Winning and Losing: Effects on Impulsive Action

In the present study, we examined the effect of wins and losses on impulsive action in gambling (Experiments 1–3) and nongambling tasks (Experiments 4–5). In each experiment, subjects performed a simple task in which they had to win points. On each trial, they had to choose between a gamble and a nongamble. The gamble was always associated with a higher amount but a lower probability of winning than the nongamble. After subjects indicated their choice (i.e., gamble or not), feedback was presented. They had to press a key to start the next trial. Experiments 1–3 showed that, compared to the nongambling baseline, subjects were faster to initiate the next trial after a gambled loss, indicating that losses can induce impulsive actions. In Experiments 4 and 5, subjects alternated between the gambling task and a neutral decision-making task in which they could not win or lose points. Subjects were faster in the neutral decision-making task if they had just lost in the gambling task, suggesting that losses have a general effect on action. Our results challenge the dominant idea that humans become more cautious after suboptimal outcomes. Instead, they indicate that losses in the context of potential rewards are emotional events that increase impulsivity.

The pattern in Figure S1 is inconsistent with the idea that the decision to gamble is always a rash decision or an impulsive act (see also e.g. Losecaat Vermeer, Boksem, & Sanfey, 2014). Stochastic accumulator models of decision making (Smith & Ratcliff, 2004) can offer a parsimonious explanation for the correlation between the choice RT and probability of gambling. Such models assume that decision making involves the accumulation of noisy information until there is enough support for a specific option. The main parameters of the selection process are the response criteria (i.e., how much information is required for an option to be selected), accumulation rate (i.e., how quickly does evidence accumulate), and the starting point (i.e. a priori bias against one or the other choice alternatives; Figure S2). The correlation between p(gamble) and choice RT can be explained by individual differences in the starting point: when subjects have a bias against gambling (i.e. they are 'gambling-averse'), the distance between the starting point and the 'gambling' boundary will be larger than the distance between the starting point and the nongambling boundary ( Figure S2, right panel). Consequently, the gambling option will be selected less frequently because the accumulated evidence in favor of it is less likely to reach the gambling boundary first. Furthermore, if the gambling boundary is reached after all, the latency of the gambling response will be (on average) longer than the latency of nongambling responses ( Figure S2, right panel). Thus, risk preference can be captured by individual differences in the starting point. Figure S2: A bias in the starting point (Z) of a sequential decision-making process can explain both the overall probability of gambling and latency differences. The left panel reflects a hypothetical 'gambling-neutral' subject; the right panel reflects a hypothetical 'gambling-averse' subject.

Does the rating influence start RT of the next trial in Experiment 2?
The p(gamble) analysis for Experiment 2 suggests that the ratings induced a reflective mode: compared with the non-gambling baseline, p(gamble) after a loss decreased on rating trials, but increased (slightly) on no-rating trials.
Here we explored if the ratings also influenced start RT on the next trial (note that the statements were presented after subjects had pressed the start key, so ratings could not influence start RT of the current trial). The results are presented in Figure S3. There was a significant main effect of trial outcome, F(2,78) = 11.532, p < . Rating trial n−1 no yes Figure S3: Start RT as a function of the outcome of the previous trial and rating properties of the previous trial (no-rating trial vs. rating trial). Error bars are 95% confidence intervals.

Does the outcome of the non-gambling task influence start RT and p(gamble) on the next trial in Experiments 4 and 5?
In Experiment 4, the outcome of the immediately preceding perceptual decision-making trial (trial n-1) influenced performance in the gambling task: subjects started the next In Experiment 5, subjects alternated between the gambling task and a stop-signal task.
The outcome of the immediately preceding stop-signal trial (trial n-1: correct go response, successful stop, unsuccessful stop) did not influence performance in the gambling task much. Subjects started the next gambling trial later after a failed stop (M = 779 ms; SD = 347) than after a successful stop (M = 741 ms; SD = 209) or a correct go (M = 725 ms; SD = 242), but these differences were not statistically significant (Table S1). Nevertheless, the numerical trends are consistent with the idea that subjects increased the priority of the stop goal after a signal trial (Bissett & Logan, 2011), and this could have counteracted the affective consequences of a negative outcome. Table S1 shows that p(gamble) were similar for trials that followed a correct go (M = . 49; SD = .17), a successful stop (M = .51; SD = .18), or an unsuccessful stop (M = .48; SD = .19), replicating our previous findings (Stevens et al., 2015). Choice latencies were also similar for trials that followed a correct go (  Note: correct = trials preceded by a correct go (no-signal) trial, SR = trials preceded by a failed stop trial (signal-respond), SI = trials preceded by a successful stop trial (signal-inhibit).

First vs. second half of the experiment
In a final set of analyses, we explored whether the sequential effect of gambling changed throughout the experiment. More specifically, we contrasted performance in the first and second half of the experiment. To increase power, we combined the data of all five experiments again. The relevant descriptive and inferential statistics appear in Tables S2 and S3, respectively. We will focus on the interaction between Trial Outcome and Part only. Trial outcome had a similar effect on start RT and p(gamble) in the first and second half of the experiment (ps > .18; Table S3). We observed a marginally significant interaction between Trial Outcome and Part in the choice RT analysis (p = .051; Table S3). Table S2 shows that choice RTs were longer after a gambled win than after a non-gamble only in the first half of the experiment. Note that choice RTs were shorter after a gambled loss than after a non-gamble in both parts. Table S2: Overview of the mean start RT, probability of gambling, and choice RT in the gambling task as a function of preceding gambling trial for the first and second part of the experiment (Part 1 vs. Part 2). The data of Experiments 1-5 are combined.