Negativity Bias: Why One Bad Review Outweighs Nine Good Ones

Negative and positive information of the same magnitude do not get the same treatment. Rozin and Royzman separated four features of the pattern in 2001: negative events have greater potency at the same objective size, their impact grows more steeply as they get closer, a negative element dominates the whole when the two are mixed, and negative states are described in a richer vocabulary than positive ones. The third of those, negativity dominance, is the one the demo above is about.

Attention follows the same asymmetry: negative items are noticed faster and hold attention longer, so they arrive at a judgment with more weight than their share of the evidence. Memory is less settled. Baumeister and colleagues report a negative advantage in particular tasks, while Rozin and Royzman treat the direction of recall as an open question and read much of the evidence as favoring positive material, so whether bad is remembered better depends on the material and the task. In impression formation this shows up as the negativity effect: a single unfavorable trait is more diagnostic than a favorable one, partly because good behavior is common and expected, so bad behavior carries more information about who someone is.

The asymmetry has an obvious use. A missed opportunity can be met again; a missed threat may not be survivable, so a system that overweighs bad news pays a small cost often and avoids a large cost occasionally. That makes negativity bias a reasonable default rather than a defect, which is worth remembering before treating a low score as a mistake.

Two limits are worth stating. The asymmetry is not a blanket rule: Rozin and Royzman point out, following work by Cacioppo and colleagues, that at low intensities the absolute value of positive affect is the higher of the two, a positivity offset, even though the slope is steeper on the negative side. And "bad is stronger than good" sits on top of a broad literature rather than a single experiment, which means the direction is well supported while the size depends heavily on the task. Being alert to bad news is also not the same as low mood, and nothing on this page is a measure of either.

The two papers that gave negativity bias its modern form are both reviews, published in 2001 within months of each other. Paul Rozin and Edward Royzman's "Negativity bias, negativity dominance, and contagion" in Personality and Social Psychology Review set out the four features above and argued that negative entities also spread more easily than positive ones: a drop of something unpleasant ruins a container of something good, and the reverse does not hold. Roy Baumeister, Ellen Bratslavsky, Catrin Finkenauer and Kathleen Vohs published "Bad is stronger than good" in the Review of General Psychology, surveying findings from close relationships, impression formation, learning, emotion and health, and arguing that the pattern was general enough to count as a basic principle.

The experimental work they drew on is older, and the account here follows the way Baumeister and colleagues report it. Norman Anderson was testing a model of how people combine information in 1965, not looking for a negativity effect, when he gave participants sets of trait adjectives with favorability values measured in advance, carefully chosen so that the positive and negative traits were equally far from the neutral point. His participants rated their overall impression of the person described. Impressions built from negative traits departed further from the midpoint than impressions built from positive traits of matching strength, and when the sets mixed the two, the unfavorable traits lowered the overall rating by more than an adding or averaging rule would predict. The favorable traits showed no equivalent extra pull. Anderson's conclusion, which ran against the model he was arguing for, was that unfavorable adjectives are more powerful in forming impressions.

Later studies in the same vein reported the same direction with different materials, including work by Hodges in 1974 and by Hamilton and Zanna in 1972, which also found that raters were more confident in a bad impression than in a good one. The reviews of 2001 are the places where those scattered results were put together.

The pattern appears wherever positive and negative evidence are combined into one verdict. The rows below are illustrations, not findings from the studies cited on this page.

The bias is in the weighting, so the counters are all ways of forcing the weights into the open.

• Count before you judge. Nine positive lines out of ten is a number; the feeling after reading them is not. Write the count down first.
• Score the parts separately, then the whole. Rating each dimension before forming an overall verdict stops one bad dimension from setting the total.
• Ask what the worst item represents. One complaint in a thousand orders and one in ten orders look identical when you read only the complaint.
• Look for the matched comparison. If a negative item moved you three points, ask what a positive item of the same size would have been worth, and whether you would give it that much.
• Put a floor under the good news. Logs, records and three-good-things style lists keep a balanced sample of the evidence in view, so the verdict is read from the record rather than from whatever comes to mind.
• Use the bias on purpose where it earns its keep. For safety, security and anything with an irreversible downside, overweighing bad news is the right policy, and the question is only whether this decision is one of those.

Negativity bias is about the weight of information, which separates it from several neighbors that sound similar.

• It is not pessimism bias, which is an expectation about the future, specifically the tendency to overestimate how likely bad outcomes are. Negativity bias is about how evidence already in hand is weighted, and it operates on judgments about the past and present as readily as about the future.
• It is not attentional bias, the tendency for some cues to capture attention at the expense of others. Attention is one of the stages at which negativity bias operates, but the weighting persists even when both items have been attended to equally.
• It is not the Von Restorff effect, in which whatever stands out from its context is remembered better. That effect is about distinctiveness regardless of valence: a single positive item among negatives also stands out, yet it does not carry the same extra weight.