Unrealistic optimism

These illustrate the pattern the research describes. How often it reflects a real bias, rather than accurate knowledge or a measurement quirk, is the disputed part.

Other people’s bikes

A student leaves their bike unlocked outside the library for a few minutes most days. The campus newsletter says dozens of bikes are stolen each term. “That’s people who leave them overnight. I’m only ever inside for ten minutes.”

This is comparative optimism. The student focuses on what lowers their risk and doesn’t ask whether the students whose bikes were taken had reasons of their own to feel safe. Researchers call this kind of one-sided comparison egocentric: you know your own precautions in detail, and “the average person” only as a vague figure.

The exam grade, weeks out

Three weeks before a statistics exam, a student expects a B+. Their homework grades have been mostly Cs. On the morning of the exam they expect a C, and on the day results come back they are braced for a D.

This is absolute optimism: the early estimate is better than the student’s own record supports. The shift is the less obvious part. Research reviewed by James Shepperd and colleagues found that optimism like this is strongest when the outcome is far off and fades, or even turns into pessimism, as the moment of truth approaches. So the bias, where it exists, depends on the situation rather than being a fixed outlook.

Good news counts more than bad

Someone guesses there’s a 10% chance their car will break down on a long road trip. A mechanic tells them it’s about 30% for cars of that age and mileage, and they revise to 12%. Their friend guessed 50% and, told the same 30%, drops straight to 30%.

This is the belief-updating version: moving your estimate a lot for good news and barely at all for bad news. Tali Sharot and colleagues proposed it as the way optimism survives contact with reality. Whether that asymmetry is real, or an artifact of how such studies are designed, is one of the sharpest disputes in this literature.

• Comparative optimism: rating your risk as lower (or your chances as better) than those of similar people. This is the version most studies measure.
• Absolute optimism: an estimate of your own risk that is rosier than an objective standard such as a base rate, a risk calculator or what actually happens later.
• Optimistic belief updating: updating more for desirable information than undesirable information.
• The planning fallacy: Shepperd and colleagues treat underestimating how long tasks will take as a case of absolute optimism (see Planning fallacy).

• When you really are lower risk. A non-smoker who exercises can correctly say their heart disease risk is below average. Comparative optimism is an error only if the comparison is wrong.
• When the risk is concentrated in a few people. If a small group carries most of the risk, most people really are below the average, so a majority rating themselves “below average” isn’t proof of bias.
• When your precautions change the odds. Someone who locks their bike, or has already fixed the problem that caused past trouble, can reasonably expect better results than the base rate.
• When hope isn’t being used as a forecast. Aiming for the best case, or staying upbeat, is fine as long as plans are built on the likely case.

The test: would you give this estimate to a stranger with exactly your circumstances?

A below-average estimate with the facts behind it

In a survey, a 30-year-old who has never smoked, has normal blood pressure and no family history of heart disease rates their risk of a heart attack by 60 as “well below average for my age”.

The answer is optimistic in direction but not unrealistic: the risk factors they list really do put them below average. Researchers can check claims like this against a risk calculator, and when they do, some people’s low estimates turn out to be accurate. The test above passes.

Most of the room says “below average”

At a workshop, 30 cyclists rate their chance of a serious crash in the next year compared with the average cyclist in their city. Twenty-two say “below average”.

That sounds impossible, but it isn’t. Serious crashes are rare and cluster among a minority of riders, so most individual cyclists really are below the mean. Adam Harris and Ulrike Hahn showed that with rare events, limited response scales and small samples, a group of perfectly accurate people can produce exactly this pattern. Showing a real bias takes checking individuals against their actual risk, not just counting how many say “below average”.

Several explanations have been proposed, and they aren’t mutually exclusive:

• Motivation. Believing bad things won’t happen to you feels better, and may reduce anxiety. Sharot’s work proposes that this shows up as discounting unwelcome information.
• Egocentric focus. When comparing yourself with “the average person”, you think in detail about your own precautions and advantages and only vaguely about everyone else’s. Shepperd and colleagues attribute much comparative optimism about rare negative events to this kind of thinking.
• Perceived control. People are more optimistic about events they believe they can control, and less optimistic about events they see as beyond their control, according to the reviews by Shepperd and colleagues (see Illusion of control).
• Statistics rather than psychology. Harris and Hahn argue that rare events, blunt response scales and regressive guesses about “the average person” can produce the classic pattern with no bias at all.

• Ask what the base rate is, then ask what really sets you apart. In studies reviewed by Shepperd and colleagues, giving people base rates reduced absolute optimism, but often didn’t eliminate it: homeowners told that 73% of homes in their area had high radon levels still put their own chance at 54% on average.
• Picture the other people in the statistic. Most of them also think they’re careful. This follows from the egocentric account above; it hasn’t been tested here as a remedy.
• Check whether the good news is moving you more than the bad. If a reassuring number would change your plans but an alarming one wouldn’t, that asymmetry deserves a second look. This follows from the updating research; it hasn’t been tested as a remedy.

Status: contested. Comparative optimism has been reported in hundreds of studies of many different events, and its defenders say it holds up. But a serious critique argues that the standard group method can’t distinguish bias from statistical artifact, the proposed mechanism (asymmetric updating) is itself disputed in both directions, and no multi-lab or preregistered replication that settles the artifact question could be found. The dispute over how much of the pattern is real fits this site’s contested label.

The foundational study. Weinstein (1980), the paper that coined “unrealistic optimism”, asked students to compare their own chances of future life events with those of the average person. They rated themselves less likely than average to experience negative events such as being fired or being sued, and, usually less strongly, more likely to experience positive ones.

The artifact critique. Harris and Hahn (2011) argued that most comparative optimism studies use a group-level test (whether the average self-rating falls below “average”) that can be failed by perfectly unbiased people. Their abstract describes how rare negative events, the response scales used and sampling constraints can together produce the classic pattern “for purely statistical reasons”, and concludes this raises questions “over the very existence of an optimistic bias about risk”. A tell-tale check is to use rare positive events: an artifact predicts people will again rate themselves below average, while real optimism predicts above. Shah and colleagues (2016) note that studies including such events found the artifact’s pattern.

The defense. Shepperd, Klein, Waters and Weinstein (2013) replied that the artifacts apply mainly to one of four kinds of measure: comparative optimism assessed at the group level. Studies that compare individuals with a risk calculator still find optimism: in a nationally representative US sample of over 14,000 women, 41.8% were unrealistically optimistic about their breast cancer risk and 13.4% unrealistically pessimistic. They also point to studies the artifacts can’t explain, such as patients in an early-phase clinical trial where the expected benefit was effectively zero for everyone, of whom 62.5% still thought they were more likely than the average participant to benefit. At the same time, they document boundary conditions: optimism shrinks as feedback nears, and people often overestimate their absolute risk of rare or highly publicized events. More recently, a multinational, partly preregistered study by Kuper-Smith and colleagues (2021) found comparative optimism about catching COVID-19 early in the pandemic, though not about getting severe symptoms.

The updating dispute.

• Sharot, Korn and Dolan (2011) had 19 participants estimate their chances of 80 adverse events, shown the average probability, and then estimate again. They updated more after better-than-expected information (an average of 11.2 percentage points) than after worse-than-expected information (7.7), and the pattern was linked to brain activity.
• Shah, Harris, Bird, Catmur and Hahn (2016) argued this pattern is also a statistical artifact. When they added genuinely positive events, the asymmetry flipped: for positive events, people updated more after bad news, which an optimism account can’t explain. Across simulations and five experiments, they found no evidence of optimistic updating once measured against a Bayesian standard.
• Garrett and Sharot (2017) reanalyzed the question, argued Shah and colleagues’ experiments had confounds, and reported an optimistic update bias for positive events too. Kuzmanovic and Rigoux (2017) fit computational models controlling for several cognitive factors and found a biased model fit their data better than an unbiased one.
• Burton, Harris, Shah and Hahn (2022), in three preregistered experiments (300 participants), found apparent “biased” updating even with emotionally neutral events, where optimism shouldn’t apply. Garrett and Sharot (2021), in a preprint, reported failing to replicate that finding.

What remains uncertain. That people rate their own risks as lower than other people’s is well established. How much of that reflects a genuine bias rather than accurate knowledge and measurement artifacts, whether the asymmetric updating mechanism is real, and how optimism about oneself affects actual behavior are all open. Shepperd and colleagues themselves note that evidence linking unrealistic optimism to what people actually do “remains thin”.