Cognitive dissonance

type: effect · origin: study · evidence: contested · themes: judgements and probabilities

Claim assessed: When you act against your attitude, especially by your own choice, the attitude moves toward what you did, to reduce the discomfort.

Published 8 October 2026, updated 8 October 2026 · how we work

In short

Cognitive dissonance is, according to Leon Festinger’s 1957 theory, the discomfort you feel when two things you think don’t fit together, for example when you do something that contradicts what you believe. To get rid of it, you would usually change your attitude to match what you did. It is one of the best-known theories in psychology. The largest test so far, with almost 4,900 students in 39 labs, found that people who write arguments against their own view do shift their attitude a little. The central prediction of the classic method, that the attitude changes more when you chose to write, wasn’t confirmed, and whether discomfort is what drives the change remains unclear. Meta-analyses of other forms of dissonance, such as hypocrisy or choosing between two options, find small to moderate effects, but they don’t show that discomfort is the driver either.

What it says

The theory. Festinger proposed that a mismatch between what you believe and what you do creates an unpleasant state that people are motivated to reduce. The most studied route is attitude change: once you have done something, the behaviour can’t be undone, so the opinion adjusts. The theory has been applied to beliefs disconfirmed by facts, to effort justification, to hypocrisy and to decisions.

The classic method. Most of the evidence comes from a single method, called induced compliance. In the first studies, like Festinger and Carlsmith’s (1959), people did something against their opinion for a small or a large reward. The small reward wasn’t enough of a justification, so the attitude changed more. Later, choice was varied instead of the reward. Participants write an essay in favour of a measure they oppose, such as raising tuition fees. Some are reminded that taking part is voluntary (high choice), others are simply told to write (low choice). Those who chose can’t blame the situation, so they should feel more dissonance and change their attitude more. Croyle and Cooper’s (1983) study, the one chosen for replication, had 10 people per group and a very large effect (d = 2.40). We haven’t read these two studies; we describe them from Vaidis et al. (2024).

The replication. Vaidis et al. (2024) repeated the study as a Registered Report: the method and analyses were approved by the journal Advances in Methods and Practices in Psychological Science before the data were collected. Thirty-nine labs in 19 countries took part, in 14 languages, with 4,898 students recruited and 3,822 left in the analyses. The students were split into three groups: an essay against their view, with choice; an essay against their view, without choice; and a neutral essay, with choice, about other topics for a future committee. In 37 labs the attitude had also been measured about two weeks earlier, on a scale from 1 to 9.

What came out. The choice manipulation worked: those reminded that taking part was voluntary felt much freer to decline. Their attitude, however, didn’t change more than that of the people without choice: the difference was practically zero (d = −0.03 on the post-essay attitude, d = 0.02 on the change). What was confirmed is something else: those who wrote arguments against their view, with or without choice, ended up less opposed than those who wrote a neutral essay. The effect is small (d = 0.26 on the post-essay attitude, 0.31 on the change). The results held after excluding those who didn’t follow the instructions and with a four-item measure instead of one, and they didn’t vary significantly from lab to lab.

The discomfort. Those who wrote against their view felt more uncomfortable (d = 0.22) and especially more conflicted (d = 0.58) than those with a neutral essay, and choice didn’t change that either. But the discomfort didn’t explain the attitude change: in the high-choice group, those who felt worse ended up more opposed, not less. The authors note that the result also fits self-persuasion: when you generate arguments for an idea yourself, it comes to seem more acceptable, without any discomfort needed.

The debate that followed. The journal published four commentaries and a reply from the authors.

• Pauer, Linne and Erb (2024) reanalysed the data using how free participants said they felt, rather than the group they had been put in. The attitude moved toward the essay only among those who felt free; among those who felt constrained, the difference from the neutral essay disappeared. They propose that reminders of voluntary participation have become a routine that no longer changes the feeling of choice. The analysis was done after the results, wasn’t planned and is correlational, which the authors acknowledge.
• Harmon-Jones and Harmon-Jones (2024) observe, according to the authors’ reply, that in the no-choice group the feeling of freedom was at the midpoint of the scale, so those participants too may have felt dissonance.
• Cyrus-Lai, Tierney and Uhlmann (2024) discuss, according to the same reply, what can be concluded from a failed replication. They note that replication results vary little from lab to lab and that researchers’ reputation doesn’t predict whether a study replicates, so the usual explanations for failures have little support.
• Sleegers et al. (2024), the authors of the replication, reply that nine published studies, with an even higher feeling of freedom in the no-choice group, were still presented as successes. The correlation Pauer et al. found may have other causes: someone with a strong opinion may feel less free and change less, for the same reason.

The older literature. In the same reply, Sleegers et al. reviewed 104 essay studies published over more than 50 years. They had an average of 16.7 participants per group (median 13.5), none was preregistered and only two had published their data. With groups that small, the large effects reported are probably overestimated.

The founding studies. Kelly (2026), in a preprint that hasn’t yet been reviewed by other researchers, checked whether the means reported in the first studies are mathematically possible, given the scale and the number of participants. In Festinger and Carlsmith (1959), students did a boring task and were then paid 1 or 20 dollars to tell the next participant it had been interesting. Those paid 1 dollar rated the task better (1.35 on a scale from −5 to 5) than those paid 20 dollars (−0.05) and the control group (−0.45). Kelly shows that some of the means reported in the article can’t come from 20 participants per group. Also, 26% of the 1-dollar group were excluded, and none of the control group; if those excluded had rated the task even slightly negatively, the difference would no longer have been significant. He finds similar problems in the study that launched the idea of effort justification (Aronson and Mills, 1959) and in an early replication of each. Transcription errors can explain some of the values, not all of them.

Other forms of dissonance. The large replication tested only the essay. For other forms there are meta-analyses, with a similar pattern of results:

• Induced hypocrisy. People publicly advocate a behaviour, such as using condoms, and then recall times when they didn’t follow it themselves. Priolo et al. (2019), across 38 studies, found that this increases intention (r = 0.35) and behaviour (r = 0.30), with no signs of inflation from unpublished studies. Discomfort, however, didn’t rise significantly (r = 0.18, an interval that includes zero, 7 studies), and simply recalling the lapses, without the public advocacy step, had a similar effect. Most studies were small, and the authors urge caution.
• Choosing between two options. After choosing between two things you like equally, you come to like the chosen one more. The classic method had a flaw: the choice could simply reveal a preference that already existed. Enisman, Shpitzer and Kleiman (2021) pooled 43 studies that correct the flaw (2,191 participants) and found an effect of d = 0.40, with no signs of publication bias. We read only the abstract.
• Vicarious dissonance. When you see someone from your group act against the group’s shared view, you would change your own attitude too. Jaubert et al. (2024), in a preregistered meta-analysis of 24 studies, found an effect that is small after correcting for unpublished studies (g ≈ 0.2). Discomfort didn’t appear, and the observed person’s choice didn’t matter.

Example

Someone is asked at work to present the advantages of a new procedure they don’t like to their colleagues. They prepare the arguments, say them out loud, and a few weeks later the procedure no longer seems quite so bad. The replication suggests this can happen even when the presentation was imposed on them, not only when they volunteered. The example is an illustration, not a study cited here.

How to apply it

The steps below are a practical way to apply it that we propose, based on the studies cited.

1. Write down your view before arguing for something you don’t believe. When you play a role, at work, in a debate or in a negotiation, note briefly what you think. Later you can see whether you have moved, and how far.
2. Don’t count on “you chose freely” to change someone’s mind. In the replication, reminding people they could decline added nothing to the effect of the essay.
3. To reconsider an idea, write arguments for the other side. The effect is small, but it appeared without choice too. A related technique is described in the entry “Consider the opposite”.
4. Don’t take discomfort as a sign that you’ve changed your mind. In the study, those who felt worse didn’t change their attitude more.

Limits and nuances

• One large replication. Only the essay method has been tested at scale and preregistered. For hypocrisy and choosing between options we have only meta-analyses of mostly small studies, and for effort justification we found no open synthesis. The authors of the replication say explicitly that they didn’t refute the theory, but showed that its main method doesn’t replicate reliably.
• The effects exist; the explanation is disputed. In all the forms above, the behaviour or the choice changes something, but the discomfort the theory predicts isn’t confirmed. Jaubert et al. warn that an attitude change doesn’t prove that a state of dissonance existed.
• Part of the change doesn’t come from the essay. The students started out strongly opposed (around 1.6 on the 1 to 9 scale), and attitudes rose in every group, including the neutral-essay one. Only the difference between groups can be attributed to the essay.
• Context. The data were collected during the pandemic, with masks and distancing, from students, mostly on the topic of tuition fees. Harmon-Jones and Harmon-Jones proposed that participants could have attributed their tension to the pandemic or to a nervous experimenter. The authors reply that the same explanation should have erased the effect of the essay too, which did appear.
• We haven’t read the meta-analyses. Kenworthy et al. (2011) and Kim et al. (2014) found medium to large effects for this method (d = 0.81 and 0.51, according to Vaidis et al.). Given the group sizes of the studies they include, they may overestimate the effect.
• What we read. Vaidis et al. (2024), from the University of Greenwich repository (the published version), and Pauer et al. (2024) and Sleegers et al. (2024), from the University of Amsterdam repository, we read in full. We couldn’t download the commentaries by Harmon-Jones and Harmon-Jones (2024) and Cyrus-Lai et al. (2024); we describe them from Sleegers et al.’s reply. The fourth commentary, by Lishner (2024), we haven’t read at all. We describe Festinger and Carlsmith (1959) and Croyle and Cooper (1983) from Vaidis et al., and the figures from Festinger and Carlsmith and the study by Aronson and Mills (1959) from Kelly (2026), a preprint we read in full. We read the Priolo et al. (2019) meta-analysis in full, as the accepted manuscript from the HAL repository; we couldn’t compare it with the published version. We read Jaubert et al. (2024) in full, from PubMed Central. For Enisman et al. (2021) we read only the abstract; we describe Chen and Risen’s (2010) study, which showed the method’s flaw, from that abstract. The article is available only behind a paywall and we found no free copy. It is the strongest evidence for the theory we found; read in full, it could change what we say about choosing between options, but not the result of the essay replication.
• Revised on 8 October 2026. We noted under “What we read” that the Enisman et al. (2021) meta-analysis is available only behind a paywall, and what it could change.

Sources

• Leon Festinger, James M. Carlsmith (1959). Cognitive consequences of forced compliance
• Robert T. Croyle, Joel Cooper (1983). Dissonance arousal: Physiological evidence
• David C. Vaidis, Willem W. A. Sleegers, Florian van Leeuwen et al. (2024). A multilab replication of the induced-compliance paradigm of cognitive dissonance
• Shiva Pauer, Roman Linne, Hans-Peter Erb (2024). From the illusion of choice to actual control: Reconsidering the induced-compliance paradigm of cognitive dissonance
• Eddie Harmon-Jones, Cindy Harmon-Jones (2024). Dissonance in the induced-compliance paradigm: A commentary on Vaidis et al. (2024)
• Wilson Cyrus-Lai, Warren Tierney, Eric Luis Uhlmann (2024). Noise versus signal: What can one conclude when a classic finding fails to replicate?
• Willem W. A. Sleegers, Florian van Leeuwen, Robert M. Ross et al. (2024). When replication fails: What to conclude and not to conclude?
• Daniel Priolo, Audrey Pelt, Roxane Saint-Bauzel et al. (2019). Three decades of research on induced hypocrisy: A meta-analysis
• Maya Enisman, Hila Shpitzer, Tali Kleiman (2021). Choice changes preferences, not merely reflects them: A meta-analysis of the artifact-free free-choice paradigm
• Sara Jaubert, Adrien Alejandro Fillon, Lionel Souchet, Fabien Girandola (2024). Vicarious dissonance: Pre-registered meta-analysis
• Thomas Kelly (2026). Impossible values in canonical cognitive dissonance studies and their early replications

See also: Confirmation bias, Consider the opposite, Sunk cost