, aiming at a tool that develops teamwork skill rather than a share of a mark. For a student team the reading is stable. Keep more than one source of evidence, set the criteria before the work begins, and offer a real route to challenge the reading. What the literature does not supply is the study showing a challengeable evidence base beating a single weighted number.

The sources this unit rests on

• Latane, B., Williams, K. and Harkins, S. (1979). Many hands make light the work: the causes and consequences of social loafing. Journal of Personality and Social Psychology, 37(6), 822-832. doi:10.1037/0022-3514.37.6.822 Had participants clap and shout alone and in groups; individual effort fell as the group grew, by more than coordination loss alone explained. It named social loafing.
• Karau, S. J. and Williams, K. D. (1993). Social loafing: a meta-analytic review and theoretical integration. Journal of Personality and Social Psychology, 65(4), 681-706. doi:10.1037/0022-3514.65.4.681 Meta-analysed 78 studies; loafing held across tasks and populations, moderated by evaluation potential, expectations of co-workers' performance, task meaningfulness and culture. It establishes the effect and makes identifiability the lever a team can pull.
• Hoegl, M. and Gemuenden, H. G. (2001). Teamwork quality and the success of innovative projects. Organization Science, 12(4), 435-449. doi:10.1287/orsc.12.4.435.10635 Tested a six-facet model of teamwork quality, balance of member contributions among them, on 575 members, leaders and managers of 145 German software teams. Teamwork quality predicted performance and satisfaction, but the strength varied with who rated it.
• Loughry, M. L., Ohland, M. W. and Moore, D. D. (2007). Development of a theory-based assessment of team member effectiveness. Educational and Psychological Measurement, 67(3), 505-524. doi:10.1177/0013164406292085 Built the Comprehensive Assessment of Team Member Effectiveness from the teamwork literature, testing items on two student surveys and refining them by factor analysis. Its 87 items cover 29 contribution types in five categories, all behavioural rather than counts.
• Kaufman, D. B., Felder, R. M. and Fuller, H. (2000). Accounting for individual effort in cooperative learning teams. Journal of Engineering Education, 89(2), 133-140. doi:10.1002/j.2168-9830.2000.tb00507.x Reports an autorating scheme in two sophomore engineering courses, where confidential self- and peer ratings became weighting factors multiplied into the team grade. It rates responsibility rather than output.
• Falchikov, N. and Goldfinch, J. (2000). Student peer assessment in higher education: a meta-analysis comparing peer and teacher marks. Review of Educational Research, 70(3), 287-322. doi:10.3102/00346543070003287 Meta-analysed 48 studies comparing peer and teacher marks; the two align more closely for global judgements on well-understood criteria than for marking several dimensions at once. Peer marks are valid in aggregate and weakest exactly where a contribution mark needs them.
• Topping, K. (1998). Peer assessment between students in colleges and universities. Review of Educational Research, 68(3), 249-276. doi:10.3102/00346543068003249 Reviewed 31 studies of peer assessment's reliability and validity; 18 reported it adequate across many applications and seven unacceptably low. Most compared marks or grades rather than the formative feedback a contribution mark needs.
• Freeman, L. and Greenacre, L. (2010). An examination of socially destructive behaviours in group work. Journal of Marketing Education, 33(1), 5-17. doi:10.1177/0273475310389150 Found that actively contributing students had difficulty distinguishing teammates who were socially loafing from those genuinely struggling with the material. A peer rating cannot see the difference the mark must price.
• Britton, E., Simper, N., Leger, A. and Stephenson, J. (2015). Assessing teamwork in undergraduate education: a measurement tool to evaluate individual teamwork skills. Assessment and Evaluation in Higher Education, 42(3), 378-397. doi:10.1080/02602938.2015.1116497 Developed and tested a tool assessing an individual's teamwork skills, aimed at formative development rather than at a share of contribution. It assesses the skill where the rest of the literature apportions the mark.
• Aggarwal, P. and O'Brien, C. L. (2008). Social loafing on group projects: structural antecedents and effect on student satisfaction. Journal of Marketing Education, 30(3), 255-264. doi:10.1177/0273475308322283 Found that loafing rose with the project's scope and the group's size, while multiple peer evaluations reduced its incidence. It locates the remedies in course design rather than in individual diligence.
• Bacon, D. R., Stewart, K. A. and Silver, W. S. (1999). Lessons from the best and worst student team experiences. Journal of Management Education, 23(5), 467-488. doi:10.1177/105256299902300503 Found that clarity of instructions, team longevity and self-selection of teammates most influenced whether a student had a good or a miserable team experience. Team size and the grade share attached to team performance did not separate the best from the worst.
• Strong, J. T. and Anderson, R. E. (1990). Free-riding in group projects: control mechanisms and preliminary data. Journal of Marketing Education, 12(2), 61-67. doi:10.1177/027347539001200208 Reviewed the causes of free-riding and set out fifteen recommendations for reducing it, with preliminary data on student perceptions of six techniques. It is the origin of the control mechanisms the unit's evidence requirement answers.
• Rasooli, A., Pitt, E., Turner, J., Varga-Atkins, T., Moindrot, W., Preston, S. and Asgari, S. (2025). Understanding students' perceptions of fairness in group work assessment. Assessment and Evaluation in Higher Education, 50(7), 1012-1024. doi:10.1080/02602938.2025.2502587 Studied fairness in group-work assessment across three large UK universities; the majority of students were concerned with individual recognition, with solutions that were mostly instructor-focused. It is the recent student-side evidence that makes the unit's thesis arguable and points to where the area is heading.
• Usher, M. (2025). Generative AI versus instructor versus peer assessments. Assessment and Evaluation in Higher Education, 50(6), 912-927. doi:10.1080/02602938.2025.2487495 Compares grades and written feedback from an AI chatbot, peers and the instructor on 76 undergraduates' group projects. Chatbot marks ran higher than either human while peer and instructor marks aligned closely. It adds a model to the proxy problem; its limit is one course and one tool.