Moral licensing

These illustrate the pattern the research claims. They are not evidence that it’s common.

The cake after the gym

Someone who is trying to eat less sugar goes for a hard run in the morning. At an office party that afternoon, they take a large slice of cake: “I’ve earned it after this morning.”

The run feels like a credit to spend, but whether it changes anything depends on the goal. If the goal is cutting sugar, the run has no bearing on the cake. This is the self-control cousin of moral licensing, about personal goals rather than right and wrong; the research reviewed below tested moral choices, not this version.

Recycling after the park cleanup

After spending Saturday morning picking up litter at a park, someone tosses a recyclable bottle in the trash on Sunday instead of rinsing it: “I did my bit for the environment this weekend.”

The cleanup doesn’t change where this bottle ends up. This is the claimed cross-domain form, where one good act licenses a lapse of a different kind. Experiments testing exactly this idea, that acting in an environmentally friendly way makes people behave less honestly afterward, found no licensing in three replication attempts with 1,274 participants in total (see Evidence).

The reliable colleague

A coworker who is known for staying late to help others leaves a team meeting early without telling anyone. Asked about it later, they say: “People know I pull my weight. No one’s going to think I was slacking.”

Here the good record isn’t used as a balance to spend, but to make an ambiguous act look innocent to others. Researchers call this licensing through moral credentials. It’s the version most tied to reputation, and a recent meta-analysis found moderate evidence of licensing only in studies where people’s choices were observed (see Evidence).

• Moral credits: good deeds build up a balance that can be spent on later misdeeds, which are still seen as wrong, like carbon offsets for moral behavior.
• Moral credentials: a good record changes how an ambiguous later act is interpreted, so that it doesn’t look wrong at all.
• Moral cleansing is the proposed mirror image: a bad act prompts a compensating good one.
• Moral consistency is the opposite prediction: doing good makes you see yourself as a good person, so you keep doing good. Some large replications found this pattern instead.

• When the earlier act changes the situation. If you’ve already given what a fundraiser needed, declining to give again isn’t licensing.
• When you’re managing a genuine budget. If the goal really is a balance, such as calories eaten against calories burned, a harder workout does change how much you can eat.
• When fairness depends on what each person has contributed. Having done more than your share of a shared chore is a fair reason to ask someone else to do the next one.
• When the later act isn’t actually wrong. Resting or treating yourself is not a lapse just because it follows a good deed.

The test: does the earlier good act change the consequences of this one, or only how you feel about doing it?

Eating for a long run

A runner training for a marathon does a 20-mile run on Sunday and eats a much bigger dinner than usual, as their training plan recommends.

The run changed the thing the dinner affects: how much energy their body needs. This is the genuine budget condition: the first act alters what the second act costs, which is what licensing lacks.

Taking turns with the dishes

Two roommates share the dishes. One has done them every night for a week while the other was busy with exams. On Saturday, the first says: “I’ve done all week. Can you take tonight?”

The earlier work does bear on the next choice, because the chore is shared and the question is who should do it. That’s the fairness condition. Licensing would be saying “I’ve done all week, so I can leave my pan in the sink for someone else” in a household where each person cleans up after themselves.

The original explanation is intrapsychic: people track a sense of themselves as moral, a good deed raises it, and a later lapse then feels less threatening to that self-image. A second explanation is interpersonal: good deeds build a reputation with others, and a good reputation makes an ambiguous act less likely to be judged harshly. The two predict different things. If licensing works mainly through reputation, it should appear when people are watched and fade when they act in private, which is what one recent meta-analysis reports.

Reviewers of the field also point out that past good behavior often leads to more good behavior, not less. A 2016 review proposed that people tend toward consistency when they think about how an act reflects their values, and toward licensing when they think about what they’ve accomplished. The licensing move resembles Special pleading: making an exception to a standard for yourself, with a past good act as the reason. It is also a flattering way to see your own choices, like the Self-serving bias.

No remedy has been tested for moral licensing specifically, and it’s not clear how often there is anything to remedy. A registered replication found no sign that thinking about a good deed in abstract terms, rather than concrete ones, changes whether people follow it with more or less good behavior.

The test above is still useful in your own decisions: when “I’ve earned it” comes to mind, ask whether the earlier act changed the consequences of the next one.

Status: contested. The evidence against a general licensing effect is strong: preregistered replications of the best-known studies found nothing or the opposite, and bias-corrected meta-analyses put the average effect near zero. It isn’t labeled failed, because that label needs large replications finding an effect near zero and recent reviews treating the effect as not established, and neither holds cleanly. A multi-lab replication of the founding study found a small effect in the predicted direction, and the most recent meta-analysis treats licensing as real when people are observed, while finding no evidence for it when they act alone. That leaves credible results on both sides, which is what “contested” means here.

The founding studies. Monin and Miller (2001) reported three experiments in which people who had first established credentials as unprejudiced were later more willing to express views that could look prejudiced. In the first, participants who had the chance to disagree with blatantly sexist statements were then more willing to say a stereotypically male job suited a man. In the second, those who had first picked a woman or a Black candidate for a job with no stereotype attached were later more likely to reject a member of that group for a job stereotyped as suited to the majority group. Many studies followed, extending the idea to charity, honesty and environmental behavior.

Meta-analyses.

• Blanken, van de Ven and Zeelenberg (2015) pooled 91 studies (7,397 participants) and estimated an average effect of d = 0.31, a small-to-medium effect. Published studies showed larger effects than unpublished ones, and the authors noted that studies would need many more participants to draw solid conclusions.
• Kuper and Bott (2019) applied corrections for publication bias to the larger of the existing data sets. One method (PET-PEESE) reduced the effect to d = −0.05, statistically indistinguishable from zero; another (a selection model) to d = 0.18, which they found signs might still be an overestimate. They concluded that both the evidence for moral licensing and its size had likely been inflated by publication bias.
• Rotella and colleagues (2026), a preregistered meta-analysis of 115 experiments (21,770 participants), found a small average effect before correction (g = 0.21), larger when participants were explicitly observed (g = 0.65) than when they were alone (g = 0.13). After correcting for publication bias with a Bayesian method, they report moderate evidence against licensing when participants were unobserved (g = −0.01), and moderate evidence for it when they were observed (g = 0.51), though the interval for the observed studies ran from 0.00 to 0.79. The overall bias-corrected estimate was at or slightly below zero. The authors conclude that licensing is driven by concern with how others see you, and that “evidence for intrapsychic motives is inconclusive”. Whether studies were observed was coded by the meta-analysts after the fact, because the original studies weren’t designed to test it.

Replications.

• Blanken and colleagues (2014) made three attempts to replicate a well-known finding that writing about your own good traits lowers charitable donations, including one with 940 participants. None found the effect.
• Many Labs 3 (Ebersole and colleagues, 2016) repeated the first Monin and Miller study across many university labs (3,134 participants). Disagreeing with blatantly sexist statements did lead to a slightly greater preference for a man for the job, a small effect (about d = 0.14, well below the original’s), and the original’s finding that only men were affected did not replicate.
• Urban, Bahník and Kohlová (2019) ran two conceptual and one close replication (1,274 participants in total) of a study reporting that buying environmentally friendly products led to more cheating, using three different cheating tasks. They found no licensing effect.
• Bahník and Vranka (2022) tested whether turning down small bribes in a laboratory task led people to take larger ones later (297 participants) and found no evidence of licensing.
• Xiao and colleagues (2024), a registered report, directly replicated Monin and Miller’s second study online with 932 participants. The effect was close to zero for the hiring scenario about ethnicity (d = 0.02 to 0.08) and went moderately in the opposite direction for the one about gender (d = −0.50 to −0.38). The authors note that their study was online and the original was in person, where participants may have felt more observed.
• Martuza and Kim (2025), a registered report with 5,091 participants, tested an earlier claim that how abstractly people think about a past moral or immoral act decides whether they then act consistently or compensate. That did not replicate. Recalling a moral act led to more prosocial intentions than recalling an immoral or neutral one, whether the act was recent or distant: the authors found evidence only for consistency, none for compensating behavior.

Other concerns. Mullen and Monin (2016), reviewing the field, noted that many studies lacked a baseline condition, for example comparing people who recalled a good deed only with people who recalled a bad one, which leaves it ambiguous whether licensing occurred at all.

What remains uncertain. The evidence for licensing under observation comes from a subset of studies classified after the fact, and has not yet been tested in a large preregistered in-person study designed for it. Whether the small Many Labs 3 effect reflects licensing or something about that particular task is unclear. The popular version, that people who do good routinely give themselves permission to behave badly afterward, goes well beyond what the evidence supports.