Dishonesty Study

Click here to access the paper

Watch a short video explaining the study


Frequently Asked Questions

What exactly did you do?

Watch a video explaining the experimental paradigm we used

What exactly did you find?

1. BEHAVIOURAL FINDING: The amount by which participants lied got larger and larger over the course of the block. We quantified the amount of dishonesty on each trial in British Pounds (£) (to learn how, watch the video above – ‘What exactly did you do?’).


Dishonesty escalation was observed only when participants lied for their own benefit, not when they did so solely for the benefit of others.

2. BRAIN FINDING: A network of brain regions associated with emotion (see below how we defined this network) responded strongly when participants lied initially. But as time went on, it would respond less and less to the same amount of lying. The greater the drop in sensitivity, the more a person increased their lying the next opportunity they got.

We conducted two analyses – here is exactly what we did:

We computed the response of this network to one unit (£) of lying every time someone lied. We then calculated the drop of this number from one trial to the next – indexing adaptation. We then used this number to predict by how much dishonesty would escalate on the next trial in each subject.

In another analysis, we calculated a term called “time-weighted dishonesty”. This is the amount a participant lied on each trial times a weight that is larger for the first few trials and decreases as a participant progresses through the block of trials. Activity in the emotional network of the brain correlated with “time weighted dishonesty” while controlling for the main effect of time and the main effect of dishonesty.

Is this the first time a slippery slope has been documented empirically?

Yes. There are many anecdotes of small lies snowballing over time. But to our knowledge this study provides the first empirical evidence that dishonesty escalates, when all else is held constant.

Previous studies have shown that dishonesty escalates when incentives are gradually increased by the experimenter (Welsh et al., 2015) and that attitudes to other peoples’ dishonesty changes over time (Gino & Bazerman, 2009).

Perhaps brain activity just decreased over time regardless of dishonesty?

No. Brain activity did not decrease over time, this is a common misunderstanding.

Pg 4: “The effects were not due to BOLD response simply decreasing over time, since the model controlled for the independent effect of time by including this as the first regressor. Furthermore, examining the effects of the time regressor (rather than time-weighted dishonesty) did not reveal significant effects in the ROI (Self-serving: t24 = 0.08, P = 0.94; one-sample t-test versus 0; Self-harming: t24 = 1.31, P = 0.20; one-sample t-test versus 0; Self-serving versus Self-harming: t24 = −0.85, P = 0.40; paired-sample t-test).”

What decreased was the sensitivity to dishonesty – this was very specific to the exact amount someone lied, and it was detected only in the emotional network, not elsewhere.

Could you get the same results if brain activity is constant or saturating, irrespective of dishonesty?

No. Our results do not reflect a general property of the BOLD response, such as it saturating over time or staying constant. We conducted a permutation test were one person’s BOLD response is used to predict another persons behavior. Because general properties of BOLD response will be similar across participants, if they were driving the effect, mismatching in this way should generate similar results to the ones we report. Yet, conducting this test 1000 times, each time mixing and matching one person’s bold with another’s behavior did not reveal significant results (see pg 4 of the article).

In addition, we ran our analysis on other regions of the brain, not related to emotion. Again, if your results were due to general response properties of BOLD response you would expect these to be similar in different brain regions and therefore to find similar results. Yet we do not find the same patters of results elsewhere in the brain.

Are there individual differences?

Yes. Some people were more dishonest than others and some people’s dishonesty escalated faster.

Does this paint a pessimistic view of humanity?

Our data also shows a positive side of humans. First, people were much more likely to lie when it also benefited someone else rather than just themselves. Second, people could have lied much more than they did, but did not, and lost money for it.

How did you define the “emotion network” in the brain?

We used an online database called neurosynth ( that compiles data from tens of thousands of studies. We entered the term “emotion” and neurosynth generated a map of brain voxels that have the strongest selective association with the term emotion across this large database. This map predominantly, but not exclusively, comprised of the amygdala, a structure known to be key for emotional processing.

Could your result be explained as adaptation to risk or reward?

Probably not. First, the incentive structure was designed such that adaptation to dishonesty could not be explained by adaptation to reward. Specifically, participants were told that the available reward would be unknown to them and vary on each trial. Thus, while on a specific trial a larger magnitude of dishonesty would likely result in greater gains, greater dishonesty on that trial relative to the previous one would not necessarily correspond to greater returns. Moreover, neither the amount available, nor the outcome of the trial, would be disclosed at any point during the experiment. In addition, participants were told that rewards would not accumulate, but rather one trial would be selected at random at the end of the experiment and the participant would be paid according to that trial. We went to great length to explain the reward structure to the participants and to ensure their comprehension by using a quiz with example trials prior to the beginning of each block of trials. Risk is also an unlikely explanation; we made it clear to participants from the outset that no feedback would be available during the experiment and no punishment either.

Could your results reflect emotional arousal?

Yes, we believe the brain activity we observe is associated with arousal.

This interpretation fits with previous studies suggesting emotional arousal plays an important role in dishonesty (Schachter & Ono, in Schachter & Latane, 1964; Dienstbier & Hunter, 1971). For example, a group of students who were given beta-blockers, which reduce arousal, were twice as likely to cheat on an exam relative to those given placebo (Shachter and Latane, 1964).

How do you know your findings are related to emotion?

The association with emotion is an inference made in the article with caution. Tying any effect to emotion is tricky. One could ask subjects directly how they feel, but in our case such introspection would likely interfere with the behaviour of interest. One could manipulate emotion/arousal (see example above). One could measure physiological (i.e. SCR) or neurological (i.e amygdala activity) responses known to be associated with emotional arousal – the later is what we did (see “How did you define the ‘emotional network’ in the brain?).

We also examined other brain regions previously associated with other aspects of dishonesty. These did not generate similar results, suggesting selectivity to the emotional network. Note, we cannot say what type, or aspect, of emotion our findings speak to.

Do your results replicate?

We have replicated the behavioural results a few times before publishing this study (two of which are reported in the article). We conducted the fMRI study only once and future replication studies are needed.