Trait impulsivity and acute stress interact to influence choice and decision speed during multi-stage decision-making

Impulsivity and stress exposure are two factors that are associated with changes in reward-related behavior in ways that are relevant to both healthy and maladaptive decision-making. Nonetheless, little empirical work has examined the possible independent and joint effects of these factors upon reward learning. Here, we sought to examine how trait impulsivity and acute stress exposure affect participants’ choice behavior and decision speed in a two-stage sequential reinforcement-learning task. We found that more impulsive participants were more likely to repeat second-stage choices after previous reward, irrespective of stress condition. Exposure to stress, on the other hand, was associated with an increased tendency to repeat second-stage choices independent of whether these choices previously led to a reward, and this tendency was exacerbated in more impulsive individuals. Such interaction effects between stress and impulsivity were also found on decision speed. Stress and impulsivity levels interacted to drive faster choices overall (again irrespective of reward) at both task stages, while reward received on the previous trial slowed subsequent first-stage choices, particularly among impulsive individuals under stress. Collectively, our results reveal novel, largely interactive effects of trait impulsivity and stress exposure and suggest that stress may reveal individual differences in decision-making tied to impulsivity that are not readily apparent in the absence of stress.

A prominent question that has emerged across animal and human models of decision-making concerns how individual differences shape the way in which reinforcement drives subsequent choice behavior. The propensity to repeat previously rewarded actions and avoid those that do not yield reward is a fundamental tenet of decision-making 1 and highlights the critical role that learning plays in the choices we make. In recent years, this tightly coupled relationship has been investigated using reinforcement learning (RL) approaches that formalize how the value of candidate actions are learned through experienced outcomes and how distinct valuation systems may contribute to decision control [2][3][4] . In addition to providing a mechanistic understanding of how one's history of reinforcement drives subsequent choice behavior, these computational approaches also afford the opportunity to examine how individual differences shape distinct features of reward-driven behavior [5][6][7][8][9] .
A growing body of work supports the notion that both trait-like individual differences and more transient changes in affective state exert observable effects on an individual's responsivity to rewarding outcomes 10,11 -and, further, how these outcomes simultaneously shape subsequent choices and the speed with which these choices are made. Two such prominent trait-and state-like factors that have garnered considerable attention in the literature are impulsivity and stress exposure, respectively. Impulsivity-a multidimension construct defined broadly as the general disposition to rash action, including, but not limited to, acting without substantial forethought or consideration of potential consequences, waiting/motor impulsivity and perseverance, and inattention [12][13][14][15] -has been shown to relate to poor financial, health, social and professional outcomes [16][17][18] , and to contribute significantly to almost all psychiatric disorders marked by pathological choice, such as substance use [19][20][21][22][23][24] and impulse control disorders 25 , obesity [26][27][28][29] , and excessive gambling 30 . The observation that impulsive behavior typically emerges in rewarding contexts suggests that an individual's trait level of impulsivity may be associated with differences in Figure 1. Structure of the Two-Stage reinforcement-learning task. In each trial, subjects chose between two initial options, leading to either of two second-stage choices (green or blue states), for different, slowly changing, chances of monetary reward. Each first-stage option preferentially led to one of the two second-stage states ("common"), however, on 30% of trials ("rare") it instead led to the other.
for examination of how immediately preceding rewards impact choices-and RTs-that never lead to immediate rewards. Here, we take advantage of the sequential nature of this task to probe how both impulsivity and stress affect two features of reward learning: the probability of choosing rewarded options and choice speed (RT), at each choice stage.
Based on prior findings using simple RL tasks, we hypothesized that impulsivity and acute stress might exert independent or joint effects on learning as indicative of enhanced reward responsivity. Specifically, we expect these factors will be associated with an increased propensity to repeat rewarded choices (and perhaps choice more generally, irrespective of reward, given purported deficits in deliberative decision-making processes), and that this may selectively emerge in second-stage choices, which-unlike first-stage choices-can yield immediate rewards. Given previous work on impulsivity and RT 63 , as well as past theoretical accounts that stress intensifies implicit (or internally-imposed) time pressure 57 , we further expect impulsivity and stress to affect choice RTs. While we have reason to believe these constructs should affect RTs similarly across both choice stages, it is possible that we may see divergence in RTs among first-stage choices that can uniquely occur after previous reward.

participants.
Fifty-six healthy individuals participated in the study (30 female, age: M = 25.67 years; SD = 7.27 years) and were paid 5 cents per rewarded trial to incentivize performance. The proportions of females in control and stress conditions were 0.50 and 0.58 respectively (see below). All research and experimental procedures were approved by the New York University Committee on Activities Involving Human Subjects and were performed in accordance with these approved Institutional Review Board guidelines and regulations. Written informed consent was obtained from all participants. Following our earlier study which utilized this dataset 49 , we identified and excluded participants who failed to meet a response deadline on more than 15 trials (n = 3), and who failed to demonstrate responsivity to reward as defined by repeating previously rewarded second-stage responses on less than 50% of trials (n = 4). impulsivity assessment. Participants were administered the BIS-11 questionnaire as a measure of trait impulsivity 64 , which consists of 30 statements, such as "I do things without thinking" and "I am more interested in the present than the future" with which participants stated their level of agreement on a four-point scale. Higher summed scores indicate higher levels of impulsivity. Total BIS-11 scores ranged from 33 to 88 (M = 58.0, SD = 10.44). Importantly, BIS-11 scores did not differ significantly between the control (M = 58.15, SD = 11.36) and stress conditions (M = 57.80, SD = 9.06) [t = 0.11, p = 0.91], described below.
Acute stress manipulation. Participants were randomly assigned to undergo a stress or control manipulation prior to the task. In the stress condition (n = 20), participants underwent the Cold Presser Task (CPT) 65 , during which they were asked to immerse their right hand up to and including the wrist for 3 min in ice water (0-5 °C). Participants in the control condition (n = 28) submerged their right hand up to and including the wrist for 3 min into room temperature water (21-30 °C). Immediately after the manipulation, participants indicated on a scale ranging from 0 ("not at all") to 10 ("very much") how unpleasant they found the immersion procedure. As reported in our original study, the CPT manipulation successfully evoked a subjective stress response: participants in the stress condition reported that the CPT was significantly more unpleasant (M = 6.68, SD = 0.54) than those in the control condition (M = 2.19, SD = 0.38) [t = 6.95, p < 0.001].
To assess physiological stress responses, saliva samples were also collected throughout to assess participants' cortisol levels, using an absorbent oral swab that participants placed under their tongues for 2 min. To control for diurnal rhythms in cortisol levels, all participants were run between 1 pm and 6 pm. Sample collection occurred at baseline after a 10 min acclimation period (s1), immediately after baseline cognitive measures (not reported) and task instructions (s2, ~25 min after s1), 10 min after CPT administration (s3, ~43 min after s1), and immediately following the task (s4, ~64 min after s1). Samples were frozen and preserved immediately after collection at −30 °C and transported frozen to a CLIA-certified analytical laboratory where cortisol concentrations were determined with high-sensitivity enzyme immunoassay kits (Salimetrics, LLC, State College, PA). Duplicate assays were conducted for each sample interval, and the average of the two values was used in our analyses.
Cortisol responses were found to peak during the task (i.e, 10 min after the stress manipulation, cf. Figure 2 of Ref. 49 ). As previously reported, we found a significant interaction between condition (stress/control) and time of cortisol measurement (F = 19.99, p < 0.0001), such that only participants in the stress group exhibited a marked increase in cortisol response. Within both stress and control groups, cortisol concentrations did not change significantly between s3 and s4 (ps > 0.54) further suggesting that cortisol concentrations remained stable throughout the RL task. Thus, to facilitate interpretability of potential interaction effects with impulsivity, our analyses of stress effects focused on condition assignment rather than participant-level cortisol response. two-step decision-making task. Participants performed 200 trials of the two-step RL task 59 (Fig. 1), originally employed in our prior study to dissociate parameterized indices of model-free and model-based strategic contributions to choice behavior. In each two-stage trial, participants first made a choice between two options (depicted as fractals; first-stage), which probabilistically lead to one of two second-stage "states" (colored green or blue). In each of these subsequent states, subjects made another choice between two options (second-stage choice), which were associated with different probabilities of monetary reward. Choosing one of the first-stage options led to one of the second-stage states most of the time (70%) and led to the other second-stage state the remaining 30% of the time. Because the second-stage reward probabilities independently change over time, decision-makers need to make trial-by-trial adjustments to their choices in order to effectively maximize payoffs. Prior to initiating the task, participants were provided with task instructions and completed 10 practice trials to familiarize themselves with the task structure and response procedure. Note that at this point, the control and stress groups were subject to identical procedures and thus differences in choice behavior cannot be attributed to the conditions under which task instructions were received. Following administration of the stress/control manipulation and cortisol sample s3, participants completed 200 trials of the two-step RL task (Fig. 1).
In the first stage, two fractal images appeared on a black background (indicating the initial state), and there was a 1.5 s response window during which participants could choose the left-or right-hand response using the "Z" or "?" keys, respectively. After a response was entered, the selected option was highlighted for the rest of the response window. The background color also changed in accordance with the second-stage state. After this transition, participants' chosen first-stage action moved to the top of the screen. Two fractal images, corresponding to the actions available in the second-stage, were displayed; participants had 1.5 s to make a response. As in first-stage choice, the selected option was highlighted for the rest of the response window. Then, either a picture of a coin (indicating that they had been rewarded on that trial) or the number zero (indicating that they had not been rewarded that trial) was shown. The probability of receiving a reward for either second-stage choice was determined by an independent drifting Gaussian random walk (SD = 0.025) with reflecting boundaries at 0.25 and 0.75. The mapping of actions to stimuli and transition probabilities were randomized across participants. Data analysis. Our analysis approach relied upon a mixed-effects regression approach similar to that used to analyze choice behavior in previous studies using this task 59,66 . Because in our current report we examined first-stage and second-stage choices separately, our regression models jointly analyzed behavior across the two stages using dummy variables specifying at which stage each effect is estimated (see Appendix for syntax used to specify the models). This also allowed us to separately estimate 'baseline' repetition rates and RTs at each stage (effectively allowing for a separate intercept term at each choice stage). At the first stage we estimated the trial-by-trial effect of the previous trial's reward and at the second stage we estimated the trial-by-trial effect of previous reward (conditioned upon the last visit to that second-stage state) and the transition type (common versus rare) that led to that second-stage state. These models were estimated using the lme4 package for the R programming language 67 .
In the model examining choices, we specified a mixed-effects logistic regression (using glmer in R) to simultaneously explain the first-stage choice on each trial (coded as stay versus switch relative to the last first-stage choice made) and the second-stage choice (coded as stay versus switch relative to the last second-stage choice made in that particular state). A second mixed-effects regression model (using lmer in R) was specified for RTs. RTs were log-transformed to remove skewness and RTs exceeding 3 SDs from a participant's mean RT were excluded from analysis in both the choice-and RT-predicting models 68 , resulting in 17,974 total observations. To account for practice effects in regressions predicting RTs, a linear predictor of trial number was additionally included. In both the choice and RT models, within-subject factors were taken as random effects across subjects, and parameter estimates and statistics reported are at the population level. Continuous covariates (BIS-11 scores and trial numbers) were entered into the regressions as z-scores. All other binary predictor variables were coded −1/1 in the case of the choice model and coded 0/1 in the case of the RT model.
Significance testing of individual regression coefficients was performed using Satterthwaite's degrees of freedom method implemented in the lmerTest package 69,70 in the case of the RT-predicting model and Wald tests, as implemented by the lme4 package, in the case of the choice-predicting model. This multilevel modeling approach yields conservative parameter estimates that preclude the need to adjust for multiple comparisons 71 . Because of  (Table 1). (2020) 10:7754 | https://doi.org/10.1038/s41598-020-64540-0 www.nature.com/scientificreports www.nature.com/scientificreports/ the partial pooling inherent in multilevel models, coefficient estimates are "shrunk" toward a population-level mean, effectively correcting for the increased risk of false positives incurred by testing all effects of interest within a single model 72 .

Results
Our previous study examined how acute stress affected the expression of model-based and model-free learning at the first stage as described in detail in Otto, Raio, et al. (2013). We note that since our originally reported effects of stress on model-based learning are reported in this previous work, and others' , they are not considered further in the current manuscript. Here, our analysis focused on possible relationships between acute stress and individual differences in impulsivity, and their interaction, on first-and second-stage choice behavior as a function of previous rewards-in effect, focusing on 'model-free' features of choice-and choice RTs as a function of previous rewards and state transitions.
First-stage choice behavior. As is typically observed in two-stage tasks of this kind, we found a significant main effect of previous reward on first-stage choice behavior such that participants were more likely to repeat previously rewarded first-stage actions (β stage 1× previous reward term , p < 0.0001; see Table 1 for full regression coefficient estimates and degrees of freedom). This main effect of previous reward is usually interpreted as the contribution of a 'model-free' RL strategy 49,59,66 . This first-stage responsivity to previous rewards did not appear to be influenced by either stress condition, impulsivity level, or their interaction (β stage 1× previous reward × stress term, p = 0.910, β stage 1× previous reward × BIS-11 term, p = 0.463, β stage 1× previous reward × stress × BIS-11 term, p = 0.892, respectively). In addition, there were no significant main or interaction effects on participants' tendency to choose the same option overall irrespective of previous reward (all p > 0.428; see Table 1).
Second-stage choice behavior. We next examined second-stage choice (which could lead to immediate reward), conditioned upon choice made on the participant's last visit to that second-stage state and as a function of reward obtained on that visit. We found a significant main effect of reward (β stage 2 × previous reward in state, p < 0.0001; Table 1), mirroring previous findings 73 . This effect of previous reward conditioned on the last visit to that state is indicative of a 'win-stay' strategy and can also be taken as an index of responsivity to recent reward feedback 46 .
Importantly, we found this win-stay like behavior increased with individuals' level of impulsivity (Fig. 2), as indicated by a significant interaction between recent reward feedback at the second-stage and impulsivity level (β stage 2× previous reward in state × BIS-11 term, p = 0.019; Table 1). However, the relationship between impulsivity level and responsivity to recent rewards did not appear to depend on participants' assigned stress condition (β stage 2× previous reward in state × BIS-11 × stress term, p = 0.850; Table 1). Instead, the overall tendency to repeat second-stage choices (that is, regardless of whether they were previously rewarded or not) was predicted by stress condition (β stage 2 × stress term, p = 0.025; Table 1), an effect qualified by a stress condition by impulsivity level interaction ( Fig. 3; β stage 2 × BIS-11 × stress term, p = 0.001). That is, stressed participants exhibited a greater tendency to select the same options regardless if they previously led to reward, and this effect was specifically present in more impulsive participants under stress. These results indicate that repeating previously rewarded choices at the second-stage was related to impulsivity-but not stress condition-while choice repetition in general was predicted by stress condition and its interactive effect with impulsivity.
first-stage Rts. We next examined whether first-stage choice RTs-irrespective of recent reward outcomes-differed by impulsivity level and acute stress. Neither impulsivity (β stage 1 × BIS-11 term, p = 0.891) nor stress condition (β stage 1 × stress term, p = 0.215) alone predicted overall first-stage RTs (see Table 2 for full regression coefficient estimates). However, we observed a negative interaction between stress and impulsivity level, such that response speeding increased in more impulsive individuals under stress (Fig. 4A,B; β stage 1 × stress × BIS-11 term, p = 0.003). In other words, acute stress selectively increased choice speed in more impulsive participants, while this relationship with impulsivity was not present under control conditions. We also observed a tendency for a reward received on the previous trial to slow subsequent first-stage RTs (β stage 1 × previous reward term, p < 0.0001). This post-reward tendency for slowing was significantly predicted by an interaction between impulsivity level and stress condition (β stage 1 previous reward × stress × BIS-11 term, p = 0.027) (Fig. 4C,D). This effect was not significant for impulsivity (β stage 1 × previous reward × BIS-11 term, p = 0.52) nor stress condition (β stage 1 × previous reward × stress term, p = 0.114) alone, suggesting that slowing after previous reward is observed only when more impulsive participants were stressed.

Second-stage Rts. A similar pattern of results emerged for RTs associated with second-stage choices.
Choice speed at the second-stage (irrespective of previous reward) was not affected by impulsivity or assigned stress condition (β stage 2 × BIS-11 term, p = 0.715 and β stage 2 × stress term, p = 0.707, respectively). However, again, the interaction between stress condition and impulsivity level on RTs was significant (β stage 2 × stress × BIS-11 term, p = 0.032), suggesting that only stressed participants who are more impulsive made faster choices overall at the second-stage (Fig. 5).
Unlike the first-stage choice RTs, we did not find that reward received on the immediately preceding trial in that state slowed subsequent second-stage RTs made in the same state (β stage 2 × previous reward in state term, p = 0.212), nor did we find these RTs to be predicted by impulsivity level (β stage 2 × previous reward in state × BIS-11 term, p = 0.231). However, we did find that previous reward in the same state tended to slow subsequent second-stage RTs in more impulsive participants under stress, although this interaction between impulsivity level and stress condition was only observed at trend level significance (β stage 2 × previous reward in state × stress × BIS-11 term, p = 0.065).
Effect of transition structure. As previous work with the same task has found that second-stage RTs are slower following rare transitions than common transitions 66,73 , we also examined how this second-stage slowing was related to participants' impulsivity level and stress. Post-transition slowing is thought to reflect surprise resulting from an uncommon event (assuming participants have knowledge of the task transition structure). We found that, as previously observed, participants' second-stage choices were slower following rare transitions than following common transitions ( Fig. 6; β stage 2 × current transition term, p < 0.001) but interestingly, this slowing effect was exacerbated by acute stress such that participants in the stress condition exhibited a larger slowing effect than control participants (β stage 2 × stress × current transition term , p = 0.01, see Table 2, Fig. 6).

Discussion
Impulsivity and exposure to acute stress are two prominent factors thought to alter reward-related learning and decision-making, yet few investigations have examined how these factors might jointly alter RL processes in humans. Leveraging a two-stage RL task, we examined how reward-contingent (and surprise-contingent) choice and decision speed are shaped by acute stress exposure and trait impulsivity.
Based on the extant literature, we hypothesized that impulsivity and acute stress might independently-or jointly-exert effects on learning that would be indicative of enhanced reward responsivity. Indeed, we observed that the tendency to repeat a previously reinforced choice increased with higher impulsivity levels, suggesting a "win-stay" or "Law of Effect" tendency prevails in more impulsive individuals. Importantly, this relationship between impulsivity and reward responsivity only emerged in second-stage choice behavior where choices result in immediate rewards, but not in first-stage choices that cannot lead directly to reward. This is consistent with a  (Table 1).

Scientific RepoRtS |
(2020) 10:7754 | https://doi.org/10.1038/s41598-020-64540-0 www.nature.com/scientificreports www.nature.com/scientificreports/ broader conceptualization of trait impulsivity as exerting a stronger effect on behavior in immediately rewarding contexts 32 . Interestingly, while impulsivity levels predicted this manifestation of reward responsivity, acute stress exposure alone yielded no such effect, nor did it interact with impulsivity to influence this behavior. This suggests that impulsivity appears to selectively render individuals more susceptible to repeat rewarded (but not unrewarded) choices, a dissociation that was revealed because an inherent feature of our task structure is that second-stage choices are immediately followed by an outcome while first-stage choices are not.
We did, however, observe a main effect of stress condition as well as an interaction between acute stress and impulsivity in overall choice tendencies, such that more impulsive participants under stress were more likely to repeat second-stage choices regardless of whether these choices were previously rewarded. This perseveration-like behavior is consistent with notions that stress and impulsivity impair deliberative decision processes [45][46][47][48][49][50] and suggest these factors may interact to render participants less sensitive or precise in their representation of the task's reward structure, such that choices are repeated even if only a subset of them are rewarded. The finding that this interaction was significant only in second-stage choices is consistent with the fact that only second-stage choices are associated with immediate rewards, while first-stage choices are not. Together, these results suggest that higher impulsivity levels increase the propensity to repeat previously rewarded actions (akin here to better learning), but that this propensity generalizes to unreinforced choices when coupled with stress exposure. This points to stress exposure as driving a potentially maladaptive form of choice perseveration or repetition-especially in more impulsive individuals-that persists independent of reward outcome. This inflexible adjustment of choice behavior to reinforcement has also been documented in previous investigations showing that acute stress exposure leads to reduced responsivity to both positive and negative feedback during the learning phase of probabilistic reward tasks 45,46 .
Not only did acute stress render impulsive participants more likely to repeat choices independent of reward, but it also made choices faster in a reward-independent manner, particularly in the first-stage. Specifically, under stress, more impulsive participants made faster choices overall, while this effect was not present under control conditions. The fact that stress increases decision speed in more impulsive participants is consistent with theoretical 57 and empirical 58 accounts that suggest stress might engender 'internal' time pressure, which may already be a trait-like feature of impulsive individuals' decision process 63 , that is most pronounced under stress.
While impulsive participants were faster to respond when under stress, a distinct RT profile emerged for responses following reward. Specifically, we found that receiving a reward on the previous trial (i.e., at the second-stage) slowed subsequent first-stage RTs. This post-reinforcement slowing was more prevalent among impulsive individuals when under acute stress. Such post-reinforcement "pausing" has been described in a broad range of reward-based choice tasks across species as slower RTs after reward receipt or 'wins' [74][75][76][77][78] as well as faster response rates after unexpected omission of reward or loss 34,36,78,79 . This post-reward slowing is generally thought to reflect greater attentional or orienting responses to receipt of reward. However, one possibility is that rewards tend to incur reward prediction errors (PEs) 59 , which has been found to slow subsequent choices in a variety of tasks 80,81 . While we do not measure PEs directly, theoretically, an enhanced responsivity to PEs in stressed individuals who are more impulsive could potentially arise from changes in expectation of reward, which is consistent with the neuromodulatory changes imposed by stress exposure, particularly dopamine-which drives PE signaling 59,82 and changes rapidly after stress exposure 56,82 . Post-reinforcement pausing may thus result from elevated orienting response to reward receipt in impulsive individuals under stress. An alternate possibility is that after reward receipt, participants experience greater conflict before making subsequent first-stage choices (given the nature of our two-stage task), an account supported by recent demonstrations of RT slowing with greater decision difficulty or conflict 83,84 . Such decision conflict may be exacerbated in impulsive individuals under stress, consistent with the well-established finding that stress decreases cognitive capacity and flexibility. Future work using joint modeling approaches-such as recent demonstrations using RL models paired with drift diffusion models 83-85 -could help clarify the interplay between these choice and RTs effects. www.nature.com/scientificreports www.nature.com/scientificreports/ Although the overall patterns of RTs observed at the second-stage mirrored that of the first-stage for the interactive effect of impulsivity and stress on overall RTs-i.e., regardless of previous reward, more impulsive subjects were faster under stress-we observed a distinct pattern between the two stages with respect to RTs following a previously rewarded choice. Specifically, unlike first-stage RTs, reward-based slowing of second-stage RTs was not modulated jointly by impulsivity and stress. This divergence may arise from differences in reward availability in each choice stage stemming from the task structure (Fig. 1). Second-stage choices can lead to immediate reward, while first-stage choices cannot. Consequently, only first-stage choices can immediately follow the receipt of reward. Thus, our ability to detect an effect of acute stress and impulsivity upon choice and RTs following feedback might require an immediately preceding rewarding event. Future model-based work examining RTs in sequential RL tasks, following recent work by Shahar and colleagues 85 , will be especially important to better understand how individual differences drive such differential effects on choice behavior at distinct stages.
Finally, consistent with previous work, we found that second-stage RTs were slower following rare transitions than common transitions 66,86 , and further, that, while not related to impulsivity, this slowing effect was exacerbated by stress. This responsivity to transition frequency has been interpreted as a reflection of knowledge of the transition structure, wherein uncommon transitions engender surprise (i.e., expectancy violation 66 ). Here, acute stress-but not trait impulsivity-appeared to intensify this response to surprising events, suggesting that acute stress could tune individuals to simple violations of expectations. Interestingly, recent work identifies surprise as a key driver of (subjective) stress response 87 , suggesting the possibility of a bidirectional, positive feedback relationship between surprise stemming from the environment and the acute stress response. stress conditions, and upon first-stage differential RTs following previous reward (versus non-reward) in the (C) control and (D) stress conditions. Median first-stage choice RTs (or post-reward -no-reward RT differences) for individual subjects are depicted on the vertical axis. The regression line is computed from the group-level (fixed-effect) estimated from the mixed regression model (Table 2).
Notably, the sequential structure of the choice task enabled us to probe simple relationships between trait impulsivity and acute stress and how choice and RT are affected by previous reinforcement in a way that could not be revealed in a standard RL task. Specifically, the sequential nature of the task allowed us to examine the serial effect of choice (e.g. how reward received following second-stage choices affected subsequent first-stage choices), which would not be possible to ascertain without this two-stage feature. This task structure further afforded the opportunity to independently characterize how choice behavior and decision speed changed in firstvs. second-stage components of the task as a function of trait impulsivity and stress exposure.
Converging lines of research have highlighted the importance of characterizing how both individual differences and affective state can shape learning and decision-making processes. Our results reveal novel interactive effects of trait impulsivity and stress exposure and suggest that rather than exerting a direct effect on reward repetition and choice speeding, stress appears to reveal choice tendencies in individuals higher in trait-impulsivity. This pattern is corroborated by work that points to stress as a factor that reveals underlying choice biases rather than exerting a purely directional effect on choice behavior 41,88 .
Our findings motivate a number of future research directions. First, while impulsivity is undoubtedly a multidimensional construct 16 , here we used the BIS-11 64 , a widely used and validated self-report measure of impulsive behavior to measure and define impulsivity. Future research may seek to test how more circumscribed forms of