Participants
We ran simulations to determine the target sample size and statistical power for our study. Finance constraints allowed us to test 150 participants. And we expected the effect of the drugs on dishonest behavior to be around 6% since Cohn et al. show that reminding prison inmates of their criminal identity increased success rates by 6% percentage points (Cohn et al. 2015). Hence, we aimed to increase statistical power to an optimum by increasing the trials on the die rolling task (i.e., reducing the measurement error, Blake and Gangestad 2020). To estimate statistical power, we performed 10,000 simulations of placebo samples, each consisting of 50 subjects rolling the die multiple times. We assumed that each die roll follows a binomial process with a success rate of 50%. Similarly, we simulated 10,000 samples for methylphenidate and atomoxetine, with 50 subjects per sample, assuming success rates ranging from 50% to 70% in 1% increments. Using rank-sum tests, we compared each pair of drug and placebo samples to determine if the difference in successful die rolls is statistically significant. The proportions of hypotheses rejected at a two-sided 5% significance level provided estimates of statistical power for different numbers of die rolls. The simulation results indicate that a sample size of 50 subjects per treatment group is sufficient to detect an increase or decrease in success rates of 6% points with 87% or 84% power, respectively, if they roll the die 25 times.
Participants (N = 156) were randomly assigned to one of three conditions: placebo (n = 53), methylphenidate (n = 50), or atomoxetine (n = 53). Four participants discontinued the study due to experiencing nausea (one in the methylphenidate condition, three in the atomoxetine condition) and an additional participant was excluded due to a computer failure (placebo condition). This resulted in a final sample of 151 participants. We did not find differences between conditions for age, gender, income, and education (see Table S1). The study had a double-blind design such that neither participants nor experimenters knew in which condition the participant was.
Before participating in the study, people interested in participating completed a survey that assessed potential exclusion factors. People were excluded from participating if they self-reported having a diagnosis of ADHD or ADD, or if they reported currently using or having a history of using ADHD drugs or any other potentially cognitive-enhancing substances. Furthermore, people were also excluded if their score on the Beck Depression Inventory – II (Beck et al. 1996) exceeded 12, if they had prior experience with experiments involving similar tasks, if they had a history of illicit drug use, if they were taking specific medications such as blood thinners, epinephrine medications for high blood pressure, diet pills, antidepressants or anti-seizure medications, if they smoked more than five cigarettes per day, if their answers on the Alcohol Use Disorders Identification Test (Reinert and Allen 2007) indicated hazardous drinking (score of eight or higher), or if they had a history of mental illness. Participants were also instructed to refrain from excessive caffeine consumption on the morning of the experiment, and to abstain from alcohol consumption the night before.
Procedure
On the day of the experiment, participants were again provided with the participant information sheet to ensure their understanding of the study’s procedures and risks, after which they signed the consent form. Subsequently, various physiological measurements were taken, including blood pressure, heart rate, weight, height, mood, and attention. Following this, the drug administration took place under the supervision of a medical attendant. Participants received three pills containing either 30 mg of methylphenidate, 60 mg of atomoxetineFootnote 1, or a placebo composed of sugar. The dosage levels were determined based on previous studies that have demonstrated cognitive effects (Chamberlain et al. 2007; Nandam et al. 2014).
After a lapse of eighty minutes from the drug administration, a second round of measurements was conducted, encompassing blood pressure, heart rate, mood, and attention. The battery of incentivized economic tasks was administered ninety minutes post-drug administration. The timing of these tasks was determined in consideration of the timeframe within which peak plasma levels are typically observed for the three drugs (90–180 min) subsequent to oral ingestion in adults (Chamberlain et al. 2006; Nandam et al. 2011, 2014; Hester et al. 2012).
Randomization, blinding and code-breaking procedures
Participants were randomly assigned to one of three groups (methylphenidate, atomoxetine, or placebo) using a pre-generated list. Gender balance was ensured by recruiting an equal number of male and female participants. The randomization lists for men and women were generated using a standard software package for random number generation in MATLAB. To maintain blinding, the researcher responsible for data collection was unaware of the treatment group assigned to each participant. Blinding was achieved by encapsulating the methylphenidate, atomoxetine, and placebo tablets in identical form. The randomization codes and schedule were held by a researcher who was not involved in drug administration or data analysis. At the end of the study, participants were informed whether or not they had received a drug, and if so, which one they were assigned to. There were no significant differences between conditions for education level, χ²(2) = 2.35, p = 0.309, income, χ²(2) = 3.05, p = 0.218, or age, F(2, 146) = 1.34, p = 0.265. (Table S1).
Measures
The instructions for every task can be found at https://osf.io/uv5ke/?view_only=ce44a28f86d543e3a5839b7fda79acb8.
Attention
To measure sustained attention, we employed a Rapid Visual Information Processing task (Sarter et al. 2001). In this task, participants had to detect specific number sequences (e.g., 1, 3, 5) within a rapid succession of single digits. A white square was displayed on the screen, within which from 1 to 9 appeared randomly at a rate of 100 digits per minute. Participants were instructed to identify and respond to target sequences of digits, such as 13 − 5, 2–4-6, 3–5-7, and 4–6-8, by pressing the space key as quickly as possible.
Mood
To assess participants’ mood, we utilized a self-report measure before and 80 min after the drug administration, consisting of 16 items adapted from (Crockett et al. 2015). Participants were asked to indicate their feelings on a slider scale from 0 to 1. The scale included opposing endpoints for each dimension, such as sociable-withdrawn or incompetent-proficient. The 16 items were factor analyzed, resulting in two factors: energy (reflecting feelings of being energetic, alert, quick-witted) and contentment (reflecting feelings of calmness, contentment, and happiness).
Dishonesty
To measure dishonesty, participants rolled a six-sided die for a total of 25 rolls. Participants were informed that they would roll a die 25 times, with a 50% chance of winning 10 tokens (equivalent to £1) and a 50% chance of winning nothing on each roll. Prior to each roll, participants were informed which outcome would yield the 10-token payoff (e.g., rolling an odd or even number). Participants then rolled the die under a cup and reported their outcome on the computer. This task was conducted privately in an isolated room to eliminate concerns about being caught while misreporting outcomes. However, dishonest behavior is detectable at the group level by comparing the percentage of reported successful die rolls to the honesty benchmark of 50% in each of the three conditions.
Participants entered the outcomes of each roll into the computer interface. In each round, half of the rolled numbers (e.g., 1, 2, 3) resulted in a payoff of 10 tokens (£1), while the remaining numbers yielded no payoff. Prior to each round, participants were informed on the computer screen which numbers would result in a payoff. Using a cup, participants physically rolled the die and then entered the resulting outcomes into the computer. To ensure comprehension of the task rules, participants also reported the corresponding payoffs. This task was conducted individually in a separate room, ensuring privacy and absence of observation. If the random computer draw at the end of the experiment indicated that the die rolling task was selected for payment, participants received the earnings accumulated from all 25 rounds.
Risk preferences
In order to assess participants’ risk preferences, we employed a task involving trade-offs between risky and certain payments (Falk, et al., 2016). Participants were presented with two tables, each containing 21 rows. Within each row, participants had to choose between a guaranteed payment and a lottery option that had a 50% chance of yielding 200 Tokens (£20) and a 50% chance of yielding nothing. The lottery option remained the same across all rows in both tables, while the safe payment varied. The safe payment decreased in increments of 10 token (£1) steps from 200 tokens (£20) to 0 token (£0). In the second set of tables, we introduced random variations by adding or subtracting up to 3 tokens to each safe payment option. If this task was randomly selected for payment at the end of the experiment, one of the participants’ choices was randomly selected for payment. The row at which participants switched from preferring the lottery to preferring the safe payment indicates their individual risk preferences.
Material self-interest
Each participant was given an endowment of 200 tokens (£20) and had to make a decision regarding the amount they wished to donate to different charities: Red Cross, UNICEF, and Doctors Without Borders. The order of presentation for these options was randomized to mitigate potential order effects.
Impulsivity
Participants made a series of binary choices, where they had to decide between receiving a specified number of tokens at a later date or obtaining an equal or smaller amount (e.g., 120 tokens or £12) at an earlier date. We implemented three different scenarios: “today vs. in 3 months”, “today vs. in 6 months”, and “in 3 months vs. in 6 months.” Within each scenario, participants were given a list with 25 choice situations involving the delayed payment and an earlier payment. The first row of the list featured an earlier payment amount equivalent to the delayed payment (e.g., 120 token). In subsequent rows, the amount of the delayed payment increased by 5 tokens, with slight variations of plus or minus one or two tokens. Participants’ level of impulsivity was determined by the average point at which they switched from selecting the earlier payment to opting for the delayed payment across all three scenarios. If this task was randomly selected for payment, the computer randomly chose one row within a scenario. If participants chose an “today” payment option in that row, they received the corresponding amount immediately after the testing session. Alternatively, if they chose the delayed payment (i.e., in 3 months or in 6 months), they could choose between receiving the amount by mail or picking it up in person.
Drug administration beliefs
Following the conclusion of the experiment, participants were presented with a series of questions aimed at gathering insights into their experiences during the study, as well as their beliefs regarding the drug manipulation and the tasks. Specifically, participants were asked whether they believed they had received a drug or a placebo, and if they believed they had received a drug, they were further queried about their perception of which specific drug they had received.
Data analysis
To examine the effects of methylphenidate and atomoxetine on cognitive performance and dishonest behavior, we analyzed data from two primary tasks: a sustained attention task and a die-rolling task. The dependent variables in the sustained attention task were participants’ reaction times (RTs) and accuracy in detecting numerical sequences. These outcomes were analyzed using repeated-measures ANOVAs to test for interaction effects between drug condition (methylphenidate, atomoxetine, placebo) and time (pre- vs. post-dosing). Follow-up pairwise comparisons were conducted using ANOVAs or t-tests, where appropriate, to explore specific differences between conditions. Dishonest behavior was assessed using a die-rolling task, where participants self-reported the outcomes of private dice rolls. The primary dependent variable was the proportion of reported successful rolls, with higher proportions indicating greater dishonesty. We first compared observed success rates to the 50% benchmark for honest reporting using one-sample t-tests. Differences in dishonesty across drug conditions were assessed using non-parametric Kruskal-Wallis tests, followed by Bonferroni-adjusted pairwise comparisons using Mann-Whitney U tests. Additional analyses used simulation-based approaches and Bayesian methods to evaluate whether observed differences could plausibly arise from random variation alone. Highest Density Intervals (HDIs) were computed to determine whether observed distributions were consistent with honest reporting.
To assess alternative explanations, including demand effects and the potential role of mood, attention, impulsivity, risk-taking, and self-interest, we conducted a series of control analyses. These included regression analyses to test whether drug-induced changes in attention (as measured by changes in RTs) predicted dishonesty, and ANCOVAs to test whether the effect of methylphenidate on dishonesty remained significant when controlling for participants’ beliefs about drug assignment, mood, and other psychological states.