Abstract

Whereas positive feedback is both rewarding and informative, negative feedback can be construed as either punishing (because it is indicative of poor performance) or informative (because it may lead to goal attainment). In this neuroimaging experiment, we highlighted the informational value of negative feedback by intermixing trials with and without feedback. When performance feedback is expected, positive feedback triggers an increase in striatal activity, whereas negative feedback elicits a decrease in striatal activity. We predicted that, in contrast, when feedback receipt is unpredictable, the striatal response to negative feedback would increase. Participants performed a paired-associate learning task during fMRI scanning. In one condition (“blocked feedback”), the receipt of feedback was predictable—participants knew whether or not they would receive feedback for their responses. In another condition (“mixed feedback”), the receipt of feedback was unpredictable—on a random 50% of trials, participants received feedback, and they otherwise received no feedback. Negative feedback in the mixed feedback condition elicited more striatal activity than negative feedback in the blocked feedback condition. In contrast, feedback omission evoked more striatal activity when feedback delivery was expected, compared to when it was unpredictable. This pattern emerged from an increase in caudate activity in response to negative feedback in the mixed feedback condition and a decrease in ventral striatal activity in response to no feedback in this condition. These results suggest that, by emphasizing the informational value of negative feedback, an unpredictable feedback context alters the striatal response to negative feedback and to the omission of feedback.

INTRODUCTION

Learning from the consequences of one's actions is critical for pursuing and attaining goals. Such feedback-based learning has been shown to engage the striatum (Liljeholm & O'Doherty, 2012). The caudate nucleus, in particular, has been consistently implicated in feedback processing (Tricomi & Fiez, 2008; Seger & Cincotta, 2005). When an outcome is perceived as rewarding, there is an increase in caudate nucleus activity, whereas punishment leads to a decrease in caudate activity (Mullette-Gillman, Detwiler, Winecoff, Dobbins, & Huettel, 2011; Delgado, Locke, Stenger, & Fiez, 2003; Delgado, Nystrom, Fissell, Noll, & Fiez, 2000). This pattern extends to intrinsic rewards, such as performance-related feedback—positive feedback leads to an increase in caudate activation, whereas negative feedback elicits a decrease in activity (DePasque Swanson & Tricomi, 2014; Dobryakova & Tricomi, 2013; Marco-Pallares, Müller, & Münte, 2007; Tricomi, Delgado, McCandliss, McClelland, & Fiez, 2006; Seger & Cincotta, 2005; Elliott, Frith, & Dolan, 1997). Accordingly, this region and surrounding striatal regions have been widely implicated in the processing of the subjective value of outcomes (Clithero & Rangel, 2014; Daniel & Pollmann, 2014; Bartra, McGuire, & Kable, 2013).

Importantly, this value signal in the striatum is sensitive to contextual influences, such as delay to reward receipt (Kable & Glimcher, 2007), task expectations (DePasque Swanson & Tricomi, 2014), sense of agency (Tricomi, Delgado, & Fiez, 2004), whether an outcome is framed as a loss or a gain (Tom, Fox, Trepel, & Poldrack, 2007), and social context (Simon, Becker, Mothes-Lasch, Miltner, & Straube, 2014). The subjective value of an outcome also depends on the other available outcomes (Lohrenz, McCabe, Camerer, & Montague, 2007; Nieuwenhuis et al., 2005; Breiter, Aharon, Kahneman, Dale, & Shizgal, 2001). For example, receiving $0 in a task may be construed as negative and lead to a dip in striatal activity, if the alternative is receiving a positive amount of money, or it may be construed as positive and lead to an increase in striatal activity, if the alternative is losing money (Nieuwenhuis et al., 2005).

This context sensitivity may emerge because the striatum is a major target of dopamine neurons, which have been shown to encode “prediction error,” or the difference between expected and actual outcomes (Bayer & Glimcher, 2005; Schultz & Dickinson, 2000; Schultz, Dayan, & Montague, 1997). After all, expectations are formed on the basis of contextual cues, including counterfactual comparisons. By signaling the deviation from expectation, dopamine links reward processing and learning—highly unexpected positive outcomes are more informative for future behavior and are also perceived as more rewarding. Indeed, the level of information that an outcome provides is an important contributor to the level of dopaminergic firing. Dopamine neurons signal advance information about outcomes, even if knowing this information has no bearing on the future outcome (Bromberg-Martin & Hikosaka, 2009). In addition, studies have found that a subset of dopamine neurons code for “motivational salience,” rather than reward prediction error: Even when feedback is inherently negative (e.g., air puff to the eye), there will be an increase in dopaminergic firing if this feedback is unexpected (Matsumoto & Hikosaka, 2009).

Most paradigms conflate the informative and affective subcomponents of reward (Smith & Delgado, 2015), because they are often coupled, but what happens when informative outcomes are not also valuable? Does the striatum prioritize informational value or valence? To answer this question, it is useful to consider the special case of negative performance-related feedback. Positive performance feedback is both rewarding and informative; therefore, it will almost invariably lead to an increase in striatal activity. Negative feedback, on the other hand, is informative, but it can be construed as either punishing (because it is indicative of poor performance) or rewarding (because it may eventually lead to goal attainment). Tricomi and Fiez (2012) examined whether the level of information that negative feedback provides influences striatal responses to negative feedback using fMRI. In their task, when there were four possible correct responses, negative feedback was relatively uninformative. When there were only two possible correct responses, negative feedback was maximally informative, because the correct answer was then evident. Negative feedback in the two-response condition elicited an increase in caudate fMRI signal relative to negative feedback in the four-response condition. This is consistent with a “goal attainment” account of caudate activity (Han, Huettel, Raposo, Adcock, & Dobbins, 2010). Bischoff-Grethe, Hazeltine, Bergren, Ivry, and Grafton (2009) found that negative performance feedback elicited more signal in the caudate than uninformative feedback, in a context in which positive feedback was not a possible outcome. That is, when traditionally “rewarding” feedback is unavailable, the striatum may process the most informative outcome (in this case, negative performance feedback) as it would process a reward.

These previous studies provide preliminary evidence that highlighting the informational aspect of negative feedback can lead to an increase in striatal activity and may also elicit a preference for negative feedback relative to no feedback. However, these studies had several constraints. By changing the number of options presented in different conditions, Tricomi and Fiez (2012) changed the nature and difficulty of the task. Bischoff-Grethe et al.'s (2009) results relied on the absence of positive feedback as a possible outcome. In the current study, we sought to underline the informational value of negative feedback with a simpler manipulation. Participants performed a paired-associate learning task (as in Tricomi & Fiez, 2008, 2012) during fMRI scanning, followed by a posttest. In one condition (the “blocked feedback” condition), the receipt of feedback was totally predictable—participants knew whether they would receive feedback for their responses or whether they would not receive feedback. In another condition (the “mixed feedback” condition), the receipt of feedback was totally unpredictable—on a random subset (50%) of trials, participants received feedback for their responses, and on the other 50% of trials, they received no feedback about whether they made a correct response. We hypothesized that, consistent with past studies, when feedback was expected, negative feedback would be perceived as a punishment and would lead to a decrease in striatal activity compared to positive feedback. In contrast, when feedback was unexpected, we hypothesized that the presence of feedback, regardless of its valence, would be preferable to no feedback (because feedback provides information that is useful for goal attainment) and would thus lead to increased striatal activity.

METHODS

Participants

Twenty-two participants were initially recruited for this study (10 men, 11 women, 1 not reported; mean age = 23.05, SD = 3.03). Two participants were excluded from analyses because of excessive motion during scanning. The final behavioral and neuroimaging analyses were conducted on 20 participants (9 men, 11 women; mean age = 22.86, SD = 2.97). Participants responded to a posted advertisement, and all gave written informed consent. The experiment was approved by the institutional review board of the University of Medicine and Dentistry of New Jersey as well as the institutional review board of Rutgers University.

Experimental Procedure

This experiment involved a paired-associate word learning task similar to the one used in our previous work (Dobryakova & Tricomi, 2013; Tricomi & Fiez, 2008, 2012). It began with a “study phase” outside the scanner, during which participants viewed a series of 220 distinct target words with two possible word matches below each one. The correct word match was highlighted in green, and each set of words (target word + two possible matches) was presented on screen for 6 sec. Participants were instructed to try their best to remember the correct matches for the target words, because they would be tested on this knowledge later. The words were matched for word length and frequency at the trial level. The words contained four to eight letters and one to two syllables, had Kučera–Francis frequencies of 20–650 words per million, and had imageability ratings of over 400 according to the MRC database (Coltheart, 1981). The words presented on the same trial had a score of less than 0.2 on the latent semantic analysis similarity matrix (Landauer, Foltz, & Laham, 1998), indicating low semantic relation, and they did not rhyme or begin with the same letter.

After the study phase, fMRI data were acquired while participants performed the “learning phase.” One hundred sixty of the target words from the study phase were presented again, but this time the correct matches were not highlighted. Participants were instructed to press Button 1 or Button 2 to indicate which word they believed was the correct match for the target word based on their memory from the study phase. After a 4-sec response period, a feedback display was shown (1 sec). The trial ended with a jittered intertrial interval (1–8 sec; Figure 1). Before each set of eight trials, participants were shown a label indicating whether the set of trials would be a “maybe feedback” block, “definite feedback” block, or “no feedback” block. In the “maybe feedback” blocks (i.e., mixed feedback condition), the feedback display consisted of veridical feedback (a picture of a green check mark for a correct answer; a red “X” for a wrong answer) for 50% of trials, and no feedback (a pound sign) for 50% of trials. Participants did not know which trials would yield feedback, but they understood that whether feedback would be presented on each trial of the “maybe feedback” blocks was randomly determined. In “definite feedback” blocks (i.e., blocked feedback condition), participants always received veridical feedback (check mark or red X) indicating whether they were correct or incorrect. Finally, in “no feedback” blocks (also in the blocked feedback condition), participants were always shown a pound sign after their response, whether it was correct or incorrect.

Figure 1. 

Example trial in learning phase. Each trial began with a screen indicating what type of trial it was (“definite feedback”; “no feedback”; “maybe feedback”; jittered 1–8 sec). Then participants saw a target word with two response options (4 sec). They responded with a button press, indicating which option they believed was paired with the target word in the study phase. Finally, they received either veridical feedback or no feedback, depending on the trial type (1 sec).

Figure 1. 

Example trial in learning phase. Each trial began with a screen indicating what type of trial it was (“definite feedback”; “no feedback”; “maybe feedback”; jittered 1–8 sec). Then participants saw a target word with two response options (4 sec). They responded with a button press, indicating which option they believed was paired with the target word in the study phase. Finally, they received either veridical feedback or no feedback, depending on the trial type (1 sec).

Eighty trials were in the mixed feedback condition (40 with and 40 without feedback), 40 were in the blocked condition with feedback, and 40 were in the blocked condition with no feedback. Therefore, in half of trials, receipt of feedback was perfectly predictable, whereas in the other half of trials receipt of feedback was unpredictable (see Figure 2 for learning phase summary). Participants were asked to use any feedback that they did receive during this phase to try to remember the correct word pairs, because they would be tested again later.

Figure 2. 

Learning phase summary. There were 160 trials in the learning phase, during which participants underwent fMRI scanning. These trials were divided into two conditions (“mixed” and “blocked” feedback conditions), so that for half of feedback trials, feedback was expected, and for the other half of feedback trials, feedback was unexpected.

Figure 2. 

Learning phase summary. There were 160 trials in the learning phase, during which participants underwent fMRI scanning. These trials were divided into two conditions (“mixed” and “blocked” feedback conditions), so that for half of feedback trials, feedback was expected, and for the other half of feedback trials, feedback was unexpected.

In this scanning phase, there were four functional runs constituting 172 repetition times each (approximately 7 min 10 sec). Each run contained either only mixed feedback trials or only blocked feedback trials, and they alternated. Thus, there were two possible orders (Mixed/Blocked/Mixed/Blocked or Blocked/Mixed/Blocked/Mixed), and the order was counterbalanced across subjects.

Immediately following the scan, participants completed a computerized posttest. Participants were shown each target word and its two possible matches, in random order, and were once again asked to press a button indicating which match they thought was correct. The posttest was self-paced, and participants made confidence judgments following each trial by choosing a number from 1 to 7 (1 = complete guess, 7 = completely sure). Participants also completed a short posttask questionnaire, containing two key questions to elicit their preferences about feedback in the mixed feedback condition: (1) Which did you prefer—getting negative (red X) feedback for a wrong answer in the “maybe feedback” phase, or getting no feedback (pound sign) for a wrong answer in the “maybe feedback” phase? (2) How did you feel when you did not get feedback in the “maybe feedback” condition?

fMRI Data Acquisition

A 3T Siemens (Erlangen, Germany) Allegra head-only scanner and a Siemens standard eight-channel head coil were used for data acquisition at the University of Medicine and Dentistry of New Jersey. Anatomical images were acquired using a T1-weighted protocol (256 × 256 matrix, 176 1-mm sagittal slices) Functional images were acquired using a single-shot gradient-echo EPI sequence (repetition time = 2500 msec, echo time = 25 msec, flip angle = 80°, field of view = 192 × 192 mm, slice gap = 0 mm). Forty-three contiguous oblique-axial slices (3 mm × 3 mm × 3 mm voxels) were acquired in an oblique orientation of 30° to the AC–PC axis, which reduces signal dropout in the ventral PFC relative to AC–PC aligned images (Deichmann, Gottfried, Hutton, & Turner, 2003).

Behavioral Data Analysis

We analyzed accuracy (proportion correct) from the learning phase and the posttest, as well as confidence data from the posttest. For learning phase accuracy, we compared the number of positive and negative feedback events between the mixed and blocked conditions using a paired t test. This was to ensure that fMRI power was roughly matched between conditions. For posttest accuracy, we performed a repeated-measures ANOVA with three factors: (1) Feedback availability in learning phase (received/omitted), (2) Condition (mixed/blocked), and (3) Learning phase accuracy (correct/incorrect). For posttest confidence, we also conducted an ANOVA including these three factors, plus an additional factor for Posttest accuracy (correct/incorrect).

fMRI Data Analysis

Analysis of imaging data was conducted using Brain Voyager QX software, version 2.0 (Brain Innovation, Maastricht, The Netherlands). The data were initially corrected for motion and slice scan time using cubic spline interpolation. Furthermore, spatial smoothing was performed using a three-dimensional Gaussian filter (6 mm FWHM), along with voxel-wise linear detrending and high-pass filtering of frequencies (three cycles per time course). Structural and functional data of each participant was then transformed to standard Talairach stereotaxic space (Talairach & Tournoux, 1988).

A random-effects analysis was performed on the functional data using a general linear model. The onsets of each feedback event were modeled as stick functions and then convolved with a canonical hemodynamic response function to create regressors of interest for the different conditions (i.e., positive, negative, and no feedback events in the mixed feedback condition and positive, negative and no feedback events in the blocked feedback condition). Regressors of no interest were also generated using the realignment parameters from the image preprocessing to further correct for residual subject motion. Regressors of no interest also included the few trials for which no subject response was recorded.

Because we were primarily interested in striatal activity in response to feedback, we selected a priori ROIs in the caudate and the ventral striatum for testing. In the caudate, cubic regions of 125 mm3 were centered at x = 12, y = 8, and z = 12 and x = −12, y = 8, and z = 12. These coordinates were used in a previous study investigating a similar question (Tricomi & Fiez, 2012), and these regions have been shown to be responsive to performance feedback. The ROIs in the ventral striatum were the same size, and were centered at x = 10, y = 8, and z = −4 and x = −10, y = 8, and z = −4. These coordinates have been used to define ventral striatal ROIs in previous research (Tricomi & Lempert, 2015; Bischoff-Grethe et al., 2009) and are based on peak activations from studies of reward processing (Breiter et al., 2001; Delgado et al., 2000). Time courses of activity were extracted from these regions for each of the six event types, and statistical analysis was performed on the average z-scored time series data in the period 6–9 sec after feedback presentation. This is the time period at which differences between feedback conditions have been found to emerge in past studies (Tricomi & Fiez, 2008; Tricomi et al., 2006). Specifically, we conducted a repeated-measures ANOVA with Feedback valence (positive/negative/no feedback) and Condition (mixed/blocked) as factors to examine differences in striatal activity between conditions.

In addition, we performed pairwise contrasts on positive, negative, and no feedback events within the mixed feedback runs and within the blocked feedback runs. We also contrasted each feedback type between conditions (positive feedback blocked > positive feedback mixed; negative feedback blocked > negative feedback mixed; no feedback blocked > no feedback mixed). We generated whole-brain statistical maps based on these contrasts. These maps were cluster-threshold corrected to a whole brain p value of .005, from an uncorrected voxelwise p value of <.005, to minimize the likelihood of Type I error. We selected this alpha-level for cluster thresholding using Bonferroni correction to account for the fact that we performed nine comparisons (0.05/9 = 0.0056).

RESULTS

Behavioral Results

Learning Phase Accuracy

The average number of positive feedback events in the mixed feedback condition was 25.9 (of 40; SD = 4.98; average number of negative feedback events = 14.1), which was not significantly different from the number of positive feedback events in the blocked feedback condition (mean = 25.85, SD = 5.46; average number of negative feedback events = 14.15; two-tailed paired t test: t(19) = 0.05; p = .96). In the learning phase overall, the average proportion correct was 0.66 (SD = 0.10), which was above chance (t(19) = 6.72; p < .001).

Preference Ratings

On the posttask questionnaire, 17 of the 20 participants reported a preference for getting negative feedback to getting no feedback in the mixed feedback condition. In describing getting no feedback in the mixed feedback condition, they reported feeling “uncertain,” “frustrated,” “unsure,” “nervous,” and “annoyed.”

Posttest Accuracy

On average, participants performed above chance on the posttest (mean proportion correct = 0.72; SD = 0.45; t(19) = 8.79; p < .001). Despite subjective reports, during the test phase, there was no main effect of Condition on accuracy (F(1, 21.35) = 0.002; p = .966). Therefore, participants were able to learn equally well from feedback in both learning contexts. There was a main effect of Feedback availability (F(1, 21.07) = 32.03; p < .001). Accuracy for feedback trials from both conditions was better than accuracy for trials in which feedback was not received (feedback received vs. no feedback: t(19) = 4.56, p < .001; feedback vs. no feedback in mixed feedback condition: t(19) = 3.06, p = .006; feedback vs. no feedback in blocked feedback condition: t(19) = 3.51, p < .001; Figure 3A). This demonstrates that participants were learning from the feedback given to them during the learning phase in the scanner.

Figure 3. 

(A) Posttest accuracy for words from the mixed (left) and blocked (right) feedback conditions. Although there were no differences in learning from feedback between conditions, participants performed better at posttest if they had received feedback during the learning phase than if they had not received feedback. They also performed better if they had previously been accurate on those words. Dotted line indicates chance performance. (B) Posttest confidence. Participants endorsed higher confidence ratings for words that they had seen in the mixed feedback context compared to those they saw in the blocked feedback context. They were also more confident on trials that were correct in the posttest as well as on those that were correct in the learning phase. Error bars indicate SEM. *p < .05; **p < .01; ***p < .001; ns = not significant.

Figure 3. 

(A) Posttest accuracy for words from the mixed (left) and blocked (right) feedback conditions. Although there were no differences in learning from feedback between conditions, participants performed better at posttest if they had received feedback during the learning phase than if they had not received feedback. They also performed better if they had previously been accurate on those words. Dotted line indicates chance performance. (B) Posttest confidence. Participants endorsed higher confidence ratings for words that they had seen in the mixed feedback context compared to those they saw in the blocked feedback context. They were also more confident on trials that were correct in the posttest as well as on those that were correct in the learning phase. Error bars indicate SEM. *p < .05; **p < .01; ***p < .001; ns = not significant.

Unsurprisingly, there was a main effect of Learning phase accuracy (F(1, 19.82) = 90.88; p < .001) and a significant interaction between Feedback availability and Learning phase accuracy (F(1, 22.93) = 37.03; p < .001) on posttest accuracy; that is, when participants received feedback in the learning phase, they were more likely to be accurate in the posttest, especially if they had been given negative feedback, which allowed errors to be corrected. No other interactions were significant (Condition × Feedback availability: F(1, 21.34) = 0.30; p = .589; Condition × Learning phase accuracy: F(1, 21.48) = 0.98; p = .33; Condition × Learning phase accuracy × Feedback availability: F(1, 21.24) = 2.04; p = .17).

Posttest Confidence Ratings

The average confidence rating during the posttest was 4.66 (scale from 1 to 7; SD = 2.22). As expected, there was a main effect of posttest accuracy on confidence (F(1, 22.38) = 70.42; p < .001); confidence was higher for accurate than for inaccurate posttest responses (t(19) = 8.62; p < .001). There was also a significant main effect of Learning phase accuracy (F(1, 25.98) = 14.5; p = .001), and a Learning phase accuracy × Posttest accuracy interaction (F(1, 26.81) = 24.05; p < .001) on confidence ratings. Confidence was higher for words that participants had gotten correct in the learning phase, especially if they were accurate on these words in the posttest as well.

Consistent with the accuracy data, there was a main effect of Feedback availability (F(1, 46.83) = 12.61; p = .001) on confidence. Participants displayed increased confidence for target words for which feedback was received during the learning phase. This effect was especially pronounced for trials that were correct at posttest; that is, there was a significant Feedback availability × Posttest accuracy interaction (F(1, 31.16) = 4.9; p = .034). Although there was no significant two-way interaction between Feedback availability and Learning phase accuracy (F(1, 26.96) = 0.44; p = .561), there was a significant three-way interaction between Feedback availability, Learning phase accuracy, and Posttest accuracy (F(1, 27.02) = 11.90; p = .002). This indicates that participants showed increased confidence for target words that were correct at the posttest and for which feedback—negative feedback, in particular—was received previously.

Notably, there was a main effect of Condition (mixed vs. blocked) on confidence ratings (F(1, 28.12) = 4.34; p = .046). Post hoc paired t tests revealed that confidence ratings for words from mixed feedback trials were significantly higher than ratings for words from blocked feedback trials (t(19) = 4.12, p < .001; Figure 3B). The interaction between Condition and Feedback availability was not significant (F(1, 31.61) = 0.48; p = .493), showing that confidence was higher when trials were seen in a mixed context, even when feedback was not previously received (t(19) = 2.23; p = .038). All remaining interactions in the ANOVA were not significant (Condition × Learning phase accuracy: F(1, 39.57) = 2.4; p = .129; Condition × Posttest accuracy: F(1, 31.86) = 0.59; p = .45; Condition × Feedback availability × Learning phase accuracy: F(1, 28.07) = 0.89; p = .353; Condition × Learning phase accuracy × Posttest accuracy: F(1, 21.87) = 0.35; p = .561; Condition × Feedback availability × Posttest accuracy: F(1, 36.63) = 0.517; p = .477; Condition × Feedback availability × Learning phase accuracy × Posttest accuracy: F(1, 17.33) = 2.048; p = .17).

fMRI Results

Caudate ROI Analysis

We extracted time courses, time-locked to feedback onset, for each event type from two regions in the caudate head, centered on the coordinates ±12, 8, 12 (Figure 4). A repeated-measures ANOVA on the average z-scored time series data in the period 6–9 sec following feedback presentation revealed a main effect of feedback valence (F(2, 38) = 4.14; p = .024) and a significant Feedback valence × Condition interaction (F(2, 38) = 4.17; p = .023) on the right side, as well as on the left side (Valence: F(2, 38) = 4.28; p = .021; Valence × Condition: F(2, 38) = 6.09; p = .005). There was no main effect of Condition on either side (right: F(1, 19) = 1.08; p = .311; left: F(1, 19) = 1.18; p = .291).

Figure 4. 

Time courses extracted from right caudate ROI in blocked (A) and mixed (B) feedback conditions, and time courses extracted from left caudate ROI in blocked (C) and mixed (D) feedback conditions. Gray boxes indicate the time period over which significance tests were performed (6–9 sec postfeedback onset). For both regions, negative feedback elicited more activity than no feedback in the mixed feedback condition, but not in the blocked feedback condition. Furthermore, response to negative feedback was greater in the mixed condition compared to the blocked condition, whereas response to no feedback was greater in the blocked condition compared to the mixed condition.

Figure 4. 

Time courses extracted from right caudate ROI in blocked (A) and mixed (B) feedback conditions, and time courses extracted from left caudate ROI in blocked (C) and mixed (D) feedback conditions. Gray boxes indicate the time period over which significance tests were performed (6–9 sec postfeedback onset). For both regions, negative feedback elicited more activity than no feedback in the mixed feedback condition, but not in the blocked feedback condition. Furthermore, response to negative feedback was greater in the mixed condition compared to the blocked condition, whereas response to no feedback was greater in the blocked condition compared to the mixed condition.

Post hoc t tests revealed that there were no differences between condition for positive feedback (right: t(19) = 0.29; p = .772; left: t(19) = 0.17; p = .871). Critically, however, activity in the caudate was greater in response to negative feedback in the mixed condition compared to negative feedback in the blocked condition (right: t(19) = 2.13; p = .046; left: t(19) = 2.45; p = .024). Moreover, the response to no feedback showed the opposite pattern (greater in the blocked than in the mixed conditions: left: t(19) = 2.53; p = .02; trending on the right side: t(19) = 1.83; p = .083).

These differences between conditions appeared to be driven by an increased response to negative feedback in the mixed feedback condition; that is, negative feedback elicited significantly more activity in the caudate compared to no feedback in the mixed feedback context (right: t(19) = 2.31; p = .03; left: t(19) = 4.02; p < .001), but this was not the case in the blocked feedback condition (there was a trend in the opposite direction; right: t(19) = −1.79; p = .08; left: t(19) = −1.19; p = .25). Similarly, positive feedback elicited more striatal activity than negative feedback in the blocked condition (right: t(19) = 2.61; p = .017; left: t(19) = 2.27; p = .035), but not in the mixed feedback condition (right: t(19) = 0.1; p = .92; left: t(19) = −0.31; p = .76). As expected, activity in the caudate was greater in response to positive feedback relative to no feedback in both conditions (mixed, right: t(19) = 2.83; p = .01; mixed, left: t(19) = 3.28; p = .004; blocked, right: t(19) = 3.46; p = .003; blocked, left: t(19) = 3.56; p = .002).

Ventral Striatum ROI Analysis

Time courses extracted bilaterally from ventral striatum (regions centered on coordinates ±10, 8, −4) are shown in Figure 5. Once again, a Valence × Condition ANOVA showed that there was a main effect of Valence (right: F(2, 38) = 14.72; p < .001; left: F(2, 38) = 19.73; p < .001), and a significant Valence × Condition interaction (right: F(2, 38) = 5.27; p = .01; left: F(2, 38) = 6.03; p = .005), but there was no main effect of Condition (right: F(1, 19) = 0.48; p = .50; left: F(1, 19) = 0.16; p = .74).

Figure 5. 

Time courses extracted from right ventral striatum ROI in blocked (A) and mixed (B) feedback conditions, and time courses extracted from left ventral striatum ROI in blocked (C) and mixed (D) feedback conditions. Gray boxes indicate the time period over which significance tests were performed (6–9 sec post feedback onset). For both regions, no feedback elicited more activity than negative feedback in the blocked feedback condition, but not in the mixed feedback condition. Just as in the caudate, response to negative feedback was greater in the mixed condition compared to the blocked condition, whereas response to no feedback was greater in the blocked condition compared to the mixed condition.

Figure 5. 

Time courses extracted from right ventral striatum ROI in blocked (A) and mixed (B) feedback conditions, and time courses extracted from left ventral striatum ROI in blocked (C) and mixed (D) feedback conditions. Gray boxes indicate the time period over which significance tests were performed (6–9 sec post feedback onset). For both regions, no feedback elicited more activity than negative feedback in the blocked feedback condition, but not in the mixed feedback condition. Just as in the caudate, response to negative feedback was greater in the mixed condition compared to the blocked condition, whereas response to no feedback was greater in the blocked condition compared to the mixed condition.

Just as in the caudate, post hoc t tests showed that ventral striatal activity was greater in response to negative feedback in the mixed condition compared to the blocked condition (right: t(19) = 2.78; p = .012; left: t(19) = 2.72; p = .014), and the response to no feedback was greater in the blocked condition relative to the mixed condition (left: t(19) = 2.72; p = .013; trending on the right: t(19) = 1.85; p = .08). Positive feedback was processed similarly in both conditions (right: t(19) = −0.04; p = .97; left: t(19) = 0.43; p = .67).

In the ventral striatum, however, these condition differences stemmed from a decreased response to no feedback in the mixed feedback condition. In this region, there was no difference in activity between negative feedback and no feedback in the mixed condition (right: t(19) = −0.74; p = .47; left: t(19) = −1.18; p = .25), but negative feedback elicited significantly less activity than no feedback in the blocked condition (right: t(19) = −5.27; p < .001; left: t(19) = −5.84; p < .001). In addition, positive feedback elicited more ventral striatal activity than negative feedback in both the mixed (right: t(19) = 2.25; p = .036; left: t(19) = 2.31; p = .032) and blocked conditions (right: t(19) = 5.87; p < .001, left: t(19) = 6.41; p < .001). Positive feedback did not elicit significantly more activity than no feedback in either condition on the right side (mixed: t(19) = 1.77; p = .09; blocked: t(19) = 1.20; p = .25), but this effect was marginally significant on the left side (blocked: t(19) = 2.16; p = .044; mixed: t(19) = 2.17; p = .043).

Whole-brain Analysis: Blocked Feedback Runs

To further explore the processing of feedback within each condition, we conducted whole-brain pairwise contrasts between each of the feedback types within the blocked feedback runs. Replicating past studies (Tricomi & Fiez, 2008; Delgado et al., 2000), a large cluster extending from the ventral striatum through the caudate was significantly activated for the contrast between positive and negative feedback (Figure 6 and Table 1; p < .005, corrected). There was also a large activation extending from the striatum through the medial frontal gyrus for the contrast between positive feedback and no feedback in these runs (Figure 7 and Table 1; p < .005, corrected). The negative feedback > no feedback contrast revealed no significant striatal activation (see Table 1 for full list of activations).

Figure 6. 

Axial views of ventral striatum (left) and caudate activation (right) in positive feedback > negative feedback contrast in the blocked feedback condition (p < .005, cluster threshold corrected).

Figure 6. 

Axial views of ventral striatum (left) and caudate activation (right) in positive feedback > negative feedback contrast in the blocked feedback condition (p < .005, cluster threshold corrected).

Table 1. 

Blocked Feedback Condition Contrasts

RegionCluster Size (mm3)Max t StatTalairach Coordinates (Peak)
Negative Feedback > No Feedback 
Bilateral anterior cingulate, BA 24 891 5.060776 2 52 42 
 1713 5.308829 −4 31 18 
 
No Feedback > Negative Feedback 
Bilateral occipital lobe, lingual gyrus, BA 18 11108 7.018979 −22 −98 −6 
R cerebellum 3060 5.540032 5 −77 −21 
 
Positive Feedback > Negative Feedback 
L inferior temporal gyrus, BA 20 1775 4.454925 −58 −8 −21 
R frontal lobe, BA 4 12124 5.489245 17 −29 54 
L occipital lobe, fusiform gyrus, BA 19 65334 7.817195 −25 −83 −12 
R cerebellum 1925 6.668442 32 −38 −21 
Bilateral ventral striatum/caudate 19233 7.788105 11 22 0 
Bilateral superior frontal gyrus, BA 10 3340 5.694795 −4 64 24 
L cingulate gyrus, BA 31 1502 4.745449 −4 −32 30 
L middle frontal gyrus, BA 10 8360 6.380085 −37 52 0 
L subcallosal gyrus, BA 34 2578 5.180820 −13 4 −15 
L superior frontal gyrus, BA 8 3015 5.577122 −19 25 48 
L inferior parietal lobule, BA 39 4977 5.153872 −49 −65 42 
 
Negative Feedback > Positive Feedback 
(None) 
 
Positive Feedback > No Feedback 
R inferior parietal lobule, BA 40 3755 5.8945 41 −41 45 
Bilateral medial frontal gyrus/striatum 32165 7.6485 2 58 15 
L cerebellum 3891 5.2189 −31 −35 −24 
Bilateral posterior cingulate, BA 23 6173 6.7949 −4 −44 24 
L thalamus 1457 4.3233 −1 −17 12 
L parahippocampal gyrus, BA 30 1071 4.631 −22 −41 3 
 
No Feedback > Positive Feedback 
(None) 
RegionCluster Size (mm3)Max t StatTalairach Coordinates (Peak)
Negative Feedback > No Feedback 
Bilateral anterior cingulate, BA 24 891 5.060776 2 52 42 
 1713 5.308829 −4 31 18 
 
No Feedback > Negative Feedback 
Bilateral occipital lobe, lingual gyrus, BA 18 11108 7.018979 −22 −98 −6 
R cerebellum 3060 5.540032 5 −77 −21 
 
Positive Feedback > Negative Feedback 
L inferior temporal gyrus, BA 20 1775 4.454925 −58 −8 −21 
R frontal lobe, BA 4 12124 5.489245 17 −29 54 
L occipital lobe, fusiform gyrus, BA 19 65334 7.817195 −25 −83 −12 
R cerebellum 1925 6.668442 32 −38 −21 
Bilateral ventral striatum/caudate 19233 7.788105 11 22 0 
Bilateral superior frontal gyrus, BA 10 3340 5.694795 −4 64 24 
L cingulate gyrus, BA 31 1502 4.745449 −4 −32 30 
L middle frontal gyrus, BA 10 8360 6.380085 −37 52 0 
L subcallosal gyrus, BA 34 2578 5.180820 −13 4 −15 
L superior frontal gyrus, BA 8 3015 5.577122 −19 25 48 
L inferior parietal lobule, BA 39 4977 5.153872 −49 −65 42 
 
Negative Feedback > Positive Feedback 
(None) 
 
Positive Feedback > No Feedback 
R inferior parietal lobule, BA 40 3755 5.8945 41 −41 45 
Bilateral medial frontal gyrus/striatum 32165 7.6485 2 58 15 
L cerebellum 3891 5.2189 −31 −35 −24 
Bilateral posterior cingulate, BA 23 6173 6.7949 −4 −44 24 
L thalamus 1457 4.3233 −1 −17 12 
L parahippocampal gyrus, BA 30 1071 4.631 −22 −41 3 
 
No Feedback > Positive Feedback 
(None) 

Activation table for pairwise contrasts between feedback types in the blocked feedback condition.

Figure 7. 

Axial view of striatal activation in positive feedback > no feedback contrast in blocked feedback condition (p < .005, cluster threshold corrected).

Figure 7. 

Axial view of striatal activation in positive feedback > no feedback contrast in blocked feedback condition (p < .005, cluster threshold corrected).

Whole-brain Analysis: Mixed Feedback Runs

Just as in the blocked feedback runs, the ventral striatum and caudate were significantly more activated when positive feedback receipt was contrasted with no feedback during the mixed feedback runs (Figure 8 and Table 2; p < .005, corrected). However, the contrast between positive and negative feedback in the mixed feedback runs yielded no striatal activity at our statistical threshold (Table 2). Critically, the contrast between negative feedback and no feedback in these unpredictable feedback runs revealed robust activity in the left ventral caudate (Figure 9 and Table 2; p < .005, corrected). This finding is consistent with the results of our ROI analysis and indicates that negative feedback does not produce a decrease in striatal activation relative to no feedback in the case where receipt of feedback is unpredictable. Rather, there is an increase in ventral caudate activation following negative feedback.

Figure 8. 

Axial view of striatal activation in positive feedback > no feedback contrast in the mixed feedback condition (p < .005, cluster threshold corrected).

Figure 8. 

Axial view of striatal activation in positive feedback > no feedback contrast in the mixed feedback condition (p < .005, cluster threshold corrected).

Table 2. 

Mixed Feedback Condition Contrasts

RegionCluster Size (mm3)Max t StatTalairach Coordinates (Peak)
Negative Feedback > No Feedback 
R inferior frontal gyrus, BA 47 1209 4.908 26 13 −15 
L medial frontal gyrus, BA 6 5025 5.403 −7 1 51 
Bilateral brainstem, midbrain, red nucleus 4384 6.9973 2 −23 −3 
L ventral striatum/caudate head 1209 5.0343 −7 13 0 
L parietal lobe, postcentral gyrus, BA 3 1814 4.97 −31 −32 57 
 
No Feedback > Negative Feedback 
Bilateral occipital lobe/cerebellum 28524 6.996 −19 −86 −24 
 
Positive Feedback > Negative Feedback 
L cerebellum 48502 7.8136 −31 −77 −24 
Bilateral occipital lobe, cuneus, BA 19 1185 4.8048 29 −80 24 
R parietal lobe, precuneus, BA 31 1750 4.7449 2 −53 30 
 
Negative Feedback > Positive Feedback 
(None) 
 
Positive Feedback > No Feedback 
R parietal lobe, precuneus, BA 31 2267 4.2914 20 −74 24 
Bilateral ventral striatum/caudate head 4403 4.7327 −13 7 3 
Bilateral medial frontal gyrus, BA 10 5170 6.0402 −4 61 15 
Bilateral parietal lobe, precuneus, BA 31 6234 6.558 −4 −50 30 
L cerebellum 2236 4.7722 −4 −56 0 
L inferior parietal lobule, BA 40 1074 4.4943 −28 −38 51 
 
No Feedback > Positive Feedback 
(None) 
RegionCluster Size (mm3)Max t StatTalairach Coordinates (Peak)
Negative Feedback > No Feedback 
R inferior frontal gyrus, BA 47 1209 4.908 26 13 −15 
L medial frontal gyrus, BA 6 5025 5.403 −7 1 51 
Bilateral brainstem, midbrain, red nucleus 4384 6.9973 2 −23 −3 
L ventral striatum/caudate head 1209 5.0343 −7 13 0 
L parietal lobe, postcentral gyrus, BA 3 1814 4.97 −31 −32 57 
 
No Feedback > Negative Feedback 
Bilateral occipital lobe/cerebellum 28524 6.996 −19 −86 −24 
 
Positive Feedback > Negative Feedback 
L cerebellum 48502 7.8136 −31 −77 −24 
Bilateral occipital lobe, cuneus, BA 19 1185 4.8048 29 −80 24 
R parietal lobe, precuneus, BA 31 1750 4.7449 2 −53 30 
 
Negative Feedback > Positive Feedback 
(None) 
 
Positive Feedback > No Feedback 
R parietal lobe, precuneus, BA 31 2267 4.2914 20 −74 24 
Bilateral ventral striatum/caudate head 4403 4.7327 −13 7 3 
Bilateral medial frontal gyrus, BA 10 5170 6.0402 −4 61 15 
Bilateral parietal lobe, precuneus, BA 31 6234 6.558 −4 −50 30 
L cerebellum 2236 4.7722 −4 −56 0 
L inferior parietal lobule, BA 40 1074 4.4943 −28 −38 51 
 
No Feedback > Positive Feedback 
(None) 

Activation table for pairwise contrasts between feedback types in the mixed feedback condition.

Figure 9. 

Axial view of striatal activation in negative feedback > no feedback contrast in the mixed feedback condition (p < .005, cluster threshold corrected).

Figure 9. 

Axial view of striatal activation in negative feedback > no feedback contrast in the mixed feedback condition (p < .005, cluster threshold corrected).

Whole-brain Analysis: Mixed versus Blocked Feedback

Finally, we compared each of the feedback types between conditions (Table 3). These contrasts did not reveal any striatal activity at our chosen threshold.

Table 3. 

Contrasts between Mixed and Blocked Feedback Conditions

RegionCluster Size (mm3)Max t StatTalairach Coordinates (Peak)
Positive Feedback Mixed > Positive Feedback Blocked 
R superior temporal gyrus, BA 39 2580 4.475 44 −50 27 
R occipital lobe, cuneus, BA 18 1267 4.182 5 −74 15 
R occipital lobe, precuneus, BA 31 807 4.849 5 −62 27 
Bilateral occipital lobe, cuneus, BA 19 773 4.574 −1 −83 36 
L middle temporal gyrus, BA 39 1031 4.416 −37 −65 27 
 
Positive Feedback Blocked > Positive Feedback Mixed 
(None) 
 
Negative Feedback Mixed > Negative Feedback Blocked 
L parahippocampal gyrus, BA 34 688 5.8485 −10 −2 −21 
 
Negative Feedback Blocked > Negative Feedback Mixed 
(None) 
 
No Feedback Mixed > No Feedback Blocked 
R superior temporal gyrus, BA 39 4539 5.1139 44 −56 30 
R superior frontal gyrus, BA 10 4785 5.342 26 58 24 
R middle frontal gyrus, BA 9 2253 5.604 35 16 30 
R medial frontal gyrus, BA 8 1195 5.129 5 49 42 
L cerebellum 1782 4.937 −13 −71 −33 
 
No Feedback Blocked > No Feedback Mixed 
(None) 
RegionCluster Size (mm3)Max t StatTalairach Coordinates (Peak)
Positive Feedback Mixed > Positive Feedback Blocked 
R superior temporal gyrus, BA 39 2580 4.475 44 −50 27 
R occipital lobe, cuneus, BA 18 1267 4.182 5 −74 15 
R occipital lobe, precuneus, BA 31 807 4.849 5 −62 27 
Bilateral occipital lobe, cuneus, BA 19 773 4.574 −1 −83 36 
L middle temporal gyrus, BA 39 1031 4.416 −37 −65 27 
 
Positive Feedback Blocked > Positive Feedback Mixed 
(None) 
 
Negative Feedback Mixed > Negative Feedback Blocked 
L parahippocampal gyrus, BA 34 688 5.8485 −10 −2 −21 
 
Negative Feedback Blocked > Negative Feedback Mixed 
(None) 
 
No Feedback Mixed > No Feedback Blocked 
R superior temporal gyrus, BA 39 4539 5.1139 44 −56 30 
R superior frontal gyrus, BA 10 4785 5.342 26 58 24 
R middle frontal gyrus, BA 9 2253 5.604 35 16 30 
R medial frontal gyrus, BA 8 1195 5.129 5 49 42 
L cerebellum 1782 4.937 −13 −71 −33 
 
No Feedback Blocked > No Feedback Mixed 
(None) 

Activation table for pairwise contrasts between conditions for the same feedback type.

DISCUSSION

Although positive feedback is both informative and rewarding, negative feedback is informative, but punishing. In the current study, we examined whether unpredictable presentation of feedback (“mixed feedback” condition) would highlight the informational value of negative feedback, leading to a change in the striatal response to negative feedback. Consistent with our hypothesis, negative feedback in the mixed feedback condition elicited significantly more striatal activity compared to negative feedback in the blocked feedback condition, in both ventral striatum and caudate nucleus ROIs. Moreover, the response to no feedback receipt showed the opposite pattern; activity in both striatal subregions increased when feedback was omitted in the blocked feedback condition compared to when it was omitted in the mixed feedback condition.

On a whole-brain level, positive feedback elicited significantly more striatal activity than negative feedback in the blocked feedback condition, but this was not the case in the mixed feedback condition. We also identified a region in the ventral caudate that showed greater activation in response to negative feedback relative to no feedback in the mixed feedback condition, but we did not see this difference when feedback was predictable. Finally, positive feedback elicited significantly more striatal activity than no feedback in both conditions.

Our manipulation altered the striatal response to both negative feedback and no feedback, albeit somewhat differently in the caudate and the ventral striatum. The caudate nucleus processed negative feedback more similarly to positive feedback when feedback presentation was unpredictable. That is, when the informational aspect of negative feedback was emphasized, in a context in which feedback may not be received at all, caudate activity in response to negative feedback increased. This is consistent with previous research (Tricomi & Fiez, 2012; Bischoff-Grethe et al., 2009). In contrast, in the ventral striatum, negative feedback did not elicit significantly more activity than no feedback in the mixed feedback condition. Instead, there was a dip below baseline in response to no feedback when feedback presentation was intermittent. This pattern is consistent with a punishment response to no feedback in this condition (Tricomi & Fiez, 2008; Tricomi et al., 2004; Delgado et al., 2000). The fact that the “no feedback” signal in ventral striatum depends on context has relevance for designing and interpreting studies of performance feedback; the omission of feedback may not be an appropriate baseline under some conditions.

This divergence in the processing of feedback between the caudate and the ventral striatum is compatible with research showing that these two parts of the striatum process different information about outcomes (e.g., Tricomi & Lempert, 2015; O'Doherty et al., 2004). Future work should more systematically vary the affective and informative subcomponents of feedback to investigate the roles of these regions further.

Previous neuroimaging studies have revealed reward prediction error signals in both the ventral striatum and caudate (e.g., O'Doherty et al., 2004; McClure, Berns, & Montague, 2003; O'Doherty, Dayan, Friston, Critchley, & Dolan, 2003), because they are target areas of dopamine neurons. The level of reward prediction error is related both to the value of the outcome and to the extent that an outcome deviates from expectation. In the blocked feedback condition, participants know that they will receive feedback that will give them full information about the correct answer (because even in the case of negative feedback, the correct answer can be determined to be the response that wasn't chosen). In the mixed feedback condition, because there was no a priori information about the type of feedback that would be received on any given trial, deviation from expectation increased, leading to elevated reward prediction error for both positive and negative feedback. Meanwhile, not receiving feedback in the mixed feedback phase may have been perceived as “worse than expected,” because there had been a chance that information would be provided. Our design was not optimized for calculating prediction error for several reasons: Feedback was deterministic, each trial was unique, and the assignment of value to the different outcomes might vary between conditions or between subjects. Nevertheless, our results are consistent with the idea that increased prediction error would boost striatal activity associated with feedback events, even if the feedback signals inaccuracy.

Another possible contributing factor to our results is that the resolution of uncertainty is intrinsically rewarding. This preference has also been shown to involve the dopaminergic system (Bromberg-Martin & Hikosaka, 2009; Eliaz & Schotter, 2007; Ahlbrecht & Weber, 1996). Whether or not having information leads to a reward, simply knowing whether or not one is correct might be valuable because it reduces uncertainty. In the mixed feedback phase, uncertainty is increased because it has two components: uncertainty about whether one is correct and uncertainty about whether one will find out if one is correct. Thus, the receipt of feedback in this condition leads to an increased reduction in uncertainty. It has been shown that dopamine neurons increase their firing rate in response to non-instrumental information (Bromberg-Martin & Hikosaka, 2009), consistent with a resolution-of-uncertainty account. Therefore, although many computational models of dopamine function (e.g., Schultz et al., 1997) do not predict that information would be rewarding unless it leads to increased reward at a later time, it is important to consider that there is an inherent preference for certainty that might emerge through dopaminergic mechanisms.

Despite the difference in the neural processing of negative feedback between the mixed and blocked feedback conditions, there were no significant differences in learning from negative feedback between these conditions. That is, accuracy at posttest did not differ between trials that had received negative feedback in the mixed feedback condition and those that had received negative feedback in the blocked feedback condition. We did, however, observe that participants endorsed more confidence on words that they had previously seen in the mixed feedback context relative to those that they saw in the blocked feedback context. This finding was not restricted to trials for which feedback was actually received previously, suggesting that it may reflect increased attention to trials at times when feedback delivery was intermittent. How the differential neural processing of feedback leads to increased confidence in the absence of increased accuracy is an open question.

In summary, our results build on previous research by showing that a simple contextual manipulation (i.e., changing the predictability of feedback receipt) can change the striatal response to negative feedback, as well as to the omission of feedback. An unpredictable feedback context served to stress the informational aspect of negative feedback, rather than the affective aspect. This interpretation of negative feedback led to an increase in caudate activation, which may facilitate learning. In the ventral striatum, the omission of feedback led to a dip in activation, only in the mixed feedback condition. These results contribute to our understanding of the role of the striatum in feedback processing. They also have implications for education. If we emphasize the informational value of performance feedback, students may become more appreciative of feedback, become more confident in their responses, and be more likely to interpret negative feedback as a learning opportunity, rather than as a punishment.

Acknowledgments

This work was supported by grants from the National Institute on Drug Abuse (R15DA029544) and the National Science Foundation (BCS1150708) awarded to E. T. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institute on Drug Abuse, the National Institutes of Health, or the National Science Foundation.

Reprint requests should be sent to Elizabeth Tricomi, Department of Psychology, Rutgers University, 353 Smith Hall, 101 Warren Street, Newark, NJ 07102, or via e-mail: etricomi@scarletmail.rutgers.edu.

REFERENCES

Ahlbrecht
,
M.
, &
Weber
,
M.
(
1996
).
The resolution of uncertainty: An experimental study
.
Journal of Institutional and Theoretical Economics
,
152
,
593
607
.
Bartra
,
O.
,
McGuire
,
J. T.
, &
Kable
,
J. W.
(
2013
).
The valuation system: A coordinate-based meta-analysis of BOLD fMRI experiments examining neural correlates of subjective value
.
Neuroimage
,
76
,
412
427
.
Bayer
,
H. M.
, &
Glimcher
,
P. W.
(
2005
).
Midbrain dopamine neurons encode a quantitative reward prediction error signal
.
Neuron
,
47
,
129
141
.
Bischoff-Grethe
,
A.
,
Hazeltine
,
E.
,
Bergren
,
L.
,
Ivry
,
R. B.
, &
Grafton
,
S. T.
(
2009
).
The influence of feedback valence in associative learning
.
Neuroimage
,
44
,
243
251
.
Breiter
,
H. C.
,
Aharon
,
I.
,
Kahneman
,
D.
,
Dale
,
A.
, &
Shizgal
,
P.
(
2001
).
Functional imaging of neural responses to expectancy and experience of monetary gains and losses
.
Neuron
,
30
,
619
639
.
Bromberg-Martin
,
E. S.
, &
Hikosaka
,
O.
(
2009
).
Midbrain dopamine neurons signal preference for advance information about upcoming rewards
.
Neuron
,
63
,
119
126
.
Clithero
,
J. A.
, &
Rangel
,
A.
(
2014
).
Informatic parcellation of the network involved in the computation of subjective value
.
Social Cognitive and Affective Neuroscience
,
9
,
1289
1302
.
Coltheart
,
M.
(
1981
).
The MRC psycholinguistic database
.
Quarterly Journal of Experimental Psychology
,
33A
,
497
505
.
Daniel
,
R.
, &
Pollmann
,
S.
(
2014
).
A universal role of the ventral striatum in reward-based learning: Evidence from human studies
.
Neurobiology of Learning and Memory
,
114
,
90
100
.
Deichmann
,
R.
,
Gottfried
,
J. A.
,
Hutton
,
C.
, &
Turner
,
R.
(
2003
).
Optimized EPI for fMRI studies of the orbitofrontal cortex
.
Neuroimage
,
19
,
430
441
.
Delgado
,
M. R.
,
Locke
,
H. M.
,
Stenger
,
V. A.
, &
Fiez
,
J. A.
(
2003
).
Dorsal striatum responses to reward and punishment: Effects of valence and magnitude manipulations
.
Cognitive, Affective, & Behavioral Neuroscience
,
3
,
27
38
.
Delgado
,
M. R.
,
Nystrom
,
L. E.
,
Fissell
,
C.
,
Noll
,
D. C.
, &
Fiez
,
J. A.
(
2000
).
Tracking the hemodynamic responses to reward and punishment in the striatum
.
Journal of Neurophysiology
,
84
,
3072
3077
.
DePasque Swanson
,
S.
, &
Tricomi
,
E.
(
2014
).
Goals and task difficulty expectations modulate striatal responses to feedback
.
Cognitive, Affective & Behavioral Neuroscience
,
14
,
610
620
.
Dobryakova
,
E.
, &
Tricomi
,
E.
(
2013
).
Basal ganglia engagement during feedback processing after a substantial delay
.
Cognitive, Affective & Behavioral Neuroscience
,
13
,
725
736
.
Eliaz
,
K.
, &
Schotter
,
A.
(
2007
).
Experimental testing of intrinsic preferences for noninstrumental information
.
American Economic Review
,
97
,
166
169
.
Elliott
,
R.
,
Frith
,
C. D.
, &
Dolan
,
R. J.
(
1997
).
Differential neural response to positive and negative feedback in planning and guessing tasks
.
Neuropsychologia
,
35
,
1395
1404
.
Han
,
S.
,
Huettel
,
S. A.
,
Raposo
,
A.
,
Adcock
,
R. A.
, &
Dobbins
,
I. G.
(
2010
).
Functional significance of striatal responses during episodic decisions: Recovery or goal attainment?
Journal of Neuroscience
,
30
,
4767
4775
.
Kable
,
J. W.
, &
Glimcher
,
P. W.
(
2007
).
The neural correlates of subjective value during intertemporal choice
.
Nature Neuroscience
,
10
,
1625
1633
.
Landauer
,
T. K.
,
Foltz
,
P. W.
, &
Laham
,
D.
(
1998
).
Introduction to latent semantic analysis
.
Discourse Processes
,
25
,
259
284
.
Liljeholm
,
M.
, &
O'Doherty
,
J. P.
(
2012
).
Contributions of the striatum to learning, motivation, and performance: An associative account
.
Trends in Cognitive Sciences
,
16
,
467
475
.
Lohrenz
,
T.
,
McCabe
,
K.
,
Camerer
,
C. F.
, &
Montague
,
P. R.
(
2007
).
Neural signature of fictive learning signals in a sequential investment task
.
Proceedings of the National Academy of Sciences, U.S.A.
,
104
,
9493
9498
.
Marco-Pallares
,
J.
,
Müller
,
S. V.
, &
Münte
,
T. F.
(
2007
).
Learning by doing: An fMRI study of feedback-related brain activations
.
NeuroReport
,
18
,
1423
1426
.
Matsumoto
,
M.
, &
Hikosaka
,
O.
(
2009
).
Two types of dopamine neuron distinctly convey positive and negative motivational signals
.
Nature
,
459
,
837
841
.
McClure
,
S. M.
,
Berns
,
G. S.
, &
Montague
,
P. R.
(
2003
).
Temporal prediction errors in a passive learning task activate human striatum
.
Neuron
,
38
,
339
346
.
Mullette-Gillman
,
O. A.
,
Detwiler
,
J. M.
,
Winecoff
,
A.
,
Dobbins
,
I.
, &
Huettel
,
S. A.
(
2011
).
Infrequent, task-irrelevant monetary gains and losses engage dorsolateral and ventrolateral prefrontal cortex
.
Brain Research
,
1395
,
53
61
.
Nieuwenhuis
,
S.
,
Heslenfeld
,
D. J.
,
Alting von Geusau
,
N. J.
,
Mars
,
R. B.
,
Holroyd
,
C. B.
, &
Yeung
,
N.
(
2005
).
Activity in human reward-sensitive brain areas is strongly context dependent
.
Neuroimage
,
25
,
1302
1309
.
O'Doherty
,
J. P.
,
Dayan
,
P.
,
Friston
,
K.
,
Critchley
,
H.
, &
Dolan
,
R. J.
(
2003
).
Temporal difference models and reward-related learning in the human brain
.
Neuron
,
38
,
329
337
.
O'Doherty
,
J. P.
,
Dayan
,
P.
,
Schultz
,
J.
,
Deichmann
,
R.
,
Friston
,
K.
, &
Dolan
,
R. J.
(
2004
).
Dissociable roles of ventral and dorsal striatum in instrumental conditioning
.
Science
,
304
,
452
454
.
Schultz
,
W.
,
Dayan
,
P.
, &
Montague
,
P. R.
(
1997
).
A neural substrate of prediction and reward
.
Science
,
275
,
1593
1599
.
Schultz
,
W.
, &
Dickinson
,
A.
(
2000
).
Neuronal coding of prediction errors
.
Annual Review of Neuroscience
,
23
,
473
500
.
Seger
,
C. A.
, &
Cincotta
,
C. M.
(
2005
).
The roles of the caudate nucleus in human classification learning
.
Journal of Neuroscience
,
25
,
2941
2951
.
Simon
,
D.
,
Becker
,
M. P.
,
Mothes-Lasch
,
M.
,
Miltner
,
W. H.
, &
Straube
,
T.
(
2014
).
Effects of social context on feedback-related activity in the human ventral striatum
.
Neuroimage
,
99
,
1
6
.
Smith
,
D. V.
, &
Delgado
,
M. R.
(
2015
).
Reward processing
. In
A. W.
Toga
&
R. A.
Poldrack
(Eds.),
Brain mapping: An encyclopedic reference
(pp.
361
366
).
Amsterdam
:
Elsevier
Talairach
,
J.
, &
Tournoux
,
P.
(
1988
).
Co-planar stereotaxic atlas of the human brain: An approach to medical cerebral imaging
.
New York
:
Thieme
.
Tom
,
S. M.
,
Fox
,
C. R.
,
Trepel
,
C.
, &
Poldrack
,
R. A.
(
2007
).
The neural basis of loss aversion in decision-making under risk
.
Science
,
315
,
515
518
.
Tricomi
,
E.
,
Delgado
,
M. R.
,
McCandliss
,
B. D.
,
McClelland
,
J. L.
, &
Fiez
,
J. A.
(
2006
).
Performance feedback drives caudate activation in a phonological learning task
.
Journal of Cognitive Neuroscience
,
18
,
1029
1043
.
Tricomi
,
E.
, &
Fiez
,
J. A.
(
2008
).
Feedback signals in the caudate reflect goal achievement on a declarative memory task
.
Neuroimage
,
41
,
1154
1167
.
Tricomi
,
E.
, &
Fiez
,
J. A.
(
2012
).
Information content and reward processing in the human striatum during performance of a declarative memory task
.
Cognitive, Affective & Behavioral Neuroscience
,
12
,
361
372
.
Tricomi
,
E.
, &
Lempert
,
K. M.
(
2015
).
Value and probability coding in a feedback-based learning task utilizing food rewards
.
Journal of Neurophysiology
,
113
,
4
13
.
Tricomi
,
E. M.
,
Delgado
,
M. R.
, &
Fiez
,
J. A.
(
2004
).
Modulation of caudate activity by action contingency
.
Neuron
,
41
,
281
292
.

Author notes

*

Current affiliation: Department of Psychology, New York University.