Popular in Psychology

Books

City Time: On Being Sentenced to Rikers Island

Independent Scholar David A. Campbell

Most Cited

The adapted mind

Dalhousie University Jerome Barkow

Experimental Psychology

Time to investigate human visual shape perception

University of Montreal Martin Arguin

Most Cited

James S. Jackson and the program for research on Black Americans

Harvard University David R. Williams

Psychology

Violence and Mental Illness: Rethinking Risk Factors and Enhancing Public Safety

Duke University Eric Elbogen

Cognitive Psychology

Cognitive remediation – where are we now and what should we do next?

King's College London Til Wykes

Does competitive winning increase subsequent cheating?

View more insights in Social Sciences - Psychology - Psychological Methods - Social Psychology

View more figures from University of Leicester

View the page of Andrew Colman

Read the Study

Andrew Colman discusses a replication of the Schurr and Ritov experiment, that winners of pairwise competitions are more likely than others to steal money in subsequent games of chance against different opponents.

Image courtesy of interviewee. May 12, 2024

Generating Workspace

Export as PDF

This video video discusses research on the relationship between competitive success and dishonest behavior, particularly focusing on a study published in 2016 that suggested winners of competitions are more likely to cheat afterward. The original study involved participants completing a challenging visual test and then playing a dice game to assess cheating based on reported results.

The findings indicated that those who had won the test reported higher numbers in the dice game than those who lost, suggesting a link between winning and increased dishonesty. However, the author presents a critique of this study, noting its small sample size and lack of a control group.

In response, the author and collaborators conducted more rigorous experiments with larger sample sizes and proper controls. Their research concluded that there was no significant difference in cheating behavior between winners, losers, or a control group, thereby challenging the original findings. The only notable result was that individuals with a strong sense of fairness—identified as having high inequality aversion—were less likely to cheat.

The discussion highlights the importance of understanding the factors that influence cheating behavior, particularly in contexts like academic dishonesty and wider societal issues of inequality and corruption. The conclusion emphasizes that the notion of winning leading to dishonest behavior lacks empirical support, and identifies inequality aversion as a relevant factor in predicting honesty.

Language:

Export as PDF

X-TIMESTAMP-MAP:LOCAL=00:00:00.000,MPEGTS=900000

00:00.295 --> 00:04.435
In 2016, in the proceedings of the National Academy

00:04.535 --> 00:07.595
of Sciences, very prestigious American Magazine,

00:08.665 --> 00:10.645
an article appeared with the title,

00:10.755 --> 00:14.045
winning a Competition Predicts Dishonest Behavior.

00:14.975 --> 00:19.235
Uh, and the idea in this, in, in this paper was

00:19.575 --> 00:21.235
to report an experiment

00:21.335 --> 00:25.195
or a series of experiments in which competitive winning

00:25.805 --> 00:27.515
tends to make people cheat.

00:27.515 --> 00:30.435
According to the authors winners

00:30.495 --> 00:34.195
of pair wide com competitions involving skill

00:35.275 --> 00:36.505
cause people or,

00:36.605 --> 00:39.785
or they're more likely if they've won that competition

00:40.245 --> 00:43.545
to steal money in subsequent games of chance

00:43.655 --> 00:46.185
with completely different opponents.

00:47.005 --> 00:49.305
And the way they did it is, is as follows.

00:49.595 --> 00:51.905
First of all, they gave everybody,

00:52.165 --> 00:54.945
all the participants in the study a visual test,

00:55.185 --> 00:57.625
a very difficult visual test

00:57.625 --> 00:59.265
and above all, one

00:59.265 --> 01:01.105
that people are completely unfamiliar with.

01:01.165 --> 01:02.465
So they wouldn't have any way

01:02.805 --> 01:05.105
of judging whether they'd done well

01:05.105 --> 01:07.305
or badly in relation to other people.

01:08.005 --> 01:09.985
And this enabled the experimenters then

01:10.005 --> 01:14.585
to tell half the participants that they'd won by doing

01:15.125 --> 01:17.025
better than a particular level

01:17.525 --> 01:21.475
and half that they'd, failed

01:21.695 --> 01:24.395
by doing worse than a particular level when they

01:24.395 --> 01:25.515
were paired with other people.

01:26.855 --> 01:28.555
And so they had winners

01:28.555 --> 01:31.635
and lu losers artificially created in this way,

01:32.655 --> 01:34.915
and they then played a dice game.

01:35.535 --> 01:39.715
Now the dice game involved paper cups exactly like this one.

01:39.715 --> 01:41.235
This is in fact one of the ones

01:41.615 --> 01:44.595
of the type we used when we replicated this,

01:44.855 --> 01:47.955
but it's almost identical to the ones Sure.

01:47.955 --> 01:51.595
And Reov used in the proceedings paper.

01:52.495 --> 01:56.715
And this paper cup had a tiny little hole,

01:57.495 --> 02:01.555
in, in the base, so that if you inverted the cup

02:01.555 --> 02:03.275
and put it over a pair of dice

02:03.615 --> 02:06.795
and then shook the dice like this, then

02:07.705 --> 02:10.995
peering down into the base through the hole in the base

02:10.995 --> 02:14.395
of the cup, you could see what the total number

02:14.415 --> 02:16.035
of Pips showing was.

02:16.335 --> 02:19.395
If you've got two dice, the lowest you can have is two,

02:19.395 --> 02:21.235
because you have two ones

02:21.575 --> 02:24.475
and the highest is 14 is 12.

02:24.535 --> 02:25.795
If you get two sixes.

02:27.015 --> 02:30.795
And the, average expected from a large number

02:30.795 --> 02:32.035
of people is seven.

02:32.735 --> 02:36.155
So in an experiment you should expect if people are not

02:36.395 --> 02:38.395
cheating, they, they were getting paid for the number

02:38.395 --> 02:39.835
of pips they reported.

02:40.015 --> 02:42.515
By the way, the, the more they reported,

02:42.615 --> 02:43.955
the more money they would get.

02:44.775 --> 02:47.515
If nobody cheated, you would expect an average of seven.

02:47.695 --> 02:50.355
In fact, what turned out to be the case is

02:50.355 --> 02:52.875
that the average was slightly higher than seven.

02:53.015 --> 02:54.595
So a small number, small

02:54.595 --> 02:56.195
but significant number

02:56.195 --> 02:59.155
of the participants must have cheated statistically

02:59.485 --> 03:00.485
Speaking.

03:00.985 --> 03:05.725
And, it turned out that, that in the group

03:05.905 --> 03:08.605
who had won the competition

03:08.825 --> 03:13.465
before the average, that they claimed the number

03:13.465 --> 03:15.145
of pips they claimed and the money they took

03:16.105 --> 03:19.795
from the opponent was higher than the,

03:19.825 --> 03:21.115
from the people who lost.

03:21.535 --> 03:23.435
So they concluded that

03:24.495 --> 03:27.385
winning a competition predicts dishonest behavior.

03:28.755 --> 03:32.375
And this was, very intriguing finding.

03:32.615 --> 03:36.215
I mean, it's in intrinsically interesting,

03:36.475 --> 03:38.215
but for me it was particularly interesting

03:38.415 --> 03:40.015
'cause I knew that there was literature

03:40.305 --> 03:41.975
suggesting exactly the opposite.

03:42.075 --> 03:45.615
Namely that, it's a sense of

03:46.615 --> 03:50.975
entitlement, victim entitlement that causes

03:51.595 --> 03:54.735
losers to, be dishonest.

03:55.105 --> 03:59.055
There was a, for example, a very well cited article

03:59.195 --> 04:02.815
by ZEC in 2010, which had found evidence for that.

04:03.195 --> 04:06.735
And there were a few others. What sure Andre of in the

04:07.685 --> 04:10.115
experiment we are talking about found was

04:10.115 --> 04:13.555
that winners not losers tended to cheat more.

04:13.855 --> 04:17.075
And it, they, they attributed it to a sense of entitlement.

04:17.615 --> 04:21.195
But the original earlier research had suggested

04:21.195 --> 04:24.395
that victim entitlement from losing sort of makes people,

04:25.205 --> 04:26.265
dishonest.

04:26.725 --> 04:31.105
In 2017 sometime Sky Television got in touch with me

04:31.885 --> 04:35.105
and invited me to participate in one

04:35.105 --> 04:37.745
of the programs in the series that they used

04:37.745 --> 04:40.735
to run called Duck Quacks.

04:40.825 --> 04:42.975
Don't echo strange name,

04:43.595 --> 04:45.735
but this is a panel game, a sort

04:45.735 --> 04:48.135
of humorous panel game in which three

04:48.805 --> 04:52.655
celebrities chameleons, used

04:52.655 --> 04:55.135
to discuss interesting facts.

04:55.795 --> 04:58.135
And in some cases they used to do experiments

04:58.195 --> 04:59.775
to check on the interesting facts.

05:00.075 --> 05:02.735
And that's where the duck, that's where the title comes from

05:02.735 --> 05:05.175
because there was a, an interesting fact,

05:05.175 --> 05:07.135
which is in fact a false fact.

05:07.555 --> 05:09.455
The duck quacks don't echo.

05:09.475 --> 05:10.855
And so they discussed it

05:10.915 --> 05:13.175
and did a little sort of experiment.

05:13.395 --> 05:16.375
So they, they, they had discovered this article by

05:17.185 --> 05:18.535
Ritov in the proceedings

05:19.555 --> 05:22.855
and they asked me whether I'd come onto the program just

05:22.855 --> 05:24.655
for a brief talking head.

05:24.855 --> 05:28.495
'cause they wanted a, a buffin to explain what was done

05:28.715 --> 05:31.495
and what the fact was that the,

05:32.045 --> 05:34.095
article had had found.

05:35.115 --> 05:40.095
And, I, I did this and the, I was the talking head

05:40.635 --> 05:41.975
and the program was made,

05:41.975 --> 05:45.255
and it was broadcast in October, 2017.

05:46.155 --> 05:50.135
And the episode was called Winning a Game makes people more

05:50.135 --> 05:54.095
likely to Cheat the Next Time, which is a similar thing to

05:54.095 --> 05:56.055
what Sure OV said.

05:57.195 --> 06:00.615
And this particular episode was presented by the singer

06:00.635 --> 06:04.335
and TV personality, Mel C or Sporty Spice.

06:04.995 --> 06:06.935
And the three participants discussed it

06:06.935 --> 06:08.015
and they decided, yes,

06:08.015 --> 06:10.295
they thought it's quite likely to be true.

06:10.835 --> 06:12.815
And then they did a little experiment,

06:13.205 --> 06:17.575
ludicrous com comedic experiment in which it did seem that,

06:17.575 --> 06:18.815
that this was the case.

06:19.515 --> 06:21.435
Now, by that time

06:22.075 --> 06:24.755
I thought more deeply about this than I thought a really,

06:24.915 --> 06:29.155
a more rigorous test than was published in the original

06:29.155 --> 06:32.355
experiment was needed for three particular reasons.

06:32.525 --> 06:34.825
First of all, the Sure

06:34.865 --> 06:37.665
and Reov experiment was very underpowered,

06:37.725 --> 06:40.425
by which I mean there were far two participants

06:40.525 --> 06:43.065
by modern standards to,

06:43.415 --> 06:46.905
have confidence in the results that were achieved.

06:47.405 --> 06:51.065
For example, they had only 23 participants in the key

06:51.075 --> 06:54.545
experiment, which established the effect that

06:54.565 --> 06:56.425
that's in the title of the article.

06:56.655 --> 06:58.945
They had 46 participants,

06:59.085 --> 07:02.675
but half of them, their data weren't analyzed

07:02.915 --> 07:04.835
'cause they were just the, the opponents

07:04.855 --> 07:07.235
of the participants that were being analyzed.

07:07.615 --> 07:10.755
So 23 is nowadays considered far too small.

07:11.135 --> 07:14.195
So an underpowered study, we need a big study.

07:14.835 --> 07:16.995
Secondly, there was no proper control group.

07:17.665 --> 07:19.195
They had winners and losers,

07:19.215 --> 07:22.835
but what they didn't have is a control group of people who,

07:22.935 --> 07:25.395
who, who, who were neither given the winning

07:25.535 --> 07:27.875
nor the losing manipulation to begin with.

07:28.535 --> 07:32.355
And in proper scientific studies, you need a control group

07:32.815 --> 07:35.155
to, to have a baseline of what it would be.

07:36.075 --> 07:39.215
If, if you're claiming that winners tend to cheat more,

07:39.395 --> 07:40.975
you need a baseline to see

07:41.485 --> 07:44.135
what the level would've been if they hadn't been winners

07:44.135 --> 07:45.295
and hadn't been losers.

07:46.785 --> 07:50.525
And thirdly, there are other possible causes

07:50.525 --> 07:53.965
of cheating apart from sense of entitlement, which was sure

07:53.965 --> 07:57.005
and OVS suggestion, for example, self-confidence.

07:57.365 --> 07:59.805
I mean maybe, maybe if you think you're a winner,

08:00.325 --> 08:02.285
'cause you've just won a competition,

08:02.455 --> 08:04.525
maybe it decreases your self-confidence.

08:05.065 --> 08:08.445
And it's reasonable to assume that people who self-confident

08:09.105 --> 08:10.365
are more likely to cheat.

08:10.385 --> 08:12.485
If they're inclined to cheat, they're more likely

08:12.505 --> 08:14.645
to cheat if they're self-confident.

08:15.585 --> 08:18.765
So, there are other, there are other reasons too,

08:18.785 --> 08:20.965
but we thought a rigorous experiment was needed.

08:21.345 --> 08:24.925
So I got together with my colleagues, briny fr current Ro

08:24.985 --> 08:28.085
and Marty Manola in my department.

08:28.465 --> 08:32.205
And we en enrolled Jeremy Miles of the University

08:32.205 --> 08:36.645
of Southern California as a further co-author.

08:37.145 --> 08:40.085
And we did a big experiment very rigorously.

08:40.705 --> 08:43.285
We did what's called a power analysis, which is

08:43.955 --> 08:46.245
statistical procedure for determining

08:46.545 --> 08:48.125
how many participants you lead

08:48.505 --> 08:51.965
to have really strong confidence in the

08:51.965 --> 08:53.205
results of an experiment.

08:53.545 --> 08:54.845
And it depends on the type

08:54.845 --> 08:56.330
Of of experiment and so on.

08:56.825 --> 08:59.285
And when we did this analysis, it turns out

08:59.285 --> 09:01.725
that we needed at least 250.

09:02.745 --> 09:04.045
So we did an experiment,

09:04.045 --> 09:08.365
which in fact had 259 participants, very high power.

09:09.265 --> 09:11.685
And we did a second experiment online

09:11.795 --> 09:15.525
because of lockdown actually, which came in the middle

09:15.525 --> 09:16.565
of all this.

09:16.945 --> 09:19.765
But that was fine. And I'll describe both.

09:20.625 --> 09:23.605
So the first experiment was supposed to be a replication

09:23.605 --> 09:25.365
of the Sure and reach of experiment,

09:25.365 --> 09:28.005
but with high power, proper control group

09:28.935 --> 09:33.425
and, more ed all round, very,

09:33.425 --> 09:34.465
very carefully done.

09:35.365 --> 09:37.745
Um, the second study, the online study,

09:38.725 --> 09:42.825
we couldn't use people shaking dice under a cup

09:42.825 --> 09:45.705
because online participants mightn't have

09:46.265 --> 09:47.745
suitable cups and dice.

09:48.725 --> 09:50.985
So we did a modification, which is fine

09:50.985 --> 09:53.785
because it's, it's useful to study a phenomenon

09:54.015 --> 09:55.425
with slightly different methods.

09:55.965 --> 09:59.665
We got all the online participants,

09:59.765 --> 10:04.345
we have 275 of them even more than in the laboratory study

10:04.965 --> 10:07.065
to toss a coin five times

10:07.845 --> 10:12.625
and announced how many 10 times, sorry, tosser coin 10 times

10:13.125 --> 10:15.465
and announced how many heads they got.

10:15.965 --> 10:17.025
And they were rewarded

10:17.025 --> 10:20.105
with Amazon vouchers depending on how many they got.

10:21.075 --> 10:25.615
And, the average statistically should be five.

10:26.435 --> 10:30.535
And in fact, as we'll see, there was cheating in both our

10:31.545 --> 10:34.815
laboratory experiment and in our online experiment.

10:35.775 --> 10:37.315
Um, statistically

10:37.315 --> 10:39.355
of course we don't know which people cheated

10:39.355 --> 10:40.675
because that's the whole point.

10:41.495 --> 10:43.875
Um, we gave them a procedure which enabled them

10:43.875 --> 10:46.355
to cheat without ever being identified.

10:46.375 --> 10:48.675
But we can see from the statistical

10:48.675 --> 10:49.915
results that they cheated.

10:50.975 --> 10:55.595
And we also did a structural equation modeling exercise

10:56.015 --> 10:59.795
for the online study in order to find other possible factors

10:59.865 --> 11:01.995
that might be related to cheating,

11:02.245 --> 11:04.035
apart from winning and losing.

11:04.655 --> 11:07.675
So we measured with psychometric scales,

11:07.975 --> 11:12.035
we measured self-confidence luckiness, which is a feeling

11:12.035 --> 11:13.035
of luckiness people.

11:13.185 --> 11:15.995
Some, some, some people feel very lucky.

11:16.095 --> 11:19.675
And all people sometimes feel luckier than at other times.

11:20.295 --> 11:22.915
And feeling of luckiness is like you feel

11:22.915 --> 11:24.195
you're on a winning streak.

11:24.505 --> 11:29.075
This might encourage people to cheat, possibly entitlement.

11:29.205 --> 11:32.955
There are psychometric stu, measures of entitlement

11:32.955 --> 11:37.795
and we got a proper scale, which never, never been,

11:38.155 --> 11:40.155
measured in the, sure and OV study.

11:40.855 --> 11:43.115
And finally, inequality aversion.

11:43.195 --> 11:47.165
This is something, beloved of,

11:47.165 --> 11:51.485
behavioral economists, which is, a finding that

11:52.365 --> 11:53.460
People generally,

11:53.625 --> 11:56.965
but particularly students, are

11:57.795 --> 11:59.965
very keen on fairness.

12:01.285 --> 12:03.945
And they want the results of experiments

12:04.005 --> 12:07.065
to reward people equally wherever possible.

12:07.245 --> 12:09.185
And this is called inequality aversion.

12:09.275 --> 12:12.265
Makes them uncomfortable if they're getting less than the

12:12.345 --> 12:15.945
other person, but it also makes many people uncomfortable if

12:15.945 --> 12:17.305
they're getting more than the other person.

12:17.685 --> 12:19.025
So we measured all these things

12:19.925 --> 12:24.665
and we wrote up a pre-registration as it's called, of,

12:24.685 --> 12:26.225
of both of these experiments.

12:27.115 --> 12:30.895
Uh, so it's a pre-registered study, which then was refered

12:30.895 --> 12:33.535
and, and we sent it to a big Jo journal.

12:34.435 --> 12:39.075
And, when they, finally approved it,

12:39.415 --> 12:42.995
we then went and, and performed both of these experiments.

12:43.695 --> 12:46.395
The sure and re offs experiment was very underpowered

12:46.395 --> 12:50.915
because at the time they did it in 2016, it was

12:50.975 --> 12:53.395
before the issue

12:53.495 --> 12:57.835
of statistical power became such a hot topic in,

12:57.945 --> 12:59.435
psychological research.

13:00.015 --> 13:02.875
And lots of people used to do underpowered experiments

13:02.875 --> 13:04.115
and you don't blame them for that.

13:04.575 --> 13:09.075
But there was a huge amount of literature about power

13:09.665 --> 13:13.685
that was, started, started being,

13:14.505 --> 13:16.325
published around that time.

13:16.745 --> 13:21.365
And from then onwards, studies with such small samples

13:21.465 --> 13:23.565
as the Sure and retail just unpublishable.

13:23.945 --> 13:27.845
So, our study with huge samples,

13:27.935 --> 13:32.365
which were very high powered, they were published

13:33.635 --> 13:37.215
in a single article in 2022 in

13:38.055 --> 13:41.055
a journal called The Royal Society Open Science.

13:42.385 --> 13:46.445
And, we expected,

13:47.585 --> 13:50.925
you know, I'm going to tell you the results now, we expected

13:52.855 --> 13:54.735
probably the losers would cheat more

13:54.735 --> 13:58.535
because the previous literature had shown that although sure

13:58.535 --> 14:00.495
and Rito had found that winners cheat more.

14:01.155 --> 14:04.135
In fact, what we found was that neither the winners

14:04.135 --> 14:06.335
nor the cheaters cheated more.

14:06.475 --> 14:09.375
And remember we had a control group, so we had a group

14:09.375 --> 14:11.215
of people who were neither winners

14:11.235 --> 14:13.335
nor losers to compare them with.

14:14.395 --> 14:17.495
And there was absolutely no difference between any

14:17.495 --> 14:18.775
of these groups in either

14:18.915 --> 14:21.615
of these experiments in our laboratory experiment

14:21.755 --> 14:23.615
or our online experiment.

14:25.035 --> 14:28.015
So it's a good job that we did the structural equation

14:28.335 --> 14:32.015
modeling study with all those psychometric tests because

14:32.015 --> 14:33.655
otherwise we'd have had absolutely

14:33.655 --> 14:35.535
nothing positive to report.

14:35.535 --> 14:38.095
It would've been very interesting to report this

14:38.775 --> 14:43.495
negative result, but we, we, we, we also looked at

14:44.395 --> 14:46.095
all those variables which might,

14:46.625 --> 14:48.575
which we thought might be related

14:48.715 --> 14:50.335
to cheating and tried to see

14:50.765 --> 14:53.925
What the statistical relationship was in our

14:54.025 --> 14:55.125
big online study.

14:56.845 --> 15:00.025
And amazingly only inequality aversion

15:01.015 --> 15:03.145
predicted cheating backwards.

15:03.165 --> 15:07.865
That's to say people who had high inequality aversion were

15:07.865 --> 15:11.345
less likely to cheat, presumably

15:11.375 --> 15:13.905
because they have a strong sense of fairness

15:14.165 --> 15:17.105
and people who have a sense of fairness are unlikely,

15:17.525 --> 15:20.065
you know, are unlikely to want to cheat.

15:21.205 --> 15:25.105
But, self-confidence, which we thought would be related,

15:25.945 --> 15:29.145
absolutely no relationship luckiness feeling lucky,

15:29.985 --> 15:31.145
absolutely no relationship

15:31.805 --> 15:35.105
and entitlement, absolutely no relationship

15:35.195 --> 15:37.805
among the winners, all the losers.

15:38.755 --> 15:42.175
And our study is very large and very rigorous.

15:42.355 --> 15:47.095
So we think, that really is, an important

15:47.805 --> 15:49.015
finding in this area.

15:49.845 --> 15:51.095
It's obviously important

15:51.275 --> 15:53.455
to understand honesty and dishonesty.

15:53.535 --> 15:55.495
I mean, these are socially important things

15:55.495 --> 15:58.895
because particularly nowadays current concerns

15:59.945 --> 16:03.375
about increasing academic dishonesty in the digital age,

16:03.375 --> 16:07.815
it's very difficult to, to do ex to do,

16:08.145 --> 16:09.775
exams with students nowadays

16:09.775 --> 16:13.055
because of the risk of cheating.

16:14.075 --> 16:17.455
Um, there's also, of course increasing problems of tax

16:18.045 --> 16:21.215
avoidance and evasion in some countries in many countries.

16:22.235 --> 16:26.815
And there's widening inequality in many developed countries,

16:27.225 --> 16:28.815
which is affecting corruption

16:28.815 --> 16:31.895
and crime, at both ends of the spectrum.

16:32.595 --> 16:34.415
And so we need to understand cheating

16:34.755 --> 16:37.295
and it's not well understood, surprisingly.

16:38.705 --> 16:40.885
Um, what we've established is clearly

16:40.885 --> 16:43.245
that winning a competition does not

16:43.715 --> 16:45.365
predict dishonest behavior.

16:45.515 --> 16:47.685
Neither, neither does losing. By the way.

16:48.025 --> 16:51.325
The only thing that we've contributed is a very small thing,

16:51.905 --> 16:54.045
but it's significant, namely

16:54.045 --> 16:57.045
that inequality aversion is relevant here.

16:57.265 --> 17:02.015
People who have strong feelings of inequality aversion,

17:02.015 --> 17:04.895
people who have strong senses of fairness, presumably

17:05.675 --> 17:07.255
are much less likely to cheat.

17:07.635 --> 17:09.575
Now this is, perhaps something

17:09.575 --> 17:11.895
that seems obvious in hindsight,

17:12.515 --> 17:15.895
but often, findings are obvious in hindsight.

17:15.895 --> 17:17.895
And it did need to be documented.

17:17.895 --> 17:19.655
And this has been documented,

17:20.035 --> 17:23.015
but of course it needs to be checked by other researchers.

17:23.835 --> 17:26.895
And finally, of course there are lots of other factors

17:27.045 --> 17:30.695
that could be related to, cheating

17:30.885 --> 17:35.055
that could be investigated and we might or might not,

17:35.115 --> 17:36.735
but other people probably will.

17:38.055 --> 17:39.555
And that's about it. Thank you.

00:00 - In 2016, in the proceedings of the National Academy of Sciences, very prestigious American Magazine, an article appeared with the title, winning a Competition Predicts Dishonest Behavior. and the idea in this, in, in this paper was to report an experiment

00:21 - or a series of experiments in which competitive winning tends to make people cheat. According to the authors winners of pair wide com competitions involving skill cause people or, or they're more likely if they've won that competition to steal money in subsequent games of chance with completely different opponents. And the way they did it is, is as follows. First of all, they gave everybody,

00:52 - all the participants in the study a visual test, a very difficult visual test and above all, one that people are completely unfamiliar with. So they wouldn't have any way of judging whether they'd done well or badly in relation to other people. And this enabled the experimenters then to tell half the participants that they'd won by doing better than a particular level

01:17 - and half that they'd, failed by doing worse than a particular level when they were paired with other people. And so they had winners and lu losers artificially created in this way, and they then played a dice game. Now the dice game involved paper cups exactly like this one. This is in fact one of the ones of the type we used when we replicated this, but it's almost identical to the ones Sure.

01:47 - And Reov used in the proceedings paper. And this paper cup had a tiny little hole, in, in the base, so that if you inverted the cup and put it over a pair of dice and then shook the dice like this, then peering down into the base through the hole in the base of the cup, you could see what the total number of Pips showing was. If you've got two dice, the lowest you can have is two, because you have two ones

02:21 - and the highest is 14 is 12. If you get two sixes. And the, average expected from a large number of people is seven. So in an experiment you should expect if people are not cheating, they, they were getting paid for the number of pips they reported. By the way, the, the more they reported, the more money they would get. If nobody cheated, you would expect an average of seven.

02:47 - In fact, what turned out to be the case is that the average was slightly higher than seven. So a small number, small but significant number of the participants must have cheated statistically Speaking. And, it turned out that, that in the group who had won the competition before the average, that they claimed the number of pips they claimed and the money they took

03:16 - from the opponent was higher than the, from the people who lost. So they concluded that winning a competition predicts dishonest behavior. And this was, very intriguing finding. I mean, it's in intrinsically interesting, but for me it was particularly interesting 'cause I knew that there was literature suggesting exactly the opposite. Namely that, it's a sense of

03:46 - entitlement, victim entitlement that causes losers to, be dishonest. There was a, for example, a very well cited article by ZEC in 2010, which had found evidence for that. And there were a few others. What sure Andre of in the experiment we are talking about found was that winners not losers tended to cheat more. And it, they, they attributed it to a sense of entitlement. But the original earlier research had suggested that victim entitlement from losing sort of makes people,

04:25 - dishonest. In 2017 sometime Sky Television got in touch with me and invited me to participate in one of the programs in the series that they used to run called Duck Quacks. Don't echo strange name, but this is a panel game, a sort of humorous panel game in which three celebrities chameleons, used to discuss interesting facts.

04:55 - And in some cases they used to do experiments to check on the interesting facts. And that's where the duck, that's where the title comes from because there was a, an interesting fact, which is in fact a false fact. The duck quacks don't echo. And so they discussed it and did a little sort of experiment. So they, they, they had discovered this article by Ritov in the proceedings

05:19 - and they asked me whether I'd come onto the program just for a brief talking head. 'cause they wanted a, a buffin to explain what was done and what the fact was that the, article had had found. And, I, I did this and the, I was the talking head and the program was made, and it was broadcast in October, 2017. And the episode was called Winning a Game makes people more likely to Cheat the Next Time, which is a similar thing to

05:54 - what Sure OV said. And this particular episode was presented by the singer and TV personality, Mel C or Sporty Spice. And the three participants discussed it and they decided, yes, they thought it's quite likely to be true. And then they did a little experiment, ludicrous com comedic experiment in which it did seem that, that this was the case. Now, by that time

06:22 - I thought more deeply about this than I thought a really, a more rigorous test than was published in the original experiment was needed for three particular reasons. First of all, the Sure and Reov experiment was very underpowered, by which I mean there were far two participants by modern standards to, have confidence in the results that were achieved. For example, they had only 23 participants in the key experiment, which established the effect that

06:54 - that's in the title of the article. They had 46 participants, but half of them, their data weren't analyzed 'cause they were just the, the opponents of the participants that were being analyzed. So 23 is nowadays considered far too small. So an underpowered study, we need a big study. Secondly, there was no proper control group. They had winners and losers, but what they didn't have is a control group of people who,

07:22 - who, who, who were neither given the winning nor the losing manipulation to begin with. And in proper scientific studies, you need a control group to, to have a baseline of what it would be. If, if you're claiming that winners tend to cheat more, you need a baseline to see what the level would've been if they hadn't been winners and hadn't been losers. And thirdly, there are other possible causes of cheating apart from sense of entitlement, which was sure

07:53 - and OVS suggestion, for example, self-confidence. I mean maybe, maybe if you think you're a winner, 'cause you've just won a competition, maybe it decreases your self-confidence. And it's reasonable to assume that people who self-confident are more likely to cheat. If they're inclined to cheat, they're more likely to cheat if they're self-confident. So, there are other, there are other reasons too, but we thought a rigorous experiment was needed.

08:21 - So I got together with my colleagues, briny fr current Ro and Marty Manola in my department. And we en enrolled Jeremy Miles of the University of Southern California as a further co-author. And we did a big experiment very rigorously. We did what's called a power analysis, which is statistical procedure for determining how many participants you lead to have really strong confidence in the results of an experiment.

08:53 - And it depends on the type Of of experiment and so on. And when we did this analysis, it turns out that we needed at least 250. So we did an experiment, which in fact had 259 participants, very high power. And we did a second experiment online because of lockdown actually, which came in the middle of all this. But that was fine. And I'll describe both.

09:20 - So the first experiment was supposed to be a replication of the Sure and reach of experiment, but with high power, proper control group and, more ed all round, very, very carefully done. the second study, the online study, we couldn't use people shaking dice under a cup because online participants mightn't have suitable cups and dice. So we did a modification, which is fine

09:50 - because it's, it's useful to study a phenomenon with slightly different methods. We got all the online participants, we have 275 of them even more than in the laboratory study to toss a coin five times and announced how many 10 times, sorry, tosser coin 10 times and announced how many heads they got. And they were rewarded with Amazon vouchers depending on how many they got. And, the average statistically should be five.

10:26 - And in fact, as we'll see, there was cheating in both our laboratory experiment and in our online experiment. statistically of course we don't know which people cheated because that's the whole point. we gave them a procedure which enabled them to cheat without ever being identified. But we can see from the statistical results that they cheated. And we also did a structural equation modeling exercise

10:56 - for the online study in order to find other possible factors that might be related to cheating, apart from winning and losing. So we measured with psychometric scales, we measured self-confidence luckiness, which is a feeling of luckiness people. Some, some, some people feel very lucky. And all people sometimes feel luckier than at other times. And feeling of luckiness is like you feel you're on a winning streak.

11:24 - This might encourage people to cheat, possibly entitlement. There are psychometric stu, measures of entitlement and we got a proper scale, which never, never been, measured in the, sure and OV study. And finally, inequality aversion. This is something, beloved of, behavioral economists, which is, a finding that People generally, but particularly students, are very keen on fairness.

12:01 - And they want the results of experiments to reward people equally wherever possible. And this is called inequality aversion. Makes them uncomfortable if they're getting less than the other person, but it also makes many people uncomfortable if they're getting more than the other person. So we measured all these things and we wrote up a pre-registration as it's called, of, of both of these experiments. so it's a pre-registered study, which then was refered

12:30 - and, and we sent it to a big Jo journal. And, when they, finally approved it, we then went and, and performed both of these experiments. The sure and re offs experiment was very underpowered because at the time they did it in 2016, it was before the issue of statistical power became such a hot topic in, psychological research. And lots of people used to do underpowered experiments and you don't blame them for that.

13:04 - But there was a huge amount of literature about power that was, started, started being, published around that time. And from then onwards, studies with such small samples as the Sure and retail just unpublishable. So, our study with huge samples, which were very high powered, they were published in a single article in 2022 in a journal called The Royal Society Open Science. And, we expected,

13:47 - you know, I'm going to tell you the results now, we expected probably the losers would cheat more because the previous literature had shown that although sure and Rito had found that winners cheat more. In fact, what we found was that neither the winners nor the cheaters cheated more. And remember we had a control group, so we had a group of people who were neither winners nor losers to compare them with. And there was absolutely no difference between any

14:17 - of these groups in either of these experiments in our laboratory experiment or our online experiment. So it's a good job that we did the structural equation modeling study with all those psychometric tests because otherwise we'd have had absolutely nothing positive to report. It would've been very interesting to report this negative result, but we, we, we, we also looked at all those variables which might,

14:46 - which we thought might be related to cheating and tried to see What the statistical relationship was in our big online study. And amazingly only inequality aversion predicted cheating backwards. That's to say people who had high inequality aversion were less likely to cheat, presumably because they have a strong sense of fairness and people who have a sense of fairness are unlikely,

15:17 - you know, are unlikely to want to cheat. But, self-confidence, which we thought would be related, absolutely no relationship luckiness feeling lucky, absolutely no relationship and entitlement, absolutely no relationship among the winners, all the losers. And our study is very large and very rigorous. So we think, that really is, an important finding in this area. It's obviously important

15:51 - to understand honesty and dishonesty. I mean, these are socially important things because particularly nowadays current concerns about increasing academic dishonesty in the digital age, it's very difficult to, to do ex to do, exams with students nowadays because of the risk of cheating. there's also, of course increasing problems of tax avoidance and evasion in some countries in many countries. And there's widening inequality in many developed countries,

16:27 - which is affecting corruption and crime, at both ends of the spectrum. And so we need to understand cheating and it's not well understood, surprisingly. what we've established is clearly that winning a competition does not predict dishonest behavior. Neither, neither does losing. By the way. The only thing that we've contributed is a very small thing, but it's significant, namely

16:54 - that inequality aversion is relevant here. People who have strong feelings of inequality aversion, people who have strong senses of fairness, presumably are much less likely to cheat. Now this is, perhaps something that seems obvious in hindsight, but often, findings are obvious in hindsight. And it did need to be documented. And this has been documented, but of course it needs to be checked by other researchers.

17:23 - And finally, of course there are lots of other factors that could be related to, cheating that could be investigated and we might or might not, but other people probably will. And that's about it. Thank you.

Faculti AI can make mistakes