2.1 Experimental procedures

Participants were invited to a classroom near the experimental laboratory. All sessions took place in the afternoon, between 12:45 and 17:05, in a sequence of three sections. Each section lasted for about two hours. Upon arrival, participants were informed about important procedural matters (particularly the functioning of the tool kit for taking saliva samples), and then they filled out a short questionnaire on their sleeping schedule on the day of the experiment, illness over the last three months, the medication they took, their alcohol consumption on the previous day, and the time at which they had lunch, as well as what they had for lunch. They were given an ID number, and they were told that the experiment was of a scientific nature, that the data were all recorded anonymously, and that payment would be made privately in cash at the end of the session. Participants were requested to follow the instructions given by the experimenters and not to talk to each other. They signed a consent form explaining their right to abandon the experiment at any moment in time.

Each experimental session had a maximum of 10 participants, randomly sorted into two groups of similar size. The maximum number of subjects in a group was five and the minimum was four. Participants in the same group moved together during the experiment following the protocol described below.

In the TSST treatment, groups were guided into a quiet classroom (near the classroom used for reception and general instructions but not visible from the control group’s room) in which a committee with two members (wearing white lab coats and showing neutral faces) and clearly visible video cameras were waiting for them. First each participant was requested to deliver a two-minute speech (summarizing past achievements and skills for a mock job interview) in front of the committee, the video cameras, and their peers after two minutes of preparation, and then to perform some mental arithmetic exercises for 80 seconds. The arithmetic exercise consisted of serial subtraction in front of the committee and peers. Participants were required to subtract 17 from a randomly assigned four-digit number and keep subtracting for 80 seconds. When an arithmetic mistake was detected, the committee instructed the participant to start over. They were told that the quality of their speech would be evaluated by a professor at Waseda University later on, that the mental arithmetic task was easy enough for elementary students, and also that their performance would be recorded. Each participant performed these tasks standing up, while others (including the committee members) were seated. The order of performance was random so that participants could not anticipate their turn. The TSST consists of the above-described tasks. It is a frequently used, reliable for inducing stress in human participants in the experimental laboratory [22]. As detailed above, we used its more time- and cost-efficient group version, the “Trier Social Stress Test for Groups” (TSST-G) introduced by von Dawans et al. [21], which allows for subsequent interaction among participants under acute psychological stress. For simplicity, we use the “TSST” and “no-TSST” labels when referring to our treatments.

Individuals assigned to the no-TSST treatment group remained in the initial classroom and were allowed to read quietly. They were not allowed to sleep, talk to each other, or use any electronic devices.

Upon completion of the TSST or a 40-minute waiting time (in the no-TSST treatment), groups were led into the computer rooms (experimental laboratory) by an assistant and were seated at computer terminals separated by walls from each other. We used two adjacent computer rooms (3-803 and 3-804 on the Waseda main campus) which allowed us to work with two groups of up to five participants with five minutes of delay between them. We carried out three pairs of parallel sessions each day in a sequence. The computer rooms were dark, and each booth had a lightbulb installed emitting white, blue, or red light. This choice of colors was inspired by studies reporting (potentially beneficial) effects of blue light on performance, general mood, and stress levels (e.g., [26, 27]). White constitutes a natural reference point, while red was chosen for control as another unnatural light color. The possible effect of light color is not of primary interest here, but we carefully controlled for it in our statistical analysis. All our sessions took place in the afternoon during the Japanese winter season in two computer laboratories with closed curtains to reduce light contamination from the outside. All bulbs in the same room were of the same color. Participants were asked to read and follow instructions received on-screen. All decisions were made and all interactions occurred through computers.

The first computerized task consisted of a simple reaction-time test implemented in PsychoPy [28]. The screen displayed a single letter (Z or Y) for 0.5 seconds in a random sequence of 60 items, and participants were required to hit a certain key (J) on the keyboard when a specific letter (Z or Y) appeared, in a maximum of one second after the stimulus, and not to hit any key when the other letter was displayed. Around half of our participants had to respond to Z’s and the other half had to respond to Y’s. This task lasted for five minutes and was not incentivized monetarily. We do not use the data recorded during this task; they will be discussed elsewhere.

The second computerized task lasted for 15 minutes and consisted of a sequence of situations in which participants were asked to make decisions for monetary gains. This part was implemented in z-Tree [29] and included the following individual and multiple-person problems:

• Cognitive Reflection Test (CRT) with three questions,
• two-person dictator game (with an endowment of ¥1000),
• five-person beauty-contest game (for a ¥500 prize),
• tasks eliciting risk attitude (an open question, three lottery-choice questions, the bomb-search risk-preference elicitation task).
• two-person social-dilemma game,
• two-person joy-of-destruction game (with a maximum harm of ¥1000).

S5 Appendix presents the original zTree screens and their English translation. All participants went through the same sequence of problems in the order as listed above. They were informed about the outcome of each problem before moving to the next one. We decided not to randomize the order of problems for two reasons. Firstly, our experimental sessions were carefully timed to ensure that tasks, questions and requests for saliva samples happen exactly at the same moment (with the administration of the stressor as a reference point) across all our sessions. Secondly, as some of the tasks involved interaction with other participants, everyone in the same session had to enter each stage in a coordinated fashion. Given that we kept the tasks relatively diverse and their number small, it is unlikely that participants experienced cognitive fatigue or could learn to make “better” decisions as moving along the sequence of tasks. More importantly, our observations for estimating the impact of acute stress on decision-making stem from a between-subject experimental design that does not involve repetition. All the relevant conclusions are based on comparing behavior in the treatment (stressed) group to behavior in the reference (not stressed) group while controlling for as many covariates as possible.

After the second task, the classroom’s lights were turned on, and participants received a short debriefing about the experiment, focusing on the nature and the importance of the TSST for our research. We apologized for the stress induced. S5 Appendix presents the text read aloud to participants as an apology. In the no-TSST treatments, the debriefing did not refer to stress but simply expressed our gratitude for their participation in the experiment.

Finally, the third computerized task lasted for 10 minutes and consisted of a long questionnaire (implemented in z-Tree) with questions targeting personality traits (BFI, PVQ) and personal well-being (CES-D for depression, STAI-T for anxiety proneness), as well as a few philosophical thought experiments on distributive justice. We use a shortened 10-item version of the Big Five Inventory (BFI) [30], the Portrait Values Questionnaire (PVQ) [31], the Center for Epidemiological Studies-Depression (CES-D) [32], and the T-anxiety scale (STAI-T) of the State-Trait Anxiety Inventory (STAI) [33]. These questionnaires are all standard, validated tools that are widely used by psychologists. None of the questions in this task was incentivized monetarily. We use the responses to the BFI, PVQ, CES-D, and STAI-T as controls for personality traits in the statistical analysis reported below but do not work with the items from the philosophical thought experiments; these will be analyzed and discussed elsewhere.

The participants were paid their earnings from the second task together with a ¥500 show-up fee in private at their computer terminal right after the third computerized task and were then allowed to leave. The average monetary earnings, excluding the show-up fee, was around ¥2200. The monetary compensation the participants received was significantly higher than the typical earnings from economics experiments on decision-making in the laboratory at Waseda. According to the official exchange rate at the time of our experimental sessions, it was equivalent to 20 USD and would be enough to buy four lunch boxes on campus. The earnings were also more volatile than usual, with a minimum of ¥200 and a maximum of ¥4000.

We measured participants’ stress levels through saliva samples and simple paper-based questionnaires administered four times during the experiment, as described below. Groups were moved through the experimental phases and tasks with a stop-watch to make sure that the samples were taken in a controlled, timely manner. Recall that all our sessions took place in the afternoon (between 12:45 and 17:05 in a sequence with three sections) during the Japanese winter season in order to control light conditions; this also helped us control for the diurnal cycles of cortisol secretion.

1. After arrival, as part of the general instructions, participants were shown and practiced how to use the saliva sample kits. A test sample was collected as they did so, but its results are not used in the statistical analysis. The transcript of these instructions is presented in S5 Appendix.
2. Five minutes before the TSST, after the two-minute speech-preparation time. This is the benchmark measurement (minute -5). Participants in the no-TSST treatment supplied an initial sample after receiving general instructions, matching the schedule of the TSST treatment. Note that the TSST is our minute 0 reference.
3. Twenty minutes after the first sample, after the TSST or after the waiting time in the no-TSST treatment.
4. Thirty minutes after the first sample, once groups moved to the computer rooms and before the computerized tasks.
5. Fifty-five minutes after the first sample, after the second computerized task and before debriefing (and the third computerized task with questionnaires).

Our experimental design includes four saliva samples, out of which the first serves as the reference point. The second and the third were meant to capture the peak in salivary cortisol concentration that typically occurs 20-30 minutes after the stressor [7, 21], while the fourth was timed to show its decay. These features of the experimental design are in line with the guideline set by Shields [34] for studies exploring stress and cognition. While a larger number of test could have offered a more accurate picture of the evolution of cortisol (and acute stress), its costs in terms of a smaller number of observations (due to our hard budget constraint) and repeated interruptions in the flow of the experimental session would probably have outweighed its benefits. Also, providing enough saliva for clinical tests more than four times could have been physically too demanding for quite a few of our participants.

Along with providing saliva samples, participants were also requested to self-report their well-being. Together with each saliva sampling toolkit, participants were given a sheet of paper with three printed 10 cm-long horizontal visual analog scale (VAS) representing a 0-to-100 scale (ranging from 0 to 100), on which participants were asked to provide ticks in answer to questions on their current stress, excitement and fatigue levels, respectively. After each saliva sampling, participants were asked to answer the following three questions on VAS: “How much stress do you feel?” “How excited are you?” “How tired are you?” Fig 1 offers a visual summary of the timing of our experimental sessions.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Timing of a typical session in section 1 with or without TSST.

https://doi.org/10.1371/journal.pone.0244881.g001

2.2 Outcome variables

We have chosen a limited number of simple decisions problems with which to analyze participants’ behavior under stress. The problems have all appeared and received extensive attention in the experimental and behavioral economics literature. The list below offers a brief description of each one and offers a reference to the standard economic model of decision-making that serves as our benchmark. Note that we refer to the tasks and the related behavioral concepts as customary in economics, in particular in game theory and behavioral economics. S5 Appendix presents the original Japanese zTree screens (and an English translation beneath each one) used for the monetarily incentivized tasks.

• Cognitive Reflection Test [35] with the three standard questions (about the baseball bat, the water lily, and the machines), used to measure participants’ tendency to override a quick but typically incorrect response and engage in reflection to find the correct answer. Questions appeared on the screen one by one, answers had to be entered in within 20 seconds and each correct answer was rewarded with ¥100. The related statistical variable counts the number of correct answers given in the 20 seconds/question timeframe. It proxies the decision-maker’s reasoning capacity, a key—albeit hidden—element in the economic model of rational decision-making which presupposes careful considerations and limitless cognitive abilities. This test has been extensively used in empirical research [36], and it has been shown to outperform other measures (e.g., the Raven test [37]).
• A two-person dictator game in which each participant had to decide how much money to keep from an exogenously assigned endowment of ¥1000 and how much to give to a partner randomly and anonymously selected and assigned from the group of participants. The amount given to the partner in the dictator game is typically interpreted as a numerical measure of altruism [38], because a decision-maker with standard self-regarding preferences would keep the entire endowment. Our variable records the amount of money kept by the dictator. Participants were required to make a decision in a maximum of three minutes. Pairs were formed in such a way that each participant acted as a dictator and as a partner (or receiver) exactly once.
• The beauty-contest game is a guessing game in which each participant simultaneously chooses a number between 0 and 100 (limits also permitted). We implemented the version in which the participant whose number lies the closest to of the average of all chosen numbers wins a fixed prize of ¥500, while all other participants receive nothing. Most groups had five participants (some had 4). This game is dominance solvable, but a large number of iterations are required due to the large number of available strategies. The unique Nash equilibrium of the game (independently of group size) is when each participant chooses the number 0. Experimental and behavioral economists interpret the chosen number as a measure of strategic sophistication, given that larger numbers correspond to dominated strategies and indicate naïve play and smaller ones indicate higher levels of iterative thinking about others’ behavior in order to maximize one’s expected gain [39]. Participants were required to make a decision within three minutes.
• Risk attitudes play a key role in describing decision-making under risk in economic models. They describe how willing or unwilling the decision-maker is to take risk. The literature does not agree on the best method of eliciting risk attitudes, so we have applied three different approaches.
  • Following Dohmen et al. [40], we requested participants to answer the following question on a scale from 0 to 10: “How do you see yourself: are you a person who is generally willing to take risks, or do you try to avoid taking risks?” Larger values for the related variable indicate more willingness to take risk. Participants were required to give an answer in a maximum of one minute.
  • Inspired by Falk et al. [19], we also offered three monetarily incentivized choices between a fixed lottery (which paid ¥50 with 50% probability and ¥650 with 50% probability) and varying sure payments (¥350, ¥300, and ¥250). Participants who choose the lottery more often (recorded as our variable) are more willing to take risk. Participants were required to make their choices within three minutes.
  • We also implemented the bomb-search risk-elicitation task [41] in which participants had to decide how many to open of the 100 boxes that appeared on the screen, one of which contained a “bomb”. Participants received ¥1 for each box they opened if the one with the bomb remained unopened and nothing otherwise. Choosing a larger number of boxes indicates more willingness to take risk. Participants were required to make a decision within three minutes.
• Participants’ cooperativeness was measured in an extreme social-dilemma situation in which they were randomly assigned into pairs and had to decide simultaneously whether to keep and pocket ¥250 or send it to their partner, in which case the money was doubled. This game is an instance of the well-known “prisoner’s dilemma”, albeit without Tucker’s original story [42], in which personal and group interests are opposed. The game played by agents with standard self-regarding preferences has an equilibrium in dominant strategies in which both agents keep their initial endowment. The equilibrium outcome however is not efficient (not socially optimal), as agents forgo the additional earnings that could have arisen from the doubling of the monetary amounts had the money been sent to the opponent. We will refer to this game as a “gift-exchange game”, even though the experimental-economics literature uses that name to refer to a more complex two-person interaction [43]. Sending one’s endowment (i.e., a gift) to one’s partner is going to be interpreted as cooperative behavior. The related variable takes a value of 1 if the participant decides to cooperate and 0 otherwise. Participants were required to make a decision within three minutes. Pairs were formed in such a way that each participant acted as a sender and as a receiver exactly once.
• Finally, our list includes a measure of nastiness (the dark side of human nature) through a game called the “joy of destruction” [44]. We offered each participant the opportunity to destroy no more than ¥1000 of another participant’s earnings. They were shown the accumulated monetary gains of a randomly chosen participant in an anonymous manner and were allowed to destroy (at no cost and at no gain) at most ¥1000 yen of it. This game is the negative counterpart of the dictator game, as the typical economic decision-maker with self-regarding preferences would not want to incur in costs that bring no personal benefits in monetary term. The amount destroyed is therefore interpreted as a numerical measure for one’s “nastiness”. Participants were required to make a decision within three minutes. Pairs were formed in such a way that each participant acted as an active and as a passive party exactly once.

The above list is far from being an exhaustive one of the behavioral traits and patterns in which experimental and behavior economists have shown interest over the past five decades. Beside the cited references above, we refer our readers to the two volumes of the Handbook of Experimental Economics [45, 46] which contain ample discussion on the dictator game (chapter 4 in volume 1, chapter 4 in volume 2), the beauty-contest game (chapters 3 and 7 in volume 2), risk attitudes (chapter 1 in volume 1), and social dilemmas (chapter 2 in volume 1).

Overall, we have tried to assemble a basic set of decision-problems that have played a prominent role in the literature and that are easy to explain relatively quickly. Recall that stress levels decay with time and that our cortisol concentration measurements imposed strict time limits and a strict schedule. As detailed in the introduction, our goal was to choose a sufficiently diverse set of tasks to be able to characterize participants’ preferences to some extent and to be able to judge whether their decisions are made as carefully and in such a strategic manner as assumed by the standard economic model of human decision-making.

3.1 Stressor validation

As shown in Fig 2, the TSST had a positive impact on participants’ cortisol levels which, on average, peaked in the sample taken 30 minutes after the stressor was administered (30’). Following Cahlíková and Cingl [7], we categorize participants as stressed when they experience a cortisol-level change of more than 2.5 nmol/l anytime during the experiment relative to the sample taken right at the beginning of the experiment (-5’). Recall that, when referring to the timing of saliva samples, we treat the TSST as minute 0.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Evolution of cortisol concentration with and without TSST.

https://doi.org/10.1371/journal.pone.0244881.g002

We summarize the validation of the stressor included in our experimental design in the following result.

Result 1 The TSST has resulted in stress (i.e., in a cortisol-concentration response of more than 2.5 nmol/l) in 75% of our participants.

Interestingly, 9% of those participants who did not participate in the TSST also experienced stress during their participation. We attribute that to the experiment itself—to spending a considerable amount of time in a tightly controlled environment, periodically producing saliva samples, and making decisions under time pressure for monetary gains. Fig 3 displays the distribution of peak cortisol concentration response, showing that TSST tends to provoke an increase that varies largely across the participant pool. Our data also reveal that the TSST results in stress significantly more often in men than in women (89% vs. 57% of the cases, respectively; p-value = 0.0003 for a large-sample z-test of proportions).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Distribution of cortisol concentration response with and without TSST.

https://doi.org/10.1371/journal.pone.0244881.g003

The color of the light under which participants were required to make decisions does not seem to have a significant impact on the impact of TSST on cortisol concentration levels. The proportion of participants that were categorized as stressed after the TSST is 78% for white light, 81% for red, and 67% for blue. Although the distribution of cortisol-concentration response levels look slightly more skewed for white light than for the other two (see Figs 1, 2 and 3 in S1 Appendix, pairwise comparisons based on the Kolmogorov-Smirnov equality-of-distributions test reveal no statistically significant differences (combined p-value = 0.27 for red and white, combined p-value = 0.68 for blue and white, combined p-value = 0.17 for red and blue). The parametric pairwise proportion tests deliver the same conclusions, with p-value = 0.76 for red and white, 0.30 for blue and white, and 0.18 for red and blue. With a p-value = 0.77 in a Spearman’s rank correlation test, we can not reject the hypothesis that color and stress are independent.

Even though the 2.5 nmol/l threshold is often used in the literature to categorize people as stressed, some researchers (e.g., [47]) consider it overly conservative and suggest using a lower limit of 1.5 nmol/l. As a robustness check, we have repeated the above-described analysis with that lower value and find that the qualitative result on the validation of the stressor does not change; only minor numerical changes appear. That is, with a threshold of 1.5 nmol/l for change in cortisol concentration, the TSST results in stress in 80% of our participants, while 15% are categorized as stressed without going through the TSST. The TSST induces stress significantly more often in men than in women (91% vs. 64% of the cases, respectively, p-value = 0.0013). The proportion of participants that were categorized as stressed after the TSST is 78% for white light, 90% for red, and 69% for blue. The distance between these percentages suggests that blue light helps the avoidance of stress caused by TSST and red light adds to it. Pairwise proportion tests show that those distances to the benchmark white color are not statistically significant (p-value = 0.17 for red vs. white, and p-value = 0.44 for blue vs white), while the difference between blue and red is (p-value = 0.04). Nevertheless, with a p-value = 0.22 in a Spearman’s rank correlation we can not reject the hypothesis that color and stress are independent. Moreover, the Kolgomorov-Smirnov test compares the cortisol-distributions across subsamples defined by color. We conclude that the color of light has no significant impact on the effect of TSST.

Regarding participants’ self-declared wellness levels measured on visual analog scales (VAS), the TSST had a significant impact on stress, excitement, and fatigue, but with different time patterns. Figs 4, 5, and 6 in S2 Appendix show the evolution of self-reported wellbeing (i.e., stress, excitement, and fatigue, respectively) in a box-plot form similar to that shown in Fig 3. The description that follows relies on binary comparisons between the TSST treatment group and the reference group via the Wilcoxon rank-sum test (or Mann-Whitney two-sample test for independent samples) reported in Table 1 in S2 Appendix. Participants who went through the TSST arrived significantly more stressed to the experimental laboratory and remained so until the last measurement (at 55’), when statistical differences vanished. In terms of excitement (or arousal), the TSST treatment group shows a similar pattern: Self-reported excitement levels significantly exceed those of the benchmark group at the beginning (for at least 20 minutes), after which they become statistically indistinguishable. Interestingly, for excitement, this equality happens around the third measurement (at 30’), after which the benchmark group reports higher levels than the TSST group. Differences in fatigue across the two groups are less pronounced but are significantly different for the first three measurements, with the TSST group reporting higher levels.

3.2 Decisions under acute stress

We now turn our attention to the impact of stress on behavior as typically measured through a number of standard problems and interactive games in the economic experimental laboratory. Table 1 reports the mean values for all our variables of interest, along with effect sizes measured through Cohen’s d, p-values for the Wilcoxon rank-sum test (also known as the “Mann-Whitney two-sample test”) for most of them, and p-values for a large-sample z-test in case of proportions. We make comparisons for the subsample of both the participants who participated in the TSST (top panel of the table) and the participants who were categorized as stressed during the experiment according to one or the other threshold for change in cortisol level (bottom two panels of the table). Our main finding is summarized in the following result.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Stress and changes in behavior (binary comparisons and treatment effects).

https://doi.org/10.1371/journal.pone.0244881.t001

Result 2 Acute stress does not have a significant impact on cognitive skills, strategic sophistication, risk attitudes, altruism, cooperativeness, or nastiness.

The qualifier “significant” refers here to both statistical significance at the usual levels and effect size. Hence, Result 1 is based on a conservative approach that also takes alternative definitions (i.e., TSST vs. cortisol response) of the treatment group into consideration.

Fig 4 offers a visual summary of Result 1 by displaying the TSST treatment effects with the help of point estimates and symmetric 95% confidence intervals around them.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. The effect of acute stress on decision-making.

TSST treatment effects visualized by point estimates and 95% confidence intervals.

https://doi.org/10.1371/journal.pone.0244881.g004

Given our sample sizes and α = 0.05, the statistical power of the applied two-tailed Wilcoxon rank-sum test to detect a medium effect size (i.e., when Cohen’s d = 0.5) is 0.89 for the split according to TSST (with subsamples of 91 and 99 observations) and stressed (1.5+) (with subsamples of 98 and 91 observations) and 0.88 for stressed (2.5+) (with subsamples of 108 and 81 observations). These numbers decrease to 0.48 and 0.47 when the target effect size is d = 0.3, and further to 0.25 and 0.24 when it is d = 0.2. We have carried out the power analysis with the help of G*Power [48]. It is important to note that these estimates for statistical power are for a single (isolated and independent) statistical test. Our results however are related to a survey-type set of questions and tasks that stem from the same respondent. While our measurements capture sufficiently different aspects of human decision-making (except for the three methods targeting risk preferences) and none of the questions was repeated, some might argue that the statistical tests that we perform are not fully independent. Although our main comparisons are across treatments (related to stress) and across participants (based on a between-subject design), the above-reported estimates are to be interpreted as upper bounds for statistical power.

When looking at the change caused by TSST or stress in average behavior, we recorded an increase in the number of questions answered correctly in the cognitive reflection test and a decrease in the number chosen in the beauty-contest game (i.e., an increase in strategic sophistication). Note that our between-subject design does not allow for looking at changes in individual behavior. It seems as if participants paid more attention to their tasks and made somewhat sharper decisions under stress. However both of these effects are small (as measured by Cohen’s d), and none is statistically significant at the usual levels. Following Cohen’s guidelines, we interpret a d with an absolute value of around 0.2 as a “small,” one around 0.5 as a “medium,” and one around 0.8 as a “large” size effect [49, p. 25-26]. In the end, our results are in line with the early finding on strategic sophistication under acute stress reported by Leder, Häusser and Mojzisch [1], as we detect no statistically significant differences between the treatment and benchmark groups in terms of the number chosen in the (first round of a) beauty-contest game. They do differ from the conclusion reached by Bendahan et al. [4] that acute stress increases decision-makers’ willingness to take risks.

To measure risk attitudes, we have used two monetarily-incentivized tasks and a self-evaluative open question. When answering the latter, post-TSST participants and those who were stressed claimed that they were somewhat less keen on taking risks in their life. In line with this claim, they also chose the safe payment more often against the risky lottery. At the same time, however, they decided to open a few more boxes in the bomb elicitation task, a behavior that signals more willingness to take risks. Except for the difference in lottery choices between the TSST group and its benchmark, none of the listed changes is large or even statistically significant at the usual levels. Our conclusions on risk-attitudes being unaffected by acute stress are in line with the results reported by Sokol-Hessner et al. [3].

In terms of altruism (money sent in the dictator game), cooperativeness (money sent in the gift-exchange game), and nastiness (money destroyed in the joy-of-destruction game), the TSST and stress on average have resulted in lower altruism, less cooperativeness, and more nastiness. Again, however, none of the listed changes is large or statistically significant at the usual levels, except for the comparison between the TSST group and its benchmark for the gift-exchange game.

3.3 Robustness check: Redefined treatment group

When considering cortisol levels, instead of having participated in TSST as a proxy for stress, none of the treatment effects is statistically significant, except for a smaller number of boxes opened (hinting at more risk-aversion) when the 2.5 nmol/l threshold is used for assigning participants into the stressed category.

Inspired by the literature on gender differences in behavior under acute stress, we also performed separate binary comparisons between the treatment groups and the corresponding benchmarks for the two self-reported gender groups. Table 2 in S3 Appendix presents all the numerical details. The results suggest that stress increases cognitive ability in men and reduces their cooperativeness (both effects are small). The only other statistically significant effect observed is that women choose smaller numbers in the beauty-contest games (hinting at increased cognitive ability) but only when cortisol response with the 2.5 nmol/l threshold is used to define the treatment category.

3.4 Robustness check: Regression analysis

We perform a robustness test for the above results by estimating the impact of the TSST and stress while controlling for a large number of variables. The controls include

• demographic variables (i.e., gender, age, height and weight, college grade, college major),
• experimental-session characteristics (i.e., date, laboratory ID, section ID, light color),
• a measure of cognitive skills (via the CRT), and
• measures for personality traits (BFI and PVQ), and
• personal well-being (CES-D and STAI-T).

S4 Appendix describes the applied scoring rules and presents a descriptive analysis of the control variables in Tables 3, 4, and 5 in S4 Appendix. It is important to point out that our reference, the standard economic model of individual decision-making, is blind to gender, age, weight, and all the other factors listed above. We nevertheless include them in the regression analysis in order to control for possible unduly-ignored effects while our main interest remains in exploring the impact of acute stress. The use these control variables has numerous precedents in the literature both in economics and medicine in spite of not appearing in the reference model of human decision-making (e.g., gender in [12, 15, 36], cognitive skills in [37], personality traits in [7, 50], personal well-being and psychiatric disorders in [51]).

Given that we observed significantly different reactions to the stressor according to the participants’ gender, our regressions include the gender and the treatment variables as an interaction terms as well. The estimation results displayed in Table 2 confirm our null result from the above-described binary comparisons, with a few exceptions summarized in the following result. The complete version of Table 2 appears in Tables 6, 7, and 8 in S4 Appendix. Note that the last two rows of the table report p-values for hypothesis tests related to treatment effects for male and female participants separately.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Stress (TSST) and changes in behavior (regression analysis).

https://doi.org/10.1371/journal.pone.0244881.t002

Result 3 Acute stress has no significant impact on cognitive skills, strategic sophistication, risk attitudes, altruism, cooperativeness, or nastiness, except that it significantly decreases men’s risk aversion (as measured by the lottery-choice risk-elicitation task).

Note that the men in the TSST treatment group chose the risky lottery significantly less often (on average, 1.4 less often in the three questions) than did the men in the reference no-TSST group. Otherwise, our regressions find no statistically significant effect on decisions or behavior. This finding seems to be in line with earlier reports on increased risk aversion induced by acute risk in men [6, 7]. Fig 5 plots the treatment effects together with 95% confidence intervals for a quick and visual summary of our results. It is noteworthy that we find no gender effects in the no-TSST group either, except that females destroy significantly more of their partner’s gains in the joy-of-destruction game. The general no-result and the statistically significant gender effect are robust to changes in the measurement of acute stress. As Tables 9 and 13 and Figs 7 and 8 in S4 Appendix show, it does not matter whether we proxy for acute stress using participation in TSST or using a cortisol concentration response of more than 1.5 nmol/l or more than 2.5 nmol/l. As in Table 2 reported in the main text, S4 Appendix contains the full versions of Tables 9 and 13 (Tables 10, 11, and 12, and Tables 14, 15, and 16, in S4 Appendix respectively) and the figures with the estimated treatment effects (Figs 7 and 8 in S4 Appendix). However, we are unable to identify any statistically significant change in risk attitudes when using the other two alternative definitions of stress.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. Predictive margins of TSST by gender with 95% CIs.

https://doi.org/10.1371/journal.pone.0244881.g005