Principles

The following five principles informed every design decision in The Fair Feedback Project. The principles are presented here not as abstract ideals but as commitments that carry practical consequences throughout the project — shaping what we recommend, what we caution against, and what we decline to promise.

1. Bias in student evaluations is a structural problem, not an individual failing.

The research makes clear that bias in student evaluations of teaching is systemic — rooted in cultural stereotypes, institutional norms, and evaluation instrument design. Individual instructors did not create this problem and cannot be expected to solve it alone.

The Fair Feedback Project provides individual-level strategies because they can be implemented immediately and the evidence suggests they can help. But it does so within an explicit framework that calls for institutional and policy-level reform. Every recommendation in the Instructor Track is accompanied by a clear acknowledgment that individual action is necessary but not sufficient, and the Institutional Track exists precisely to support the structural changes that the research calls for.

We urge institutions to adopt the recommendations of the American Sociological Association's 2019 Statement on Student Evaluations of Teaching, endorsed by nearly two dozen scholarly organizations: use student evaluations as one component of a holistic assessment of teaching, not as a standalone metric; frame evaluation instruments as opportunities for student feedback rather than ratings of teaching effectiveness; and exercise caution in the use of evaluation data in personnel decisions.

2. All recommendations are tied to specific research findings.

Every strategy, every piece of generated language, and every recommendation in The Fair Feedback Project is grounded in peer-reviewed scholarship. Where the evidence is strong, we say so. Where it is mixed, limited, or context-dependent, we say that too. We do not present contested findings as settled, and we do not extrapolate beyond what the data support.

This commitment shapes the project in concrete ways. When the Instructor Track generates anti-bias language for evaluation preambles, it draws on the specific messaging approaches that Peterson et al. (2019), Genetin et al. (2022), and Boring and Philippe (2021) found to be effective — and it steers instructors away from the normative and explicitly gendered framing that Boring and Philippe (2021) and Ayllón and Zamora (2025) found to be ineffective or counterproductive. When the project describes the likely effects of a strategy, it reports the range of findings across studies, including null results such as Owen, De Bruin, and Wu (2025) and variable results such as Mitchem et al. (2025), rather than cherry-picking only the positive outcomes.

Throughout the project, citations are provided so that university instructors, staff, and administrators can consult the original studies and make informed decisions for their own contexts.

3. The Fair Feedback Project is designed as a professional best practice, not an additional burden.

We are acutely aware that the faculty most likely to seek out this project are those most harmed by biased evaluations — women, faculty of color, non-native English speakers, and other marginalized instructors. We have designed the project so that engaging with it feels empowering rather than exhausting, and so that the labor of mitigating bias is framed as a shared professional responsibility rather than the individual burden of those most affected.

This principle has several practical implications. The project is designed for use by all instructors, including those who may benefit from current evaluation practices. Bias mitigation is framed as something a thoughtful professional does — comparable to inclusive syllabus design or accessible course materials — not as a defensive measure taken by the structurally vulnerable. We encourage centers for teaching and learning and faculty leaders to promote The Fair Feedback Project broadly and to incorporate its resources into programming, faculty development workshops, and departmental conversations about evaluation practice, so that engagement with bias mitigation becomes part of the ordinary fabric of teaching rather than an act of individual self-protection.

The project is also designed to respect peoples' time. An instructor should be able to generate useful, evidence-based materials in a single focused session. Depth is available for those who want it — the research base, the nuances of framing, the institutional resources — but it is never required to reach a practical outcome.

4. Transparency about limitations is a feature, not a flaw.

The Fair Feedback Project cannot eliminate bias in student evaluations. We are direct about this because instructors deserve honesty, and because overpromising would ultimately undermine the credibility of the broader reform effort.

Here is what the evidence supports and what it does not. Well-designed messaging interventions can meaningfully reduce bias in quantitative ratings for some instructors in some contexts. Effects vary by discipline, course level, student demographics, institutional culture, and the specific framing of the intervention. There is less evidence that messaging interventions reduce bias in qualitative comments, where abusive and prejudicial language is most prevalent (Owen, De Bruin, & Wu, 2025). Self-affirmation exercises have shown promise, but operate through a different mechanism — reducing inflated ratings for male professors rather than raising ratings for female professors (Hoorens, Dekkers, & Deschrijver, 2021). And no individual-level strategy can compensate for an institutional policy that treats raw evaluation scores as a reliable measure of teaching quality.

We present these limitations not to discourage action but to calibrate expectations. An instructor who uses The Fair Feedback Project and sees modest improvement in their evaluations has achieved something real. An instructor who uses it and sees no change has not failed — they have encountered the limits of individual action against a structural problem, and that experience can itself become fuel for advocacy.

5. Effective messaging requires care and nuance.

The intervention research reveals that the content and framing of anti-bias messaging is critical — and that well-intentioned but poorly designed interventions can do more harm than good. This principle governs the language The Fair Feedback Project generates and recommends.

Three categories of findings inform our approach. First, informational messages work better than normative ones. Messages that describe how bias operates, share research findings, and invite students to reflect on their judgments have shown positive effects (Boring & Philippe, 2021; Genetin et al., 2022; Peterson et al., 2019). Messages that simply instruct students not to discriminate have generally been ineffective (Boring & Philippe, 2021). Second, implicit-bias framing works better than explicit gender-bias framing. A study testing debiasing videos found that a video about implicit bias generally reduced the gender gap in evaluations, but a video explicitly focused on gender bias triggered backlash — male students rated female instructors even lower than the control group (Ayllón & Zamora, 2025). Third, the same message can affect different students differently. An Australian replication found that male students responded to bias messaging by raising ratings of disadvantaged instructors, while female students responded by lowering ratings of advantaged instructors (Kim, Williams, Johnston, & Fan, 2024). The net effect was reduced disparity, but through divergent mechanisms.

The Fair Feedback Project steers instructors toward messaging approaches that the evidence supports and away from approaches the evidence suggests are ineffective or counterproductive. We explain these choices throughout the project so that instructors understand not just what to say, but why — and so that they can make informed adaptations for their own contexts rather than treating any generated language as a script to be followed uncritically.