Cognitive load promotes honesty

Abstract

In three experiments, we examined the cognitive underpinnings of self-serving dishonesty by manipulating cognitive load under different incentive structures. Participants could increase a financial bonus by misreporting outcomes of private die rolls without any risk of detection. At the same time, they had to remember letter strings of varying length. If honesty is the automatic response tendency and dishonesty is cognitively demanding, lying behavior should be less evident under high cognitive load. This hypothesis was supported by the outcome of two out of three experiments. We further manipulated whether all trials or only one random trial determined payoff to modulate reward adaptation over time (Experiment 2) and whether payoff was framed as a financial gain or loss (Experiment 3). The payoff scheme of one random or all trials did not affect lying behavior and, discordant to earlier research, facing losses instead of gains did not increase lying behavior. Finally, cognitive load and incentive frame interacted significantly, but contrary to our assumption gains increased lying under low cognitive load. While the impact of cognitive load on dishonesty appears to be comparably robust, motivational influences seem to be more elusive than commonly assumed in current theorizing.

Fig. 1

Mean reported outcome for each letter string condition in Experiment 1. The bold horizontal line depicts the chance value of 3.5. Error bars represent the 95% confidence intervals of the individual means (CI_M)

Fig. 2

A Mean reported outcome for each string length in Experiment 2. The bold horizontal line depicts the chance value of 3.5. Error bars represent confidence intervals of the individual means (CI_M). B Mean reported outcome for each combination of incentive structure and block in Experiment 2. The bold horizontal line depicts the chance value of 3.5 and the bold numbers indicate the block within incentive structure. Error bars represent the confidence intervals for paired differences, calculated separately for each incentive structure (CI_PD; Pfister & Janczyk, 2013)

Fig. 3

Mean transformed reported outcome for each letter string condition and incentive structure in Experiment 3. The bold horizontal line depicts the chance value of 0. Error bars represent the 95% confidence intervals of the individual means (CI_M)

Fig. 4

Mean reported outcome for each combination of letter string length, incentive structure and block in Experiment 2. The bold horizontal line depicts the expected value of 3.5 and the bold numbers represent the block within incentive structure. Error bars represent the confidence intervals of the paired differences calculated separately for each combination of letter string length and incentive structure (CI_PD)

Fig. 5

Mean reported outcome for each combination of letter string length, incentive structure and block in Experiment 3. The bold horizontal line depicts the expected value of 0 and the bold numbers represent the block within incentive structure. Error bars represent the confidence intervals of the paired differences calculated separately for each combination of letter string condition and incentive structure (CI_PD)

Fig. 6

Root mean square error for our approximation of the probability distribution for the sum of up to 84 die reports. As a benchmark, we used exact probabilities computed in R 4.0.3 (R Core Team, 2018) using the dice package (v1.2; Arena, 2014). We did not make computations for higher numbers of die reports because computation for exact probabilities of 84 rolls already took excessive computation time. In our studies, the number of die reports was 90 (Experiment 1), 180 (Experiment 2) and 120 (Experiment 3)

Fig. 7

Distribution of (transformed) reported outcomes for all experiments. Each circle represents one participant. The curve at the right side of each plot shows the normal distribution of reported outcomes for this experiment. Non-outliers are shown in black, outliers (sum of reported outcomes exceeded the 97.5%-percentile or fell below the 2.5%-percentile of this normal distribution) are shown in light gray. The upper and lower horizontal lines represent the corresponding cut off values and the middle horizontal lines represent the expected values