Can Honesty Oaths, Peer Interaction, or Monitoring Mitigate Lying?

Abstract

We introduce several new variants of the dice experiment by Fischbacher and Föllmi-Heusi (Journal of the European Economic Association 11(3):525–547, 2013) to investigate measures to reduce lying. Hypotheses on the relative performance of these treatments are derived from a straightforward theoretical model. In line with previous research, we find that groups of two subjects lied at least to the same extent as individuals—even in a novel treatment where we assigned to one subject the role of being the other’s monitor. However, we find that our participants hardly lied if they do not benefit and only others do, even if they were in a reciprocal relationship. Thus, we conclude that collaboration on lying mostly happens for personal gain. To mitigate selfish lying, an honesty oath which aims to increase moral awareness turned out to be effective.

Appendix

Factor Approximation in Our Theoretical Model of the Utility of Lying

In our “Results” section we found evidence to confirm some of the major predictions we made for behavioral differences between treatments based on our theoretical model of the utility of lying. Ex post, we can use the model for a more detailed interpretation of our findings. Thus, we estimate the most essential key factors of the model by using the observed mean values of payoffs per treatment.

In each treatment, the mean value of payoffs under complete honesty ($\bar {r}$) is expected to approximate €2.50. We assume: $\bar {r}={\sf C}\!\!\!\!\raise.8pt\hbox{=}\,2.50$ for each treatment.

Baseline

The observed mean value of payoffs in the baseline treatment is ${\bar {p}_{{\text{baseline}}}}={\sf C}\!\!\!\!\raise.8pt\hbox{=}\,3.66$. Assuming that on average all players engaged in the optimal amount of lying, we get:

$${\left( {\frac{1}{{2{{\bar {\delta }}_{{\text{baseline}}}}}}} \right)^2} - \bar {r}={\bar {p}_{{\text{baseline}}}} - \bar {r}$$

with ${\bar {\delta }_{{\text{baseline}}}}$ being the average dislike of lying for oneself in the baseline treatment.

Solving this equation for ${\bar {\delta }_{{\text{baseline}}}}$, we get:

$${\bar {\delta }_{{\text{baseline}}}}=\frac{1}{{2\sqrt {{{\bar {p}}_{{\text{baseline}}}}} }}=0.26$$

Moral Awareness

The observed mean value of payoffs in our moral awareness treatment is ${\bar {p}_{{\text{moral}}}}={\sf C}\!\!\!\!\raise.8pt\hbox{=}\,2.66$. Again, assuming that on average all players engaged in the optimal amount of lying, we get:

$${\left( {\frac{1}{{2{{\bar {\alpha }}_{{\text{moral}}}}{{\bar {\delta }}_{{\text{moral}}}}}}} \right)^2} - \bar {r}={\bar {p}_{{\text{moral}}}} - \bar {r}$$

with ${\bar {\delta }_{{\text{moral}}}}$ being the average dislike of lying for oneself, and ${\bar {\alpha }_{{\text{moral}}}}$ being the factor to which the honesty oath increased moral awareness on average compared to the baseline treatment.

Solving this equation for ${\bar {\alpha }_{{\text{moral}}}}$, we get:

$${\bar {\alpha }_{{\text{moral}}}}=\frac{1}{{2{{\bar {\delta }}_{{\text{moral}}}}\sqrt {{{\bar {p}}_{{\text{moral}}}}} }}$$

Since we already capture the difference in moral awareness between Baseline and Moral Awareness by using the factor ${\bar {\alpha }_{{\text{moral}}}}$, we assume that the average dislike of lying for oneself between those two treatments does not change (${\bar {\delta }_{{\text{moral}}}}={\bar {\delta }_{{\text{baseline}}}}$).

Now we can calculate ${\bar {\alpha }_{{\text{moral}}}}$:

$${\bar {\alpha }_{{\text{normal}}}}=\frac{1}{{2{{\bar {\delta }}_{{\text{baseline}}}}\sqrt {{{\bar {p}}_{{\text{moral}}}}} }}=1.17$$

The interpretation of this factor is that the honesty oath increased the average of moral awareness by about 17%.

Monitoring

The observed mean value of payoffs in the monitoring treatments is ${\bar {p}_{{\text{monitor}}}}={\sf C}\!\!\!\!\raise.8pt\hbox{=}\,3.78$. Again, assuming that on average all players engaged in the optimal amount of lying, we get:

$${\left( {\frac{1}{{{{\bar {\alpha }}_{{\text{monitor}}}}{{\bar {\delta }}_{{\text{monitor}}}}}}} \right)^2} - \bar {r}={\bar {p}_{{\text{monitor}}}} - \bar {r}$$

with ${\bar {\delta }_{{\text{monitor}}}}$ being the average dislike of lying for oneself, and ${\bar {\alpha }_{{\text{monitor}}}}$ being the factor to which monitoring increased moral awareness on average compared to the baseline treatment.

Solving this equation for ${\bar {\delta }_{{\text{monitor}}}}$, we get:

$${\bar {\delta }_{{\text{monitor}}}}=\frac{1}{{{{\bar {\alpha }}_{{\text{monitor}}}}\sqrt {{{\bar {p}}_{{\text{monitor}}}}} }}$$

Furthermore, we reason that $1 \leq {\bar {\alpha }_{{\text{monitor}}}} \leq {\bar {\alpha }_{{\text{moral}}}}$. With this, we can estimate ${\bar {\delta }_{{\text{monitor}}}}$:

$$0.44=\frac{1}{{{{\bar {\alpha }}_{{\text{moral}}}}\sqrt {{{\bar {p}}_{{\text{monitor}}}}} }} \leq {\bar {\delta }_{{\text{monitor}}}} \leq \frac{1}{{\sqrt {{{\bar {p}}_{{\text{monitor}}}}} }}=0.51$$

Comparing this to the baseline treatment, we get:

$$1.69 \times {\bar {\delta }_{{\text{baseline}}}}{{{\text{}}}} \leq {\bar {\delta }_{{\text{monitor}}}} \leq 1.98 \times {\bar {\delta }_{{\text{baseline}}}}$$

This means that the average dislike of lying for oneself increased by at least 69% due to the presence of a monitor (while already considering possible effects of an increased moral awareness due to monitoring). Both effects, however, are narrowly overshadowed by the division of moral costs between both players (n = 2), since:

$$\frac{{{{\bar {\alpha }}_{{\text{monitor}}}}{{\bar {\delta }}_{{\text{monitor}}}}}}{n}=\frac{{1.98 \times {{\bar {\delta }}_{{\text{baseline}}}}}}{2}<{\bar {\delta }_{{\text{baseline}}}}$$

Reciprocity

Since we find no significant difference between both reciprocity treatments, we define ${\bar {p}_{{\text{reci}}}}={\sf C}\!\!\!\!\raise.8pt\hbox{=}\,2.89$ as the observed mean value of payoffs in the pooled reciprocity treatments.

Again, assuming that on average all players engaged in the optimal amount of lying, we get:

$${\left( {\frac{{{{\bar {\beta }}_{{\text{reci}}}}}}{{2{{\bar {\delta }}_{{\text{reci}}}}}}} \right)^2} - \bar {r}={\bar {p}_{{\text{reci}}}} - \bar {r}$$

with ${\bar {\delta }_{{\text{reci}}}}$ being the average dislike of lying for another person, and ${\bar {\beta }_{{\text{reci}}}}$ being the factor to which players care about others on average.

Solving this equation for ${\bar {\beta }_{{\text{reci}}}}$, we get:

$${\bar {\beta }_{{\text{reci}}}}=2{\bar {\delta }_{{\text{reci}}}}\sqrt {{{\bar {p}}_{{\text{reci}}}}}$$

(3)

Group

The observed mean value of payoffs in the group treatment is ${\bar {p}_{{\text{group}}}}={\sf C}\!\!\!\!\raise.8pt\hbox{=}\,4.14$. Again, assuming that on average all players engaged in the optimal amount of lying, we get:

$${\left( {\frac{{1+{{\bar {\beta }}_{{\text{group}}}}}}{{{{\bar {\delta }}_{{\text{group}};i}}+{{\bar {\delta }}_{{\text{group}};j}}}}} \right)^2} - \bar {r}={\bar {p}_{{\text{group}}}} - \bar {r}$$

with ${\bar {\delta }_{{\text{group}};i}}$ being the average dislike of lying for oneself, ${\bar {\delta }_{{\text{group}};j}}$ being the average dislike of lying for another person, and ${\bar {\beta }_{{\text{group}}}}$ being the factor to which players care about others on average.

Solving this equation for ${\bar {\beta }_{{\text{group}}}}$, we get:

$${\bar {\beta }_{{\text{group}}}}=\left( {{{\bar {\delta }}_{{\text{group}};i}}+{{\bar {\delta }}_{{\text{group}};j}}} \right)\sqrt {{{\bar {p}}_{{\text{group}}}}} - 1$$

(4)

Furthermore, we can assume that the factor to which players care about others and their dislike of lying for another person do not change between treatments ($\bar {\beta }:={\bar {\beta }_{{\text{reci}}}}={\bar {\beta }_{{\text{group}}}}$ and ${\bar {\delta }_j}:={\bar {\delta }_{{\text{reci}}}}={\bar {\delta }_{{\text{group}};j}}$), since we already consider the number of other players affected by lying (m) and the number of players participating in the decision (n) separately.

This implies that we can equate (3) with (4):

$$2{\bar {\delta }_j}\sqrt {{{\bar {p}}_{{\text{reci}}}}} =\left( {{{\bar {\delta }}_{{\text{group}};i}}+{{\bar {\delta }}_j}} \right)\sqrt {{{\bar {\delta }}_{{\text{group}}}}} - 1$$

Since we argue that the dislike of lying for oneself is higher than the dislike of lying for another person (${\bar {\delta }_{{\text{group}};i}} \geq {\bar {\delta }_j}$), we can reason that:

$$2{\bar {\delta }_j}\sqrt {{{\bar {p}}_{{\text{reci}}}}} \leq 2{\bar {\delta }_{{\text{group}};i}}\sqrt {{{\bar {p}}_{{\text{group}}}}} - 1$$

Solving this inequality for ${\bar {\delta }_j}$, we get:

$${\bar {\delta }_j} \leq {\bar {\delta }_{{\text{group}};i}} \times \sqrt {\frac{{{{\bar {p}}_{{\text{group}}}}}}{{{{\bar {p}}_{{\text{reci}}}}}}} - \frac{1}{{2\sqrt {{{\bar {p}}_{{\text{reci}}}}} }}$$

Here we can only guess ${\bar {\delta }_{{\text{group}};i}}$. However, if we approximate the dislike of lying for oneself in the group treatment with the corresponding value from the baseline treatment (${\bar {\delta }_{{\text{group}};i}}\approx{\bar {\delta }_{{\text{baseline}}}}$) we get:

$${\bar {\delta }_j} \leq {\bar {\delta }_{{\text{baseline}}}} \times \sqrt {\frac{{{{\bar {p}}_{{\text{group}}}}}}{{{{\bar {p}}_{{\text{reci}}}}}}} - \frac{1}{{2\sqrt {{{\bar {p}}_{{\text{reci}}}}} }}=0.26 \quad \times \sqrt {\frac{{{{\bar {p}}_{{\text{group}}}}}}{{{{\bar {p}}_{{\text{reci}}}}}}} - \frac{1}{{2\sqrt {{{\bar {p}}_{{\text{reci}}}}} }}=0.017$$

If this approximation is somewhat correct, this inequality has a very intuitive interpretation: The dislike of lying for another person was extraordinary low (the dislike of lying for oneself was at least 15 times higher than the dislike of lying for another person).

Furthermore, this indicates that the average factor to which players care about others ($\bar {\beta }$) was low as well, since:

$$\bar {\beta }={\bar {\beta }_{{\text{reci}}}}=2{\bar {\delta }_{{\text{reci}}}}\sqrt {{{\bar {p}}_{{\text{reci}}}}} =2{\bar {\delta }_j}\sqrt {{{\bar {p}}_{{\text{reci}}}}} =3.4 \times {\bar {\delta }_j} \leq 3.4 \times 0.017=0.058$$

Summary

Following our theoretical model of the utility of lying we can interpret our findings in more detail:

• The honesty oath increased the moral awareness of our participants by about 17% (≈ 20%) on average.

• The dislike of lying for oneself was at least 15 times higher than the dislike of lying for another person.

• This was due to the fact that the dislike of lying for another person was extraordinarily low.

• The degree to which players care about others in the experiment was relatively low.

• Monitoring increased the dislike of lying for oneself by at least 69% (≈ 70%) (already considering possible effects of increased moral awareness due to monitoring). However, these two monitoring effects combined were still narrowly overshadowed by the division of moral costs between both players.