BioMed Research International

Volume 2015, Article ID 726798, 11 pages

http://dx.doi.org/10.1155/2015/726798

## The Context Matters: Outcome Probability and Expectation Mismatch Modulate the Feedback Negativity When Self-Evaluation of Response Correctness Is Possible

^{1}Institute of Psychology, University of Kiel, 24118 Kiel, Germany^{2}Institute of Psychology, University of Bonn, 53111 Bonn, Germany

Received 27 August 2015; Revised 7 November 2015; Accepted 8 November 2015

Academic Editor: Antonino Vallesi

Copyright © 2015 Anja Leue et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

#### Abstract

Individuals typically evaluate whether their performance and the obtained feedback match. Previous research has shown that feedback negativity (FN) depends on outcome probability and feedback valence. It is, however, less clear to what extent previous effects of outcome probability on FN depend on self-evaluations of response correctness. Therefore, we investigated the effects of outcome probability on FN amplitude in a simple go/no-go task that allowed for the self-evaluation of response correctness. We also investigated effects of performance incompatibility and feedback valence. In a sample of participants, outcome probability was manipulated by means of precues, feedback valence by means of monetary feedback, and performance incompatibility by means of feedback that induced a match versus mismatch with individuals’ performance. We found that the 100% outcome probability condition induced a more negative FN following no-loss than the 50% outcome probability condition. The FN following loss was more negative in the 50% compared to the 100% outcome probability condition. Performance-incompatible loss resulted in a more negative FN than performance-compatible loss. Our results indicate that the self-evaluation of the correctness of responses should be taken into account when the effects of outcome probability and expectation mismatch on FN are investigated.

#### 1. Introduction

Individuals often use external information in order to adapt their performance and their behavior in social and economic settings [1–3]. To this end, individuals typically evaluate whether their expectations, their performance, and the obtained feedback match. Some of these evaluation processes may occur relatively fast across time. That is why event-related potentials (ERP) of the electroencephalogram (EEG) are promising parameters, because these measurements provide important insights into fast changes of neural activity during feedback-related processes across time [4].

Miltner et al. found that feedback about the accuracy of performance elicits a negative deflection occurring about 250 ms postfeedback at frontocentral electrode sites [5]. This negative deflection has been originally named as error-related negativity to feedback (feedback-ERN or fERN) [5, 6] and later as feedback negativity (FN) because the FN is not restricted to error feedback. In tasks that provided positive feedback following correct responses and negative feedback following incorrect responses, a more positive FN amplitude has been observed for positive feedback, whereas the FN had a negative deflection following negative feedback [7–9]. It has, therefore, been proposed that the FN is driven by the positivity of outcome rather than by its negativity. According to Foti et al., the FN tracks the relative outcome valence in the context of the alternative outcomes, suggesting that FN depends on one’s expectations immediately before the outcome [7].

In their influential work, Holroyd and Coles investigated the response-locked ERN as an internal signal of outcome expectation and the FN as an external signal of outcome expectation in a probabilistic learning task [6]. Their investigation of the ERN and FN within one task demonstrates that these ERPs are related to the same neurological systems. In their probabilistic learning task, six stimuli were associated with different outcome probability and feedback conditions. By performing the task, participants learned the association of stimulus, correct response, and feedback. When the association of stimulus, response, and feedback has been learned, the stimulus becomes a precue for the expected response-feedback association. Holroyd and Coles compared the response-ERN and the FN and demonstrated that in the 50% outcome probability condition the response-ERN was less pronounced (more positive) than the FN [6]. In the 50% condition, correct responses could be followed by reward but also by punishment. Thus, participants could not learn from the obtained feedback so that a self-evaluation of the correctness of their responses was impossible. Consequently, participants could not evaluate whether the feedback matched or mismatched their performance.

In contrast to the 50% outcome probability condition of the probability learning task [6], the correctness of responses could be learned in the 100% outcome probability condition, because the feedback was compatible with the responses. Accordingly, in the 100% outcome probability condition, the response-ERN was more pronounced than the FN. This finding suggests that the response served as an internal signal of outcome expectation when participants had learned the association between stimuli and required responses. The work of Holroyd and Coles [6] provided important insights into the relation between ERN and FN in a probabilistic learning task. Particularly, their model can explain the common neurological basis of these ERPs and the amplitude of ERN and FN. Both ERN and FN can represent a negative reinforcement learning signal that is conveyed to the functioning of the ACC. Nevertheless, several conditions determining FN remain to be investigated in order to improve our understanding of the neural mechanisms of FN.

Knowledge of the correctness of a response was not a priori given in Holroyd and Coles [6] and could be learned in the 100% outcome probability condition, but it could not be learned in the 50% outcome probability condition. Accordingly, low outcome probability was associated with unknown correctness of responses, and high outcome probability was associated with knowledge of the correctness of response. Therefore, the effects of outcome probability and the effect of self-evaluation of the correct responses on the FN could not be disentangled in Holroyd and Coles [6]. It is therefore possible that the effects attributed to outcome probability were not effects of the outcome probability alone but effects of different degrees of knowledge concerning the (self-evaluated) correctness of the responses. In consequence, it is impossible to infer from Holroyd and Coles [6] whether outcome probability (without any effect of outcome probability on knowledge of result) has an effect on FN.

Moreover, Hajcak et al. [10] applied three different precues in order to manipulate the predictability of outcome (named as outcome probability in Holroyd and Coles [6]) in a gambling task. The numbers 1, 2, or 3 were presented on a screen to inform participants how many doors would be associated with a win feedback. Based on this information, participants were asked to guess whether they would win or lose money. Because the number of doors and the doors that were associated with monetary win or loss changed from trial to trial, participants could not have any knowledge on the correctness of their responses and they could not learn an association of correctness of responses and feedback. This might be the reason why precues in Hajcak et al. [10] did not become a salient stimulus of outcome probability so that no significant FN differences were observed between predictable and less predictable outcome.

Briefly, when outcome probability varies in a context that allows for a self-evaluation of correctness of one’s performance, outcome probability may affect feedback processing and FN. One reason for an effect of outcome probability on FN could be that it provides a frame of reference for the evaluation of outcome valence. For example, when two outcomes with a different valence follow a correct response and when the outcome probability is at chance level, individuals will not know whether a more negative outcome or a more positive outcome will follow their correct response. Since two outcomes can follow the correct response when outcome probability is at chance level, the more positive outcome is likely to be evaluated against the more negative outcome. In contrast, when the outcome following a correct response is completely predictable (i.e., 100% outcome probability), only one outcome is possible following correct responses. When no alternative outcomes are possible, no comparative outcome evaluation is possible. Therefore, an outcome probability at chance level may induce another relative outcome valence for correct responses compared to a 100% outcome probability, which implies a complete certitude of the outcome. Such differences of relative outcome valence induced by different outcome probability can occur even when the absolute valence of the outcome is the same in the two outcome probability conditions. However, the certitude that no other outcome is possible in the 100% condition can only occur, when individuals have perfect knowledge of the correctness of response related to a given outcome. Therefore, we aimed at investigating the effect of outcome probability on FN under a condition of constant knowledge of result. One method to investigate the effect of outcome expectations on the FN amplitude is to define outcome probability by means of precues in a simple task that allows for knowledge of the correctness of response.

To sum up, we expect an effect of outcome probability on feedback valence. The FN has been initially suggested to represent an emotional evaluation of feedback valence on a good-bad dimension with a more negative FN amplitude following negative compared to positive feedback [10–12]. Holroyd et al. have demonstrated that feedback valence varies depending on context and that outcome is evaluated in relation to other possible outcomes in a specific context ([13]; see also [14–16]). Accordingly, the aim of our study was to investigate whether the context of possible outcomes is affected by outcome probability when knowledge of correct responses is given.

Beyond outcome probability and feedback valence, we aim at introducing a second concept during feedback processing that we name performance-compatibility. To be more concrete, when the correct response is a priori known and the task is pretty simple, individuals have an internal representation of the correctness of their responses and might establish an expectation of the feedback valence that should derive from their performance. According to the first-indicator hypothesis [17], which has been derived from the reinforcement-learning-theory (RLT [6]) to predict variations of feedback processing depending on an individual’s performance, performance monitoring might induce an “internal signal” of outcome expectation. In simple tasks with predefined correct and incorrect responses that are known to the participants, they are usually able to evaluate whether they responded correctly or incorrectly. When an internal performance-based outcome expectation corresponds to the received feedback, the FN should not indicate an expectation mismatch. In contrast, when individuals responded correctly and obtained negative feedback, internal outcome expectation and received feedback do not match. Similarly, a mismatch between an internal performance-based signal of outcome expectation and external feedback might occur after responding incorrectly and receiving a feedback that does not correspond to the erroneous response (i.e., no-loss or even win). In these examples, external feedback is incompatible with an individual’s internal signal of outcome expectation that is due to performance monitoring in a simple task. Thus, when performance tasks allow for a self-evaluation of the correctness of the responses, self-evaluation of performance might serve as an internal signal of outcome expectation. In the following, an external feedback that is compatible with the performance-based signal of outcome expectation will be named “performance-compatible feedback,” whereas external feedback that is incompatible with the performance-based outcome expectation will be named “performance-incompatible feedback.”

In sum, we aimed at investigating effects of outcome probability (2 levels: 100% versus 50% condition) and feedback valence (2 levels: no-loss versus loss) on the FN. Outcome probability and feedback valence were independently manipulated in a task condition with constant a priori knowledge of the correct response and performance-compatible outcome. More specifically, we expect outcome probability at chance level to establish a more complex context for the evaluation of feedback valence than 100% outcome probability. The 100% outcome probability condition will result in the complete certitude of the resulting outcome/feedback when knowledge of result is available. Accordingly, the relative valence of a no-loss feedback following correct responses will be considered being less negative in the 50% condition when loss feedback is also possible following correct responses compared to the 100% condition, when only no-loss feedback is possible for correct responses. Therefore, the FN is expected to be less pronounced (less negative) for no-loss following correct responses in the 50% condition than for no-loss following correct responses in the 100% condition (hypothesis a). In contrast, incorrect responses are followed by loss in the 100% condition and can be followed by no-loss or loss in the 50% condition. Accordingly, the relative valence of loss should be more negative in the 50% condition (when no-loss may also occur following incorrect responses) than in the 100% condition where no-loss is the only outcome to be expected (hypothesis b). Moreover, we aimed at investigating effects of expectation mismatch on the FN when performance-compatibility (2 levels: performance-compatible versus incompatible outcome) and feedback valence (2 levels: no-loss versus loss) were independently manipulated in a task condition with a priori knowledge of the correct response and an outcome probability at chance level (50%). In the 50% outcome probability condition, performance-compatible no-loss and performance-compatible loss were thought not to induce an expectation mismatch. However, performance-incompatible no-loss should induce a better-than-expected mismatch and performance-incompatible loss should evoke a worse-than-expected mismatch. Accordingly, we expected performance-incompatible feedback to result in a more negative FN than performance-compatible feedback (hypothesis c).

#### 2. Materials and Method

##### 2.1. Sample

A total of students ( male) of the University of Bonn, Germany, participated voluntarily in this study (age: years, , and range: 19–28 years). Participants of this study were selected from a larger project if they had at least 20 artefact-free EEG epochs per task condition (cf. EEG recording and quantification). At the beginning of this study, we obtained a written informed consent from all participants according to the Declaration of Helsinki. The ethical board of the German Foundation of Psychologists provided a positive evaluation of the experimental protocol of the present study. All participants were right-handed according to the handedness inventory of Oldfield [18] and had normal or corrected-to-normal vision.

##### 2.2. Precue Go/No-Go Task

All participants performed a go/no-go task comprising a total of 512 trials. Reinforcement-related versions of this go/no-go task were previously tested [19, 20]. However, the precued version of the present go/no-go task has been newly developed. Go and no-go stimuli were presented with equal frequency (i.e., 256 go stimuli and 256 no-go stimuli). Go and no-go stimuli were white colored geometric forms consisting of a square and a circle, respectively. In the present go/no-go task we used two feedback precues representing outcome probability. One precue (indexed by “#”) signaled an outcome probability of one (100% condition). That is, participants knew that correct responses to go and no-go stimuli always resulted in a no-loss feedback (i.e., 0 Cents). In contrast, when participants responded too slowly to go stimuli (500 to 1,000 ms poststimulus) or incorrectly to no-go stimuli, they always received a monetary loss feedback (i.e., −2 Cents or −4 Cents; Table 1). Feedback was always performance-compatible in the 100% outcome probability condition. For ethical reasons and in order to avoid irritating the participants, no performance-incompatible feedback was introduced in the 100% outcome probability condition (Table 1). Providing performance-incompatible feedback in the 100% outcome probability condition would have meant that participants obtain monetary loss in each case of correct responses and no-loss in each case of incorrect responses (cf. Table 1). In consequence, participants would have often been punished although they responded in accordance with the instruction. In order to avoid these problems, performance-incompatible feedback was exclusively provided in trials with an outcome probability at chance level (50% condition indexed by “?” as a precue).