RETRACTED ARTICLE: Local and global effects of motivation on cognitive control
Motivation has been found to enhance cognitive control, but the mechanisms by which this occurs are still poorly understood. Cued motivational incentives (e.g., monetary rewards) can modulate cognitive processing locally—that is, on a trial-by-trial basis (incentive cue effect). Recently, motivational incentives have also been found to produce more global and tonic changes in performance, as evidenced by performance benefits on nonincentive trials occurring within incentive blocks (incentive context effect). In two experiments involving incentivized cued task switching, we provide systematic evidence that the two effects are dissociable. Through behavioral, diffusion-modeling, and individual-differences analyses, we found dissociations between local and global motivational effects that were linked to specific properties of the incentive signals (i.e., timing), while also ruling out alternative interpretations (e.g., practice and speed—accuracy trade-off effects). These results provide important clues regarding the neural mechanisms by which motivation exerts both global and local influences on cognitive control.
KeywordsCognitive control Motivation Reward Context Individual differences Task switching
Cognitive control refers to the process by which cognition and behavior are directed toward the fulfillment of internally generated goals (Miller & Cohen, 2001). These internally generated goals or intentions then create bias signals whose aggregate effect is to establish the proper mappings between inputs, internal states, and outputs in order to promote task-appropriate responding. Cognitive control is exerted as a means of overcoming interference from the environment, default action tendencies, or even other internal goals that are competing for behavioral priority. Underlying this theoretical idea is that the success of cognitive control is dependent upon goal activation strength, a view that has long been prominent in the goal pursuit literature (Eccles & Wigfield, 2002; Fishbach, Shah, & Kruglanski, 2004; Koo & Fishbach, 2008; Kruglanski et al., 2002). Specifically, top-down biasing effects can only occur from strongly activated goals, while the presence of weakly activated task goals may not be sufficiently strong to overcome distractions from the environment. Why might goals vary in activation strength or priority level? It seems somewhat intuitive that motivation might serve a prioritization function, by modulating the activation strength of task goals in relationship to the motivational value placed on the goal outcome. Thus, a natural theoretical hypothesis is that the efficacy of cognitive control will be directly linked to the motivational value of task goals.
A growing number of studies have begun to address the question of whether motivation can enhance behavioral performance by specifically influencing cognitive control (Pessoa, 2009; Pessoa & Engelmann, 2010). The predominant approach in these studies has been to manipulate the motivational value of task performance on a trial-by-trial basis. Utilizing categories of motivational incentives such as money, food/liquid, or social praise as rewards (or punishments), these studies have demonstrated that cognitive performance is enhanced on trials with high motivational value (Aarts, 2008; Bijleveld, Custers, & Aarts, 2011; Dreisbach, 2006; Engelmann, Dmaraju, Padmala, & Pessoa, 2009; Markman, 2005; Muller et al., 2007; Savine, Beck, Edwards, Chiew, & Braver, 2010; Veling & Aarts, 2010). Moreover, in studies using neuroimaging methods, these behavioral enhancements have been linked to parallel enhancements of activity in brain regions implicated in executive function (Jimura, Locke, & Braver, 2010; Krawczyk, Gazzaley, & D’Esposito, 2007; Leon & Shadlen, 1999; Locke & Braver, 2008; Pochon et al., 2002; Savine & Braver, 2010; Taylor et al., 2004; Watanabe, Hikosaka, Sakagami, & Shirakawa, 2002).
In our own prior work, we have provided direct evidence for motivation-related enhancements of cognitive control on individual incentive-cued versus non-incentive-cued trials (incentive cue effect). Specifically, local incentive cueing eliminated the robust behavioral costs associated with task switching (Savine et al., 2010), and these behavioral indices of local motivational enhancement were correlated with increased transient activity in the lateral prefrontal cortex (Savine & Braver, 2010). Consequently, we obtained evidence supporting the hypothesis that incentive cues increase the activation strength of the relevant task goals that must be updated on a trial-by-trial basis during task switching. Yet, a potentially more striking aspect of our prior results was the observation of a second, distinct incentive-linked performance effect that appeared to index a global motivational process (Savine et al., 2010). That is, when contrasting non-incentive-cued trials that occurred within instructed incentive task blocks against non-incentive-cued trials that occurred in incentive-naive task blocks, a separate global and state-like performance benefit was observed for the nonincentive trials that occurred within the salient motivational context (incentive context effect). These effects were not found just in cued task switching, but also generalized to other cognitive control tasks (e.g., working memory; Savine et al., 2010). Our prior findings thus suggested specific and dissociable mechanisms by which motivation influenced cognition and goal-directed behavior.
Nevertheless, in our prior work, our experimental design had important limitations, which left open alternative interpretations of the local and global incentive effects, particularly the incentive context effect. Thus, we conducted two experiments to provide further support for the dissociability of the motivational effects and to better determine their unique impacts on cognitive control. To foreshadow, Experiment 1 focused on the incentive context effect, to more firmly establish its validity and mechanisms. We hypothesized that the incentive context effect would be rapid, state-like, and global in influence. Thus, it should have an immediate onset when incentive contingencies are changed due to the change in goal structure. We tested this prediction by extending the cued task-switching design and employing a larger data set to more firmly establish that the global motivational context exerts a unique influence on task performance.
In Experiment 2, we focused on the specific mechanisms underlying the incentive cue effect and their relationship to the incentive context effect. In particular, we hypothesized that the local motivational effects of incentive cues would depend on the integration of the motivational information with task-relevant information, such as that conveyed by task cues. The resulting integrated goal representation should enhance how task-relevant information is represented and maintained in working memory. To explore this issue, we decoupled the time available for preparation based on task cue information (i.e., the cue–target interval) from the temporal proximity of task and incentive cues, by manipulating the timing of incentive cues while holding the cue—target interval constant. By doing so, we were able to examine whether changes in incentive timing have differential effects on the cue and context effects. Moreover, in this study, we also made use of two additional methodological tools to understand the mechanisms underlying incentive cue and context effects: (1) diffusion model analyses of decision-relevant parameters and (2) individual-differences analyses based on trait sensitivity to reward incentives.
In this task-switching experiment, we aimed to rigorously determine whether the global motivational context exerts a unique influence on task performance, independent of local effects related to the trial incentive value. In our prior work, we found evidence that performance was significantly faster during incentive blocks than during baseline blocks, even on the trials that participants knew had no potential reward value themselves (i.e., nonincentive trials; Savine et al., 2010). However, we could not rule out two potential alternative interpretations of that initial finding: (1) that the effect actually was due to task practice or sequential order, rather than motivation, since participants always performed the incentive condition following the baseline condition; and (2) that the context effect was actually due to local carryover from performing nonincentive trials in close temporal proximity to incentive trials.
In the present experiment, we directly addressed these two alternative interpretations in multiple ways. First, to examine issues related to sequential order and task practice, we modified the experimental design to include both preincentive and postincentive baseline blocks, as well as a control group of participants who ran an extended series of matched baseline blocks, but with no intervening incentive block. We predicted that the incentive context effect would be robust even when controlling for effects of practice and sequential order. Second, to examine issues related to local carryover effects, we examined participants in two separate incentive sessions, one in which incentives were offered on a high proportion of trials (75 %), and the other in which incentives were offered on a low proportion of trials (25 %). If local carryover effects from incentive trials were the source of enhanced performance on the intermixed nonincentive trials in these blocks, this would predict reduced incentive context effects in the low-incentive-proportion condition relative to the high-incentive-proportion condition. In contrast, we propose that the incentive context effect reflects a motivation-induced state change in the global task goals; thus, such a state change should not be sensitive to manipulations of incentive proportion. Finally, we examined the onset and decay of the context effect through time-course analyses.
A group of 40 participants (19 male, 21 female), 18–26 years of age, were recruited from Washington University to participate in the study; of these, 24 performed a two-session incentive task-switching experiment (incentive group), while the other 16 served as a nonincentive control group. Written informed consent was obtained, in accordance with the Washington University Medical Center Human Subjects Committee. All participants were right-handed, native English speakers, had corrected-to-normal vision, and were free from psychiatric or neurological disorders. Participants received reimbursement for their participation ($10/h). Additionally, the participants in the incentive group received an additional monetary bonus due to the reward incentives. Although these participants believed otherwise until the end of the experiment, the bonus was a fixed amount per incentive session ($5, slightly larger than the maximum possible reward), independent of task performance.
Visual stimuli were presented using E-Prime software (Schneider, Eschman, & Zuccolotto, 2002) running on version 2.0 software on a Dell Inspiron desktop computer. Bivalent stimuli (faces with superimposed words) were used for two different classification tasks: gender discrimination (male or female) for faces, and syllable judgment (two-syllable or not) for words. The tasks and stimuli were analogous to those used in Savine et al. (2010). The stimulus combinations were created randomly from a bank of 144 faces (male and female), 76 two-syllable words, 38 one-syllable words, and 38 three-syllable words (thus yielding 21,888 possible unique stimulus combinations). The faces were stripped of hair and distorted with noise to equate perceptual difficulty between the words and faces. Responses were indicated and recorded via buttonpresses on a Cedrus buttonbox interfaced with E-Prime. The same two buttons were used for each task. Because the two different tasks involved both bivalent stimuli and overlapping response buttons, the resulting stimulus–response ambiguity necessitated the demand for cognitive control (Meiran, 2000). The mappings of stimulus category (male/female, two-syllable/not-two-syllable) to responses were varied and counterbalanced across participants.
After a cue—target interval (CTI) of 2,000 ms, participants responded to the face/word target stimulus. Trial responses were followed by informative visual feedback (2,000 ms). The response—cue interval (RCI) was held at a constant 2,500 ms in all trials, since this is a duration at which the impact of passive decay from the prior trial on switch costs appears to be minimal (~1 ms per 100 ms of additional RCI; Meiran, 2000). Thus, the long fixed RCI minimized the residual effects of previous trials on performance and enabled better isolation of the effects of preparatory time on task-switching performance.
All participants in both the incentive group and the nonincentive control group performed 12 task blocks. Some of these were single-task blocks, in which the task cues were all of a single type (“Attend Word” or “Attend Face”), while others were mixed-task blocks, in which the two task cues were randomly intermixed, leading to equal probabilities of task-switch and task-repeat trials. The stimulus combinations (face and word) that constituted the target probes were always unique (randomly generated without replacement), with no repeats for each participant.
For the control group of participants, a single within-subjects experimental session was performed, composed of only nonincentive blocks (eight mixed-task, four single-task; 64 trials in each). In contrast, the incentive group of participants performed two separate within-subjects testing sessions (spaced 2–7 days apart). Each session consisted of three phases in sequential order: preincentive baseline (nonincentive), incentive, and postincentive baseline (nonincentive). Each phase consisted of four 48-trial blocks (two single-task and two mixed-task, randomly intermixed within each phase). Block order was pseudorandomly counterbalanced across participants in both the control and experimental conditions, with the constraint that each phase included at least one single-task and one task-switching block. Across the two sessions, the incentive blocks differed in their frequencies of incentive and nonincentive cued trials. In the high-incentive blocks, 75 % of the trials were incentive cued and 25 % were nonincentive cued. In the low-incentive blocks, the frequencies were reversed (25 % incentive cued, 75 % nonincentive cued). Thus, incentive-group participants performed high-incentive blocks in one session and low-incentive blocks in the other session, with session assignment counterbalanced across participants.
In the incentive blocks, participants were instructed that they had the potential to obtain rewards on incentive trials, but only if their responses were correct and faster than a prespecified cutoff. This cutoff was individually set for each participant on the basis of their own asymptotic baseline performance in the first nonincentive phase (using the 3rd and 4th blocks, to enable performance to stabilize). The criterion was calculated as the median reaction times (RT) of correct responses in the task blocks occurring after the performance asymptote. Thus, the performance criteria were demanding and equated across participants, ensuring that the obtained incentive rates were below ceiling and dependent on optimized performance (emphasizing both speed and accuracy). However, if this value was slower on the second session, the cutoff value from the first session was used, as a way to correct for potential “deliberate” slowing in the second session. Postresponse visual feedback on incentive trials indicated whether or not the incentive was obtained (rewarded = green rectangle with exclamation point; nonrewarded [correct but slow] = green rectangle; incorrect = red triangle). On the nonincentive and baseline trials, only correct and incorrect visual feedback messages were provided. In all blocks, whether incentive, nonincentive, or in the control group, participants were always instructed to respond as quickly as possible while still maintaining high accuracy.
Dependent variables and data analysis
In the context of this experiment, we were interested in examining the behavioral effects of cognitive control demands, the behavioral effects of reward motivation, and how reward motivation interacted with cognitive control demands to influence task performance. As such, the analyses focused on a set of dependent variables, based on RTs and error rates, identified from prior work as being useful for examining such effects.
Cognitive control demands were quantified in terms of the differences in performance between three different task-switching trial types: single-task trials, task-repeat trials during task-switch blocks, and task-switch trials. These three trial types reflected a progressive increase in cognitive control demands: task switch > task repeat > single task. In the task-switching literature, the classic measure of cognitive control demand is the local task-switch cost (task switch−task repeat), computed for both RTs and error rates. Positive values reflect increased cognitive control demands, but higher values indicate that this demand is not met optimally. In other words, evidence of optimal cognitive control function would be an elimination of task-switching costs on behavioral performance.
Motivational influences on behavioral performance were quantified in terms of two dependent variables identified in our prior work (Savine et al., 2010)—the incentive context effect and incentive cue effect—that appear to track global and local motivational effects, respectively. The incentive context effect is defined as the difference in performance between nonincentive trials that occurred within incentive blocks (i.e., within the incentive context) and nonincentive trials that occurred in the perceptually identical but incentive-naive baseline blocks (after baseline performance had asymptoted); that is, incentive context = baseline trial performance−nonincentive trial performance. The incentive context effect can be computed for both RT and error rate, with positive values indicating improved performance (i.e., faster RTs or reduced error rates) during the incentive block relative to the baseline block. Because the incentive context effect is defined for the nonincentive trials, we interpret it as reflecting a global and sustained motivational effect present on all trials within the incentive block, even those not explicitly rewarded, purely by virtue of occurring within a more salient motivational context. The incentive cue effect is defined as the difference in performance between the randomly intermixed incentive and nonincentive trials that occur within incentive blocks; that is, incentive cue = nonincentive trial performance−incentive trial performance. The incentive cue effect can also be computed for both RT and error rate, with positive values indicating improved performance (i.e., faster RTs or reduced error rate) on incentive trials relative to nonincentive trials.
The incentive cue and context effects were computed for each of the three task-switching trial types described above. It is important to note that incentive cue and context effects are statistically orthogonal to cognitive control demands, as defined by trial-type effects and local task-switching costs. For example, changes in the magnitudes of incentive cue and context effects could occur independently of changes in the magnitude of switch costs. However, it is possible that the two factors (cognitive control and motivation) interact. Such an effect would be observed in terms of either increased cue and context effects on trial types that produce higher cognitive control demands (i.e., task-repeat and/or task-switch trials), or alternatively in terms of reduced switch costs associated with incentive context or incentive cues. These two effects are equivalent, and reflect alternative ways of expressing interactions between motivation and cognitive control demands. Consequently, in presentations of the results, we will use both formats for description.
A primary focus of the study was to determine the source of the incentive context effect, in terms of motivational and nonmotivational factors. Thus, the analyses were first geared toward answering this question. A second goal was to more firmly establish whether the incentive context and cue effects are in fact dissociable. Subsequent analyses were geared toward addressing this question. For the interested reader, we report additional analyses in the supplementary materials that focus on sequential effects of incentive trials and their interaction with incentive frequency.
Task-switching performance (Exp. 1)
Switch Costs (Switch – Repeat)
Nonincentive Control Group
Response time (ms)
Baseline (No Knowledge of Future Incentives)
Response time (ms)
Nonincentive Trials Within Incentive Blocks
25 % Incentive Cues
Response time (ms)
75 % Incentive Cues
Response time (ms)
25 % Incentive Cues
Response time (ms)
Proportion rewarded trials
75 % Incentive Cues
Response time (ms)
Proportion rewarded trials
Baseline task-switching effects
In the nonincentive baseline blocks for the experimental group, the results of one-way ANOVAs across trial types (single, repeat, or switch) replicate the common finding in the literature that RTs and error rates increase with increasing cognitive control load [for RT, F(1, 23) = 5.33, p < .05; for error rate, F(1, 23) = 4.95, p < .05; see Table 1].1 Follow-up t tests also demonstrated that local task-switching costs were present when contrasting the task-switch and task-repeat trials for both RTs [t(23) = 2.27, p < .05] and error rates [t(23) = 2.02, p < .05]. The nonincentive control group exhibited analogous results in trial-type one-way ANOVAs [for RT, F(1, 23) = 5.09, p < .05; for error rate, F(1, 23) = 5.40, p < .05; Table 1] and in follow-up t tests examining task-switching costs [for RT, t(23) = 2.18, p < .05; for error rate, t(23) = 2.37, p < .05]. These results suggest that the present face/word task-switching paradigm (1) produces results consistent with previous task-switching work and (2) provides sensitive measures of cognitive control demands and task-switching costs that can be probed via the subsequent incentive manipulations.
Proportion of rewarded trials
The average reward rates on incentive-cued trials in the 25 %-frequency condition were 80 % (range: 64 %–91 %) and 77 % (range: 62 %–93 %) in the 75 %-incentive-cue-frequency condition. No differences existed between the conditions (t < 1). These reward rates were much higher than the predicted rate of 50 % obtained rewards that was based on baseline performance, indicating that motivation generally enhanced behavioral performance.
Incentive context effects are not related to practice or sequential order
We examined the primary question of whether practice and sequential order contribute to the incentive context effect, by first directly contrasting the performance profiles of the control and incentive groups across the experimental session. For this analysis, we included only the first session of the incentive-group participants (collapsing across the incentive proportion manipulation) and only nonincentive trials (i.e., selecting only the subset of nonincentive trials performed within the incentive phase, Blocks 5–8). Performance on nonincentive trials across the two groups was contrasted in a mixed ANOVA, with Group as a between-subjects factor (control, incentive) and Block (1–12) as a within-subjects factor. A highly significant interaction was observed [F(11, 29) = 56.3, p < .001]. We conducted follow-up mixed ANOVAs within the individual phases (2 Group × 4 Block) to elucidate the source of the primary interaction.
The observed patterns in RTs clearly demonstrate that the change in nonincentive trial performance during the incentive phase could not be attributable to residual practice effects, but instead reflected a global shift in performance caused by the change in motivational context. This performance shift was long-lasting, with a slow return to baseline in the postincentive phase. These points are made apparent by three additional analyses. First, in both groups, practice effects were apparent and equivalent in the first two blocks [2 Group × 2 Block ANOVA: block effect, F(1, 23) = 4.9, p < .05; Block × Group, F = 1.4], but then were stable by the third and fourth blocks (block effect: F < 1). Second, the change in incentive context produced an instantaneous effect on performance in the incentive group, as indicated by a fine-grained transition analysis that contrasted the last five trials of the fourth block against the first five trials of the fifth block (2 Block × 5 Trial repeated measures ANOVA; see Fig. 2b). Performance was dramatically changed even by the first trial of the incentive block, with no additional trial-by-trial adjustments [block main effect, F(1, 23) = 24.6, p < .001; Block × Trial interaction, F < 1]. In contrast, the performance asymptote was clear in the control group, because no effect occurred at the comparable transition point (2 Block × 5 Trial ANOVA: Fs < 1). Lastly, the incentive context effects could not be due purely to a change in the task instructions, since there was no equivalent transition effect when the incentive phase ended, which was also communicated to participants in exactly the same manner as when it began (Fig. 2c). Indeed, here the two groups looked identical, in the lack of a 2 Block × 5 Trial transition effect (Fs < 1). Thus, the differential performance of the incentive group in the postincentive phase appears to reflect a carryover of the incentive context. This suggests that the state-like effects of motivation on task performance have a relatively slow decay, even when participants are explicitly instructed regarding the change in incentive context.
The incentive context effect is not an artifact of local carryover from intermixed incentive trials
The next analysis directly tested the possibility that the incentive context effects observed on nonincentive trials within the incentive block might actually reflect a carryover effect from the locally cued and intermixed incentive trials. For this analysis, we compared the RT incentive context effects (nonincentive trials within the incentive phase vs. nonincentive phases) for the two sessions performed by the incentive-group participants. One session contained a high incentive proportion (75 %) and the other a low incentive proportion (25 %). A local carryover hypothesis would predict greater context benefits in the high-proportion condition than in the low-proportion condition, because in the high-proportion condition most nonincentive trials would be performed in close temporal proximity to an incentive trial. Instead, the analyses revealed absolutely no effect of incentive proportion on the magnitude of the incentive context effect in the experimental group (Phase 1−Phase 2 nonincentive trials) on overall RTs collapsed across trial types (high proportion, 194 ms; low proportion, 188 ms; t < 1).
Incentive context effects are dissociable from incentive cue effects
The two effects also dissociated in terms of the influence of incentive proportion (see Table 1). In a 2 Proportion × 3 Trial Type repeated measures ANOVA, incentive proportion did not impact the context effect (F < 1); however, it did influence the cue effect [Proportion × Trial Type interaction: F(2, 22) = 4.36, p < .05]. Although on single-task and task-repeat trials there were no proportion effects (ts < 1), on task-switch trials there was a stronger cue effect for RTs in the low-proportion condition [t(23) = 2.06, p < .05], but a trend toward a stronger effect for error rates in the high-proportion condition [t(23) = 1.69, p = .1]. Similarly, when examining task-switch costs, these were statistically eliminated for RTs only in the low-proportion condition (t < 1), but for error rates only in the high-proportion condition (t < 1). We prefer not to interpret these effects, because they are complex and were not predicted, but they are nevertheless consistent with our primary hypothesis that the incentive cue and context effects reflect distinct motivational mechanisms.
Finally, the incentive context and incentive cue effects were dissociable in their offset. While the context effect decayed slowly across multiple postincentive blocks before returning to baseline performance levels, the cue effect decayed in the first five trials of the postincentive phases. In a fine-grained analysis (5 Trial × 3 Phase ANOVA), a significant Trial × Phase interaction [F(4, 20) = 4.7, p < .01] was observed, which was statistically extinguished by the end of the first postincentive block (F < 1). Furthermore, the previously reduced local switch costs observed during the incentive phase quickly returned to baseline levels in the postincentive phase. Although there was only an 11-ms cost on the first task-switch trial of the postincentive phase, by the third task-switch trial, this cost had increased to 56 ms [t(23) = 1.97, p < .05]. Taken together, these results provide evidence that the specific cognitive control benefits associated with the incentive cue effect are lost very rapidly when the potential to earn incentives is taken away, while the nonspecific motivational benefits of global motivational context persist long afterward.
Experiment 1 confirmed the validity of the incentive context effect as a distinct index of motivational influences on cognitive task performance, rather than a confounding effect of task practice or local carryover from temporally proximal incentive trials. In particular, we still observed a robust context effect even when directly assessing and controlling for these confounds, by including (1) a nonincentive control group, (2) a postincentive baseline, (3) a manipulation of incentive proportions, and (4) analyses testing for local sequence effects. Crucially, the incentive context effect was observed instantaneously, on the very first trial of the incentive phase (i.e., just after participants were informed of the change in motivational incentives). This suggests that the effect manifested through a change in the task goals maintained in working memory. Interestingly, however, the context effects dissipated rather slowly when incentives were removed during the postincentive phase, taking a number of blocks to return to baseline. This pattern indicates a relatively tonic, state-like mechanism that only gradually returns back to baseline levels when motivational salience is decreased. The slow return to baseline might reflect the residual activation of task performance goals. However, an alternative possibility is that it reflects a residual state of arousal caused by maintaining the incentivized context for an extended period. One means to adjudicate between these ideas would be to examine whether the performance gains due to motivational context transfer to a different task setting or are restricted to continued performance with the previously incentivized task.
A second key aspect of the results is that they replicate, and thus more firmly establish, the dissociability of the incentive cue and context effects. The nature of the dissociation provides clues as to the underlying mechanisms that give rise to each effect. In particular, the incentive context effect was primarily observed as a global speeding of RTs that impacted all trials, with no reliable effects on accuracy, and that was independent of task switching and the associated cognitive control demands. Thus, the context effect may reflect a global motivational mechanism of generalized readiness, which could result in more efficient encoding and response selection. In contrast, the incentive cue effect was associated with robust improvements in both RT and error rate that were amplified under task-switching conditions. These patterns indicate that incentive-cued trials did not lead to a simple adjustment of the speed–accuracy trade-off function (i.e., attempting to go faster at a potential cost to accuracy), but instead produced an optimization of cognitive control that was reflected in reduced, or even eliminated, task-switching costs. Consequentially, the cue effect appears to reflect a more local motivational mechanism that potentiates the trial-relevant goals activated by task cues.
However, our hypotheses regarding the dissociable mechanisms that underlie the incentive cue and context effects require further experimental evidence to support their characterization. In particular, we predicted that incentive cue effects should depend on the temporal association and integration of incentive signals with local task information. Thus, if incentive cue information cannot be linked with task cue information, we predicted that the incentive cue effects would be reduced or eliminated. In contrast, if incentive signals were more closely associated with target stimuli than with task cues (via temporal proximity), we predicted that the magnitude of the incentive context effect would increase via an enhancement of target-linked and response selection processes. In the next experiment, we directly tested this idea by manipulating the timing of incentive signals while holding all other aspects of the task constant.
In Experiment 2, we explicitly examined the hypothesis that expressions of the incentive cue and context effects are dependent on the temporal association of incentive signals with specific task events. Specifically, we predicted that incentive cues would enhance task encoding and preparatory processing. However, such effects are likely to be dependent on the incentive cues being present during the time window in which task encoding and preparation occurs. Thus, we hypothesized that local motivational influences would only be dominant when incentive cues are presented in close temporal proximity to task cues, as indexed by enhanced utilization of task cue information and effective preparatory control (i.e., reduction of switch costs). In contrast, when incentive cues were not temporally linked to the task cue, we predicted that local motivational effects would be reduced. Conversely, we predicted that the global motivational goals associated with incentive context would operate primarily in terms of enhanced target-related processing (i.e., stimulus encoding and motor processes). One interesting possibility is that when incentive signals occur close in time to target events, rather than to task cues, that these global motivational effects might be amplified. Thus, we predicted that incentive context effects would be largest under conditions in which the incentive cues were more strongly target-linked than task-cue-linked.
In addition to testing these hypotheses with standard analyses of error rate and RT, we also analyzed the data using the well-known diffusion model of evidence accumulation during decision making (Ratcliff & McKoon, 2008; Wagenmakers, van der Maas, & Grasman, 2007). The diffusion model is a process model that enables estimation of three unobserved parameters related to decision making. It is based on the combination of RT and error rate data, thus providing an alternative summary measure of performance. Prior studies have shown success in applying the diffusion model to cued task-switching paradigms (Karayanidis et al., 2009; Karayanidis et al., 2010; Madden et al., 2010; Mansfield, Karayanidis, Jamadar, Heathcote, & Forstmann, 2011).
In the present application, we hypothesized that the parameters estimated from the diffusion model could provide a more powerful means of dissociating the motivational mechanisms underlying the incentive cue and context effects. In particular, if the incentive context effect reflects the influence of global motivational goals on stimulus encoding and motor responses, we would expect that such effects would be reflected in the two process estimates that reflect these stimulus and response factors: namely, nondecision time and response caution. In contrast, the local motivational influence of incentive cues on preparatory task rule activation should be associated with a faster accumulation of the information needed to make task decisions, and thus should be reflected in the drift rate parameter, which indexes this effect. Moreover, we predicted that the processes most impacted by (or implicated in) preparatory processing or proactive control, such as evidence accumulation (drift rate), would be strongest when the experimental conditions would favor that type of processing (when the local task and incentive cues were closely temporally linked). Processes that would be most impacted by (or implicated in) target and response processing and reactive control, such as response caution, were predicted to be strongest when the experimental conditions would favor target-related incentive processing (such as if the incentive signals were closely temporally linked to the stimulus targets).
Finally, we also tested the validity and utility of the diffusion model estimates by grounding them in terms of individual-differences effects. Specifically, we focused on the well-established personality trait of reward sensitivity (Carver & White, 1994), which we predicted would moderate the degree to which incentive effects enhanced behavioral performance (i.e., highly reward-sensitive individuals would show larger incentive effects). If the diffusion model parameters reflected dissociable processes associated with incentive cue and context effects, we might expect these dissociations to also show up in terms of differential correlations with trait reward sensitivity. Specifically, we predicted that trait reward sensitivity might predict the drift rate effects associated with incentive cueing, but also potentially response caution and/or nondecision time effects associated with the incentive context. Moreover, if our incentive-timing manipulation was effective in further pulling apart these effects, the reward sensitivity effects might also show differential interactions with incentive timing (i.e., correlations with the incentive cue effect would be strongest when incentive cues were temporally linked to task cues, while correlations with incentive context would be strongest when the cues were instead linked to target events).
A group of 24 participants (mean age 20.2 years; 14 female, 10 male), 18–26 years of age, were recruited from Washington University to participate in the study. Written informed consent was obtained, in accordance with the Washington University Medical Center Human Subjects Committee. All participants were right-handed, native English speakers, had corrected-to-normal vision, and were free from psychiatric or neurological disorders. The participants received reimbursement for participation ($10/h), plus an additional monetary bonus due to the reward incentives. Although participants were not informed of this until the end of the experiment, the bonus was a fixed amount per incentive session ($5, slightly larger than the maximum possible reward), independent of task performance.
The same basic materials were used as in Experiment 1.
Participants again performed cued task switching in single- and mixed-task blocks of 48 trials each under baseline and incentive conditions. The general task procedure, task cues, feedback displays, RCI, and incentive criteria were the same as in Experiment 1. Incentive-cued and non-incentive-cued trials were presented in equal proportions (i.e., 50 % incentive frequency). Also, individuals were again only rewarded if their responses were correct and faster than the prespecified cutoff that was individually set for each participant on the basis of their asymptotic baseline performance in the initial, nonincentive phase (using the 3rd and 4th blocks, to enable performance to stabilize).
The manipulations of I-CTI timing varied the relationship of the incentive cue to the task cue and target events within the trial. In the 2,000- and 2,750-ms I-CTI conditions (hereafter referred to as cue-linked conditions), the incentive cue was closely linked to the task cue by occurring either simultaneously with it (2,000 I-CTI) or closely preceding it (2,750 I-CTI). In the 500- and –100-ms I-CTI conditions (hereafter referred to as target-linked conditions), the incentive cue was instead closely linked to the task target, by either occurring closely preceding it (500 I-CTI) or right after it (–100 I-CTI). In the remaining two I-CTI conditions (1,600- and 1,000-ms I-CTI), the incentive cue was linked, but potentially more weakly, to both task events, since it occurred after the task cue, but long before the target.
Two task-switching blocks and two single-task blocks were performed for each of the target-linked I-CTI timings. Two task-switching blocks and one single-task block (48 trials each) were performed for the remaining I-CTI timings.4
Dependent variables and data analyses
As in Experiment 1, our analyses focused on local task-switch costs (task-switch – task-repeat trials) for both RT and error rate, as a measure of cognitive control effects. Incentive context and cue effects on both RT and error rate were utilized as measures of motivational influence. Interactions of context and cue effects with trial type (task-switch, task-repeat) and on switch cost were used to test for interactions of motivation and cognitive control. These effects were examined across the different I-CTI conditions to determine whether I-CTI timing modulated motivational influences on task performance, and on cognitive control, more specifically. For these analyses, investigations of I-CTI effects focused on the direct comparison of the cue-linked and target-linked conditions to reduce complexity and increase interpretability, by binning across the relevant I-CTI conditions. Additional analyses used diffusion model estimates rather than RT and error rate, and also incorporated individual-difference measures of reward sensitivity, as described below.
Diffusion model analysis
The diffusion model assumes that observable RT and error rate measures reflect the contributions of three primary parameters that underlie performance in two-alternative forced choice decision tasks. These parameters include: (1) the drift rate (v), which indexes how quickly perceptual evidence is integrated toward a decision, thus reflecting decision-making efficiency; (2) the response criterion (a), which indexes how much evidence is required to generate a decision, thus reflecting response caution or conservativeness; and (3) the nondecision time (t0), which indexes the time periods that precede or follow the decision phase itself (e.g., stimulus encoding and response execution; hence, the other variable name often used: Ter), thus reflecting factors that may shorten or lengthen the duration of such periods.
For the primary analyses reported below, we estimated these parameters from the data using the EZ-diffusion algorithm described by Wagenmakers et al. (2007) and used in prior diffusion analyses of cued task switching (Karayanidis et al., 2009; Karayanidis et al., 2010; Madden et al., 2010). This algorithm enables direct analytic computation of v, a, and Ter based on three parameters of the observable data: correct-RT mean and variance, plus accuracy.5 We computed the three diffusion model parameters separately for each participant, focusing on just the cue-linked and target-linked conditions (binning across I-CTIs) to increase power. Primary analyses examined the data for each parameter in terms of incentive cue and context effects using repeated measures ANOVAs with Trial Type (switch, repeat) and I-CTI Condition (cue-linked, target-linked) as factors.
Assessment of reward sensitivity
Participants completed the behavioral inhibition/approach system (BIS/BAS; Carver & White, 1994) individual-difference questionnaire prior to task engagement on the first session of the experiment. The BIS/BAS scale measures the reactivity of individuals to reward and punishment. It contains questions such as “It would excite me to win a contest,” to quantify how interested an individual was in obtaining a potential reward. Highly reward-sensitive people would find gratification in these rewards and expend more effort to achieve them. The BIS portion measures how sensitive people are to potential punishments, using questions such as “If I think something unpleasant is going to happen, I usually get pretty worked up.” Analyses focused on whether individual differences in these measures explained between-subjects variability in incentive context and cue effects related to the I-CTI manipulation. For these analyses, we used the diffusion model estimates (rather than simple RT and error rate) as the dependent measures.
Results and discussion
Task-switching behavioral performance (Exp. 2)
Switch Costs (Switch – Repeat)
Response time (ms)
Nonincentive Trials Within Incentive Blocks
Inc Cue 2,750 Prior
Response time (ms)
Inc Cue 2,000 Prior (Concurrent)
Response time (ms)
Inc Cue 1,600 Prior
Response time (ms)
Inc Cue 1,000 Prior
Response time (ms)
Inc Cue 500 Prior
Response time (ms)
Inc Cue 100 After
Response time (ms)
Inc Cue 2,750 Prior
Response time (ms)
Proportion rewarded trials
Inc Cue 2,000 Prior (Concurrent)
Response time (ms)
Proportion rewarded trials
Inc Cue 1,600 Prior
Response time (ms)
Proportion rewarded trials
Inc Cue 1,000 Prior
Response time (ms)
Proportion rewarded trials
Inc Cue 500 Prior (Concurrent)
Response time (ms)
Proportion rewarded trials
Inc Cue 100 After
Response time (ms)
Proportion rewarded trials
Baseline task-switching effects
In the nonincentive baseline blocks, the results of one-way ANOVAs across trial types (single, repeat, or switch) robustly demonstrated that RTs and error rates increased with increasing cognitive control load [RT, F(1, 23) = 6.89, p < .05; error rate, F(1, 23) = 5.52, p < .05; Table 2]. Follow-up t tests again identified significant local task-switching costs for both RT [t(23) = 2.27, p < .05] and error rate [t(23) = 2.02, p < .05], suggesting that the present task-switching paradigm provides robust cognitive control indices for subsequent examinations under incentivized conditions.
Proportions of rewarded trials
Participants were largely successful in obtaining possible rewards across all incentive cue timings (Table 2: range: 77 %–67 %). In a one-way ANOVA of overall reward rates across the I-CTI factor, excluding the –100-ms I-CTI condition—the only condition providing no incentive preparation time—reward rates were statistically equivalent across I-CTIs [F(4, 20) = 1.2, n.s.]. When adding the –100-ms I-CTI factor into the ANOVA, the I-CTI effect became significant [F(5, 19) = 3.68, p < .01], due to the lower rate of rewards in this condition (even though reward rates were still ≥67 %).
Incentive timing differentially impacts cue and context effects
For the incentive context effect, the data were analyzed with analogous repeated measures ANOVAs: two 2 Trial Type (repeat, switch) × 2 Incentive (nonincentive, baseline) × 6 I-CTI Timing (2,750, 2,000, 1,600, 1,000, 500, –100). We replicated the primary findings of Experiment 1, in which incentive context effects were expressed primarily in RTs [274-ms speeding; F(1, 23) = 39.7, p < .05], rather than error rates [2 % increase; F(1, 23) = 3.86, p = .06], without interacting with trial type (Fs < 1). But more importantly, there was also a significant interaction of incentive context with I-CTI in which incentive-context-related RT effects were largest at the shortest I-CTI conditions [F(5, 19) = 3.74, p < .05], with no I-CTI effects on error rates [F(5, 19) = 1.3, n.s.].
We followed up these patterns of results with post-hoc 2 I-CTI (cue-linked, target-linked) × 2 Trial Type (repeat, switch) ANOVAs testing the difference measures of the incentive cue effect and the incentive context effect. For RT, the incentive cue effect (Fig. 6b) was largest in the cue-linked conditions (repeat = 102 ms, switch = 147.5 ms), but absent in the target-linked conditions (repeat = –5.5 ms, switch = –2.5 ms) as indexed by a robust I-CTI main effect [F(1, 23) = 17.1, p < .001]. The same main effect was observed in error rates, as the significant improvements were selectively observed in the cue-linked conditions [repeat = 2 %, switch = 5 %; F(1, 23) = 4.55, p < .05; Fig. 7a]. In the target-linked conditions, error rate effects were eliminated, and in fact were numerically reversed (repeat = –1 %, switch = –1 %; F < 1.3). A trend for the opposite pattern emerged in regard to the incentive context effect, as this effect was larger for the target-linked conditions (repeat = 277 ms, switch = 282.5 ms) than for the cue-linked conditions (repeat = 223.5 ms, switch = 239 ms) [I-CTI main effect: F(1, 23) = 3.56, p = .07]. There was also a trend-level error rate increase due to incentive context as I-CTI shortened [F(1, 23) = 3.08, p = .091]. Taken together, these data suggest that when the incentive cue was likely to be integrated with task information (i.e., cue-linked), motivation primarily impacted performance via incentive cue-related enhancements of preparatory processing. In contrast, if conditions were not favorable for the integration of incentive signals with task cue information (i.e., target-linked), then the effects of motivation were more likely to be global and contextual (i.e., occurring on all trials) and occurred primarily through response speeding.
Incentive cue benefits on preparatory control depend on incentive timing
We further examined the effect of incentive timing on preparatory control by focusing on how incentive cues modulated task-switch costs. Recall that we held T-CTI constant and long at 2,000 ms. Thus, any differential effects of I-CTI on switch costs would have to be due to incentive-triggered modulation of the preparation stage. Indeed, incentive cue effects on switch costs strongly interacted with I-CTI (Figs. 6a & 7a). In the 2,000-ms I-CTI condition—in which the incentive and task cues were presented concurrently—we again observed that the robust local switch costs in RTs present on nonincentive trials (54 ms) were eliminated on incentive trials (5 ms) [t(23) = –2.55, p < .05], with a further reduction of error rate switch costs [.07 to .04; t(23) = 2.17, p < .05]. In the other cue-linked condition (2,750-ms I-CTI), RT switch costs were also eliminated [41 to 5 ms; t(23) = 2.02, p < .05], with a trend toward reducing error rate switch costs [.05 to .03; t(23) = 1.76, p = .09]. However, as the incentive cue became progressively more decoupled from the task cue, incentive signals became progressively less effective in attenuating task switch costs overall. Beginning at the 1,600-ms I-CTI condition, RT switch costs were still eliminated [65 to –1 ms; t(23) =−2.94, p < .01], but error rates actually increased numerically [.04 to .06; t(23) = 1.27, n.s.]. As the I-CTI continued to decrease, switch costs were no longer attenuated significantly on incentive trials for both RT [at the 1,000-ms I-CTI, 31 to 13 ms, t(23) = –0.68, n.s.; at the 500-ms I-CTI, 52 to 42 ms, t(23) = –0.53, n.s.; at the –100-ms I-CTI, 63 to 67 ms, t(23) = 0.09, n.s.] and error rate [at the 1,000-ms I-CTI, .02 to .05, t(23) = 2.22, p < .05; at the 500-ms I-CTI, .06 to .06, t(23) = –0.21, n.s.; at the –100-ms I-CTI, .04 to .04, t(23) = 0.074, n.s.].
These findings replicate our earlier results that incentive cues can enhance cognitive control robustly. Our preferred interpretation of the results is that the enhancements depend on whether local motivational goals can be easily integrated with task cue information in order to activate the relevant task representations during the preparatory stage. Nevertheless, an alternative hypothesis remains, in which the temporal integration of incentive and task cue information is not what drives the effects seen in our timing manipulation, but instead the total time given to prepare after an incentive cue prior to target presentation. Adjudicating between these alternatives would require further manipulations that controlled for preparation time following the incentive cue, while varying the interval between the task and incentive cue presentations (e.g., using a 4,000-ms CTI, so that the incentive cue could be presented 2,000 ms after the task cue, but also 2,000 ms before target presentation). This would enable an explicit test of the preparation time versus the cue integration account.
Diffusion model analysis supports dissociation of the incentive cue and context effects
Task-switching diffusion parameters (Exp. 2)
v (Drift Rate)
a (Response Caution)
Ter (Nondecision Time)
Baseline (No Knowledge of Future Incentives)
Nonincentive Trials Within Incentive Blocks
The response caution parameter (a) is more global and strategic, and may reflect the attempt to optimize a speed–accuracy trade-off (Forstmann et al., 2008). A highly significant incentive context effect was observed for this parameter [F(1, 23) = 48.3 p < .001], which indicated reduced response caution on nonincentive relative to baseline trials (Fig. 8b). The effect did not interact with trial type (F < 1) and remained significant even when restricting analyses to task-repeat trials [incentive context main effect: F(1, 23) = 11.5, p < .001]. In contrast, we did not find a statistically significant main effect of incentive cue on response caution (Fs < 2.4, n.s.) or any interaction with experimental conditions (Fs < 2.2, n.s.). A direct statistical test verified that the response caution effect was significantly greater for the incentive context than for the incentive cue effect [t(23) = 2.8, p < .01; Fig. 8b]. Taken together, these results support a double dissociation, in which the incentive cue effect is selectively associated with a drift rate increase, while the incentive context effect is preferentially linked with a response caution decrease. This pattern was statistically confirmed by the presence of a significant 2 Incentive Effect (context, cue) × 2 Diffusion Parameter (drift rate, response caution) crossover interaction [F(1, 23) = 17.0, p < .001].
The diffusion model produces another parameter—Ter—that has been taken to reflect aspects of processing that occur prior to, or following, the decision stage (Wagenmakers et al., 2007; Wagenmakers, Ratcliff, Gomez, & McKoon, 2008). These processes are assumed to reflect perceptual encoding and response execution stages (Ratcliff, 1978; Ratcliff & Rouder, 1998; Ratcliff & Tuerlinckx, 2002). Consistent with this hypothesis, prominent reductions in nondecision time were selectively linked to the incentive context effect [main effect of incentive context: F(1, 23) = 21.2, p < .001; Fig. 8c]. Moreover, we found a trend toward an Incentive × I-CTI interaction, indicating that these effects were largest in the target-linked condition [F(1, 23) = 3.67, p = .10]. In contrast, the effects of incentive cueing on nondecision time were somewhat variable across conditions, and thus not statistically reliable (Fs < 3.2, p > .05). Thus, the Ter effects both clearly reinforce the dissociability of the cue and context effects and demonstrate that the context effect is more complex than just a response caution decrease or a simple speed–accuracy trade-off, since it additionally appeared to co-occur with a substantial reduction in stimulus encoding and response execution time.
One potentially surprising aspect of the findings is that none of the diffusion parameter effects exhibited a statistically robust group effect of incentive timing, although the numerical patterns appeared consistent with the hypotheses (i.e., the incentive cue effect was largest for drift rates in the cue-linked condition, while the incentive context effect was largest for response caution in the target-linked condition). The lack of a group effect could reflect a lack of statistical power, given substantial between-subjects variability. Consequently, we sought to capitalize on this variability to provide support for our incentive-timing hypotheses by directly examining individual-difference effects. Specifically, we predicted that the most prominent effects of incentives would be observed for individuals showing the highest trait reward sensitivity (as indexed by the BAS self-report scale). If this hypothesis is correct, the magnitudes of both the incentive cue and context effects should be positively correlated with reward sensitivity, but this correlation might also reflect the constraints of incentive timing, with the correlations being most prominent for the incentive cue effect in the cue-linked conditions and for the incentive context effect in the target-linked conditions.
This work not only adds to the growing literature examining how trial-by-trial modulation of reward motivation (via incentive cues) modulates ongoing cognitive processing, it also points to a reconceptualization of such effects as reflecting a local motivational influence that is dissociable from the more global motivational influences on task performance that can occur purely from performing tasks within a motivationally incentivized context. The first key set of findings demonstrated that local (i.e., trial-by-trial) motivational influences selectively increased cognitive control (reducing switch costs), had an immediate onset and quick decay, enhanced decision quality (drift rate increases), and appeared strongest when local motivational cues were temporally linked to the task cues. In contrast, the second key set of findings demonstrated that global (i.e., contextual) motivational influences were valid and robust, dissociable from practice effects, and also displaying an immediate onset but a more gradual decay. Moreover, in contrast to local motivational effects, the global effects were insensitive to cognitive control demands, reduced response caution in decision making, and were strongest when the incentive cue was not temporally linked to the task cue. Finally, both global and local motivational effects showed sensitivity to individual differences, in being linked to trait reward sensitivity. Next we discuss implications of these results for (1) further study of motivation and cognition interactions and (2) the neuroscience of motivated cognition.
Mechanisms and characteristics of the incentive context effect
One of the most surprising, and potentially counterintuitive, aspects of the results is the finding that motivation exerted a distinct contextual mechanism of effect that was dissociable from the effects of local motivational cueing. This incentive context effect was indexed by the performance changes on nonincentive trials occurring within incentive blocks relative to those on trials performed in nonincentive baseline blocks. These performance changes were not subtle. Indeed, during task performance within an incentive context, RTs were faster by ~150–200 ms, marking a 20 %–25 % speed-up in responding, with little change in error rates. These effects are surprising because they indicate that participants adjusted performance significantly even on trials in which they knew no rewards were possible. Furthermore, because at least in some conditions the incentive value was indicated even before the trial started, and in all conditions it was known prior to when responses were generated, the observed incentive context effects must have reflected the heightened global motivational context, independent of trial-specific motivational value. The identification of the incentive context effect raises a number of interesting issues regarding its characteristics and underlying mechanisms. The present work addresses many of these, although a number of potential implications remain for further study.
The findings from Experiment 1 conclusively rule out an initial interpretation that the incentive context effect is related to practice effects by demonstrating, in three ways, that practice cannot explain the effect. First, block-related changes in control participants who received extensive task-switching practice did not follow performance patterns similar to those of participants who did receive incentive instructions. Second, the context effect was still observed relative to nonincentive blocks performed after incentive blocks. Finally, the effect was abrupt, indicating a strong discontinuity in performance at the incentive context transition point—a quality that is not at all characteristic of practice effects (which are gradual).
Additionally, the present study demonstrates that the incentive context effect is not a direct carryover effect from the arousal, affect, or motivational salience associated with recently performed incentive trials. Such a carryover could manifest as either a “bleed over” of incentive benefits to subsequent nonincentive trials, or instead as a dampening effect on nonincentive trials related to the disappointment, frustration, or interference occurring when nonincentive trials followed incentives. The latter interpretation predicts that nonincentive trials that immediately followed incentive trials would be performed worse than baseline, whereas the former interpretation would predict that nonincentive trials would be improved following an incentive trial, or in a context that had a higher frequency of incentive trials (relative to a low-frequency incentive context). Yet neither of these patterns was observed in the data. The incentive context effect was identical, whether performed in a low-frequency incentive context or a high-frequency context. Moreover, there was no effect of incentive lag in either context: Performance was equivalent on nonincentive trials, whether they immediately followed an incentive trial or occurred more than three trials since the last incentive trial had been experienced. Thus, it appears that the incentive context effect is not directly tied to the immediate presence of local incentives. Nevertheless, it is still possible that boundary conditions may constrain the effect. On the basis of our data, we cannot say whether the context effect would still be of the same magnitude, or even still present at all, if incentive opportunities were very sparse (e.g., 5 % or less of incentive trials). But, at the minimum, the incentive context effect appears relatively insensitive to the actual presence of locally cued incentive trials.
A third interpretation is that the incentive context effect emerges as a function of experience, performance feedback, or reinforcement-based learning during incentive blocks. This interpretation was also not supported by the data. Indeed, we observed that the incentive context effect had an instantaneous onset following the presentation of the block instructions and appeared to be at its asymptotic level on the very first trial within the incentive block. Thus, the effect appeared to reflect an immediate instruction-based shift in motivational state, rather than a gradual emergent process tied to task performance or reinforcement-related feedback. On the other hand, providing instructions that removed the incentive context did not elicit the same instantaneous return to baseline performance levels. Instead, the effect extinguished slowly, only returning to baseline performance levels after multiple blocks. This long-term carryover effect of incentive context is suggestive of an arousal-like process that scales with motivational salience, with higher salience leading to a longer time course of decay. Thus, an interesting question for future work would be to determine whether the residual effects of performing in an incentive context are nonspecific—a marker of a truly arousal-like process—and so would be exhibited even when transferring to a changed task context (i.e., incentive task switching to nonincentive working memory task). Our preferred interpretation is that the context effects reflect a modification of an overall goal structure, and as such are tied to that structure, so that context effects would only be observed when performing additional blocks within the same task goal structure. Of course, this is an empirical question that can only be addressed with additional data.
A final issue of interpretation concerns whether the incentive context effect merely reflects a strategic change in a speed-versus-accuracy set that is directly induced by the task instructions, an effect that would duplicate decades of cognitive research demonstrating that participants can change their performance criteria on the basis of speed-stress versus accuracy-stress task instructions (Bogacz, Wagenmakers, Forstmann, & Nieuwenhuis, 2010; Forstmann et al., 2008; Meyer et al., 1990; Strayer & Kramer, 1994; van Veen, Krug, & Carter, 2008; Wickelgren, 1977). In fact, the results indicate that a speed–accuracy account only fits a portion of the results obtained. First of all, the instructional set provided to participants was not identical to standard speed-stress versus accuracy-stress instructions, in that participants were only rewarded for trials on which they were both fast and accurate. That is, error trials were just as undesirable as slow trials, and as such the instructions should not have encouraged a primarily speed- or accuracy-based strategy. Instead, such an instructional set might facilitate an optimization of speed–accuracy trade-off parameters (Balci et al., 2011). Indeed, in both experiments we found that the incentive context led to a substantial speed-up of RTs, while error rate effects were minimal, and some conditions even showed a numerical decrease rather than increase. Finally, the diffusion model analyses not only revealed a reduction in the response caution parameter, which would be expected under a speed–accuracy shift, but also an extremely large reduction in nondecision time. In prior studies that have used the diffusion model to examine speed–accuracy manipulations, the only change that has been found is a reduction in response caution, with no additional changes to nondecision time (Dutilh et al., 2010; Forstmann et al., 2008). The nondecision time effects of incentive context can be interpreted as a change in goal structure and prioritization that enhances sustained attention to task-relevant processing, and also may prime motor responding. Sustained attention to the task should increase the efficiency of both stimulus encoding and the implementation of response execution processes, as well as impacting the decision stage itself.
A last key point to make regarding the potentially surprising discovery of the context effect is that it may be somewhat elusive to observe, being potentially masked by other effects. Indeed, our ability to detect, isolate, and characterize the incentive context effect in the present study was dependent on the utilization of an experimental design that manipulated motivational value in both a block-based and a trial-specific manner. Only through such a design could sustained effects of context be distinguished from more transient ones. In designs that use only block-based or only trial-by-trial motivational manipulations (i.e., the standard approach in the literature), it is impossible to determine whether findings are due to incentive context or incentive cue effects. Therefore, we recommend that future work in this area attend closely to the nature of the motivational effects observed, via the types of designs utilized in the present investigation.
Local incentive enhancement of cognitive control
A second key aspect of the results is that they help to more precisely conceptualize the impact of local (i.e., trial-specific) incentive signals. Although in the present work we only examined motivational effects within cued task switching, we believe that the findings extend more generally to other paradigms containing a similar cue–target structure. Specifically, we postulate that a specific benefit of local incentive signals is that they may enhance proactive forms of cognitive control, through the activation of trial-specific task representations, task rules, or appropriate stimulus–response mappings. These effects will be most prominent when local incentive signals can be integrated with task-specific cues that provide information regarding relevant contextual, attentional, or task information needed for optimal performance. Indeed, the present results regarding local incentive cueing are consistent with the results from prior published studies involving a range of tasks and experimental paradigms that have utilized similar cue–target structures, including the AX-CPT context-processing task (Locke & Braver, 2008), cued spatial attention (Engelmann et al., 2009; Engelmann & Pessoa, 2007; Small et al., 2005), cued task-switching versions of the Stroop (Veling & Aarts, 2010), and working memory tasks (Bijleveld, Custers, & Aarts, 2009; Jimura et al., 2010; Taylor et al., 2004).
A strong converse claim of the present work is that in task situations for which there is no means to improve performance via the utilization of preparatory control (i.e., no cue–target structure), the impact of local motivational incentives on cognitive control indices should be minimal. For example, in recent work, it was found that in an uncued flanker task, monetary rewards shifted the speed–accuracy trade-off function to the left, suggesting earlier information accumulation, providing benefits complementary to the drift rate effects found in the present study (Hübner & Schlösser, 2010). However, the magnitude of the conflict effect was unchanged, suggesting no specific effect on cognitive control processes. In our own laboratory, we have replicated these basic findings, but we also found that under flanker conditions involving informative preparatory cues, significantly reduced conflict effects were present (Chiew & Braver, 2011). This suggests that the ability of motivational signals to reduce conflict-related interference is dependent on their use in a preparatory or proactive manner, rather than in a reactive one. Indeed, in other recent work, it was found that during a response inhibition task (stop signal), which strongly depends on reactive control, motivational incentives actually impaired, rather than improved, inhibitory performance (Padmala & Pessoa, 2010). Further work will be needed to determine whether trial-based incentive cues can sometimes enhance reactive forms of cognitive control, or whether the effects are restricted to proactive or preparatory control situations.
Relationship between the incentive cue and context effects
A primary goal of the present study was to establish the dissociability of the incentive cue and context effects. This naturally leads to the interpretation that the two effects are fully independent. Nevertheless, although we have pulled apart the two effects through experimental contrasts (i.e., nonincentive vs. baseline trials and incentive vs. nonincentive trials), it is important to note that the effects do work together in terms of performance outcomes. Participants were rewarded on the basis of their performance during incentive trials, but the reward criteria were based on performance in the baseline blocks. This made reward rate attainment dependent on how much performance improved on nonincentive trials relative to baseline, as well as on the additional improvement on incentive trials relative to the nonincentive trials occurring within the incentive block. In other words, reward rates were related to the magnitudes of both the incentive context effect and the incentive cue effect (reward rate = total incentive benefit = incentive context + incentive cue). This aspect of the design is critical, because it seems clear that reward rate is the quantity that is most salient to participants, and the one being optimized via performance shifts.
On the basis of these considerations, an alternative conceptualization of the cue and context effects is that, under conditions that minimize the strength of one effect, the other might be increased in a compensatory manner, in order to maintain a high reward rate. Indeed, in the incentive-timing manipulations, the incentive cue effect was minimal or eliminated when incentive cues were linked to target presentation and the incentive context effect was largest. However, rates of obtained reward were relatively constant across the incentive timing conditions, even though the incentive cue and context effects fluctuated significantly. This pattern is suggestive of a compensatory regulation pattern, which is also consistent with the finding that trait reward sensitivity predicted both the incentive cue and context effects. If trait reward sensitivity was primarily associated with increased motivation to optimize reward rates, this could be accomplished most effectively by enhancing the incentive cue effect when conditions favored preparatory processing (i.e., cue-linked incentives), but conversely by a compensatory enhancement of the incentive context effect when incentives cues had low utility for preparation (i.e., target-linked incentives). Under this account, then, trait reward sensitivity actually might reflect an increased ability to regulate the cue and context effects most effectively in the service of reward rate optimization.
A compensatory perspective on the cue and context effects would follow most naturally from a capacity-limited resource model, in which motivational value may govern how cognitive resources are allocated toward task processing. Under this perspective, the cue and context effects would naturally trade off against each other. For example, when resources had to be devoted primarily to processing incentive trials, there might have been fewer resources available for processing on nonincentive trials. Yet our data provide no evidence for these kinds of effects, and instead show the opposite pattern: Under conditions with a high proportion of incentive trials, the magnitude of the incentive context effect was not reduced (as compared to the low-proportion conditions), as would be expected under a resource-depletion-type model. However, it may be that the opposite type of resource allocation problem exists, such that it may be difficult to modulate resources dramatically on a trial-by-trial basis, due to “startup” costs. Thus, an optimal resource allocation pattern might be to enhance performance more globally, so that smaller additional resource expenditures are needed on individual incentive trials. In the present study, we were not completely able to adjudicate the extent to which this type of compensation might be occurring, as contextual motivational information and local motivational cueing were both present within all incentive blocks. Future work could test whether compensatory strategies occurred during task performance by directly modulating the reward value of incentive trials as a function of nonincentive-trial performance. For example, if reward value was higher on incentive trials when recent nonincentive-trial performance was low, it would encourage a strategy of more dramatic incentive > nonincentive performance fluctuations, whereas the reverse valuation policy (increasing incentive-trial reward value when recent nonincentive-trial performance was high) would encourage a strategy of maintaining a stable level of high performance across all trials. Testing whether or not either or both compensatory strategies could be implemented would provide important clues as to whether incentive cue effects can regulate incentive context effects, and vice versa.
Implications regarding neural mechanisms
The data generated from the present study converge with prior cognitive neuroscience research in providing insights into the potential neural mechanisms that underlie local and global motivational influences. With regard to the trial-by-trial enhancements of cognitive control associated with the incentive cue effect, the most likely neural locus is the frontoparietal brain control network, which includes lateral prefrontal cortex (PFC), anterior cingulate cortex, and posterior parietal cortex. In a number of neuroimaging studies showing incentive-related modulation of cognition, activation in all of these regions has been found to be reliably enhanced by motivational value (Beck, Locke, Savine, Jimura, & Braver, 2010; Locke & Braver, 2008; Padmala & Pessoa, 2011; Pessoa & Engelmann, 2010; Pochon et al., 2002; Taylor et al., 2004). More specifically, in a prior neuroimaging study using this incentive task-switching paradigm, we found that left lateral PFC was the most likely site of motivational and cognitive control integration (Savine & Braver, 2010). This region showed both task-switching and incentive-cue-related increases in activation, which were correlated with performance enhancements on both a between-subjects and a trial-by-trial basis. The finding of lateral PFC involvement in this domain is highly consistent with a wealth of experimental findings and theoretical models indicating that the lateral PFC is centrally involved in successful goal maintenance (Miller & Cohen, 2001). Indeed, the results of this prior work suggested that motivational signals increase task goal activation strength, which is the primary functional mechanism of enhanced cognitive control in task switching and similar paradigms.
What is the neural source of the motivational signals that lead to lateral-PFC-based enhancements of cognitive control? A likely hypothesis is that phasic signals indicating the motivational salience of incentive cues are generated in the midbrain dopamine (DA) system (i.e., the ventral tegmental area, VTA; or the substantia nigra). A wealth of evidence supports the role of this system in signaling motivationally relevant information (Aarts et al., 2010; Berridge, 2004, 2007; Cools, 2008), with clear evidence of phasic responses occurring to the presence of reward-predicting cues (Schultz, 2000, 2002). Moreover, DA release in PFC is known to be associated with enhancements in the active maintenance capabilities of this region (Cohen, Braver, & Brown, 2002; Durstewitz, Kelc, & Güntürkün, 1999; Murphy, Arnsten, Goldman-Rakic, & Roth, 1996; Vijayraghavan, Wang, Birnbaum, Williams, & Arnsten, 2007). Finally, we have postulated that phasic DA activity in lateral PFC is a necessary precondition for preparatory or proactive control, in that such signals enable updating and robust maintenance based on contextual cues (Braver & Cohen, 2000; Braver, Gray, & Burgess, 2007). In our prior work, we have provided preliminary support for this idea (Savine & Braver, 2010), in that incentive cues during task switching were associated with increased activation in the dopaminergic midbrain. Dynamic causal modeling has also linked activations in the VTA with enhanced brain activation increases in PFC under incentivized conditions, yet in this case the directional influence was reversed, with incentive cues increasing the modulatory influence of PFC on VTA (Ballard et al., 2011). A role for phasic DA release in mediating local motivational influences on performance would also be consistent with the rapid onset and rapid offset of the local incentive-cueing effects observed in the present experiments. However, more direct support is still needed for the hypothesis that motivational incentives are mediated by direct DA release in lateral PFC.
There is less consensus regarding the candidate neural mechanisms underlying the incentive context effect. Considering its sluggish temporal dynamics of offset and its task-general nature, our interpretation is that the incentive context effect reflects increases of sustained attention or cognitive effort within a motivated context (Sarter, Gehring, & Kozak, 2006). Interestingly, in recent work we have shown that the incentive context effect, during working memory, is related to sustained increases in the activation of right-lateralized lateral PFC and parietal cortex (Jimura et al., 2010). This pattern is consistent with classic characterizations of sustained attentional processes being housed within right frontoparietal cortex (Posner & Petersen, 1990). Most recently, we have confirmed and extended this pattern, demonstrating that the incentive context effect was associated with sustained activation in right dorsolateral PFC, while the incentive cue effect was associated with transient activation of left dorsolateral PFC (Savine & Braver, 2012).
Yet, a critical question is whether the incentive context effect may be driven in part by arousal-type processes, as well. The neural specification of the sources of arousal is still not well understood, but progress has been made in recent computational and cognitive-neuroscience models in providing greater specificity as to the linkage between arousal, and in constructs such as sustained attention and response speed/vigor, in terms of the operation of specific brain neuromodulators (e.g., acetylcholine, norepinephrine [NE], and DA; Aston-Jones & Cohen, 2005; Furey, 1997; Niv et al., 2007; St. Peters, Demeter, Lustig, Bruno, & Sarter, 2011). One potential candidate mechanism has been the locus coeruleus norepinephrine (LC-NE) system, as recent work has specifically postulated a role for this system in monitoring task-related utility (Aston-Jones & Cohen, 2005). Under states of high utility, tonic activation of the LC-NE system is suppressed, enabling stronger phasic signals. These phasic signals increase the gain on sensory, decision, and response processes, thus optimizing the rate at which they are carried out. Such a process would be consistent with the role of incentive context in enhancing nondecision-time components of task performance, as well as in reducing response caution, so as to produce a strategic-like shift in the speed–accuracy function.
Another possibility might be that incentive context effects are mediated not by phasic DA or NE, but instead by tonic DA release. A recent account by Niv et al. (2007) has suggested that changes in tonic DA that occur as the environment increases in utility might be important for heightening the general “vigor” or speed of all task-relevant behaviors equivalently. According to this account, while phasic DA may serve a specific directional focus—increasing attention and action toward the specific outcomes that are of highest reward value—tonic DA might instead serve as an activational focusing agent, which increases the engagement of all activities in general, as a way of reducing the “opportunity costs” to achieve effortful behaviors (Salamone, Correa, Farrar, & Mingote, 2007).
These theoretical ideas will of course require further research to test how well they can account for the data that we have obtained. In particular, one empirical constraint is that the two motivational mechanisms appear to operate in a complementary and synergistic manner. How these effects are regulated or controlled is thus an important question. It is well-accepted that tonic and phasic modes of neurotransmitter release (i.e., tonic vs. phasic DA) may be regulated by a number of interacting factors (Grace, Floresco, Goto, & Lodge, 2007), and likewise, interactions between transmitters (i.e., NE and DA) have been frequently hinted at in the literature (Tassin, 1997). Nevertheless, it will be useful to provide direct evidence of complementary regulation of the two effects. In this regard, the most useful experimental technique will be to combine methods that enable monitoring of activity at cortical sites (e.g., fMRI imaging or single-cell recording) with those that enable monitoring or manipulation of neuromodulatory transmitters such as NE or DA (e.g., pharmacological challenge or positron emission tomography radioligand-binding methods). To our knowledge, no such studies have yet been conducted to examine motivation effects on cognitive control, but similar types of approaches have been employed in other domains related to DA and reward processing (Pessiglione, Seymour, Flandin, Dolan, & Frith, 2006; Schott et al., 2008; Schott et al., 2007).
The experiments presented here provide support for a novel global–local conceptualization of how motivation can impact cognitive control. One mechanism (indexed by the incentive cue effect) appears to operate transiently, by enhancing preparatory cognitive control and decision quality on the current trial, while the other mechanism (indexed by the incentive context effect) appears to operate in a more sustained fashion, by increasing the speed of stimulus encoding and response execution while strategically optimizing speed–accuracy trade-offs. These mechanisms work in a complementary manner to optimize task performance and reward attainment across a wide range of task conditions. The present results demonstrate the power and potential obtained from the systematic experimental control and manipulation of motivational variables during task performance, revealing new insights regarding how such variables can dynamically modulate human cognition and associated control processes.
Consistent with prior work using the face/word task-switching paradigm, error rates were higher globally (~15 % to 20 %) as compared to other task-switching paradigms, such as the letter/digit task (~5 % to 10 %). The slightly higher error rate reflects the difficulty of the overall task, which is higher than other standard task-switching paradigms, and provides a greater sensitivity to detect changes in error rate due to the motivational manipulation (if present).
A potential concern with this between-groups analysis is that the control group slightly differed from the experimental group in terms of the numbers of mixed-task blocks (eight vs. six) and of trials per block (64 vs. 48). To address this concern, we conducted a parallel analysis in which only the first mixed-task block in each phase, and only the first 48 trials of each block, were included in the analysis. Even though this analysis was conservative and underpowered, we nevertheless obtained the same Group × Block (phase) interaction [F(2, 38) = 14.5, p < .01], due to significant group differences only in the incentive phase (second block) [t(39) = 3.64, p < .01; first/third phases: ts < 1.27]. This suggests that the slight design differences between the two groups were not a strong contributor to the observed results.
Posterror trials were not removed from the analyses reported. Because the error rate was substantial, removing the posterror trials would have (1) resulted in a large data loss and (2) interacted with reward/nonreward feedback, such that if posterror trials were removed, the analyses tested would have been artificially biased on the trial level by one type of reward feedback (earned incentive). However, for exploratory purposes, we did conduct analyses after excluding posterror trials. The results were largely the same as those reported in the text for the incentive context effect. The only substantive difference was that the incentive cue effects were numerically larger when the posterror trials were removed in the 25 %-incentive-cued condition and numerically smaller in the 75 %-incentive-cued condition. Switch cost attenuation remained relatively unchanged.
Single-task comparisons in this experiment are not included in the subsequent analyses, due to the different numbers of trials present in the different I-CTI conditions. For illustrative purposes, the single-task trials in the experiment are reported in Table 3; the data suggest that these trials followed the same patterns as other conditions and manipulations reported in this study.
In supplementary materials we provide an in-depth discussion of diffusion modeling of cued task-switching data, and comparable results from analyses using a different diffusion modeling algorithm (fast-dm).
This work was supported by NIH Grant No. 2R56MH066078 awarded to T.S.B. and predoctoral NRSA fellowship 1F31DA032152-01A1 awarded to A.C.S. We thank Samuel Moore for his help with portions of this project.
- Aarts, H., Custers, R., & Veltkamp, M. (2008). Goal priming and the affective-motivational route to nonconscious goal pursuit. Social Cognition, 26(5), 555–577.Google Scholar
- Aarts, E., Roelofs, A., Franke, B., Rijpkema, M., Fernández, G., Helmich, R. C., . . . Cools, R. (2010). Striatal dopamine mediates the interface between motivational and cognitive control in humans: Evidence from genetic imaging. Neuropsychopharmacology, 35, 1943–1951. doi:10.1038/npp.2010.68 Google Scholar
- Balci, F., Simen, P., Niyogi, R., Saxe, A., Hughes, J. A., Holmes, P., et al. (2011). Acquisition of decision making criteria: reward rate ultimately beats accuracy Attention, Perception, & Psychophysics, 73(2), 640–657.Google Scholar
- Ballard, I. C., Murty, V. P., Carter, R. M., MacInnes, J. J., Huettel, S. A., & Adcock, R. A. (2011). Dorsolateral prefrontal cortex drives mesolimbic dopaminergic regions to initiate motivated behavior. Journal of Neuroscience, 31, 10340–10346. doi:10.1523/JNEUROSCI.0895-11.2011 PubMedCrossRefGoogle Scholar
- Beck, S. M., Locke, H. S., Savine, A. C., Jimura, K., & Braver, T. S. (2010). Primary and secondary rewards differentially modulate neural activity dynamics during working memory. PLoS One, 5, 1–13.Google Scholar
- Braver, T. S., & Cohen, J. D. (2000). On the control of control: The role of dopamine in regulating prefrontal function and working memory. In S. Monsell & J. Driver (Eds.), Attention and Performance XVIII (pp. 713–738). Cambridge: MIT Press.Google Scholar
- Braver, T. S., Gray, J. R., & Burgess, G. C. (2007). Explaining the many varieties of working memory variation: Dual mechanisms of cognitive control. In A. R. A. Conway, C. Jarrold, M. J. Kane, A. Miyake, & J. N. Towse (Eds.), Variation in working memory (pp. 76–106). New York: Oxford University Press.Google Scholar
- Chiew, K. S., & Braver, T. S. (2011). Monetary incentives improve performance, sometimes: speed and accuracy matter, and so might preparation. Frontiers in Psychology, 2, 325. doi:10.3389/fpsyg.2011.00325
- Dutilh, G., Wagenmakers, E.-J., Visser, I., & Van der Maas, H. L. J. (2010). A phase transition model for the speed-accuracy trade-off in response time experiments. Cognitive Science, 34, 211–250.Google Scholar
- Forstmann, B. U., Dutilh, G., Brown, S. D., Neumann, J., von Cramon, D. Y., Ridderinkhof, K. R., . . . Wagenmakers, E.-J. (2008). Striatum and pre-SMA facilitate decision-making under time pressure. Proceedings of the National Academy of Sciences, 105, 17538–17542. doi:10.1073/pnas.0805903105
- Furey, M. L., Pietrini, P., Haxby, J. V., Alexander, G. E., Lee, H. C., vanMeter, J., . . . Freo, U. (1997). Cholinergic stimulation alters performance and task-specific regional cerebral blood flow during working memory. Proceedings of the National Academy of Sciences, 94, 6512–6516.Google Scholar
- Karayanidis, F., Mansfield, E. L., Galloway, K. L., Smith, J. L., Provost, A., & Heathcote, A. (2009). Anticipatory reconfiguration elicited by fully and partially informative cues that validly predict a switch in task. Cognitive, Affective, & Behavioral Neuroscience, 9, 202–215. doi:10.3758/CABN.9.2.202 CrossRefGoogle Scholar
- Madden, D. J., Costello, M. C., Dennis, N. A., Davis, S. W., Shepler, A. M., Spaniol, J., . . . Cabeza, R. (2010). Adult age differences in functional connectivity during executive control. NeuroImage, 52, 643–657.Google Scholar
- Mansfield, E. L., Karayanidis, F., Jamadar, S., Heathcote, A., & Forstmann, B. U. (2011). Adjustments of response threshold during task switching: A model-based functional magnetic resonance imaging study. Journal of Neuroscience, 31, 14688–14692. doi:10.1523/JNEUROSCI.2390-11.2011 PubMedCrossRefGoogle Scholar
- Markman, A. B., Maddox, W. T., & Baldwin, G. C. (2005). The implications of advances in research on motivation for cognitive models. Journal of Experimental & Theoretical Artificial Intelligence, 14(4), 371–384.Google Scholar
- Meiran, N. (2000). Reconfiguration of stimulus task sets and response task sets during task switching. In S. Monsell & J. Driver (Eds.), Control of cognitive processes: Attention and performance XVIII (pp. 377–400). Cambridge: MIT Press.Google Scholar
- Meyer, D. E., Keith-Smith, J. E., Kornblum, S., Abrams, R. A., & Wright, C. E. (1990). Speed-accuracy tradeoffs in aimed movements: toward a theory of rapid voluntary action. In M. Jeannerod (Ed.), Attention and performance 13: motor representation and control (pp. 173–226). Hillsdale: Lawrence Erlbaum Associates, Inc.Google Scholar
- Müller, J., Dreisbach, G., Goschke, T., Hensch, T., Lesch, K. P., & Brocke, B. (2007). Dopamine and cognitive control: The prospect of monetary gains influences the balance between flexibility and stability in a set-shifting paradigm. European Journal of Neuroscience, 26, 3661–3668. doi:10.1111/j.1460-9568.2007.05949.x PubMedCrossRefGoogle Scholar
- Niv, Y., Daw, N. D., Joel, D., & Dayan, P. (2007). Tonic dopamine: opportunity costs and the control of response vigor. Psychopharmacology, 191, 507–520.Google Scholar
- Pochon, J. B., Levy, R., Fossati, P., Lehéricy, S., Poline, J. B., Pillon, B., . . . Dubois, B. (2002). The neural system that bridges reward and cognition in humans: An fMRI study. Proceedings of the National Academy of Sciences, 99, 5669–5674. doi:10.1073/pnas.082111099
- Savine, A. C., & Braver, T. S. (2012) Two roads to success: global and local motivation enhance goal pursuit via dissociable neural mechanisms. (In Preparation)Google Scholar
- Schneider, W., Eschman, A., & Zuccolotto, A. (2002). E-Prime user’s guide. Pittsburgh: Psychology Software Tools.Google Scholar
- Schott, B. H., Minuzzi, L., Krebs, R. M., Elmenhorst, D., Lang, M., Winz, O. H., . . . Bauer, A. (2008). Mesolimbic functional magnetic resonance imaging activations during reward anticipation correlate with reward-related ventral striatal dopamine release. Journal of Neuroscience, 28, 14311–14319. doi:10.1523/JNEUROSCI.2058-08.2008 Google Scholar