Participants
One hundred and eighty five participants (79 men) took part in the study. The sample contained two groups: adolescents (N = 90, mean age = 14.4, SD = 0.68, range = [13, 16], 40 boys) and young adults (N = 95, mean age = 23.05, SD = 1.51, range = [20, 28], 39 men, university students and people with vocational education). Adolescents were recruited via parent-teacher conferences in local schools, while adults were recruited via online advertisements. Parental consent was obtained for all underage participants. Participation was rewarded with vouchers (for a clothing store, a sporting goods store, a bookstore, or a movie theatre) valued from $5 to $15 (mean $10, equivalents in local currency), depending on the level of individual performance in the two rewarded tasks (see below). Participants could win $5, $7.5, $10, $12.5 or $15, depending on the points they scored. The point ranges for each amount were based on the results of the pilot study, which was conducted on 60 adolescents and young adults.
Procedure
The research was conducted in schools (adolescents) or at a university (adults). Before the experiment, all subjects were informed about the general purpose of the study and the experimental procedure. They were assured anonymity and confidential storage of the collected data. They were also informed that they could withdraw their participation at any time and could receive performance feedback after completing the study. The experiment lasted about 80 minutes with a ten-minute break in the middle of the session. Participants performed a set of tasks measuring risk-taking, reward sensitivity, cognitive control, and impulsivity. The order of tasks was randomized. Each task was preceded by a practice session. The amount of the reward was based on performance in two selected tasks (measuring risk-taking and reward sensitivity). The task measuring risk-taking was performed twice: in the rewarded and the non-rewarded conditions (in random order). Only the result from the former was included in the overall reward. At the end of the experiment, participants completed three questionnaires measuring everyday risk-taking, reward sensitivity, and self-control, in a fixed order.
Measures
Spaceride. The risk-taking measure resembled a racing game with obstacles. The task was similar to the Stoplight task (Chein et al., 2011). The objective was to fly a spaceship through outer space in order to reach the destination in the shortest possible time (see Figure 1). At any time during the task, participants could press the accelerate or brake key, or not press any of the keys. When participants chose any of the options, the spaceship moved at a constant speed (accordingly, fast, slow or medium). On the route there were 10 danger zones in which the spaceship could collide with a passing asteroid. In each danger zone there was 50% probability that the asteroid would appear, and the probability of its appearance was distributed uniformly over the whole of each danger zone. Entry into each danger zone was signaled by auditory and visual stimuli (sound signal, a light on a "radar panel", and the appearance of distant, non-threatening, asteroids in the background). When entering a danger zone, the participants could press the accelerate key (or not press anything), thereby risking a collision, or press the brake key to avoid the asteroid. Collisions led to temporary immobilization of the spaceship and consequently prolonged the time of the ride. Our goal was to not encourage participants to use either of the two possible strategies (safe or risky), so the expected time of flight when the safe strategy (using the brake key through the entire danger zone) was chosen equaled the expected time of flight when the risky strategy (pressing the accelerate button) was chosen. The former was 17 seconds, while the latter could be 7 seconds if no asteroid was hit (50% chance) or 26 seconds (7 seconds of flight plus 19 seconds of punitive immobilization) if an asteroid was hit (50% chance).
The measure of risk-taking was the proportion of collisions to the number of danger zones in which the asteroid appeared (i.e. the maximum possible number of collisions).
Incentivized Visual Search Task (IVS). We created a new task to measure reward sensitivity, operationalized as an increase in visual search performance along with the growing value of rewards. A 12×10 matrix of five-, six- and seven-pointed stars was presented on the screen. The goal was to search for and mark six-pointed stars by scanning the field from left to right, row by row (see Figure A in the Supplement). Participants had 30 seconds to find all six-pointed stars. After this time they were asked to mark the location of the last scanned star (signal or noise) so we were able to measure the scanned area and estimate the speed of their processing. The task consisted of 16 blocks, each of which varied in the amount of points the participants could get for correct detection. Depending on the block, the reward was 0, 1, 2 or 5 points for each correctly marked target (out of 10 in every block). There were four blocks for each reward value and the order of the blocks was randomized (except for the first turn, which was always scored 1 point to establish a reference point). Additionally, to prevent the strategy of marking all stars (correct and wrong) to maximize the score, at each turn participants lost 1 point for marking a wrong star. The value of the reward in points appeared before each block on the screen; the number of points scored was displayed after each session.
The difference between correctly indicated stimuli and false alarms was a basic performance measure. To assess participants’ individual reward sensitivity, we fitted a mixed linear model for which basic performance in the IVS task was assumed to be logarithmically dependent on the reward value. In the model this dependence included the random effect of participants, so we obtained an individual estimation of the performance increase parameter over the (log) reward value for every participant (F[1,1058.5] = 30.39, p < .001, marginal R2 = .011, conditional R2 = .3). These individual estimations were used as a measure of reward sensitivity.
Go/No-go Task. This task is a popular measure of response inhibition (e.g. Logan, 1994). In "go" trials, participants categorized numbers appearing on the screen as even or odd, but in "no-go" trials (i.e. when a specific number, indicated in the instructions, appeared on screen) they had to refrain from reacting. The version of the task used in this study consisted of 10 blocks, each containing 10 trials (on average 8.33 "go" and 1.67 "no go" trials). The display time of stimuli was 1 second; the criterion according to which it was necessary to react changed every block.
The measure of participants’ cognitive control was d’; this is a complex measure based on signal detection theory that takes into account both hits and false alarms.
Matching Familiar Figures Test (MFF). We measured impulsivity using Kagan’s MFF Test (Matczak & Kagan, 1992). The test consisted of 12 trials, in each of which participants were shown a black-and-white reference picture (e.g. a house, scissors, a telephone or a tree) and six visually similar test pictures that differed in small details. The goal in each trial was to find the drawing that was identical to the reference picture. Both the reference picture and all optional pictures were visible all the time, so the task did not involve visual memory. Both response time and accuracy were recorded. There are two simple strategies which can be used to solve the task. The first (impulsive – fast and incorrect) is to answer as soon as one sees a picture which matches the reference picture. The second (deliberate – slow and accurate) is to delay answering until one achieves a high certainty that there are no differences between the model and the match.
Since there are two simple measures of this impulsivity (speed and accuracy), both of which are equally theoretically valid, a complex measure of this variable was used in the task. As they were correlated (r = -.45, 95% CI = [-.56, -.33]), we used the main principal component of these variables as a unified measure of impulsivity.
Risk Behavior Questionnaire. This is a new measure based on the Adolescent Risk-Taking Questionnaire (ARQ; Gullone, Moore, Moss, & Boyd, 2000), in which participants assess the frequency of risky behaviors using a 5-point scale from 1 – “have never done it” to 5 – “[I do it] very often”. As the ARQ was developed to study adolescents aged 11–18 years, we modified it in the pilot study (in 197 adolescents and young adults) to make it applicable for all our age samples by adding items describing risk behaviors that are typical of young adults. After the pilot study’s data analysis, we removed items with the lowest reliability. The final version included 29 risky behaviors (we present it in Table A in the Supplement). The Cronbach’s α of the questionnaire was .86, 95% CI = [.83, .89]. The mean response to all items was the measure of everyday risk-taking.
Short Form of the Sensitivity to Punishment and Sensitivity to Reward Questionnaire (SPSRQ-SF). We used the Cooper and Gomez questionnaire adapted by Wytykowska, Białaszek, & Ostaszewski (2014). This questionnaire consists of 24 yes-no statements and measures individual sensitivity of the Behavioral Inhibition System (BIS) and the Behavioral Approach System (BAS) with separate subscales. The Cronbach’s α of the Sensitivity to Punishment subscale was .81, 95% CI = [.77, .85], and for the Sensitivity to Reward subscale it was .71, 95% CI = [.65, .77]. The mean response to all items in the Sensitivity to Reward subscale was the measure of reward sensitivity. The Sensitivity to Punishment subscale was not analyzed in this study.
Self-Knowledge New Sheet (NAS-50). A questionnaire created by Nęcka et al. (2016) for assessment of self-control. The scale consists of 50 items divided into 5 subscales: goal maintenance, proactive control, initiative and persistence, switching and flexibility, inhibition and adjournment. The answers are assessed on a 5-point scale from 1 – “definitely not” to 5 – “definitely yes”. Cronbach’s α in the present investigation was .87, 95% CI = [.84, .89]. The mean response to all items was the measure of self-control.
Statistical analyses
The analyses of the data were conducted using R (R Core Team, 2019). The effects of reward sensitivity and cognitive control on risk-taking have been tested on two levels: in behavioral tasks and in self-report measures. In the first case, we fitted a linear model for which risk-taking in the both rewarded and non-rewarded conditions of the Spaceride task depended on reward sensitivity (measured by the IVS task), cognitive control (in the Go/No-go task), impulsivity (in the MFF test), age group, and interactions between the age group and reward sensitivity and cognitive control. Additionally, we controlled for gender, group relative age (difference between age and the mean age for the age group) since some variance in age remained after dichotomization to adolescents and adults, and braking (duration of pressing the brake key in danger zones with the asteroid in the Spaceride task) since it could have a large impact on the dependent variable, i.e. frequency of crashing. On the self-report level we fitted a similar linear model to assess the relationship between everyday risk-taking, measured with the Risk Behavior Questionnaire (dependent variable), and reward sensitivity (measured with SPSRQ-SF), self-control (measured with NAS-50), age group, and interactions between SPSRQ-SF, NAS-50 and age group. For all significant effects we provide partial R2 as a measure of effect size.