The influence of reward and loss outcomes after free- and forced-tasks on voluntary task choice

The capacity to regulate behavior in environments characterized by multiple sources of information is crucial for adaptive functioning. On the one hand, effective behavior necessitates the ability to maintain stable information processing aligned with the current task goal. On the other hand, it is equally vital to continuously monitor both one's own behavior and the environment for cues that signal the potential benefits of switching to an alternative task goal. Unraveling the cognitive mechanisms that enable us to navigate these contrasting demands effectively represents a significant challenge in cognitive psychology (e.g., Dreisbach & Fröber, 2019; Eppinger et al., 2021; Goschke, 2000).

In most cases, the achievement of a task goal entails engaging in demanding cognitive processes, such as studying to pass an exam. In the field of cognitive psychology, the concept of goals as mental representations that guide our behavior has typically been investigated in laboratory settings through the notion of task-sets, which encompass the rules that participants need to follow in order to successfully link stimuli to responses (e.g., Kiesel et al., 2010; Koch et al., 2018; Meiran, 2010; Monsell, 2003; Schumacher & Hazeltine, 2016). For example, participants may be instructed to perform a letter task, wherein they determine whether a letter is a vowel or consonant and respond accordingly. Even the execution of such seemingly straightforward tasks requires cognitive effort, and research indicates that individuals weigh the costs and benefits associated with expending cognitive resources when deciding how to perform a task (e.g., how much effort to invest in studying for the exam or completing the letter task) (e.g., Shenhav et al., 2013, 2021). When deciding which of multiple tasks to choose, individuals often exhibit a bias towards perseveration, wherein they favor repeating the same task (e.g., Arrington & Logan, 2004; Koch et al., 2018; Kool et al., 2010). Based on the notion that people choose the task with the highest value, this indicates that switch task goals are less valued than the repetition task goals, because of the temporal costs (e.g., Mittelstädt et al., 2019, 2021) and the involved cognitive effort (e.g., Kool et al., 2010; Mendl & Dreisbach, 2022) when deciding to switch tasks.

In the present study, we are investigating whether and how outcomes (rewards and losses) after choosing and performing a task can further modulate the value of task goals to bias subsequent task choice behavior. Many studies have demonstrated that task performance-related control signals can be modulated by receiving performance-contingent or -independent rewards to bias subsequent task processing (e.g., Braem et al., 2012; Stürmer et al., 2011; van Steenbergen et al., 2009; Yamaguchi & Nishimura, 2019), presumably because receiving rewards can strengthen recently activated task goals (e.g., Umemoto & Holroyd, 2015), aligning with Thorndike's (1927) law of effect. Moreover, several studies have shown that when people choose between different options (e.g., different cards in gambling-like situations), they are more likely to repeat the same option following a reward and switch to another option following a loss (e.g., Elston et al., 2021; Worthy et al., 2013). Thus, although we are not aware of any studies investigating the effects of recent outcomes when people decide between different cognitive tasks rather than different options in gambling-like situations, it is probably not surprising if receiving a reward (compared to no reward) also provides choice-related signals that bias subsequent task choices. Specifically, we hypothesize that when people receive rewards for correct performance, they will be less likely to switch away from the current task in the subsequent trial compared to a condition without rewards. Conversely, receipt of losses (compared to no losses) will increase the likelihood of switching to the other task.

While previous studies have investigated the influence of prior reward cues and learning about specific behavior-reward associations on voluntary task choice behavior (e.g., Braun & Arrington, 2018; Fröber & Dreisbach, 2016; Spitzer et al., 2024), to our knowledge, no study to date has examined the immediate effects of outcome reception on subsequent voluntary task choices without any explicit reward cues or prior learning phases. For example, Fröber and Dreisbach (2016) have examined the motivating effects of reward anticipation on voluntary behavior by providing cues signaling the possibility of low or high (task-unspecific) rewards before each trial. Moreover, Braem (2017) demonstrated that participants who received greater rewards for task switches compared to task repetitions during a learning phase exhibited higher levels of switching behavior in a subsequent test phase, suggesting reinforcement learning of more abstract (task-unspecific) components related to cognitive flexibility.

Importantly, the present study goes beyond these prior studies by considering that people may differentially value free-choice compared to forced-choice task goals. In general, the investigation of voluntary task choice behavior can be approached using various experimental methods (e.g., Arrington & Logan, 2004, 2005; Brosowsky & Egner, 2021; Brüning et al., 2020; Dreisbach & Jurczyk, 2022; Jurczyk et al., 2019, 2021; Mittelstädt et al., 2022, 2023; Mittelstädt et al., 2018a, b; Spitzer et al., 2022; Wong et al., 2022). Critically, in the present study, we employed the hybrid free-forced choice paradigm introduced by Fröber and Dreisbach (2017). This paradigm involves the intermixing of free and forced choice trials, where participants are presented with either one task stimulus (forced choice) or two task stimuli (free choice) (see Fig. 1). In a modified version of this paradigm, Qiao et al. (2022) reported additional performance costs between free and forced-choice trials, even when repeating the same task, suggesting that free and forced-task goals may be differentially internally represented. Moreover, they reported a larger task repetition bias following free-choice trials compared to forced-choice trials, suggesting that participants are less inclined to disengage from a task goal that was freely chosen. This could indicate that choosing a task is itself rewarding and, hence, that a differential valuation of freely versus forced-chosen task goals can bias subsequent choice behavior. While Mittelstädt et al. (2024) have recently shown that the findings by Qiao et al. (2022) cannot be explained by the specific cues used, the reciprocal relationship between processing modes and behavioral outcomes was not investigated.¹

Specifically, while we generally predict that people are more likely to switch away from a task when experiencing negative outcomes (no reward or loss) than positive outcomes (reward or no loss), and when they have not chosen the task themselves, it is also possible that these two choice biases interact rather than being additive. Drawing on theoretical proposals and studies from various research fields (e.g., Cockburn et al., 2014; Mühlberger et al., 2017; Leotti et al., 2010; Yeung et al., 2005; Zhou et al., 2010), it seems possible that individuals pay more attention to the outcomes of tasks they have freely chosen compared to those that were imposed on them. Moreover, and not mutually exclusive, people may inflate the value of outcomes associated with freely-chosen task goals as they might believe their personal choice led to the improved outcome, even though positive and negative outcomes were randomly delivered for correct performance. For example, Yeung et al. (2005) observed increased neural outcome monitoring in a simple gambling task (i.e., selecting between different colored circles) when participants made active choices rather than when a choice was made for them. Although we are not aware of any studies investigating the interplay between task processing mode (free vs. forced) and outcomes (e.g., reward vs. no reward) on subsequent voluntary task choice behavior, these previous studies suggest that outcome manipulations might have a greater influence on switching behavior following free- compared to forced-task conditions.

In this experiment, participants freely chose between two tasks (letter and color) in half of the trials (i.e., two stimuli were presented), whereas in the other half of trials, they were forced to perform a task (i.e., only one stimulus was presented). As is common in this and other voluntary task-switching paradigms, each task was assigned to a specific hand, allowing task choice to be identified based on the response made (e.g., Arrington & Logan, 2004; Fröber & Dreisbach, 2016; Liu & Yeung, 2020; Mittelstädt et al., 2022). Critically, participants randomly received reward or no reward after correctly performing a task. Our focus was to examine which task participants voluntarily selected in free choice trials (i.e., the task associated with repetitions or switches) as a function of previous task mode (free versus forced) and previous trial feedback (reward versus no-reward). We hypothesized that participants will be more likely to switch after a forced-chosen as opposed to a freely-chosen goal and that they will be more likely to switch after having received no-reward compared to reward (i.e., main effects of previous mode and feedback). Furthermore, we tested whether participants’ sensitivity to obtaining reward (versus no reward) is larger after a freely-chosen as compared to a forced-chosen task.

Experiment 1 showed on average fewer task switches following a reward compared to when no reward was received. Additionally, participants showed a general reluctance to switch away from a freely chosen task goal, indicating a preference for tasks that were freely selected. Furthermore, the influence of the reward manipulation was more pronounced (and only significant) following freely-chosen tasks compared to forced-chosen tasks, suggesting that participants placed greater emphasis on receiving rewards for tasks they had freely chosen. The goal of Experiment 2 was to determine whether these findings would replicate when using a manipulation involving losses (versus no losses) as outcomes.

Experiment 2 revealed that not only reward versus no-reward, but also loss versus no-loss outcomes can influence subsequent task choice behavior by showing more task switches after loss compared to no loss. Furthermore, participants showed again a general preference for freely-chosen task goals over forced-chosen task goals, as they were more likely to switch away from a forced-chosen task goal. Contrary to Exp. 1, there was no evidence that the influence of the loss manipulation differentially impacted freely-chosen and forced-chosen tasks. While this suggests that rewards and losses, at least partially, differentially influence task choice behavior, one may also argue that the interaction in Exp. 1 reflects a false positive (Type 1 error). As this interaction seems theoretically interesting, we decided to conduct another experiment with a reward manipulation to see whether we would replicate the pattern observed in Experiment 1. Thus, Experiment 3 was similar to Experiment 1, except that we additionally varied the time after feedback presentation (intertrial interval short vs long).

The intertrial interval (ITI) is often varied in task-switching studies, and previous research (without outcome manipulations) has demonstrated increased switching behavior after a long compared to a short SOA (e.g., Arrington & Logan, 2005; Mittelstädt et al., 2019). Thus, observing a similar effect in the present study would more strongly connect the current task environment to the previous ones. Moreover, it also allows one to gain insights into the temporal dynamics of previous trial characteristics influencing subsequent choice behavior. Specifically, we wanted to explore whether the impact of the previous mode and previous feedback decays over time or remains relatively stable. Previous task-switching studies have suggested that the activation of previously applied task goals decays over time (e.g., Altmann & Gray, 2002). Additionally, several decision-making studies using neural networks have demonstrated that the influence of previous decisions decreases over time due to a decay of neural activation (e.g., Bonaiuto et al., 2016; Rustichini & Padoa-Schioppa, 2015; Urai et al., 2019). Since both positive outcomes and freely chosen tasks may enhance task goal values, their effects may also decay over time. However, participants may employ additional control processes to counteract this decay, enabling them to use this information in subsequent trials when facing another potentially effortful decision between tasks.

Experiment 3 replicated the pattern observed in Experiment 1: Participants specifically preferred to repeat freely-chosen tasks compared to forced-chosen tasks and they specifically biased their task choice behavior based on reward versus no-reward outcomes for freely-chosen tasks. These influences on free-choice behavior were, if any, little influenced little by the time after feedback was presented. The goal of Experiment 4 was to determine whether participants would be similarly affected by the reward manipulation and previous mode when separating task choice from task processing. Specifically, in contrast to the previous experiments, participants in Experiment 4 could first choose which of two tasks they wanted to perform (two task cues) or were forced to choose (one task cue) before the corresponding task stimulus was presented (double-registration procedure, see Fig. 3). It is difficult to predict whether the previous pattern will replicate or not, as there are good arguments for both alternatives. For example, the reinforcing effects of freely-choosing tasks and obtaining rewards (as well as their interaction) could also be observed with this approach if these effects primarily arise due to changes in higher-level goal representations rather than lower-level perceptual and/or motor processes associated with task stimuli and responses. However, without seeing the actual task they had to perform with a double registration procedure and by temporally separating choice and task processing, participants might feel less attached to task goals and perceive they have less control over outcomes—a prerequisite that may be critical to observe biases in choice behavior.

As the belief in control over the (random) reward manipulation might influence choice behavior, we also explicitly asked participants to which extent they believed that the reward was contingent on their task choices and additionally measured via a questionnaire their individual locus of control, that is, how strongly they attribute outcomes in life to their own actions rather than external forces (e.g., Rotter, 1966; Stolz et al., 2020).

The results of all four experiments demonstrate that participants are less likely to switch away from a free-choice task compared to a forced-choice task (i.e., main effects of previous mode on switch rates). One possible explanation for this preference is that participants have an overall preference for one of the two tasks. Consequently, they might tend to select that particular task more frequently in free-choice trials. In situations where they encounter two consecutive free-choice trials, they may opt for the same task both times due to their overarching preference for it, thereby increasing repetition rates. On the contrary, when the preceding trial involved a forced-choice task, it could have been either the preferred or the less favored task. As a result, choosing the preferred task in the next trial would lead to smaller repetition rates. To investigate this possibility, we categorized each participant as either having an “overall color preference” or an “overall letter preference.” This categorization allowed us to classify the previous task response in each trial as either the “preferred task” or the “nonpreferred task.” We hypothesized that if participants are still more likely to switch away after a forced-choice trial than a free-choice trial when controlling for the fact that the previous task for both modes was either the preferred or nonpreferred one, an overall task preference account cannot explain the specific free-choice repetition bias.

Indeed, when the previous task was the nonpreferred one, participants were in all experiments more likely to switch away after a forced-choice trial than after a free-choice one (Exp. 1: 47.4% vs 39.6%, p = 0.092; Exp. 2: 44.8% vs 33.6%, p = 0.001; Exp. 3: 39.3% vs 28.5%, p = 0.002, Exp. 4: 48.5% vs 34.1%, p < 0.001). Moreover, in Exp. 2–Exp. 4, when the previous task was the preferred one, participants were at least descriptively more likely to switch away after a forced-choice trial than after a free-choice one (Exp. 2: 11.8% vs 9.7%, p = 0.084; Exp. 3: 15.6% vs 13.0%, p = 0.127; Exp. 4: 21.1% vs 17.5%, p = 0.008). Thus, while an overall stable preference for one task certainly contributes to the free-task repetition bias, it seems not fully explanatory.

Finally, we also explored how the preference for one of the two tasks in free-choice trials relates to task performance in forced-choice trials, where both tasks had to be performed equally often. Across all four experiments, participants were generally faster in forced-choice trials when responding to the preferred task compared to the non-preferred task (Exp. 1: 664 vs 793 ms, p < 0.001; Exp. 2: 664 vs 783 ms, p < 0.001; Exp. 3: 713 vs 771 ms, p < 0.001; Exp. 4: 620 vs. 646 ms, p = 0.004).

In the present experiments, we examined whether voluntary choosing between two tasks (as measured by switching behavior) depends on the previously obtained outcomes (Exp. 1, Exp. 3, Exp. 4: reward vs no reward; Exp. 2: loss vs no loss) and the previously applied processing mode (free versus forced). Generally, we hypothesized that the most recently applied task goal is usually the most highly valued, leading individuals to prefer task repetitions over switches. However, the reception of positive versus negative outcomes may additionally bias how people value task goals, influencing subsequent task choices. Indeed, positively reinforcing task goals after correct performance by random rewards or no-losses generally resulted in fewer subsequent switches. Furthermore, in all experiments, participants generally showed a stronger attachment to a freely-chosen task compared to a forced-chosen task, as reflected in more switches when the previous task was forced rather than free. Interestingly, additional evidence supporting the idea that people differentially value free and forced-choice task goals comes from the finding that participants' switching behavior was more strongly influenced by reward (versus no-reward following freely-chosen tasks compared to forced-chosen tasks) in Exp. 1 and Exp. 3. While this finding suggests that people place a stronger emphasis on outcomes associated with freely chosen task goals, this inflationary effect on free choice behavior seems to be limited to rewards and situations where task choice and task performance more strongly overlap. Specifically, there was no evidence that participants' switching behavior was differentially influenced by losses (versus no loss) following both free and forced task choices in Exp. 2, and also not by reward (versus no reward) when task choice and task performance were separated via a double registration procedure in Exp. 4.

While the effects of rewards on task choice behavior have also been observed in previous voluntary task-switching studies (e.g., Braem, 2017; Fröber & Dreisbach, 2016), to our knowledge, the present study provides the first direct evidence that the immediate impact of recent outcomes can bias subsequent task choice while considering both gain and loss outcomes. Thus, the finding that people tend to switch tasks more often following a loss (versus no loss) and repeat tasks more often following a reward extends previous studies that have shown a similar effect when choosing between different options in gambling-like situations to situations involving the choice between different cognitive tasks (e.g., Elston et al., 2021; Worthy et al., 2013). Assuming that people tend to choose the task with the highest value (e.g., Braun et al., 2018; Spitzer et al., 2024), the present findings suggest that outcome-monitoring processes quickly update the value of different task goals to guide subsequent behavior. As distinct stimuli and responses were used for the two tasks, one could speculate that the reinforcing effect may have also influenced lower-level task-specific processes, such as task-specific response effectors. However, given that outcomes also influence task choices when task performance was separated from task choice in Exp. 4, it seems that outcome-related changes in biasing free task choice behavior depend at least partially on changes in the value of higher-level task representations rather than only on changes in the value of task-specific perceptual input (stimuli) and output (responses).

Moreover, the present study also demonstrates that participants are less likely to switch away from a task when they have the opportunity to choose it compared to when they are forced to choose (cf. Mittelstädt et al., 2024; Qiao et al., 2022). Going beyond previous studies, this specific choice bias remains evident when separating task choice from task performance. Thus, it seems that the act of choosing a task leads to a stronger valuation of that task, increasing the likelihood that its representation will guide subsequent behavior. This finding aligns with research from various domains, consistently demonstrating that options that are freely chosen are more highly valued than those that are not (e.g., Cockburn et al., 2014; Leotti & Delgado, 2014; Patall et al., 2008; Stolz et al., 2020). Moreover, the finding that free-task choices result in more stable behavior, as indicated by fewer voluntary switches, extends previous findings showing that task choice enhances stability by protecting task processing from distracting environmental influences (e.g., Falk et al., 2022; Gendolla et al., 2021).

As outlined in the additional exploratory task choice analyses section, we also investigated whether an overall preference for one of the two tasks could explain why participants tend to repeat a task more often after free-choice compared to forced-choice trials (e.g., in consecutive free-choice trials, participants may consistently select the same task due to an inherent preference for it). The results suggest that stable task preferences alone cannot fully explain why participants show a stronger attachment to a freely-chosen task compared to a forced-chosen task, but these preferences certainly contribute to this effect. Therefore, it appears worthwhile to more directly investigate the role of individual task preferences in voluntary task choice behavior. For example, it is plausible that task preferences vary across trials, suggesting that dynamic changes in task preferences may also partly contribute to the free-task repetition bias. Furthermore, it also seems interesting to consider what drives participants to prefer one task over the other in the first place. Considering that the exploratory analyses also revealed that participants were generally faster when performing their overall preferred task in forced-choice trials, it seems possible that performance optimization might play a critical role (cf. Mittelstädt et al., 2019).

Another novel finding of the present study is that Exp. 1 and Exp. 3 revealed a larger reward-related choice bias after free tasks compared to forced tasks, suggesting that people value more strongly rewarding outcomes of tasks they have freely chosen, as opposed to those that were forced upon them. This finding appears in line with previous studies that have demonstrated increased neural activity, as measured by the feedback-related negativity (FRN, Gehring & Willoughby, 2002), in association with free choice behavior (Mühlberger et al., 2017; Yeung et al., 2005; Zhou et al., 2010). However, in these previous studies, gambling-like choices between options rather than between choice tasks have been investigated and the impact on subsequent behavior has often been neglected (but see Gehring & Willoughby, 2002, for more risky choices after losses). Thus, it remains to be investigated whether similar heightened outcome monitoring on a neuronal level is present when evaluating the outcomes of free versus forced task choices, and to what extent this outcome-related neural activity can be linked to subsequent behavior.

Moreover, the specific causes of this free choice reward bias remain unclear and there are several, not mutually exclusive possibilities. First, one might speculate that people perceive higher efficacy in choosing a rewarding task. Specifically, as participants might be more motivated to engage in a task they have freely chosen rather than one imposed on them (e.g., Falk et al., 2022; Gendolla et al., 2021), they may pay more attention to free choice trials and weigh the associated outcomes more heavily. Second, the requirement to choose freely puts people in a state of uncertainty, making them more keen to determine whether they made the “right” choice. As confidence is generally known to affect choice history biases (e.g., Braun et al., 2018; Sanders et al., 2016), receiving a reward for a freely chosen task might particularly increase confidence in that specific task. Third, people may generally value rewards more strongly when obtained from more effortful tasks (cf. Inzlicht et al., 2018; Zentall, 2010), and positive outcomes may be more rewarding when people choose for themselves because choosing a task itself can be a demanding cognitive process (e.g., Kiesel & Dignath, 2017; Schwartz, 2000; Vohs et al., 2018). Fourth, stronger memory representations might be formed for outcomes of free compared to forced choices (e.g., Chambon et al., 2020; Katzman & Hartley, 2020; Murty et al., 2015). Assuming that people guide their task choices based on memory representations, this could also explain why participants rely more heavily on rewarding previous free choice trials when making subsequent free task choices.

When investigating the causes of this reward-specific free-choice bias in future studies, it also seems useful to consider the boundary conditions of this effect. For example, while research on decision-making suggests that people generally weigh avoiding losses more strongly than obtaining gains (loss aversion, see e.g., Chen et al., 2020; Tversky & Kahneman, 1991; Yechiam & Hochman, 2013), it is not clear how such loss versus gain outcome specificity can explain the similar influences of loss versus no-loss on choice behavior in Exp. 2 as a function of free versus forced choice processing.

Furthermore, we can only speculate why, even with rewarding outcomes, Exp. 4 did not reveal evidence for a reward-specific free-choice bias when task choice and task performance screens were separated. For example, one might argue that when separating choice and task processing, participants were better at evaluating that the reward manipulation was not contingent on their behavior, as correct performance was randomly rewarded. The questionnaire data in this experiment align with the idea that the majority of participants realized they had no control over the outcomes.⁵ Furthermore, it is interesting to note that exploratory analyses revealed that participants with lower forced switch costs in Experiment 4 did not exhibit higher switch rates, unlike in the other three experiments.⁶ This suggests that task choices in this experiment were not related to actual performance. One might speculate that reward-specific free-choice bias only emerges in environments where task choices are at least partially linked to performance, as performance efficacy—that is, the potential for increased performance to lead to better outcomes—might play a critical role in perceived control over outcomes. Thus, in future studies, it may be useful to either induce some perceived control (e.g., via a cover story), some actual choice-reward contingencies or (partial) performance-based rewards to better understand how free- and forced-task goal outcomes are evaluated.

Finally, as illustrated by the additional raincloud plots in Appendix B, there was generally quite a bit of individual variability in task choice behavior in the specific conditions. Thus, it also seems useful to consider interindividual differences in task choice behavior as a function of outcome (and processing mode) in future research. For example, Mittelstädt et al. (2021) showed in a purely voluntary task switching environment that some people rather preplan their task choices across several trials, while others decide spontaneously in each trial which task to choose. One might suspect that individuals following spontaneous task choice strategies are more sensitive to environmental cues, such as the present outcome manipulation.

In sum, this study began with the premise that people choose tasks based on their perceived value and it offers new insights into how task values are updated by outcomes and processing modes, thereby influencing subsequent voluntary task-switching behavior. Specifically, the findings indicate that people tend to repeat tasks more frequently following positive outcomes rather than negative ones and they show a preference for tasks they have chosen themselves over those imposed upon them. This suggests that both positive outcomes and free choice enhance the perceived value of task goals. Furthermore, individuals may sometimes overestimate the value of outcomes linked to tasks they have freely chosen compared to those imposed upon them. This was evident in Experiments 1 and 3, where outcomes had a stronger impact on task-switching behavior following freely chosen tasks than forced tasks. Given that we can currently only interpret the findings against a rather vague theoretical assumption that task values drive task choice behavior, it will be crucial for future studies to investigate how outcome evaluation for freely chosen and forced task goals can be conceptualized using more detailed theoretical, ideally computational, models. Considering that tasks performed in many real-world situations often differ in their degree of freedom, understanding how choices along a continuum from self-determined to externally-determined influence our behavior might also be useful for applied contexts.