Accent bias distorts student evaluations even when learning is unchanged
Updated Jul 15, 2026
Students can understand a lecture perfectly well and still rate the lecturer less favourably because of how they speak. That is the uncomfortable message of Esther Sosa-Herrera, Kat Silaj, Mary J. Keushkerian and Melissa Paquette-Smith's Assessment & Evaluation in Higher Education paper, "Evaluating the effectiveness of an awareness-based intervention on reducing accent bias in students' evaluations of instructors". For universities using student voice to review teaching, support staff development, or make quality decisions, that matters because a teaching evaluation score can look neutral even when it is not.
Context and research question
Student evaluations of teaching, or SETs, are often used as if they are straightforward indicators of instructional quality. Yet this sits uneasily with wider evidence that student evaluations of teaching are only moderately stable even before bias is considered. Accent bias adds another layer of risk. If students penalise instructors for sounding unfamiliar or non-native, universities can end up confusing bias with evidence about teaching quality.
This paper asks a practical question: can a brief awareness-based intervention reduce accent-based bias in students' evaluations of instructors? The authors tested that question in two controlled laboratory experiments with undergraduate students at UCLA. In both experiments, participants were randomly assigned to watch a short lecture narrated either by an instructor speaking English with a Mandarin accent or by an instructor speaking with an American accent. Students then completed a multiple-choice assessment of learning and an evaluation form. In Experiment 1, the final analysed sample was 100 students. In Experiment 2, the final analysed sample was 240 students, using different lecture materials and a more diverse participant group.
Key findings
The most important finding is that learning and evaluation did not move together. In both experiments, the multiple-choice assessments showed no statistically significant differences in learning by instructor accent. Students understood the lecture equally well, but still rated the Mandarin-accented instructor and lecture less favourably. For UK universities, that is the core warning: lower evaluation scores do not automatically mean weaker teaching.
The paper states the point plainly:
"participants in both experiments showed evidence of bias"
Bias was strongest in overall judgements, not only in specific teaching behaviours. In Experiment 1, students in the no-intervention condition rated the American-accented instructor and lecture more highly overall than the Mandarin-accented equivalent, despite seeing the same content. The intervention then narrowed that gap. In practical terms, this suggests that global questions such as "overall rating of the instructor" may be especially vulnerable to bias because they invite a broad impression rather than a tightly defined judgement.
The intervention showed promise, but not consistency. In Experiment 1, a short narrated prompt reminding students not to let bias shape their evaluation reduced the disparity between the two instructors on the overall items. That matters because it suggests a small design change can influence how students score teaching. But the result did not fully hold when the authors tested it again under tougher conditions.
Experiment 2 is what makes the paper especially useful for institutional teams. The authors repeated the design with a different lecture on attachment styles and a larger, more linguistically diverse sample. Accent bias still appeared, with a significant main effect of accent on overall instructor ratings. This time, however, the intervention did not significantly reduce the gap. The communication item showed the largest bias, and the intervention did not remove it. That is a strong reminder that awareness prompts are not a complete fix.
The broader conclusion is that brief anti-bias messaging may help under ideal conditions, but it is not enough on its own. The discussion suggests several reasons why the effect weakened in the second study, including the possibility that harder or less engaging material makes students more likely to misattribute difficulty to the instructor's accent. The paper also notes that accent can trigger overlapping assumptions about race, age, or gender. For universities, the implication is clear: evaluation bias is not a one-variable problem.
Practical implications
The first implication for UK higher education teams is to stop treating SET scores as self-explanatory evidence of teaching quality. If module evaluations are used in review, promotion, probation, or enhancement discussions, institutions should ask whether lower scores could partly reflect accent bias rather than weaker teaching. That leads to fairer interpretation and reduces the risk of building people processes on distorted evidence.
Second, universities should treat survey design and sample quality as part of the same governance problem. A short awareness prompt may be worth piloting, especially on overall evaluation items, but the paper shows that one sentence will not remove bias reliably. Teams should also check who is responding, because bias in judgements can combine with non-response bias in evaluations to distort the picture further. That gives institutions a more defensible basis for acting on the results.
Third, institutions should separate global impressions from specific teaching evidence. Questions about communication, organisation, clarity, and value should not simply be collapsed into one overall story about a lecturer. This is where open-text analysis helps. Student Voice Analytics fits naturally here because it can group comments on clarity, fairness, communication, and support at scale, helping teams see whether students are describing a genuine teaching issue or reproducing a biased impression.
Finally, universities should make bias review part of normal comment-analysis governance rather than a one-off EDI concern. If free-text comments and module-evaluation scores feed into formal decisions, teams need a repeatable method for checking bias risks, documenting limitations, and triangulating with peer review or other evidence. A clear student comment analysis governance checklist helps institutions do that consistently. The benefit is simple: more credible student feedback, and fairer use of it.
FAQ
Q: How should a university test whether accent bias is affecting its own module evaluations?
A: Start with like-for-like comparisons. Check whether particular staff groups are receiving systematically lower overall ratings despite similar modules, learning outcomes, and peer-review evidence. Then look separately at overall items, specific items such as communication or organisation, and open comments. If the strongest differences sit in broad impression-based questions rather than in concrete teaching issues, that is a sign the evaluation process may be picking up bias as well as experience.
Q: Does this paper show that awareness prompts should be added to every teaching evaluation form?
A: Not automatically. The evidence is mixed. In Experiment 1, the prompt helped reduce the gap in overall ratings. In Experiment 2, using different lecture content and a larger, more diverse sample, the effect was no longer statistically significant. The sensible response is to pilot a prompt locally, assess the impact carefully, and treat it as one safeguard among several rather than a complete solution.
Q: What does this change about student voice practice more broadly?
A: It strengthens the case for more careful interpretation, not less listening. Student feedback still matters, but universities need to distinguish between what students experienced and how bias may shape the way they score or describe that experience. Student voice becomes more useful when it is analysed with clear governance, compared alongside other evidence, and used to surface patterns rather than to confirm first impressions.
References
[Paper Source]: Esther Sosa-Herrera, Kat Silaj, Mary J. Keushkerian, Melissa Paquette-Smith "Evaluating the effectiveness of an awareness-based intervention on reducing accent bias in students' evaluations of instructors" DOI: 10.1080/02602938.2026.2691496
Request a walkthrough
Book a free Student Voice Analytics demo
See all-comment coverage, sector benchmarks, and reporting designed for OfS quality and NSS requirements.
- All-comment coverage with HE-tuned taxonomy and sentiment.
- Versioned outputs with TEF-ready reporting.
- Benchmarks and BI-ready exports for boards and Senate.
UK-hosted · No public LLM APIs · Same-day turnaround