Negativity bias

The one comment in the review

An employee’s annual review lists eight specific things she did well and one suggestion: “Could share project updates with the wider team more often.” Her manager sums it up as a strong year. She spends the evening replaying the suggestion and tells a friend the review “didn’t go well”.

This is the pattern the negativity bias describes. The suggestion is minor, and by the manager’s own summary the review was strongly positive, yet it outweighs eight comparable pieces of praise in her overall judgment. Whether such reactions reflect a general rule of the mind or depend on the situation and the person is part of the dispute (see Evidence).

The trip with the lost suitcase

A family takes a ten-day trip. On the first day the airline loses a suitcase, which turns up the next evening. The other nine days go well. A year later, they remember it as “the vacation where everything went wrong”.

The less obvious form: a mixed experience is judged more negatively than its parts add up to. One bad day colors the memory of the whole, even though, day by day, the good clearly outnumbered the bad. Rozin and Royzman call this pattern negativity dominance.

The neighbor who was short with you once

For two years a neighbor has taken in parcels, lent tools and chatted over the fence. One morning, in a hurry, he answers a question curtly. “Honestly, I don’t think he’s a very nice person.”

A single brusque moment on a busy morning says little about someone’s character, especially set against two years of friendliness. Treating it as decisive gives one bad interaction far more weight than many good ones of the same kind. Compare the first look-alike below, where one bad act really does say more.

Rozin and Royzman describe four ways the tendency is claimed to show up:

• Negative potency: a bad thing has a bigger effect than an equivalent good thing (the review).
• Steeper negative gradients: bad things loom more quickly than good things as they approach in time or space.
• Negativity dominance: mixtures of good and bad are judged worse than their parts add up to (the trip).
• Negative differentiation: bad things are described in more varied and complex ways and prompt a wider range of responses.

Specific research areas have their own versions: in forming impressions of people (the neighbor), in attention to threatening faces or words, in how strongly news stories arouse people, and in young children’s use of other people’s emotional reactions.

• When the bad thing is more informative. In a world where most people are honest most of the time, one dishonest act says more about a person than one honest act does.
• When the stakes are lopsided. Ignoring a danger once can be catastrophic; missing an opportunity once usually isn’t. Paying more attention to possible harms is then good judgment.
• When the bad thing is actually bigger. A serious loss outweighing a trivial gain isn’t a bias; the comparison has to be between things of the same size, which is harder to establish than it sounds.
• When the bad thing is news. Where good things are routine, a bad one may be the only thing that tells you something has changed.

The test: would a good thing of the same size, reliability and informativeness get the same weight?

Three years and one theft

A shop owner has had a reliable cashier for three years. One evening the till is short, and the security camera shows the cashier pocketing the money. The owner moves him to a job away from cash.

One bad act outweighs three years of good ones, which looks exactly like the neighbor example. But here the bad act is far more informative: people who never steal are honest at the till every day, and so, most days, are people who sometimes steal. Three years of honest shifts can’t tell those two apart; one theft on camera can.

The washed-out footbridge

Planning a hike, someone reads nine recent trail reports saying the route is in fine shape and one, posted two days ago, saying a footbridge over a river has washed out. They choose a different route.

The single negative report wins. That’s reasonable because the stakes are lopsided (a wasted morning against an unsafe river crossing) and because the negative report is also the newest, so it may describe a change the others predate.

Baumeister and colleagues proposed an evolutionary account: an organism that ignores a danger even once may be injured or killed, while one that misses an opportunity can usually try again, so responding more strongly to bad things would have aided survival. Rozin and Royzman suggest that part of what makes negative things dominant is that they are more “contagious” than positive ones.

Other researchers argue that many findings come from features of information that happen to go with badness. Negative events are rarer than positive ones in most people’s lives, which makes each one more informative, and Hans Alves and colleagues argue that positive things are also more similar to one another than negative things are, so each negative one stands out. On these accounts, part of the bias is not a response to badness as such. The same pull toward the striking case feeds the Availability heuristic, and in judging people it overlaps with the negative side of the Halo effect. Distinctiveness-based illusory correlations, in which a small group gets linked with a rare behavior, have been found stronger when the rare behavior is negative.

Baumeister and colleagues count Loss aversion among the evidence for their principle. This site treats loss aversion as a separate, narrower entry: it concerns decisions, and compares losses and gains of equal size measured from a reference point, and it has its own contested evidence.

• Check whether the good and bad are really the same size. If not, weighting them differently may be exactly right.
• Ask what the bad thing tells you that the good things don’t. If the answer is “not much”, it probably deserves no more weight than they do.
• Count before summarizing. Listing the good and bad parts of a review, a trip or a relationship before forming an overall verdict makes it harder for one item to stand in for the whole.

These follow from the definition; they haven’t been tested as remedies in the studies cited here.

Status: contested. Many specific negativity effects are well supported, including a preregistered replication and very large meta-analyses in particular domains. But the claim this entry names is the broad principle that bad is stronger than good, and its generality is seriously disputed: several lines of research find no negativity bias, or a positivity bias, in areas where the principle predicts one, and critics argue that “equivalent” good and bad are rarely defined well enough to test it.

The foundational reviews. Baumeister, Bratslavsky, Finkenauer and Vohs (2001) reviewed findings across everyday and major life events, close relationships, feedback, learning, impression formation and health, and concluded that bad is stronger than good “in a disappointingly relentless pattern”, reporting that “hardly any exceptions” could be found. Rozin and Royzman (2001) proposed a general bias, rooted in both innate predispositions and experience, in animals and humans, with the four forms listed under Variants. Vaish, Grossmann and Woodward (2008) argued that the bias is present in infancy, especially in how infants use adults’ emotional reactions to guide their own behavior.

Support for specific versions.

• News. Soroka, Fournier and Nir (2019) measured physiological arousal in 1,156 people in 17 countries as they watched real news stories. On average, negative stories aroused people more, but there was considerable variation between individuals, and some country samples would not on their own have shown a significant bias. Dubèl and colleagues (2024) confirmed the central arousal finding in a well-powered preregistered replication in the Netherlands, while also finding that people reported feeling more aroused by positive news. (Based on the abstract.)
• Contact between groups. Paolini and colleagues (2024) meta-analyzed 238 samples (about 153,000 people) and found that negative contact with members of another social group was associated with more prejudice by a larger amount than positive contact was associated with less. (Based on the abstract.)

Evidence against a general principle.

• Attention. Bar-Haim and colleagues (2007), in a meta-analysis of 172 studies, found a reliable attentional bias toward threat in anxious people (d = 0.45) that was not observed in nonanxious people. Becker and colleagues (2011) reviewed the widely cited finding that angry faces “pop out” of crowds, found the evidence inconsistent and possibly due to visual confounds, and in seven experiments found happy faces, not angry ones, detected more efficiently. Pool and colleagues (2016) meta-analyzed 243 studies (9,120 participants) and found attention drawn to positive stimuli as well, compared with neutral ones (g = 0.26). (Based on the abstracts.)
• Learning likes and dislikes. Sperlich and Unkelbach (2025) reanalyzed meta-analytic data on evaluative conditioning (coming to like or dislike something through pairing with pleasant or unpleasant things) and found no negativity bias. In five preregistered experiments, a negativity bias appeared when negative pairings were rare, and a positivity bias when positive ones were rare, supporting the view that rarity rather than badness drives the asymmetry. (Based on the abstract.)
• Direction depends on expectations. Paolini and McIntyre (2019), meta-analyzing experiments on how an encounter with one member of a group generalizes to the whole group, found that negative encounters generalized more for stigmatized groups, but positive encounters generalized more for admired ones. (Based on the abstract.)
• Conceptual criticism. Philosopher Jennifer Corns (2018) argued that the principle should be rejected as stated: “good” and “bad” and “stronger” are used in several different senses, making claims of equal size hard to test, and many supporting results have alternative explanations, such as bad information being more informative or more momentous. She notes that Rozin and Royzman themselves acknowledged the difficulty of equating negative and positive events. Alves, Koch and Unkelbach (2017) likewise offer an account of many valence asymmetries that doesn’t depend on people’s emotional reactions to badness.

What remains uncertain. In which domains, for which people, and under what conditions bad outweighs equivalent good; how much of the pattern reflects rarity, informativeness or intensity rather than badness itself; and whether a single general principle is the right description at all. That bad events and information often have larger effects than good ones in particular settings is well documented.