Loss aversion
Claim assessed: A loss weighs more than a gain of the same size.
Published · how we work
In short
Loss aversion means a loss affects us more than a gain of the same size. Its size is expressed by a coefficient, written λ: at 1, losses and gains weigh the same, and at 2, a loss weighs as much as a gain twice as large. The classic figure, 2.25, comes from a study of 25 students. Recent meta-analyses show that, on average, people do avoid losses, but less than the classic figure says, and the result varies a lot from one study to another.
What it says
The idea and the classic figure. In prospect theory, Kahneman and Tversky (1979) proposed that people judge outcomes as gains or losses relative to a reference point, not by their total wealth. Losses weigh more, which would explain why most people turn down a coin-toss bet in which they can win or lose the same amount. Tversky and Kahneman (1992) estimated the coefficient from 25 graduate students making hypothetical choices, with no real money, and got a median value of 2.25. We have not read these studies; we describe them from the two meta-analyses below.
The largest meta-analysis. Brown, Imai, Vieider and Camerer (2024) collected 607 estimates of the coefficient, from 150 articles published between 1992 and 2017 in economics, psychology, neuroscience and other fields. The estimated mean was 1.96, with an interval of 1.82–2.11. Almost 94% of the estimates were above 1. The authors also found signs that small estimates get published less often; for those who take this into account, they suggest a value of 1.73–1.74. Few study characteristics explained the differences between results. Field experiments and general-population samples gave somewhat higher values than samples of students.
A reanalysis of the raw data. Walasek, Mullett and Stewart (2024) asked authors for the raw data of studies with bets in which participants could both win and lose. They received 19 datasets, from 17 articles, and computed the coefficient the same way for all of them. The mean was 1.31, with an interval of 1.10–1.53. Across studies, values ranged from 0.65 to 3.45, and in 12 of the 19 datasets the interval included the value 1, meaning no aversion.
How much it depends on measurement. Walasek et al. describe studies in which loss aversion appeared or disappeared depending on how the questions were built. In one, the same choice between a cautious bet and a risky one showed aversion when the other options on the list mostly offered large gains. When the list was balanced, people chose the two bets equally often. The authors also note that in most experiments participants cannot lose money out of their own pocket, only from the amount given to them at the start, for ethical reasons. Gal and Rucker (2018), whom they cite, observe that there is no clear psychological theory of what causes the effect.
Example
Someone offers to toss a coin with you: heads, you lose 100; tails, you win 150. With a coefficient of 2, the loss would weigh as much as a 200 gain, so you would turn the bet down. With one of 1.3, you would accept. The calculation is simplified and is an illustration, not a study cited here.
How to apply it
The steps below are a practical approach we propose, based on the studies cited.
- Notice what you’re measuring against. The same outcome can look like a gain or a loss, depending on the reference point you choose.
- When you turn down a favourable bet, ask why. If you can repeat it many times and the loss doesn’t seriously hurt you, the refusal may come only from the loss weighing more.
- Don’t use 2.25 as a fixed rule. The average in the literature is lower and varies a lot between studies.
- When someone builds an argument on loss aversion, look at how it was measured. The way the questions are built can change the result a lot.
Limits and nuances
- The coefficient depends on the model. Walasek et al. show that the same tendency to avoid losses can be attributed to other parameters of the model, and estimates from different models cannot be compared directly.
- The two meta-analyses don’t measure the same thing. Brown et al. pool the figures reported by authors, including from field data; Walasek et al. reanalyse only the raw data available from bets, from few studies, many with poor-quality data.
- The studies are fairly old. Walasek et al. cover studies published up to 2014, and Brown et al. up to 2017.
- Riskless situations are separate. The gap between selling and buying prices is discussed under the endowment effect.
- What we read. We read Walasek et al. (2024), which is open access, in full. Of Brown et al. (2024) we read the version submitted for publication, from Ghent University’s repository: the abstract, the method, the main results, the analysis of selective publication and the discussion, but not all the appendices. We have not read Kahneman and Tversky (1979), Tversky and Kahneman (1992) or Gal and Rucker (2018); we describe them from the two meta-analyses.
Sources
- Daniel Kahneman, Amos Tversky (1979). Prospect theory: An analysis of decision under risk
- Amos Tversky, Daniel Kahneman (1992). Advances in prospect theory: Cumulative representation of uncertainty
- David Gal, Derek D. Rucker (2018). The loss of loss aversion: Will it loom larger than its gain?
- Alexander L. Brown, Taisuke Imai, Ferdinand M. Vieider, Colin F. Camerer (2024). Meta-analysis of empirical estimates of loss aversion
- Lukasz Walasek, Timothy L. Mullett, Neil Stewart (2024). A meta-analysis of loss aversion in risky contexts
See also: Endowment effect and status quo bias, Framing effect, Negativity bias