Do Students Who Skip Class Rate Teaching Differently? Attendance as a Confound in Course Evaluations

In brief: Attendance and course evaluations are entangled in two ways. Students who attend more tend to perform better and feel more positively, which colours the scores they give; and attendance shapes who is even present to complete an in-class evaluation. Burns and Ludlow (2005) found that students' rating of the need to attend a class explained about 5.3% of the variance in perceived teaching excellence after controlling for class size, instructor availability and small-group interaction - evidence that attendance-related perceptions carry independent weight in SET.

What the research says

The anchor study is Shaun M. Burns and Larry H. Ludlow (2005), "Understanding student evaluations of teaching quality: The contributions of class attendance," published in the Journal of Personnel Evaluation in Education (18(2), 127-138; the journal is now Educational Assessment, Evaluation and Accountability). They examined how students' perceptions of class attendance relate to their evaluations of teaching. Their central result: the perceived need to attend class was a significant predictor of rated teaching excellence, uniquely accounting for roughly 5.3% of the variance even after adjusting for class size, instructor availability and the extent of small-group interaction. In other words, when students feel that attending matters - that the sessions add value beyond the readings or slides - they rate the teaching more highly, and this association is not merely a by-product of class size or contact structure.

Two strands of corroborating evidence make the confound concrete. First, attendance is a robust correlate of achievement: students who attend more tend to earn better grades, and better grades are one of the strongest predictors of the evaluation scores students give. So attendance can influence SET indirectly through performance and satisfaction. Second, and often overlooked, is who responds. When evaluations are collected in class, absentees are structurally excluded, producing a self-selected respondent pool skewed toward more engaged students; the resulting scores may not represent the whole cohort. This is the same non-response and selection dynamic documented for response-rate and self-selection effects in SET.

The confound is also bidirectional, which is what makes it interesting. Good teaching plausibly causes attendance (students come because the sessions are worth attending), so part of the attendance-rating link is a legitimate teaching signal. But attendance is also driven by factors that have nothing to do with the instructor - timetable slot, commuting distance, competing deadlines, part-time work, prior interest and student conscientiousness - and those construct-irrelevant drivers then feed back into the rating. Disentangling the two is the analytic challenge. Broader work on how student, teacher and course characteristics jointly shape SET (for example, the "grading leniency to the edge" analyses in the same journal tradition) reinforces that raw scores blend genuine teaching effects with student-composition effects.

Why it matters for course evaluation in practice

For quality assurance the attendance confound creates several distinct risks:

1. Respondent bias in in-class collection. If you gather evaluations in a live session, you sample the students who showed up. A course with erratic attendance may look better than it is because its most disengaged students never filled in the form - or worse than it is if only aggrieved regulars respond. The mean is conditioned on attendance.
2. Attribution errors. A low score in a poorly attended course may reflect a timetable or logistics problem (why students stopped coming) rather than teaching. Conversely, mandatory-attendance courses concentrate a captive, sometimes resentful audience whose ratings reflect coercion as much as quality.
3. Comparability across delivery models. Flipped, seminar and lecture formats create very different "need to attend" perceptions. Comparing their raw SET means without accounting for this is comparing apples and oranges - the Burns and Ludlow result suggests the need-to-attend perception alone can move scores by several percentage points of variance.
4. Feedback that hides its own gap. Because absentees are exactly the students whose experience most needs surfacing, attendance-conditioned evaluations can systematically miss the signal a programme most needs to act on.

The remedy is to know your respondents' attendance context and to collect in ways that do not silently exclude the disengaged.

Limitations and honest caveats

• Correlational core. Burns and Ludlow (2005) is a variance-partitioning study, not a randomised design. A 5.3% unique contribution is meaningful but modest, and it establishes association, not a clean causal effect of attendance on ratings.
• "Need to attend" is a perception, not attendance. The study measured students' rated need to attend rather than logged attendance, so it speaks to how attendance-related beliefs relate to evaluations - a subtly different construct from raw presence.
• Endogeneity in both directions. Because good teaching can cause attendance and attendance can cause better ratings, part of the association is a valid teaching signal you would not want to "correct" away. Naive adjustment for attendance risks removing real quality differences.
• Vintage and setting. The anchor study is two decades old and single-context; delivery has since shifted toward blended and recorded formats that change the meaning of attendance entirely. The confound persists, but its shape has moved.
• Confounder tangle. Attendance is correlated with prior ability, interest, timetable slot and conscientiousness, so isolating its independent role is genuinely hard and no single study settles it.

Honest framing: attendance is a real and multi-channel confound in SET - part valid signal, part construct-irrelevant selection - and its biggest practical bite is in who responds, which institutions can control directly.

How Koji incorporates this

Koji is an AI-native course-evaluation platform, and it tackles the attendance confound mainly at the collection and interpretation layers.

Reach absentees, not just the room. Because Koji evaluations are conducted through asynchronous, link-based AI-moderated interviews rather than a show-of-hands in the last ten minutes of a lecture, students who were not physically present can still be invited to respond. That directly counters the in-class self-selection that conditions scores on attendance, widening the respondent base toward the disengaged students whose feedback matters most. Response-rate and non-response context can be recorded and reported so a reader knows how representative the sample is.

Capture attendance context as a covariate. Attendance frequency, delivery format and perceived need to attend can be collected as structured scale, single_choice or yes_no items, letting QA officers segment - comparing regular attenders with occasional ones, or flipped courses with lecture courses - rather than pooling scores that differ systematically in who is behind them.

Probe the reason for disengagement. Koji's conversational moderator can gently ask a low-attendance respondent why they stopped attending, and its automatic thematic analysis separates construct-irrelevant reasons (timetable clashes, commuting, recorded lectures made attendance optional) from teaching-related ones (sessions added little beyond the slides). That distinction is precisely what turns an ambiguous low score into an actionable - and correctly routed - finding, without overclaiming that Koji removes the confound.

Koji surfaces these patterns for human judgement and closing-the-loop action tracking rather than auto-adjusting anyone's score. The same AI-moderated interview engine powers Koji's core research platform at koji.so for product and customer research, where reaching non-responders is just as decisive for a representative result.

A practical protocol for quality-assurance teams

Because the attendance confound bites hardest through who responds, most of the remedy is in collection design:

1. Do not collect only in class. In-class administration structurally excludes absentees. Offer an asynchronous window so students who missed a session can still contribute, and record the response rate so readers can judge representativeness.
2. Capture attendance context. Ask a simple, non-judgemental attendance-frequency item and record delivery format. This lets you segment regular from occasional attenders instead of pooling them.
3. Probe disengagement, do not just count it. A low score from a rarely-attending student is only actionable if you know why they stopped coming. Distinguish logistics reasons (timetable, commuting, recorded lectures) from teaching reasons before drawing conclusions.
4. Resist adjusting scores for attendance. Because good teaching can cause attendance, correcting for it can erase a real quality signal. Segment and interpret transparently instead.

The recorded-lecture era changes the confound, not its existence

When Burns and Ludlow published in 2005, "attendance" meant physical presence at a scheduled session. Today, recorded and blended delivery means a student can be highly engaged with a module while attending few live sessions - or can drift away precisely because recordings make attendance feel optional. This decouples attendance from engagement in ways the original study could not anticipate, and it makes in-class evaluation collection even more skewed, because the live audience is an ever-smaller and more self-selected slice of the cohort. For European institutions expanding flexible and hybrid provision, the practical lesson is that representativeness now depends less on who shows up to a room and more on how deliberately the evaluation reaches the whole enrolled cohort. Attendance remains a genuine confound; the modern task is to stop it silently determining your respondent pool.

Related resources

• /docs/selection-bias-course-evaluations
• /docs/online-vs-paper-course-evaluation-response-rates
• /docs/raising-course-evaluation-response-rates-incentives-evidence
• /docs/grading-leniency-student-evaluations-teaching
• /docs/student-as-consumer-effect-course-evaluations-academic-performance
• /docs/interpreting-reporting-student-ratings-responsibly

References

• Burns, S. M., & Ludlow, L. H. (2005). Understanding student evaluations of teaching quality: The contributions of class attendance. Journal of Personnel Evaluation in Education (now Educational Assessment, Evaluation and Accountability), 18(2), 127-138. https://doi.org/10.1007/s11092-006-9002-7
• Spooren, P., Brockx, B., & Mortelmans, D. (2013). On the validity of student evaluation of teaching: The state of the art. Review of Educational Research, 83(4), 598-642. https://doi.org/10.3102/0034654313496870
• Uttl, B., White, C. A., & Gonzalez, D. W. (2017). Meta-analysis of faculty's teaching effectiveness: Student evaluation of teaching ratings and student learning are not related. Studies in Educational Evaluation, 54, 22-42. https://doi.org/10.1016/j.stueduc.2016.08.007