Preferences for Truth-telling

The decision whether to lie or to tell the truth is at the heart of many everyday activities. Just imagine a self-employed shopkeeper reporting her income to the tax authorities or an applicant describing his skills in a job interview. It is very difficult to find out how much people lie in such situations, since the truth is not known. Recently, laboratory experiments have been used by researchers in economics, psychology and sociology to find out.

This website documents what people do in one particular experimental paradigm, introduced by Fischbacher and Föllmi-Heusi (2013): participants privately roll a die or flip a coin such that the outcome is only observable to them. Participants then report the outcome to the researcher, who pays them a monetary payoff equal to the number they report.

In this experiment, participants have a clear monetary incentive to lie up and no individual report can be identified as truthful or not. At the same time, the researcher can judge the reports of a group of participants and can thus draw indirect conclusions about participants' behavior.

The data on this website come from our research paper (Abeler, Nosenzo, Raymond, published in Econometrica 2019). In this paper we combine data from 429 treatments (or experiments) across 90 papers involving more than 44000 participants across 47 countries. We find the following:

Finding 1: Participants forgo a large part of the potential gains from lying.

In Figure 1 each circle shows the average report of one treatment. If a circle is further up, it means that the participants in this experiment reported higher outcomes. The size of the circle indicates the number of participants in that experiment. You can get more information about the experiments by hovering over the graph or by clicking on a circle.

Due to the amount of data displayed, the graphs can take a little time to load.

Since some studies use dice-rolling and others coin-flipping, we need to make the studies comparable. We thus map all reports into a “standardized report” which is -1 if a participant reports the lowest-payoff outcome (e.g., 1 on a die) and +1 if they report the highest-payoff outcome (e.g., 6 on a die). If everyone reported truthfully, the standardized report would be 0. This standardized report is shown on the vertical axis of Figure 1. On the horizontal axis, the maximal monetary gains from lying are shown.

If participants had no concerns about lying, they would all report the maximal outcome and all circles would be at the top of the graph, at +1. In contrast, we find that the average standardized report is only 0.234. This means that participants forego about three-quarters of the potential gains from lying. This shows that people have a strong aversion to lying, even if this means that they have to leave quite a bit of money on the table. We discuss the possible motivations driving this behavior here.

Figure 1 also shows that participants continue to refrain from lying maximally when the monetary incentives to lie are increased. From left to right in the figure, the potential payoff from misreporting ranges from a few cents to 50 USD, i.e., a 500-fold increase. The reporting behavior, however, does not change. This means that participants lie roughly similarly when they can obtain small or large amounts of money from lying.

The second tab of figure 1 shows that also the distribution of reports is unaffected when the monetary incentives to lie are increased.

Finding 2: Some non-maximal-payoff states are reported more often than their true likelihood.

Figure 2 shows the distribution of reports for all experiments using six-sided die rolls. You can again get more information by clicking on the graph or using the filters.

Each line corresponds to one experiment. The higher the circle, the more participants make this report. If the line is at zero, it means that no participant made this report. The dashed line indicates the truthful distribution: each outcome occurs with 1/6 chance. As you can see, almost no line is at zero. This means that all possible reports are made in almost all experiments – even the reports that lead to very low monetary payoff.

Interestingly, some reports that do not yield the maximal payoff are reported more often than their truthful probability of 1/6. Take the second highest report, i.e., the second circle from the right. This would be number 5 in experiments where participants roll a six-sided die. This report is above the truthfulness line for almost all experiments. This means that some participants lie – but not all the way to the maximal outcome. There are many different reasons why this could be. Participants might think that reporting a 6 on a die could lead to suspicion, and so report a 5 instead. Or maybe participants didn’t want to lie too far away from their true number.

The other tabs of Figure 2 show the distributions of coin flips and other experiments with uniform distributions of 3 or 10 potential outcomes.

Finding 3: Men report higher numbers than women.

Figure 3 shows the effect of gender. The graph shows the distribution by gender of all die-rolling experiments. Men are generally less likely to report lower outcomes and more likely to report higher outcomes.

The second tab of Figure 3 shows data across studies. On the horizontal axis is the average report of the male participants and on the vertical axis is the average report of the female participants. Each circle thus represents one experiment. If a circle is below the 45° line, it means that men reported higher numbers than women in a given experiment. The majority of circles is below the 45° line, indicating that female participants report lower numbers than male participants. On average, women report 0.06 less than men (compared to an overall average of 0.234). More details on the gender effect can be found in Appendix A of our research paper.

Finding 4: Behavior is surprisingly similar across countries.

Figure 4 plots the average reports by country. The world map shows that this kind of experiment has been conducted around the world. The darker the color of a country is, the higher the average report in this country. The second tab of Figure 4 shows more details. The country average is marked by a cross. In no country do all participants lie maximally. The standardized report is maximally around 0.5. This means that even in the countries with the biggest amount of lying, people still forgo half of the potential gains from lying.

Finding 5: Behavior does not change if participants make many reports repeatedly.

In Figure 5, the horizontal axis shows how often participants have to report an outcome and how the reports of participants evolve over the course of the experiment. In most experiments, participants report their die roll or cointoss only once, these are all the circles at “1”. But in many experiments, participants make a report repeatedly, up to 60 times. The average behavior does not change with repetition of the task. This means that learning about the experiment and experience with reporting do not affect how people report.

Finding 6: In repeated experiments, only very few participants always report the payoff-maximzing outcome.

So far, we could not judge the behavior of any individual participant. If they report a 6 on a die, maybe they are indeed honest and just got lucky. However, if one participants reports many times, it is increasingly unlikely that they got "lucky" every time. For repeated experiments, we can thus judge the behavior of individual participants to some extent.

In Figure 6 we focus on experiments in which participants repeatedly report the outcome of a coin flip. We add up the number of times a participant reported the winning side of the coin. To make experiments with different numbers of rounds comparable, on the horizontal axis we plot the share of the potential high-payoff reports made by a subject. For example, if a participant never reported the winning side of the coin, they would be at the very left end of the horizontal axis. If they reported the winning side of the coin in half of the rounds, they would be in the middle of the horizontal axis; and so on. On the vertical axis, we show the difference between the observed distribution and the truthful distribution. Circles above the dashed line mean that "too many" subjects report the winning side this often. If a participant had no concern about lying or about being seen as a liar, they should always report the winning side of the coin. They would thus be at the very right end of the horizontal axis and that circle should be above the dashed line. It is extremely unlikely that such a reporting pattern could have resulted from truth-telling and pure luck: for some of the experiments shown, the chance of always tossing the winning side is less than 0.0000000001%.

As one can see in the figure, some participants do always report the high-payoff state: the circles at the very right of the graph are above the truthfulness line. Overall however, this is done by fewer than 1 in 20 participants. The size of the circles again indicates the number of participants. The vast majority of circle area is in the middle of the graph: participants report a little more than their true number or are honest and only very few participants always (or even very often) report the winning side.