Negativity Bias: Why a Few Harsh Comments Outweigh Dozens of Kind Ones

The short answer: Negativity bias — the robust finding that "bad is stronger than good" — means a small number of harsh student comments exert disproportionate influence on how faculty experience evaluations and how committees interpret them. This is a feature of human cognition, not a flaw in particular readers, and it has two costs: it harms faculty wellbeing, and it distorts decision-making by letting outliers dominate the signal. The fix is not to hide negative feedback but to structure how feedback is collected and read so that the representative weight of comments is visible, rather than whichever comment stings most. Below: the evidence, the mechanism, and the practical correction.

The cognitive law behind the sting

In 2001, Roy Baumeister and colleagues published a landmark review with a deliberately blunt title: "Bad Is Stronger Than Good." Across everyday events, major life events, close relationships, learning, and impression formation, they found the same asymmetry — negative information is processed more thoroughly, weighted more heavily, and remembered longer than equivalent positive information (Baumeister, Bratslavsky, Finkenauer & Vohs, 2001, Review of General Psychology). Bad impressions form faster and resist disconfirmation; bad feedback has more impact than good. The same year, Rozin and Royzman formalised the related principle of negativity dominance: when positive and negative elements are combined, the resulting whole is more negative than the sum of its parts.

This is why a course evaluation containing thirty-eight warm comments and two contemptuous ones does not feel like a 95% positive result. It feels like the two. The instructor is not being thin-skinned; they are being human in exactly the way the literature predicts.

What it does to course evaluation

Negativity bias quietly corrupts the evaluation process in two distinct ways.

It harms the people being evaluated. A single abusive or personal comment can overshadow an entire dataset and inflict real distress — an effect we examine in our desk research on abusive, non-constructive student comments and faculty wellbeing. Because bad is stronger than good, the harm of one cruel comment is not offset by twenty kind ones. Over time this is corrosive: it teaches faculty to fear, resent, and distrust the entire evaluation system, which is a large part of why faculty distrust course evaluations in the first place.

It distorts interpretation. When a head of department or promotion committee skims a set of comments, negativity bias means the harshest lines are the ones that anchor the reading — even when they are statistical outliers unrepresentative of the cohort. A reviewer can come away with a "concerning" impression of a course that the great majority of students experienced positively. This is the qualitative cousin of a problem we have flagged on the quantitative side: treating a small mean difference as meaningful. Here, a small number of extreme comments is treated as the headline.

"But shouldn't we take negative feedback seriously?"

This is the strongest objection, and it is right — up to a point. Negative feedback is often the most actionable: students who are struggling tell you precisely what to fix, and a culture that flinches from criticism learns nothing. Nothing here argues for suppressing, softening, or averaging away critical comments. A genuine pattern of complaints about, say, unclear assessment criteria is signal, and ignoring it would be its own failure.

The distinction that matters is representativeness, not valence. The problem is not that negative comments are read; it is that a handful of them are read as though they were the whole. The corrective is to make the weight of each theme visible — how many students raised it, in what proportion, with what intensity — so that an outlier reads as an outlier and a genuine pattern reads as a pattern. You take negative feedback seriously precisely by seeing it in proportion, not by letting the most vivid example stand in for the cohort.

Reading feedback against the bias

Several practical disciplines help committees and faculty counteract the asymmetry:

1. Quantify themes before reading verbatims. Knowing that "pace" was raised by 4 of 60 students before you read the four sharply worded comments about it reframes them as a minority view, not a verdict.
2. Separate constructive criticism from abuse. Comments that identify a fixable problem are data; comments that attack a person's appearance or identity are not feedback and should be filtered out of the evidence entirely, not weighed.
3. Report proportions, not anecdotes. Institutional summaries should lead with how widely a concern was shared, so a reader's negativity bias has a denominator to work against.
4. Give faculty the positive signal too. Because good is weaker than bad, positive themes need to be surfaced deliberately or they will be psychologically discounted.

How Koji helps read feedback in proportion

Koji for Education is designed around exactly this representativeness problem. Its automatic thematic analysis groups open-text feedback into themes and shows how many students raised each one, so a concern voiced by three respondents is never mistaken for the consensus of three hundred — every theme is tied back to the verbatim comments that produced it, but framed by its prevalence. Quality scoring helps separate substantive, constructive feedback from low-content or abusive remarks, so the latter do not silently anchor a reading. And because Koji's AI moderation is standardised and bias-aware, the same calibrated approach is applied to every response rather than depending on which comment a tired reviewer happens to land on. To be precise about the claim: Koji mitigates and surfaces the distortion negativity bias causes; it does not abolish a cognitive law wired into human attention. What it can do is give readers the denominator — the proportion and pattern — that turns a stinging outlier back into one voice among many. The same thematic-analysis engine powers the main Koji platform for customer and user research, where a few loud detractors create the identical reading problem.

Supporting the faculty who read their own feedback

Negativity bias is not only an institutional reporting problem; it is a personal one for every academic who opens their evaluations alone, at the end of a long term, and reads the worst line first. Teaching-and-learning centres can do a great deal to blunt the harm without hiding anything.

The most effective intervention is mediated feedback: rather than handing raw comments to staff unfiltered, an educational developer or peer mentor helps the lecturer read them in context — establishing the proportion of positive to critical themes up front, normalising that one or two harsh comments appear in almost every dataset, and turning genuinely constructive criticism into a concrete development plan. This reframes evaluations from a verdict to be survived into a resource to be used. Evidence on formative, consultative feedback shows it is precisely this interpretive support — not the raw scores — that drives teaching improvement.

Three further practices help. Time the release so staff are not reading evaluations in isolation during an exhausted, low-resource moment. Separate development from judgement: feedback intended to help someone teach better should not arrive through the same channel, or at the same time, as feedback feeding a promotion decision, or the negativity bias compounds with career anxiety. And train the readers — heads of department and committee members who interpret evaluations should be explicitly briefed that a handful of extreme comments are not a representative signal, the same discipline we ask of anyone interpreting student ratings responsibly. The goal throughout is not to shield academics from criticism, which would waste the most useful feedback they receive, but to ensure the criticism is read at its true weight rather than its emotional weight.

The bottom line

"Bad is stronger than good" is one of the most reliable findings in psychology, and course evaluation is a near-perfect trap for it: emotionally charged feedback, read quickly, by people with a stake in the result. The answer is not to protect faculty from criticism or to discount what students say — it is to read feedback in proportion. Quantify how widely each theme is shared, strip out abuse before it anchors anyone, and surface the positive signal that the mind discounts by default. Negativity bias cannot be switched off. But with the right structure around how feedback is collected and presented, it can be stopped from turning two comments into the whole story.