- Email: [email protected]
The influence of context condition on outcome evaluation in experimental conditions: Even vs. neutral
International Journal of Psychophysiology 141 (2019) 28–36
Contents lists available at ScienceDirect
International Journal of Psychophysiology journal homepage: www.elsevier.com/locate/ijpsycho
The influence of context condition on outcome evaluation in experimental conditions: Even vs. neutral Shuqing Zhu, Yuyang Wang, Shuqian Gao, Shiwei Jia
T
⁎
School of Psychology, Shandong Normal University, Jinan, China
ARTICLE INFO
ABSTRACT
Keywords: Feedback-related negativity (FRN) Outcome evaluation Context condition Context dependent effect Monetary incentive delay (MID) task
Feedback-related negativity (FRN) is an event-related brain potential that is elicited during outcome evaluation. Studies have found that FRN reflects a good vs. bad classification; more importantly, FRN reflects this classification in a context-dependent manner, which means that the outcome evaluation is obviously influenced by its embedded context. In the current study, we examined how the context conditions of even (i.e., the feedback was +4 or −4) and neutral (i.e., the feedback was always 0) affect the outcome evaluation in experimental conditions (gain and loss). The experimental conditions of gain (i.e., the feedback was +4 or 0) and loss (i.e., the feedback was 0 or −4) were presented with the even condition as the context in one section and with the neutral condition as the context in another section. The ERP (event-related potential) results of the two experimental conditions showed that the unfavorable outcome evoked more negative FRN than the favorable outcome in both even and neutral sections, however, the amplitude difference between unfavorable and favorable outcomes was greater in neutral section than in even section. Furthermore, the favorable outcomes evoked more positive FRN in the neutral section than in the even section. These results indicate that the context condition modulates outcome evaluation, in neutral context, the discrimination between favorable and unfavorable outcome is better, which might due to the facilitated identification of favorable outcomes in neutral context.
1. Introduction The evaluation of the consequences of our previous choices plays a key role in later decision making, which is also fundamental to adaptive behavior. Thus, an increasing amount of research has shed light upon the neural mechanisms of feedback evaluation, commonly employing the event-related potential (ERP) technique. Feedback-related negativity (FRN) is a fronto-centrally distributed negative-going ERP component that peaks at approximately 200–300 ms following external feedback presentation (Miltner et al., 1997; Gehring and Willoughby, 2002; Hajcak et al., 2006; Holroyd and Coles, 2002; Walsh and Anderson, 2012; Proudfit, 2015; Sambrook and Goslin, 2015). FRN is more negative to unfavorable outcomes, such as incorrect performance (Holroyd and Krigolson, 2007; Miltner et al., 1997) or monetary loss (Gehring and Willoughby, 2002; Hajcak et al., 2005; Yeung and Sanfey, 2004), than to favorable outcomes, such as correct performance or monetary gain. Besides, some scholars have also claimed that FRN is sensitive to violations of outcome expectations rather than outcome valence (Alexander and Brown, 2011; Oliveira et al., 2007; Pfabigan et al., 2011). According to the reinforcement learning theory (RL-Theory; ⁎
Holroyd and Coles, 2002; Nieuwenhuis et al., 2004), FRN reflects a reward prediction error signal that originates from the mid-brain dopaminergic system and is further projected to the anterior cingulate cortex (ACC). When outcomes are worse than expected, a phasic decrease in the dopaminergic system and an activity increase in the ACC will be induced; then, more negative FRN will be evoked (Holroyd and Coles, 2002; Sambrook and Goslin, 2015). It has thereafter repeatedly been reported that outcome processing contains a good vs. bad evaluation, which is demonstrated by FRN amplitude variations (Gehring and Willoughby, 2002; Hajcak et al., 2006; Yeung and Sanfey, 2004; for reviews see Nieuwenhuis et al., 2004; Walsh and Anderson, 2012; Proudfit, 2015; Sambrook and Goslin, 2015). In addition, according to the RL theory, the FRN amplitude could be modulated by the outcome prediction (Hajcak et al., 2007; Holroyd et al., 2003), for instance, the unexpected negative feedback evokes more negative FRN than the expected negative feedback does. Recently, researchers proposed the predicted response-outcome (PRO) model, which considers that FRN is evoked by the violation of outcome prediction, regardless of whether the outcome is positive or negative (Alexander and Brown, 2011; Oliveira et al., 2007). One unresolved question in terms of outcome evaluation revolves
Corresponding author at: School of Psychology, Shandong Normal University, No. 88, East Wenhua Rd., Lixia, Jinan 250014, China. E-mail address: [email protected] (S. Jia).
https://doi.org/10.1016/j.ijpsycho.2019.05.001 Received 4 September 2018; Received in revised form 14 April 2019; Accepted 5 May 2019 Available online 06 May 2019 0167-8760/ © 2019 Elsevier B.V. All rights reserved.
International Journal of Psychophysiology 141 (2019) 28–36
S. Zhu, et al.
around how the outcomes are evaluated and what factors affect outcome evaluation. Research reveals that the FRN amplitudes are not only influenced by the properties of the feedback stimulus itself, such as the size (Pfabigan et al., 2015a) and the similarity between favorable and unfavorable outcomes (Liu et al., 2014), but also modulated by its embedded context (Goyer et al., 2008; Holroyd et al., 2004; Osinsky et al., 2012; Pfabigan et al., 2015b). Previous studies have revealed many kinds of contexts, such as the outcome range under a specific condition (Holroyd et al., 2004; Pfabigan et al., 2015b), the overall probability of favorable and unfavorable outcomes (Holroyd et al., 2003; Hajcak et al., 2007), and the outcome sequence order (Goyer et al., 2008; Osinsky et al., 2012), can modulate the FRN amplitudes. How do the embedded contexts affect outcome evaluation, even though they do not convey outcome valence? According to the aforementioned theories, feedback evaluation reflects an encoding of the reward prediction error (Holroyd and Coles, 2002) or the prediction violation (Alexander and Brown, 2011). These theories mean that, before the outcome onset, the brain should generate the outcome prediction first (Bismark et al., 2013). The prediction generation is based not only on trial and error leaning but also on the embedded context. To explain how context information affects outcome processing, Osinsky et al. (2014) assumed a reference system that combines feedback and environmental information continuously to generate predictions for the next trial. In the current study, we aimed to explore whether and how context conditions, work as the feedback evaluation environment, would affect outcome evaluation in the experimental conditions (gain and loss). There were two sections in this study: one section with the even condition as context condition and another section with the neutral condition as context condition. The task manipulations were adopted from past studies (Angus et al., 2017; Holroyd et al., 2004; Pfabigan et al., 2015b). In each trial, participants finished a time estimation task and then received feedback for the estimation. At the beginning of each trial, an incentive cue indicating the possible outcomes in this trial was given first. The cue included four kinds: gain, loss, even and neutral. Under the gain (possible outcomes: +4 or 0), the loss (possible outcomes: 0 or −4), and the even (possible outcomes: +4 or −4) conditions, a favorable outcome and an unfavorable outcome would be given according to participants' performances (i.e., the favorable and unfavorable outcomes were given to correct and incorrect estimations, respectively); under the neutral condition, the outcomes were always 0, regardless of whether the time estimation was accurate. In this study, the gain and loss conditions were experimental conditions, and the even and neutral conditions were employed as context conditions. The whole experiment included two sections: in one section, the experimental conditions of gain and loss and the context condition of even were presented trial by trial in random; in another section, the same two experimental conditions (gain and loss) were presented with the neutral condition as the context condition. For data analysis, we focused on the two experimental conditions only. It is important to explore the effects of even and neutral conditions on feedback processing in other conditions, because many studies have adopted neutral (Angus et al., 2017; Broyd et al., 2012; Knutson and Greer, 2008; Novak and Foti, 2015; Pfabigan et al., 2015b; Santesso et al., 2012) or even (Holroyd et al., 2004) conditions as experimental conditions or as baseline conditions, and the results in other conditions (usually the gain and the loss conditions) were quite different. In one study (Holroyd et al., 2004) that explored the context-dependent effect of FRN, the even, gain, and loss conditions were presented block by block. The results demonstrated that FRN reflects a relatively good vs. bad evaluation according to the cued context in both the gain and loss conditions. With the monetary incentive delay (MID) task, one influential task for exploring the neural mechanisms of monetary
anticipation and consumption (Angus et al., 2017; Broyd et al., 2012; Knutson and Greer, 2008; Novak and Foti, 2015; Pfabigan et al., 2015b; Santesso et al., 2012), the gain, loss, and neutral conditions were presented trial by trial. However, the results in those studies were inconsistent with each other, and most of them were different from the results of Holroyd et al. (2004). For instance, with the MID task, Pfabigan et al. (2015b) found that in the gain condition, 0 evoked a larger FRN than +4; in the loss condition, 0 evoked a larger FRN than −4. In the gain condition, 0 was the relatively unfavorable outcome; on the contrary, in the loss condition, 0 was the relatively favorable outcome. Thus, the authors explained that FRN was not sensitive to feedback valence but was sensitive to expectation deviation. The studies of Santesso et al. (2012) and Novak and Foti (2015) found that the unfavorable outcomes evoked larger FRN than the favorable outcomes in the gain condition, while FRN for the favorable and unfavorable outcomes was not different in the loss condition. However, Angus et al. (2017) found that in both gain and loss conditions, the Reward positivity (RewP, Proudfit, 2015) for the favorable outcomes was more positive than that for the unfavorable outcomes; these results were in line with those of Holroyd et al. (2004). Thus, for the studies that used the MID task (Angus et al., 2017; Novak and Foti, 2015; Pfabigan et al., 2015b; Santesso et al., 2012), the potentials evoked by feedback processing showed inconsistent results, which needs further study for confirmation. Additionally, between the studies that designed the even condition (Holroyd et al., 2004) and other studies that used the neutral condition (Novak and Foti, 2015; Pfabigan et al., 2015b; Santesso et al., 2012), the results in the gain and loss conditions were quite different. The manipulations in the gain and loss conditions were similar in those studies (Holroyd et al., 2004; Novak and Foti, 2015; Pfabigan et al., 2015b; Santesso et al., 2012), such that the context condition (even vs. neutral) might be one critical factor that led to results discrepancies in the gain and loss conditions. Among these studies that used the MID task, the study of Angus et al. (2017) obtained quite similar results to the study of Holroyd et al. (2004); however, they used principal component analysis (PCA) and analyzed the principle component of RewP. In the current study, we replicated the manipulations of Angus et al. (2017) in one section (i.e., the neutral section) and analyzed the data with the grand average method to further test the FRN effect in the MID task. We also conducted one even section with even condition as the context condition to replicate the results of Holroyd et al. (2004) and also for a direct comparison with the neutral section. For outcome evaluation in experimental conditions, we assumed that being embedded in the neutral context should be different to being embedded in the even context. One possibility was that, in the neutral condition, the outcomes were always 0, irrespective of the performance accuracy. The outcome was meaningless for behavioral adjustment, and it weakened participants' motivation in the neutral section (Holroyd and Yeung, 2012). Thus, we predicted that the FRN amplitudes would decrease for the experimental conditions in the neutral context condition. Another possibility was that, in the neutral context, the frequencies of +4 and −4 outcomes were less than that in the even context. Less frequency caused unprediction of the +4 and −4 outcomes, which would evoke more negative FRN in the neutral context (Hajcak et al., 2007; Holroyd et al., 2003; Talmi et al., 2013). Additionally, for the good vs. bad classification, the prediction could not be made clearly. According to the study of Angus et al. (2017), we predicted that the unfavorable outcome would evoke larger FRN than the favorable outcome in both the even and neutral sections. However, according to other studies (Novak and Foti, 2015; Pfabigan et al., 2015b; Santesso et al., 2012), the results, especially those in the loss condition, were difficult to predict; thus, the current study will examine these results.
29
International Journal of Psychophysiology 141 (2019) 28–36
S. Zhu, et al.
Fig. 1. The timeline of a trial. In each trial, the cue (gain, loss, even, or neutral) that indicated the possible outcomes in this trial was presented first. Then, participants finished a time estimation task. Finally, participants received feedback.
2. Methods
(+ 4 ¥; even-favorable) and a loss of 4 ¥ (−4 ¥; even-unfavorable), respectively. In the neutral condition, feedback was always 0 ¥, irrespective of the accuracy of the time estimation. The participants were initially endowed 30 ¥ and were given a bonus according to their performance. The accuracy was determined by a tolerant time window. To ensure that participants' accuracies were approximately 50%, we used a sliding criterion to decide the tolerant time window (Mars et al., 2004; Miltner et al., 1997). The time window was initialized at 1000 ± 120 ms, which required participants to respond between 880 and 1120 ms following the presentation of the gold star in order to receive correct feedback on the first trial. The time window was shortened by 20 ms (subtracted 10 ms from the upper time limit and added 10 ms to the lower time limit) after a correct response and was lengthened by 20 ms after too fast or too slow of a response. The same adaptive algorithm was used continuously in the whole experiment, irrespective of the presented incentive cue. As Table 1 shows, the experiment had two sections: one section contained a gain condition, a loss condition, and an even condition; another section contained a gain condition, a loss condition, and a neutral condition. The gain and loss conditions were experimental conditions, and the even and neutral conditions were context conditions. For data analysis, this study focused on the experimental conditions only, which were the gain and loss conditions in the even section (even-gain and even-loss) and the gain and loss conditions in the neutral section (neutral-gain and neutral-loss). In addition, there were two possible outcomes (favorable and unfavorable) under each condition, which yielded 8 conditions in total. Before the experiment, the participants learned the main structure of the task through the instruction. In the even section, participants first conducted 12 practice trials, including 4 gain trials, 4 loss trials, and 4 even trials. Then, participants completed 80 gain trials, 80 loss trials, and 80 even trials. A total of 240 trials were presented randomly, resulting
2.1. Participants Twenty-one1 volunteers (7 males; 22.14 ± 2.54 years old) participated in the experiment. All participants were right-handed, had normal or corrected-to-normal vision and reported no prior or current psychiatric disorders. Written informed consent was obtained prior to the experiment. After the experiment, 30 ¥ and a bonus were given to each participant for their participation. The study was approved by the ethics committee of Shandong Normal University. 2.2. Apparatus and procedure Participants sat comfortably about 1 m in front of a computer monitor in an electromagnetically shielded room and performed a cued time estimation task. As Fig. 1 shows, each trial started with a centrally presented asterisk as the fixation point (800 ms). Prior to the time estimation, a cue that indicated the condition of the current trial was presented for 1000 ms. After the cue, a blank screen was presented for 1000–1500 ms at random. During this blank screen interval, participants were required to rehearse the cue in their mind and to prepare their motor response. Then, the target stimulus of a star (visual angle of 2.3°, gold color) was presented, and participants needed to press the space bar with their right index finger when they believed that 1 s had elapsed. Then, participants received feedback that appeared 1000 ms (visual angle of 2.3°) after a blank screen of 400–600 ms in random. The inter-trial interval (ITI) was 400–600 ms randomly. The task was presented by a CRT monitor (Dell M782) with the refresh rate of 75 Hz. The cue included four kinds: gain (a circle surrounding a “+”), loss (a circle surrounding a “−”), even (an empty circle) and neutral (also an empty circle); all cues had a 2.3° visual angle (Pfabigan et al., 2015b; Angus et al., 2017). In the gain condition, accurate and inaccurate responses resulted in gain of 4¥ (+ 4 ¥; gain-favorable) and gain omission (0 ¥; gain-unfavorable), respectively. In the loss condition, accurate and inaccurate responses resulted in loss avoidance (0 ¥; loss-favorable) and loss of 4 ¥ (−4 ¥; loss-unfavorable), respectively. In the even condition, accurate and inaccurate responses resulted in a gain of 4 ¥
Table 1 The conditions and the corresponding stimuli of the study. Even section Cue
1
The sample size was calculated by the G*Power 3.1 (Faul et al., 2007) with large effect size (f = 0.4) according to past studies (Gu et al., 2011; Meadows et al., 2016). We set an alpha of 0.05 and a correlation among repeated measures of 0.5. The 2 (context condition: even, neutral) × 2 (experimental condition: gain, loss) × 2 (valence: favorable, unfavorable) within-subject design determined that the number of group was 1 and the number of measurement was 8. Finally, the estimated sample size was 17. Also referring to the sample size of previous studies (Angus et al., 2017; Holroyd et al., 2004; Pfabigan et al., 2015b), 21 was the intended sample size.
Outcome
Favorable Unfavorable
Neutral section
even
gain
loss
+4 ¥ −4 ¥
+4 ¥ 0¥
0¥ −4 ¥
neutral 0¥ 0¥
gain
loss
+4 ¥ 0¥
0¥ −4 ¥
Note: the favorable outcome means the outcome for a correct response in a specific condition; the unfavorable outcome means the outcome for an incorrect response in a specific condition. In the even section, the gain and loss conditions were the experimental conditions, and the even condition was the context condition; in the neutral section, the same two experimental conditions were presented, with the neural condition as the context condition. 30
International Journal of Psychophysiology 141 (2019) 28–36
S. Zhu, et al.
apparent, the amplitude was set to 0 μV, which occurred in 2.98% of all cases, mostly after favorable feedback (Pfabigan et al., 2011). Finally, the averaged amplitude across the three electrodes was used as the amplitude of FRN. In addition, we measured the FRN amplitude with the formula of FRN-(P2 + P3)/2 from the Fz, FCz, and Cz, and for this measurement the P3 was defined as the most positive point 300–400 ms after feedback onset. Also, the averaged amplitude across the three electrodes was used as the FRN amplitude. By this measurement, on the one hand, we could remove the possible temporal overlap of the FRN with both P2 and P3; on the other hand, with multiple scoring methods we wanted to get more reliable results. Three-way ANOVAs were conducted on the FRN amplitude, with the factors of context condition 2 (even, neutral), experimental condition 2 (gain, loss), and outcome valence 2 (favorable, unfavorable).
in approximately 40 trials for each condition of gain-favorable, gainunfavorable, loss-favorable, and loss-unfavorable. Short breaks (5 min) were provided every 40 trials, and long breaks (10–15 min) were provided every 80 trials. The arrangements of the neutral section were similar to those of the even section. The order of the two sections was counterbalanced across participants. The whole experiment lasted for approximately 120 min, including the break time for approximately 60 min. The experiment was conducted by using E-Prime 2.0 (Psychology Software Tools, Inc., Sharpsburg, PA). 2.3. EEG acquisition and analyses The electroencephalogram (EEG) was recorded from 64 Ag-AgCl electrodes that were mounted in an elastic cap according to the 10–20 system using the Brain Products system (Brain Products GmbH, Munich, Germany), with a bandpass of 0.0531–80 Hz and a sampling rate of 500 Hz. All electrodes were online referenced to the FCz electrode, with the AFz electrode as the common ground. The electrode impedances were kept below 10 kΩ during recording. The vertical and horizontal electrooculograms (EOGs) were recorded from the outer canthi of the left eye and the inferior orbit of the right eye, respectively. Brain Vision Analyzer 2.0 (Brain Products GmbH, Munich, Germany) was used to conduct the EEG processing. Data were re-referenced to the linked mastoids, and the data from the FCz electrode were reinstated. Then, a 0.1–20 Hz bandpass filter with 24-dB roll-offs was applied. The epoch time window was −200 ms to 800 ms relative to feedback onset, with a −200–0 ms time window as the baseline. Ocular artifacts were corrected with an eye-movement correction algorithm. Trials with EEG voltages exceeding thresholds of ± 80 μV were considered artifacts and were excluded from the analysis. At last, EEGs were averaged for the eight conditions to generate ERPs: evengain-favorable (39.43 ± 5.21 trials survived, 1.78% trials were discarded), even-gain-unfavorable (39.29 ± 4.60 trials survived, 1.77% trials were discarded), even-loss-favorable (39.48 ± 4.64 trials survived, 1.40% trials were discarded), even-loss-unfavorable (39.19 ± 5.58 trials survived, 2.28% trials were discarded), neutralgain-favorable (40.05 ± 4.74 trials survived, 0.89% trials were discarded), neutral-gain-unfavorable (39.29 ± 4.14 trials survived, 0.64% trials were discarded), neutral-loss-favorable (39.00 ± 4.28 trials survived, 1.06% trials were discarded), and neutral-loss-unfavorable (40.29 ± 4.39 trials survived, 0.58% trials were discarded). The ERP amplitudes are susceptible to component overlap, which is often observable in the feedback-related potentials. Therefore, we chose the peak-to-peak measuring method for FRN. The peak-to-peak voltage difference (FRNeP2) between the most negative peak 200–300 ms after feedback onset and the preceding most positive peak 150–250 ms after feedback onset were measured from the Fz, FCz, and Cz electrodes (Holroyd et al., 2003; Pfabigan et al., 2011). If no negative peak was
2.4. Behavioral data analyses To test whether the trial numbers were comparable in all conditions, we compared the number of correct trial and incorrect trial in each experimental condition (gain and loss) with t-tests. Additionally, absolute erroneous amount was calculated using the formula of |estimated time − 1000| for all conditions. To examine the effects of the context condition, experimental condition (gain and loss), and outcome valence of the preceding trial on the absolute erroneous amount of current trial, an ANOVA with the context condition 2 (even, neutral), experimental condition 2 (gain, loss), and outcome valence (the preceding trial) 2 (favorable, unfavorable) as factors was conducted on the absolute erroneous amount. When necessary, the Greenhouse-Geisser correction for repeated measures was applied for all ANOVAs. The SPSS 17.0 was used for all data analyses. 3. Results 3.1. Behavioral results The overall accurate rate was 50.12 ± 0.92. The trial number of the correct and incorrect trials was not different in any of the conditions: for the even-gain condition (favorable trials: 40.05 ± 4.55; unfavorable trials: 39.90 ± 4.56), t(20) = 0.07, p = .943; for the evenloss condition (favorable trials: 40.24 ± 4.36; unfavorable trials: 39.86 ± 4.39), t(20) = −0.38, p = .707; for the neutral-gain condition (favorable trials: 39.57 ± 4.86; unfavorable trials: 40.38 ± 4.88), t(20) = 0.20, p = .844; and for the neutral-loss condition (favorable trials: 39.71 ± 4.50; unfavorable trials: 40.00 ± 4.44), t (20) = −0.15, p = .884). These results suggest that the trial numbers in all conditions were comparable. Fig. 2. The absolute erroneous amount of time evaluation in all conditions. The absolute erroneous amount was calculated with the formula of |estimated time − 1000|. Results showed that participants performed better after favorable outcomes than after unfavorable outcomes.
31
International Journal of Psychophysiology 141 (2019) 28–36
S. Zhu, et al.
Fig. 3. The grand-average FRN waveforms for all conditions at the Fz, FCz, and Cz electrodes.
Fig. 2 shows the absolute erroneous amount of time evaluation in all conditions. This absolute erroneous amount was influenced by outcome valence in the previous trial, F(1,20) = 19.49, p < .001, ηp2 = 0.49. After unfavorable outcomes (161.14 ± 54.81 ms), the absolute erroneous amount was larger than after favorable outcomes (131.44 ± 40.75 ms). The absolute erroneous amount was not affected by the context condition, F(1, 20) = 0.21, p = .654, or the experimental condition (gain and loss), F(1, 20) = 0.33, p = .572. There was no significant interaction, ps. ≥.152. These results indicate that participants performed better after favorable feedback than after unfavorable feedback, which was not modulated by context condition and experimental condition.
FRN amplitudes in each condition with the FRN-P2 measuring method. The ANOVA revealed a significant main effect for the context condition, F(1, 20) = 6.89, p = .016, ηp2 = 0.26. More negative FRN was evoked for the even (−5.31 ± 4.10 μV) context compared with that for the neutral (−4.24 ± 3.51 μV) context. The main effect of outcome valence was significant, F(1,20) = 73.20, p < .001, ηp2 = 0.79, and FRN for the unfavorable outcome (−6.53 ± 3.94 μV) was more negative than that for the favorable outcome (−3.02 ± 2.83 μV). The main effect of the experimental condition (gain and loss) was not significant (gain: −4.93 ± 3.93 μV; loss: −4.61 ± 3.78 μV), F(1,20)=1.09, p = .309. Significant interactions were observed between the context condition and outcome valence, F(1,20) = 4.73, p = .042, ηp2 = 0.19. Simple effect analysis showed that unfavorable outcomes, regardless of whether even or neutral was used as the context condition, elicited more negative FRN than favorable outcomes (ps. < .001). The difference
3.2. ERP results Fig. 3 shows the FRN waveforms in all conditions. Fig. 4 shows the
Fig. 4. The FRN peak-to-peak amplitudes in each condition with the FRN-P2 measuring method. 32
International Journal of Psychophysiology 141 (2019) 28–36
S. Zhu, et al.
Fig. 5. The outcome valance effects in even and neutral contexts. This figure shows the interaction between outcome valence and context condition. The difference waves (a) were calculated by subtracting favorable outcome evoked potential from unfavorable outcome evoked potential for even and neutral contexts, respectively. The topographies (b) show the FRN peaks in the time window of 200–300 ms.
between favorable and unfavorable outcomes tended to be larger for the neutral context (−4.07 ± 3.81 μV) than that for the even context (−2.94 ± 4.25 μV), t(41) = 1.97, p = .055, Cohen's d = 0.30. Furthermore, for favorable outcomes, more negative FRN was observed in the even context than that in the neutral context (p < .001); however, for unfavorable outcomes, FRN in the even context and neutral context was not different (p = .389). The remaining interactions were not significant, all p values ≥ .162. Fig. 5 shows the topographies of outcome valence in even (left of b) and neutral (right of b) contexts, respectively. The difference waves (a) were created by subtracting favorable outcome evoked potential from unfavorable evoked potential for even context and neutral context, respectively. Since the ANOVA showed that the valence effect was not modulated by experimental condition, for the difference waves and the topographies, the gain and loss experimental conditions were collapsed. This figure suggests that the difference wave (valence effect) was larger for neutral context than for the even context, and the FRN mainly distributes on the fronto-central scalp. Fig. 6 shows the results with the FRN-(P2 + P3)/2 measuring method. The ANOVA revealed a significant main effect of outcome valence, F(1,20) = 43.36, p < .001, ηp2 = 0.68. More negative FRN was evoked for unfavorable (−9.22 ± 4.20 μV) compared to favorable (−5.57 ± 3.34 μV) outcomes. The main effect of experimental condition (gain and loss) was significant, F(1,20) = 32.99, p < .001, ηp2 = 0.62, More negative FRN for gain (−8.24 ± 4.02 μV) condition
compared to loss (−6.55 ± 4.24 μV) condition has been found. Significant interaction was observed for context condition and outcome valence, F(1,20) = 5.29, p = .032, ηp2 = 0.21. Simple effect test indicated that unfavorable outcomes, regardless of using even or neutral as context, elicited more negative FRN than favorable outcomes (ps < .001). And the difference between favorable and unfavorable outcomes was larger in neutral context (−4.65 ± 5.41 μV) than that in even context (−2.63 ± 4.33 μV), t(41) = 2.65, p = .011, Cohen's d = 0.41. For favorable outcomes, more negative FRN was observed when even was used as context than when neutral was used as context. (p = .008); but for unfavorable outcomes, the FRN was not different between even as context and neutral as context (p = .470). 4. Discussion In this study, to examine whether and how the context condition would modulate outcome evaluation in other conditions, we presented experimental conditions of gain and loss to participants, with even as the context in one section and with neutral as the context in another section. For the behavioral results, we compared the number of correct and incorrect trials in each condition, and the results showed no difference in all experimental conditions (gain and loss), which guaranteed that the averaging trial number in each condition was comparable. The ANOVA on the absolute erroneous amount of time evaluation showed that after unfavorable feedback, participants performed worse than after
Fig. 6. The FRN base-to-peak amplitudes in each condition with the FRN-(P2 + P3)/2 measuring method. 33
International Journal of Psychophysiology 141 (2019) 28–36
S. Zhu, et al.
favorable feedback, which suggests that participants learned the task through the replication of correct response rather than the correction of the error response. It is possible that the feedback for incorrect response was too simple, without the incorrect direction and magnitude. Future studies should provide detailed outcome information to test how participants used negative feedback to learn the time estimation task (Mars et al., 2004). The analyses of FRN showed that the context condition modulated the FRN amplitudes in the experimental conditions, and FRN was more negative when even was used as the context than when neutral was used as the context when the outcome was favorable. In other words, the favorable outcomes evoked more positive brain potentials (might derive from the reward positivity, RewP) in neutral context than in even context. In addition, the binary classification of favorable and unfavorable outcomes in the experimental conditions was not affected by the context condition, and FRN evoked by unfavorable outcomes was more negative than that evoked by favorable outcomes in both the even and neutral sections, however, the difference between favorable and unfavorable outcomes was larger for the neutral context than that for the even context. Besides, the favorable vs. unfavorable classification was not affected by the experimental condition; under both gain and loss conditions, the unfavorable outcomes evoked more negative FRN than the favorable outcomes.
sections. Therefore, the FRN difference between favorable and unfavorable outcomes was larger in neutral context than in even context. 4.2. Implications for the MID task The MID task is a commonly used task for exploring the underlying mechanisms of reward expectation and consumption. Originally, this task was designed by Knutson et al. (2000) and Knutson and Greer (2008) and was mainly conducted in combination with functional magnetic resonance imaging (fMRI). Recently, some researchers were interested in the brain potentials that are evoked during reward expectation and consumption (Angus et al., 2017; Broyd et al., 2012; Novak and Foti, 2015; Pfabigan et al., 2015b; Santesso et al., 2012). Many components can be evoked by the MID task, such as the N2, contingent negative variation (CNV), and P300 during the cue stage as well as the FRN, RewP, and P300 during the feedback stage. However, the results were inconsistent among these components. By analyzing the feedback evoked component, FRN, the present study retested the evoked components during the consumption stage. In the current study, FRN was sensitive to feedback valence, and the unfavorable outcomes evoked more negative FRN than the favorable outcomes in both the gain and loss conditions. These results are consistent with those of Angus et al. (2017) and are in line with the RL theory, which supports that FRN reflects reward prediction error and thus is sensitive to feedback valence (Holroyd and Coles, 2002). In contrast to some studies (Novak and Foti, 2015; Pfabigan et al., 2015b; Santesso et al., 2012), this result supports the RL theory in both the gain and loss conditions. Pfabigan et al. (2015b) proposed that FRN is evoked by prediction violation and reflects a motivational salient signal. Novak and Foti (2015) and Santesso et al. (2012) found a valence effect of FRN in the gain condition rather than in the loss condition, which suggested that FRN is more sensitive to reward signals and generates reward positivity (Holroyd et al., 2008; Pfabigan et al., 2015b). The above analyses suggests that discrepancies still exist for the reward consumption stage in the MID task, and the same situation is also true for cue evoked components (Novak and Foti, 2015; Angus et al., 2017), which need further studies for clarification. Notably, this study found a valence effect for FRN in both the neutral section and the even section, which suggests that the decreased motivation in the neutral section did not impair the valence effect and thus removes the possible interference from the motivation factor, further confirming the valence effect of FRN in both the gain and loss conditions for the MID task.
4.1. The context condition affects outcome evaluation in experimental conditions In the introduction part, we had given two predictions. The first one was that, since all outcomes were delivered according to participants' performances in the even context, the motivation level would be greater in the even section compared with those in the neutral section (Hajcak et al., 2005, 2006; Holroyd and Yeung, 2012), and we predicted that more negative FRN would be evoked in the even section than in the neutral section. The results did not support this prediction for unfavorable outcomes, such that the motivational hypothesis was not fully verified. Another prediction was that, the +4 and −4 outcomes were less frequent in neutral section, such that the two outcomes were unpredicted and would evoke more negative FRN than that in the even section (Hajcak et al., 2007; Holroyd et al., 2003; Talmi et al., 2013), which prediction was also not verified. For the effect of context condition on experimental conditions (gain and loss), the favorable outcomes evoked more positive brain potentials in neutral context than in even context; the brain potentials that evoked by unfavorable outcomes were not different between in the neutral context and in the even context. These results suggest that the context condition modulates outcome evaluation of other conditions through favorable outcome processing rather than unfavorable processing. In other words, the context condition affects RewP evoked by the positive outcomes rather than FRN evoked by the negative outcome (Proudfit, 2015). Moreover, although the binary categorization was not modulated by the context condition, the discrimination between favorable and unfavorable outcomes was better in neutral section than in even section. In neutral section, since all outcomes were 0 in neutral condition, the +4 in gain condition and the −4 in loss condition were more infrequent and easier to be identified, and thus the classification of favorable and unfavorable outcomes in gain and loss conditions were easier in neutral context, and the amplitude difference between favorable and unfavorable outcomes in neutral section were larger than that in even section. In conclusion, the context of even and neutral condition modulates outcome evaluation in other conditions through valence evaluation, rather than the processes of prediction violation or motivational involvements. The +4 and −4 were less frequent in neutral context, which facilitated the favorable outcome processing and evoked more positive RewP in neutral section than in even section; while the unfavorable outcomes evoked comparable FRN in neutral and even
4.3. The context-dependent effect in the outcome evaluation The context effect in this section is constrained as the effects of the gain and loss cues on the binary evaluation in the experimental conditions (Holroyd et al., 2004; Kujawa et al., 2013), and it does not involve the effect of context conditions on the experimental conditions in the above sections. To explore how our brain classifies favorable and unfavorable outcomes, Holroyd et al. (2004) studied the context-dependent effect of outcome processing. In terms of the context, these authors defined it as the outcome range indicated by introduction or condition cues. In this seminal study, the authors designed the even, gain, and loss conditions, and the results showed that the outcomes were not evaluated as their absolute value but as their relative favorable or unfavorable value compared to other possible outcomes. With the MID paradigm, Angus et al. (2017) presented neutral, gain, and loss conditions, and the results also demonstrated that the outcomes were evaluated as relatively favorable or unfavorable according to possible outcomes. In contrast to those described in Holroyd et al. (2004), the three contexts described in Angus et al. (2017) were presented trial by trial rather than block by block; thus, this study extended the context effect from the global level to the local level. Obviously, the results in the 34
International Journal of Psychophysiology 141 (2019) 28–36
S. Zhu, et al.
current data.
current study also showed the context effect at the local level. Although the context effect of outcome classification has often been reported (Goyer et al., 2008; Holroyd et al., 2004; Kujawa et al., 2013; Pfabigan et al., 2015b), whether this effect can be extended to the local level is still debated. With the same MID task, some studies reported binary evaluation only in the gain condition (Novak and Foti, 2015; Santesso et al., 2012), and Pfabigan et al. (2015b) even found the opposite results in the loss condition. Similar to Holroyd et al. (2004), Kujawa et al. (2013) found that the cue at the local level could not be referred to in the outcome evaluation. Similarly, with the simple gambling task, Osinsky et al. (2014) and Gehring and Willoughby (2002) reported that the local alternative outcomes did not affect the outcome evaluation. Osinsky et al. (2014) claimed that the global context rather than the local context could affect the outcome evaluation. Thus, why did the current study find a context effect at the local level? We infer that the task is an important factor. Both the study of Angus et al. (2017) and the current study used the time estimation task. Compared to the simple gambling task (Kujawa et al., 2013; Osinsky et al., 2014), the time estimation task is learnable, and the outcomes are given according to participants' performances. Participants are more active in completing the task and care more about the outcomes. Thus, the good vs. bad classification could be found even at the local level in our study and in the study of Angus et al. (2017), but it disappeared in the studies of Kujawa et al. (2013) and Osinsky et al. (2014). The same conclusions have also been drawn by Heydari and Holroyd (2016), who found the valence effect over the prediction effect on FRN in an active task; these results failed to be found with the passive task (Talmi et al., 2013).
References Alexander, W.H., Brown, J.W., 2011. Medial prefrontal cortex as an actionoutcome predictor. Nat. Neurosci. 14, 1338–1344. https://doi.org/10.1038/nn.2921. Angus, D.J., Latham, A.J., Harmon-Jones, E., Deliano, M., Balleine, B., Braddon-Mitchell, D., 2017. Electrocortical components of anticipation and consumption in a monetary incentive delay task. Psychophysiology 54 (11), 1686–1705. https://doi.org/10. 1111/psyp.12913. Bismark, A.W., Hajcak, G., Whitworth, N.M., Allen, J.J., 2013. The role of outcome expectations in the generation of the feedback-related negativity. Psychophysiology 50, 125–133. https://doi.org/10.1111/j.1469-8986.2012.01490.x. Broyd, S.J., Richards, H.J., Helps, S.K., Chronaki, G., Bamford, S., Sonuga-Barke, E.J., 2012. An electrophysiological monetary incentive delay (e-MID) task: a way to decompose the different components of neural response to positive and negative monetary reinforcement. J. Neurosci. Methods 209 (1), 40–49. https://doi.org/10. 1016/j.jneumeth.2012.05.015. Faul, F., Erdfelder, E., Lang, A.G., Buchner, A., 2007. G* Power 3: a flexible statistical power analysis program for the social, behavioral, and biomedical sciences. Behav. Res. Methods 39 (2), 175–191. https://doi.org/10.3758/BF03193146. Gehring, W.J., Willoughby, A.R., 2002. The medial frontal cortex and the rapid processing of monetary gains and losses. Science 295, 2279–2282. https://doi.org/10. 1126/science.1066893. Goyer, J.P., Woldorff, M.G., Huettel, S.A., 2008. Rapid electrophysiological brain responses are influenced by both valence and magnitude of monetary rewards. J. Cogn. Neurosci. 20, 2058–2069. https://doi.org/10.1162/jocn.2008.20134. Gu, R., Lei, Z., Broster, L., Wu, T., Jiang, Y., Luo, Y., 2011. Beyond valence and magnitude: a flexible evaluative coding system in the brain. Neuropsychologia 49, 3891–3897. https://doi.org/10.1016/j.neuropsychologia.2011.10.006. Hajcak, G., Holroyd, C.B., Moser, J.S., Simons, R.F., 2005. Brain potentials associated with expected and unexpected favourable and unfavourable outcomes. Psychophysiology 42 (2), 161–170. https://doi.org/10.1111/j.1469-8986.2005. 00278.x. Hajcak, G., Moser, J.S., Holroyd, C.B., Simons, R.F., 2006. The feedback-related negativity reflects the binary evaluation of good versus bad outcomes. Biol. Psychol. 71, 148–154. https://doi.org/10.1016/j.biopsycho.2005.04.001. Hajcak, G., Moser, J.S., Holroyd, C.B., Simons, R.F., 2007. It's worse than you thought: the feedback negativity and violations of reward prediction in gambling tasks. Psychophysiology 44 (6), 905–912. https://doi.org/10.1111/j.1469-8986.2007. 00567.x. Heydari, S., Holroyd, C.B., 2016. Reward positivity: reward prediction error or salience prediction error? Psychophysiology 53 (8), 1185–1192. https://doi.org/10.1111/ psyp.12673. Holroyd, C.B., Coles, M.G.H., 2002. The neural basis of human error processing: reinforcement learning, dopamine, and the error-related negativity. Psychol. Rev. 109, 679–709. https://doi.org/10.1037/0033-295X.109.4.679. Holroyd, C.B., Krigolson, O.E., 2007. Reward prediction error signals associated with a modified time estimation task. Psychophysiology 44 (6), 913–917. https://doi.org/ 10.1111/j.1469-8986.2007.00561.x. Holroyd, C.B., Yeung, N., 2012. Motivation of extended behaviors by anterior cingulate cortex. Trends Cogn. Sci. 16 (2), 122–128. https://doi.org/10.1016/j.tics.2011.12. 008. Holroyd, C.B., Nieuwenhuis, S., Yeung, N., Cohen, J.D., 2003. Errors in reward prediction are reflected in the event-related brain potential. Neuroreport 14 (18), 2481–2484. https://doi.org/10.1097/01.wnr.0000099601.41403.a5. Holroyd, C.B., Larsen, J.T., Cohen, J.D., 2004. Context dependence of the event-related brain potential associated with reward and punishment. Psychophysiology 41 (2), 245–253. https://doi.org/10.1111/j.1469-8986.2004.00152.x. Holroyd, C.B., Pakzad-Vaezi, K.L., Krigolson, O.E., 2008. The feedback correct-related positivity: Sensitivity of the event-related brain potential to unexpected positive feedback. Psychophysiology 45 (5), 688–697. https://doi.org/10.1111/j.1469-8986. 2008.00668.x. Knutson, B., Greer, S.M., 2008. Anticipatory affect: neural correlates and consequences for choice. Philos. Trans. R. Soc. B 363 (1511), 3771–3786. https://doi.org/10.1098/ rstb.2008.0155. Knutson, B., Westdorp, A., Kaiser, E., Hommer, D., 2000. FMRI visualization of brain activity during a monetary incentive delay task. NeuroImage 12 (1), 20–27. https:// doi.org/10.1006/nimg.2000.0593. Kujawa, A., Smith, E., Luhmann, C., Hajcak, G., 2013. The feedback negativity reflects favorable compared to nonfavorable outcomes based on global, not local, alternatives. Psychophysiology 50 (2), 134. https://doi.org/10.1111/psyp.12002. Liu, Y., Nelson, L.D., Bernat, E.M., Gehring, W.J., 2014. Perceptual properties of feedback stimuli influence the feedback-related negativity in the flanker gambling task. Psychophysiology 51 (8), 782–788. https://doi.org/10.1111/psyp.12216. Mars, R.B., De Bruijn, E.R.A., Hulstijn, W., Miltner, W.H.R., Coles, M.G.H., 2004. What if I told you: ‘you were wrong’? Brain potentials and behavioral adjustments elicited by feedback in a time-estimation task. In: Ullsperger, M., Falkenstein, M. (Eds.), Errors, Conflicts, and the Brain. Current Opinions on Performance Monitoring. MPI of Cognitive Neuroscience, Leipzig, pp. 129–134. Meadows, C.C., Gable, P.A., Lohse, K.R., Miller, M.W., 2016. The effects of reward magnitude on reward processing: an averaged and single trial eventrelated potential study. Biol. Psychol. 118, 154–160. https://doi.org/10.1016/j.biopsycho.2016.06. 002. Miltner, W.H.R., Braun, C.H., Coles, M.G.H., 1997. Eventrelated brain potentials
4.4. Limitations Firstly, for the two types of context condition in this study, the main difference between the even and neutral contexts might be outcome frequency, and we also mainly used outcome frequency to explain the ERP results. Except valence evaluation, outcome frequency probably affects outcome prediction, which effect should be dissociated from valence processing. By using more convincing dada analyzing methods (e.g., the PCA and the time-frequency analysis) or more appropriate experimental design, we might resolve this issue in the future. Secondly, for the adaptive time window, one staircase procedure was applied for all incentive cues, which made behavioral adaptation occurred on the trial level, irrespective of the incentive cues. One adaptive algorithm for each cued condition would be more appropriate for this study. At last, except the even and neutral sections, including a control section, in which only the experimental conditions were conducted, would be better for revealing the context effect. 5. Conclusions The context of even or neutral conditions affects FRN amplitudes in experimental conditions (gain and loss). FRN in the even context was larger than that in the neutral context when outcomes were favorable. Although the unfavorable outcome evoked larger FRN than the favorable outcome in both even and neutral contexts, the amplitude difference between unfavorable and favorable outcomes was greater in neutral context than in even context. The discrimination between favorable and unfavorable outcome is better in neutral context, which derives from the better processing of favorable outcomes in neutral context. Acknowledgements This study was supported by the National Natural Science Foundation of China (31200784). The funding sources had no role in the study design, data collection, analysis, or interpretation of the 35
International Journal of Psychophysiology 141 (2019) 28–36
S. Zhu, et al. following incorrect feedback in a time-estimation task: evidence for a “generic” neural system for error detection. J. Cogn. Neurosci. 9, 788–798. https://doi.org/10. 1162/jocn.1997.9.6.788. Nieuwenhuis, S., Holroyd, C.B., Mol, N., Coles, M.G.H., 2004. Reinforcement-related brain potentials from medial frontal cortex: origins and functional significance. Neurosci. Behav. Rev. 28 (4), 441–448. https://doi.org/10.1016/j.neubiorev.2004. 05.003. Novak, K.D., Foti, D., 2015. Teasing apart the anticipatory and consummatory processing of monetary incentives: an eventrelated potential study of reward dynamics. Psychophysiology 52, 1470–1482. https://doi.org/10.1111/psyp.12504. Oliveira, F.T., Mcdonald, J.J., Goodman, D., 2007. Performance monitoring in the anterior cingulate is not all error related: expectancy deviation and the representation of action-outcome associations. J. Cogn. Neurosci. 19 (12), 1994–2004. https://doi.org/ 10.1162/jocn.2007.19.12.1994. Osinsky, R., Mussel, P., Hewig, J., 2012. Feedback-related potentials are sensitive to sequential order of decision outcomes in a gambling task. Psychophysiology 49, 1579–1589. https://doi.org/10.1111/j.1469-8986.2012.01473.x. Osinsky, R., Walter, H., Hewig, J., 2014. What is and what could have been: an ERP study on counterfactual comparisons. Psychophysiology 51 (8), 773–781. https://doi.org/ 10.1111/psyp.12221. Pfabigan, D.M., Alexopoulos, J., Bauer, H., Sailer, U., 2011. Manipulation of feedback expectancy and valence induces negative and positive reward prediction error signals manifest in event-related brain potentials. Psychophysiology 48, 656–664. https:// doi.org/10.1111/j.1469-8986.2010.01136.x. Pfabigan, D.M., Sailer, U., Lamm, C., 2015a. Size does matter! Perceptual stimulus
properties affect event-related potentials during feedback processing. Psychophysiology 52, 1238–1247. https://doi.org/10.1111/psyp.12458. Pfabigan, D.M., Seidel, E.M., Paul, K., Grahl, A., Sailer, U., Lanzenberger, R., Lamm, C., 2015b. Context-sensitivity of the feedback-related negativity for zero-value feedback outcomes. Biol. Psychol. 104, 184–192. https://doi.org/10.1016/j.biopsycho.2014. 12.007. Proudfit, G.H., 2015. The reward positivity: from basic research on reward to a biomarker for depression. Psychophysiology 52, 449–459. https://doi.org/10.1111/psyp. 12370. Sambrook, T.D., Goslin, J., 2015. A neural reward prediction error revealed by a metaanalysis of ERPs using great grand averages. Psychol. Bull. 141, 213–235. https://doi. org/10.1037/bul0000006. Santesso, D.L., Bogdan, R., Birk, J.L., Goetz, E.L., Holmes, A.J., Pizzagalli, D.A., 2012. Neural responses to negative feedback are related to negative emotionality in healthy adults. Soc. Cogn. Affect. Neurosci. 7 (7), 794–803. https://doi.org/10.1093/scan/ nsr054. Talmi, D., Atkinson, R., Elderedy, W., 2013. The feedback-related negativity signals salience prediction errors, not reward prediction errors. J. Neurosci. 33 (19), 8264–8269. https://doi.org/10.1523/jneurosci.5695-12.2013. Walsh, M.M., Anderson, J.R., 2012. Learning from experience: event-related potential correlates of reward processing, neural adaptation, and behavioral choice. Neurosci. Behav. Rev. 36 (8), 1870–1884. https://doi.org/10.1016/j.neubiorev.2012.05.008. Yeung, N., Sanfey, A.G., 2004. Independent coding of reward magnitude and valence in the human brain. J. Neurosci. 24, 6258–6264. https://doi.org/10.1523/jneurosci. 4537-03.2004.
36
Contents lists available at ScienceDirect
International Journal of Psychophysiology journal homepage: www.elsevier.com/locate/ijpsycho
The influence of context condition on outcome evaluation in experimental conditions: Even vs. neutral Shuqing Zhu, Yuyang Wang, Shuqian Gao, Shiwei Jia
T
⁎
School of Psychology, Shandong Normal University, Jinan, China
ARTICLE INFO
ABSTRACT
Keywords: Feedback-related negativity (FRN) Outcome evaluation Context condition Context dependent effect Monetary incentive delay (MID) task
Feedback-related negativity (FRN) is an event-related brain potential that is elicited during outcome evaluation. Studies have found that FRN reflects a good vs. bad classification; more importantly, FRN reflects this classification in a context-dependent manner, which means that the outcome evaluation is obviously influenced by its embedded context. In the current study, we examined how the context conditions of even (i.e., the feedback was +4 or −4) and neutral (i.e., the feedback was always 0) affect the outcome evaluation in experimental conditions (gain and loss). The experimental conditions of gain (i.e., the feedback was +4 or 0) and loss (i.e., the feedback was 0 or −4) were presented with the even condition as the context in one section and with the neutral condition as the context in another section. The ERP (event-related potential) results of the two experimental conditions showed that the unfavorable outcome evoked more negative FRN than the favorable outcome in both even and neutral sections, however, the amplitude difference between unfavorable and favorable outcomes was greater in neutral section than in even section. Furthermore, the favorable outcomes evoked more positive FRN in the neutral section than in the even section. These results indicate that the context condition modulates outcome evaluation, in neutral context, the discrimination between favorable and unfavorable outcome is better, which might due to the facilitated identification of favorable outcomes in neutral context.
1. Introduction The evaluation of the consequences of our previous choices plays a key role in later decision making, which is also fundamental to adaptive behavior. Thus, an increasing amount of research has shed light upon the neural mechanisms of feedback evaluation, commonly employing the event-related potential (ERP) technique. Feedback-related negativity (FRN) is a fronto-centrally distributed negative-going ERP component that peaks at approximately 200–300 ms following external feedback presentation (Miltner et al., 1997; Gehring and Willoughby, 2002; Hajcak et al., 2006; Holroyd and Coles, 2002; Walsh and Anderson, 2012; Proudfit, 2015; Sambrook and Goslin, 2015). FRN is more negative to unfavorable outcomes, such as incorrect performance (Holroyd and Krigolson, 2007; Miltner et al., 1997) or monetary loss (Gehring and Willoughby, 2002; Hajcak et al., 2005; Yeung and Sanfey, 2004), than to favorable outcomes, such as correct performance or monetary gain. Besides, some scholars have also claimed that FRN is sensitive to violations of outcome expectations rather than outcome valence (Alexander and Brown, 2011; Oliveira et al., 2007; Pfabigan et al., 2011). According to the reinforcement learning theory (RL-Theory; ⁎
Holroyd and Coles, 2002; Nieuwenhuis et al., 2004), FRN reflects a reward prediction error signal that originates from the mid-brain dopaminergic system and is further projected to the anterior cingulate cortex (ACC). When outcomes are worse than expected, a phasic decrease in the dopaminergic system and an activity increase in the ACC will be induced; then, more negative FRN will be evoked (Holroyd and Coles, 2002; Sambrook and Goslin, 2015). It has thereafter repeatedly been reported that outcome processing contains a good vs. bad evaluation, which is demonstrated by FRN amplitude variations (Gehring and Willoughby, 2002; Hajcak et al., 2006; Yeung and Sanfey, 2004; for reviews see Nieuwenhuis et al., 2004; Walsh and Anderson, 2012; Proudfit, 2015; Sambrook and Goslin, 2015). In addition, according to the RL theory, the FRN amplitude could be modulated by the outcome prediction (Hajcak et al., 2007; Holroyd et al., 2003), for instance, the unexpected negative feedback evokes more negative FRN than the expected negative feedback does. Recently, researchers proposed the predicted response-outcome (PRO) model, which considers that FRN is evoked by the violation of outcome prediction, regardless of whether the outcome is positive or negative (Alexander and Brown, 2011; Oliveira et al., 2007). One unresolved question in terms of outcome evaluation revolves
Corresponding author at: School of Psychology, Shandong Normal University, No. 88, East Wenhua Rd., Lixia, Jinan 250014, China. E-mail address: [email protected] (S. Jia).
https://doi.org/10.1016/j.ijpsycho.2019.05.001 Received 4 September 2018; Received in revised form 14 April 2019; Accepted 5 May 2019 Available online 06 May 2019 0167-8760/ © 2019 Elsevier B.V. All rights reserved.
International Journal of Psychophysiology 141 (2019) 28–36
S. Zhu, et al.
around how the outcomes are evaluated and what factors affect outcome evaluation. Research reveals that the FRN amplitudes are not only influenced by the properties of the feedback stimulus itself, such as the size (Pfabigan et al., 2015a) and the similarity between favorable and unfavorable outcomes (Liu et al., 2014), but also modulated by its embedded context (Goyer et al., 2008; Holroyd et al., 2004; Osinsky et al., 2012; Pfabigan et al., 2015b). Previous studies have revealed many kinds of contexts, such as the outcome range under a specific condition (Holroyd et al., 2004; Pfabigan et al., 2015b), the overall probability of favorable and unfavorable outcomes (Holroyd et al., 2003; Hajcak et al., 2007), and the outcome sequence order (Goyer et al., 2008; Osinsky et al., 2012), can modulate the FRN amplitudes. How do the embedded contexts affect outcome evaluation, even though they do not convey outcome valence? According to the aforementioned theories, feedback evaluation reflects an encoding of the reward prediction error (Holroyd and Coles, 2002) or the prediction violation (Alexander and Brown, 2011). These theories mean that, before the outcome onset, the brain should generate the outcome prediction first (Bismark et al., 2013). The prediction generation is based not only on trial and error leaning but also on the embedded context. To explain how context information affects outcome processing, Osinsky et al. (2014) assumed a reference system that combines feedback and environmental information continuously to generate predictions for the next trial. In the current study, we aimed to explore whether and how context conditions, work as the feedback evaluation environment, would affect outcome evaluation in the experimental conditions (gain and loss). There were two sections in this study: one section with the even condition as context condition and another section with the neutral condition as context condition. The task manipulations were adopted from past studies (Angus et al., 2017; Holroyd et al., 2004; Pfabigan et al., 2015b). In each trial, participants finished a time estimation task and then received feedback for the estimation. At the beginning of each trial, an incentive cue indicating the possible outcomes in this trial was given first. The cue included four kinds: gain, loss, even and neutral. Under the gain (possible outcomes: +4 or 0), the loss (possible outcomes: 0 or −4), and the even (possible outcomes: +4 or −4) conditions, a favorable outcome and an unfavorable outcome would be given according to participants' performances (i.e., the favorable and unfavorable outcomes were given to correct and incorrect estimations, respectively); under the neutral condition, the outcomes were always 0, regardless of whether the time estimation was accurate. In this study, the gain and loss conditions were experimental conditions, and the even and neutral conditions were employed as context conditions. The whole experiment included two sections: in one section, the experimental conditions of gain and loss and the context condition of even were presented trial by trial in random; in another section, the same two experimental conditions (gain and loss) were presented with the neutral condition as the context condition. For data analysis, we focused on the two experimental conditions only. It is important to explore the effects of even and neutral conditions on feedback processing in other conditions, because many studies have adopted neutral (Angus et al., 2017; Broyd et al., 2012; Knutson and Greer, 2008; Novak and Foti, 2015; Pfabigan et al., 2015b; Santesso et al., 2012) or even (Holroyd et al., 2004) conditions as experimental conditions or as baseline conditions, and the results in other conditions (usually the gain and the loss conditions) were quite different. In one study (Holroyd et al., 2004) that explored the context-dependent effect of FRN, the even, gain, and loss conditions were presented block by block. The results demonstrated that FRN reflects a relatively good vs. bad evaluation according to the cued context in both the gain and loss conditions. With the monetary incentive delay (MID) task, one influential task for exploring the neural mechanisms of monetary
anticipation and consumption (Angus et al., 2017; Broyd et al., 2012; Knutson and Greer, 2008; Novak and Foti, 2015; Pfabigan et al., 2015b; Santesso et al., 2012), the gain, loss, and neutral conditions were presented trial by trial. However, the results in those studies were inconsistent with each other, and most of them were different from the results of Holroyd et al. (2004). For instance, with the MID task, Pfabigan et al. (2015b) found that in the gain condition, 0 evoked a larger FRN than +4; in the loss condition, 0 evoked a larger FRN than −4. In the gain condition, 0 was the relatively unfavorable outcome; on the contrary, in the loss condition, 0 was the relatively favorable outcome. Thus, the authors explained that FRN was not sensitive to feedback valence but was sensitive to expectation deviation. The studies of Santesso et al. (2012) and Novak and Foti (2015) found that the unfavorable outcomes evoked larger FRN than the favorable outcomes in the gain condition, while FRN for the favorable and unfavorable outcomes was not different in the loss condition. However, Angus et al. (2017) found that in both gain and loss conditions, the Reward positivity (RewP, Proudfit, 2015) for the favorable outcomes was more positive than that for the unfavorable outcomes; these results were in line with those of Holroyd et al. (2004). Thus, for the studies that used the MID task (Angus et al., 2017; Novak and Foti, 2015; Pfabigan et al., 2015b; Santesso et al., 2012), the potentials evoked by feedback processing showed inconsistent results, which needs further study for confirmation. Additionally, between the studies that designed the even condition (Holroyd et al., 2004) and other studies that used the neutral condition (Novak and Foti, 2015; Pfabigan et al., 2015b; Santesso et al., 2012), the results in the gain and loss conditions were quite different. The manipulations in the gain and loss conditions were similar in those studies (Holroyd et al., 2004; Novak and Foti, 2015; Pfabigan et al., 2015b; Santesso et al., 2012), such that the context condition (even vs. neutral) might be one critical factor that led to results discrepancies in the gain and loss conditions. Among these studies that used the MID task, the study of Angus et al. (2017) obtained quite similar results to the study of Holroyd et al. (2004); however, they used principal component analysis (PCA) and analyzed the principle component of RewP. In the current study, we replicated the manipulations of Angus et al. (2017) in one section (i.e., the neutral section) and analyzed the data with the grand average method to further test the FRN effect in the MID task. We also conducted one even section with even condition as the context condition to replicate the results of Holroyd et al. (2004) and also for a direct comparison with the neutral section. For outcome evaluation in experimental conditions, we assumed that being embedded in the neutral context should be different to being embedded in the even context. One possibility was that, in the neutral condition, the outcomes were always 0, irrespective of the performance accuracy. The outcome was meaningless for behavioral adjustment, and it weakened participants' motivation in the neutral section (Holroyd and Yeung, 2012). Thus, we predicted that the FRN amplitudes would decrease for the experimental conditions in the neutral context condition. Another possibility was that, in the neutral context, the frequencies of +4 and −4 outcomes were less than that in the even context. Less frequency caused unprediction of the +4 and −4 outcomes, which would evoke more negative FRN in the neutral context (Hajcak et al., 2007; Holroyd et al., 2003; Talmi et al., 2013). Additionally, for the good vs. bad classification, the prediction could not be made clearly. According to the study of Angus et al. (2017), we predicted that the unfavorable outcome would evoke larger FRN than the favorable outcome in both the even and neutral sections. However, according to other studies (Novak and Foti, 2015; Pfabigan et al., 2015b; Santesso et al., 2012), the results, especially those in the loss condition, were difficult to predict; thus, the current study will examine these results.
29
International Journal of Psychophysiology 141 (2019) 28–36
S. Zhu, et al.
Fig. 1. The timeline of a trial. In each trial, the cue (gain, loss, even, or neutral) that indicated the possible outcomes in this trial was presented first. Then, participants finished a time estimation task. Finally, participants received feedback.
2. Methods
(+ 4 ¥; even-favorable) and a loss of 4 ¥ (−4 ¥; even-unfavorable), respectively. In the neutral condition, feedback was always 0 ¥, irrespective of the accuracy of the time estimation. The participants were initially endowed 30 ¥ and were given a bonus according to their performance. The accuracy was determined by a tolerant time window. To ensure that participants' accuracies were approximately 50%, we used a sliding criterion to decide the tolerant time window (Mars et al., 2004; Miltner et al., 1997). The time window was initialized at 1000 ± 120 ms, which required participants to respond between 880 and 1120 ms following the presentation of the gold star in order to receive correct feedback on the first trial. The time window was shortened by 20 ms (subtracted 10 ms from the upper time limit and added 10 ms to the lower time limit) after a correct response and was lengthened by 20 ms after too fast or too slow of a response. The same adaptive algorithm was used continuously in the whole experiment, irrespective of the presented incentive cue. As Table 1 shows, the experiment had two sections: one section contained a gain condition, a loss condition, and an even condition; another section contained a gain condition, a loss condition, and a neutral condition. The gain and loss conditions were experimental conditions, and the even and neutral conditions were context conditions. For data analysis, this study focused on the experimental conditions only, which were the gain and loss conditions in the even section (even-gain and even-loss) and the gain and loss conditions in the neutral section (neutral-gain and neutral-loss). In addition, there were two possible outcomes (favorable and unfavorable) under each condition, which yielded 8 conditions in total. Before the experiment, the participants learned the main structure of the task through the instruction. In the even section, participants first conducted 12 practice trials, including 4 gain trials, 4 loss trials, and 4 even trials. Then, participants completed 80 gain trials, 80 loss trials, and 80 even trials. A total of 240 trials were presented randomly, resulting
2.1. Participants Twenty-one1 volunteers (7 males; 22.14 ± 2.54 years old) participated in the experiment. All participants were right-handed, had normal or corrected-to-normal vision and reported no prior or current psychiatric disorders. Written informed consent was obtained prior to the experiment. After the experiment, 30 ¥ and a bonus were given to each participant for their participation. The study was approved by the ethics committee of Shandong Normal University. 2.2. Apparatus and procedure Participants sat comfortably about 1 m in front of a computer monitor in an electromagnetically shielded room and performed a cued time estimation task. As Fig. 1 shows, each trial started with a centrally presented asterisk as the fixation point (800 ms). Prior to the time estimation, a cue that indicated the condition of the current trial was presented for 1000 ms. After the cue, a blank screen was presented for 1000–1500 ms at random. During this blank screen interval, participants were required to rehearse the cue in their mind and to prepare their motor response. Then, the target stimulus of a star (visual angle of 2.3°, gold color) was presented, and participants needed to press the space bar with their right index finger when they believed that 1 s had elapsed. Then, participants received feedback that appeared 1000 ms (visual angle of 2.3°) after a blank screen of 400–600 ms in random. The inter-trial interval (ITI) was 400–600 ms randomly. The task was presented by a CRT monitor (Dell M782) with the refresh rate of 75 Hz. The cue included four kinds: gain (a circle surrounding a “+”), loss (a circle surrounding a “−”), even (an empty circle) and neutral (also an empty circle); all cues had a 2.3° visual angle (Pfabigan et al., 2015b; Angus et al., 2017). In the gain condition, accurate and inaccurate responses resulted in gain of 4¥ (+ 4 ¥; gain-favorable) and gain omission (0 ¥; gain-unfavorable), respectively. In the loss condition, accurate and inaccurate responses resulted in loss avoidance (0 ¥; loss-favorable) and loss of 4 ¥ (−4 ¥; loss-unfavorable), respectively. In the even condition, accurate and inaccurate responses resulted in a gain of 4 ¥
Table 1 The conditions and the corresponding stimuli of the study. Even section Cue
1
The sample size was calculated by the G*Power 3.1 (Faul et al., 2007) with large effect size (f = 0.4) according to past studies (Gu et al., 2011; Meadows et al., 2016). We set an alpha of 0.05 and a correlation among repeated measures of 0.5. The 2 (context condition: even, neutral) × 2 (experimental condition: gain, loss) × 2 (valence: favorable, unfavorable) within-subject design determined that the number of group was 1 and the number of measurement was 8. Finally, the estimated sample size was 17. Also referring to the sample size of previous studies (Angus et al., 2017; Holroyd et al., 2004; Pfabigan et al., 2015b), 21 was the intended sample size.
Outcome
Favorable Unfavorable
Neutral section
even
gain
loss
+4 ¥ −4 ¥
+4 ¥ 0¥
0¥ −4 ¥
neutral 0¥ 0¥
gain
loss
+4 ¥ 0¥
0¥ −4 ¥
Note: the favorable outcome means the outcome for a correct response in a specific condition; the unfavorable outcome means the outcome for an incorrect response in a specific condition. In the even section, the gain and loss conditions were the experimental conditions, and the even condition was the context condition; in the neutral section, the same two experimental conditions were presented, with the neural condition as the context condition. 30
International Journal of Psychophysiology 141 (2019) 28–36
S. Zhu, et al.
apparent, the amplitude was set to 0 μV, which occurred in 2.98% of all cases, mostly after favorable feedback (Pfabigan et al., 2011). Finally, the averaged amplitude across the three electrodes was used as the amplitude of FRN. In addition, we measured the FRN amplitude with the formula of FRN-(P2 + P3)/2 from the Fz, FCz, and Cz, and for this measurement the P3 was defined as the most positive point 300–400 ms after feedback onset. Also, the averaged amplitude across the three electrodes was used as the FRN amplitude. By this measurement, on the one hand, we could remove the possible temporal overlap of the FRN with both P2 and P3; on the other hand, with multiple scoring methods we wanted to get more reliable results. Three-way ANOVAs were conducted on the FRN amplitude, with the factors of context condition 2 (even, neutral), experimental condition 2 (gain, loss), and outcome valence 2 (favorable, unfavorable).
in approximately 40 trials for each condition of gain-favorable, gainunfavorable, loss-favorable, and loss-unfavorable. Short breaks (5 min) were provided every 40 trials, and long breaks (10–15 min) were provided every 80 trials. The arrangements of the neutral section were similar to those of the even section. The order of the two sections was counterbalanced across participants. The whole experiment lasted for approximately 120 min, including the break time for approximately 60 min. The experiment was conducted by using E-Prime 2.0 (Psychology Software Tools, Inc., Sharpsburg, PA). 2.3. EEG acquisition and analyses The electroencephalogram (EEG) was recorded from 64 Ag-AgCl electrodes that were mounted in an elastic cap according to the 10–20 system using the Brain Products system (Brain Products GmbH, Munich, Germany), with a bandpass of 0.0531–80 Hz and a sampling rate of 500 Hz. All electrodes were online referenced to the FCz electrode, with the AFz electrode as the common ground. The electrode impedances were kept below 10 kΩ during recording. The vertical and horizontal electrooculograms (EOGs) were recorded from the outer canthi of the left eye and the inferior orbit of the right eye, respectively. Brain Vision Analyzer 2.0 (Brain Products GmbH, Munich, Germany) was used to conduct the EEG processing. Data were re-referenced to the linked mastoids, and the data from the FCz electrode were reinstated. Then, a 0.1–20 Hz bandpass filter with 24-dB roll-offs was applied. The epoch time window was −200 ms to 800 ms relative to feedback onset, with a −200–0 ms time window as the baseline. Ocular artifacts were corrected with an eye-movement correction algorithm. Trials with EEG voltages exceeding thresholds of ± 80 μV were considered artifacts and were excluded from the analysis. At last, EEGs were averaged for the eight conditions to generate ERPs: evengain-favorable (39.43 ± 5.21 trials survived, 1.78% trials were discarded), even-gain-unfavorable (39.29 ± 4.60 trials survived, 1.77% trials were discarded), even-loss-favorable (39.48 ± 4.64 trials survived, 1.40% trials were discarded), even-loss-unfavorable (39.19 ± 5.58 trials survived, 2.28% trials were discarded), neutralgain-favorable (40.05 ± 4.74 trials survived, 0.89% trials were discarded), neutral-gain-unfavorable (39.29 ± 4.14 trials survived, 0.64% trials were discarded), neutral-loss-favorable (39.00 ± 4.28 trials survived, 1.06% trials were discarded), and neutral-loss-unfavorable (40.29 ± 4.39 trials survived, 0.58% trials were discarded). The ERP amplitudes are susceptible to component overlap, which is often observable in the feedback-related potentials. Therefore, we chose the peak-to-peak measuring method for FRN. The peak-to-peak voltage difference (FRNeP2) between the most negative peak 200–300 ms after feedback onset and the preceding most positive peak 150–250 ms after feedback onset were measured from the Fz, FCz, and Cz electrodes (Holroyd et al., 2003; Pfabigan et al., 2011). If no negative peak was
2.4. Behavioral data analyses To test whether the trial numbers were comparable in all conditions, we compared the number of correct trial and incorrect trial in each experimental condition (gain and loss) with t-tests. Additionally, absolute erroneous amount was calculated using the formula of |estimated time − 1000| for all conditions. To examine the effects of the context condition, experimental condition (gain and loss), and outcome valence of the preceding trial on the absolute erroneous amount of current trial, an ANOVA with the context condition 2 (even, neutral), experimental condition 2 (gain, loss), and outcome valence (the preceding trial) 2 (favorable, unfavorable) as factors was conducted on the absolute erroneous amount. When necessary, the Greenhouse-Geisser correction for repeated measures was applied for all ANOVAs. The SPSS 17.0 was used for all data analyses. 3. Results 3.1. Behavioral results The overall accurate rate was 50.12 ± 0.92. The trial number of the correct and incorrect trials was not different in any of the conditions: for the even-gain condition (favorable trials: 40.05 ± 4.55; unfavorable trials: 39.90 ± 4.56), t(20) = 0.07, p = .943; for the evenloss condition (favorable trials: 40.24 ± 4.36; unfavorable trials: 39.86 ± 4.39), t(20) = −0.38, p = .707; for the neutral-gain condition (favorable trials: 39.57 ± 4.86; unfavorable trials: 40.38 ± 4.88), t(20) = 0.20, p = .844; and for the neutral-loss condition (favorable trials: 39.71 ± 4.50; unfavorable trials: 40.00 ± 4.44), t (20) = −0.15, p = .884). These results suggest that the trial numbers in all conditions were comparable. Fig. 2. The absolute erroneous amount of time evaluation in all conditions. The absolute erroneous amount was calculated with the formula of |estimated time − 1000|. Results showed that participants performed better after favorable outcomes than after unfavorable outcomes.
31
International Journal of Psychophysiology 141 (2019) 28–36
S. Zhu, et al.
Fig. 3. The grand-average FRN waveforms for all conditions at the Fz, FCz, and Cz electrodes.
Fig. 2 shows the absolute erroneous amount of time evaluation in all conditions. This absolute erroneous amount was influenced by outcome valence in the previous trial, F(1,20) = 19.49, p < .001, ηp2 = 0.49. After unfavorable outcomes (161.14 ± 54.81 ms), the absolute erroneous amount was larger than after favorable outcomes (131.44 ± 40.75 ms). The absolute erroneous amount was not affected by the context condition, F(1, 20) = 0.21, p = .654, or the experimental condition (gain and loss), F(1, 20) = 0.33, p = .572. There was no significant interaction, ps. ≥.152. These results indicate that participants performed better after favorable feedback than after unfavorable feedback, which was not modulated by context condition and experimental condition.
FRN amplitudes in each condition with the FRN-P2 measuring method. The ANOVA revealed a significant main effect for the context condition, F(1, 20) = 6.89, p = .016, ηp2 = 0.26. More negative FRN was evoked for the even (−5.31 ± 4.10 μV) context compared with that for the neutral (−4.24 ± 3.51 μV) context. The main effect of outcome valence was significant, F(1,20) = 73.20, p < .001, ηp2 = 0.79, and FRN for the unfavorable outcome (−6.53 ± 3.94 μV) was more negative than that for the favorable outcome (−3.02 ± 2.83 μV). The main effect of the experimental condition (gain and loss) was not significant (gain: −4.93 ± 3.93 μV; loss: −4.61 ± 3.78 μV), F(1,20)=1.09, p = .309. Significant interactions were observed between the context condition and outcome valence, F(1,20) = 4.73, p = .042, ηp2 = 0.19. Simple effect analysis showed that unfavorable outcomes, regardless of whether even or neutral was used as the context condition, elicited more negative FRN than favorable outcomes (ps. < .001). The difference
3.2. ERP results Fig. 3 shows the FRN waveforms in all conditions. Fig. 4 shows the
Fig. 4. The FRN peak-to-peak amplitudes in each condition with the FRN-P2 measuring method. 32
International Journal of Psychophysiology 141 (2019) 28–36
S. Zhu, et al.
Fig. 5. The outcome valance effects in even and neutral contexts. This figure shows the interaction between outcome valence and context condition. The difference waves (a) were calculated by subtracting favorable outcome evoked potential from unfavorable outcome evoked potential for even and neutral contexts, respectively. The topographies (b) show the FRN peaks in the time window of 200–300 ms.
between favorable and unfavorable outcomes tended to be larger for the neutral context (−4.07 ± 3.81 μV) than that for the even context (−2.94 ± 4.25 μV), t(41) = 1.97, p = .055, Cohen's d = 0.30. Furthermore, for favorable outcomes, more negative FRN was observed in the even context than that in the neutral context (p < .001); however, for unfavorable outcomes, FRN in the even context and neutral context was not different (p = .389). The remaining interactions were not significant, all p values ≥ .162. Fig. 5 shows the topographies of outcome valence in even (left of b) and neutral (right of b) contexts, respectively. The difference waves (a) were created by subtracting favorable outcome evoked potential from unfavorable evoked potential for even context and neutral context, respectively. Since the ANOVA showed that the valence effect was not modulated by experimental condition, for the difference waves and the topographies, the gain and loss experimental conditions were collapsed. This figure suggests that the difference wave (valence effect) was larger for neutral context than for the even context, and the FRN mainly distributes on the fronto-central scalp. Fig. 6 shows the results with the FRN-(P2 + P3)/2 measuring method. The ANOVA revealed a significant main effect of outcome valence, F(1,20) = 43.36, p < .001, ηp2 = 0.68. More negative FRN was evoked for unfavorable (−9.22 ± 4.20 μV) compared to favorable (−5.57 ± 3.34 μV) outcomes. The main effect of experimental condition (gain and loss) was significant, F(1,20) = 32.99, p < .001, ηp2 = 0.62, More negative FRN for gain (−8.24 ± 4.02 μV) condition
compared to loss (−6.55 ± 4.24 μV) condition has been found. Significant interaction was observed for context condition and outcome valence, F(1,20) = 5.29, p = .032, ηp2 = 0.21. Simple effect test indicated that unfavorable outcomes, regardless of using even or neutral as context, elicited more negative FRN than favorable outcomes (ps < .001). And the difference between favorable and unfavorable outcomes was larger in neutral context (−4.65 ± 5.41 μV) than that in even context (−2.63 ± 4.33 μV), t(41) = 2.65, p = .011, Cohen's d = 0.41. For favorable outcomes, more negative FRN was observed when even was used as context than when neutral was used as context. (p = .008); but for unfavorable outcomes, the FRN was not different between even as context and neutral as context (p = .470). 4. Discussion In this study, to examine whether and how the context condition would modulate outcome evaluation in other conditions, we presented experimental conditions of gain and loss to participants, with even as the context in one section and with neutral as the context in another section. For the behavioral results, we compared the number of correct and incorrect trials in each condition, and the results showed no difference in all experimental conditions (gain and loss), which guaranteed that the averaging trial number in each condition was comparable. The ANOVA on the absolute erroneous amount of time evaluation showed that after unfavorable feedback, participants performed worse than after
Fig. 6. The FRN base-to-peak amplitudes in each condition with the FRN-(P2 + P3)/2 measuring method. 33
International Journal of Psychophysiology 141 (2019) 28–36
S. Zhu, et al.
favorable feedback, which suggests that participants learned the task through the replication of correct response rather than the correction of the error response. It is possible that the feedback for incorrect response was too simple, without the incorrect direction and magnitude. Future studies should provide detailed outcome information to test how participants used negative feedback to learn the time estimation task (Mars et al., 2004). The analyses of FRN showed that the context condition modulated the FRN amplitudes in the experimental conditions, and FRN was more negative when even was used as the context than when neutral was used as the context when the outcome was favorable. In other words, the favorable outcomes evoked more positive brain potentials (might derive from the reward positivity, RewP) in neutral context than in even context. In addition, the binary classification of favorable and unfavorable outcomes in the experimental conditions was not affected by the context condition, and FRN evoked by unfavorable outcomes was more negative than that evoked by favorable outcomes in both the even and neutral sections, however, the difference between favorable and unfavorable outcomes was larger for the neutral context than that for the even context. Besides, the favorable vs. unfavorable classification was not affected by the experimental condition; under both gain and loss conditions, the unfavorable outcomes evoked more negative FRN than the favorable outcomes.
sections. Therefore, the FRN difference between favorable and unfavorable outcomes was larger in neutral context than in even context. 4.2. Implications for the MID task The MID task is a commonly used task for exploring the underlying mechanisms of reward expectation and consumption. Originally, this task was designed by Knutson et al. (2000) and Knutson and Greer (2008) and was mainly conducted in combination with functional magnetic resonance imaging (fMRI). Recently, some researchers were interested in the brain potentials that are evoked during reward expectation and consumption (Angus et al., 2017; Broyd et al., 2012; Novak and Foti, 2015; Pfabigan et al., 2015b; Santesso et al., 2012). Many components can be evoked by the MID task, such as the N2, contingent negative variation (CNV), and P300 during the cue stage as well as the FRN, RewP, and P300 during the feedback stage. However, the results were inconsistent among these components. By analyzing the feedback evoked component, FRN, the present study retested the evoked components during the consumption stage. In the current study, FRN was sensitive to feedback valence, and the unfavorable outcomes evoked more negative FRN than the favorable outcomes in both the gain and loss conditions. These results are consistent with those of Angus et al. (2017) and are in line with the RL theory, which supports that FRN reflects reward prediction error and thus is sensitive to feedback valence (Holroyd and Coles, 2002). In contrast to some studies (Novak and Foti, 2015; Pfabigan et al., 2015b; Santesso et al., 2012), this result supports the RL theory in both the gain and loss conditions. Pfabigan et al. (2015b) proposed that FRN is evoked by prediction violation and reflects a motivational salient signal. Novak and Foti (2015) and Santesso et al. (2012) found a valence effect of FRN in the gain condition rather than in the loss condition, which suggested that FRN is more sensitive to reward signals and generates reward positivity (Holroyd et al., 2008; Pfabigan et al., 2015b). The above analyses suggests that discrepancies still exist for the reward consumption stage in the MID task, and the same situation is also true for cue evoked components (Novak and Foti, 2015; Angus et al., 2017), which need further studies for clarification. Notably, this study found a valence effect for FRN in both the neutral section and the even section, which suggests that the decreased motivation in the neutral section did not impair the valence effect and thus removes the possible interference from the motivation factor, further confirming the valence effect of FRN in both the gain and loss conditions for the MID task.
4.1. The context condition affects outcome evaluation in experimental conditions In the introduction part, we had given two predictions. The first one was that, since all outcomes were delivered according to participants' performances in the even context, the motivation level would be greater in the even section compared with those in the neutral section (Hajcak et al., 2005, 2006; Holroyd and Yeung, 2012), and we predicted that more negative FRN would be evoked in the even section than in the neutral section. The results did not support this prediction for unfavorable outcomes, such that the motivational hypothesis was not fully verified. Another prediction was that, the +4 and −4 outcomes were less frequent in neutral section, such that the two outcomes were unpredicted and would evoke more negative FRN than that in the even section (Hajcak et al., 2007; Holroyd et al., 2003; Talmi et al., 2013), which prediction was also not verified. For the effect of context condition on experimental conditions (gain and loss), the favorable outcomes evoked more positive brain potentials in neutral context than in even context; the brain potentials that evoked by unfavorable outcomes were not different between in the neutral context and in the even context. These results suggest that the context condition modulates outcome evaluation of other conditions through favorable outcome processing rather than unfavorable processing. In other words, the context condition affects RewP evoked by the positive outcomes rather than FRN evoked by the negative outcome (Proudfit, 2015). Moreover, although the binary categorization was not modulated by the context condition, the discrimination between favorable and unfavorable outcomes was better in neutral section than in even section. In neutral section, since all outcomes were 0 in neutral condition, the +4 in gain condition and the −4 in loss condition were more infrequent and easier to be identified, and thus the classification of favorable and unfavorable outcomes in gain and loss conditions were easier in neutral context, and the amplitude difference between favorable and unfavorable outcomes in neutral section were larger than that in even section. In conclusion, the context of even and neutral condition modulates outcome evaluation in other conditions through valence evaluation, rather than the processes of prediction violation or motivational involvements. The +4 and −4 were less frequent in neutral context, which facilitated the favorable outcome processing and evoked more positive RewP in neutral section than in even section; while the unfavorable outcomes evoked comparable FRN in neutral and even
4.3. The context-dependent effect in the outcome evaluation The context effect in this section is constrained as the effects of the gain and loss cues on the binary evaluation in the experimental conditions (Holroyd et al., 2004; Kujawa et al., 2013), and it does not involve the effect of context conditions on the experimental conditions in the above sections. To explore how our brain classifies favorable and unfavorable outcomes, Holroyd et al. (2004) studied the context-dependent effect of outcome processing. In terms of the context, these authors defined it as the outcome range indicated by introduction or condition cues. In this seminal study, the authors designed the even, gain, and loss conditions, and the results showed that the outcomes were not evaluated as their absolute value but as their relative favorable or unfavorable value compared to other possible outcomes. With the MID paradigm, Angus et al. (2017) presented neutral, gain, and loss conditions, and the results also demonstrated that the outcomes were evaluated as relatively favorable or unfavorable according to possible outcomes. In contrast to those described in Holroyd et al. (2004), the three contexts described in Angus et al. (2017) were presented trial by trial rather than block by block; thus, this study extended the context effect from the global level to the local level. Obviously, the results in the 34
International Journal of Psychophysiology 141 (2019) 28–36
S. Zhu, et al.
current data.
current study also showed the context effect at the local level. Although the context effect of outcome classification has often been reported (Goyer et al., 2008; Holroyd et al., 2004; Kujawa et al., 2013; Pfabigan et al., 2015b), whether this effect can be extended to the local level is still debated. With the same MID task, some studies reported binary evaluation only in the gain condition (Novak and Foti, 2015; Santesso et al., 2012), and Pfabigan et al. (2015b) even found the opposite results in the loss condition. Similar to Holroyd et al. (2004), Kujawa et al. (2013) found that the cue at the local level could not be referred to in the outcome evaluation. Similarly, with the simple gambling task, Osinsky et al. (2014) and Gehring and Willoughby (2002) reported that the local alternative outcomes did not affect the outcome evaluation. Osinsky et al. (2014) claimed that the global context rather than the local context could affect the outcome evaluation. Thus, why did the current study find a context effect at the local level? We infer that the task is an important factor. Both the study of Angus et al. (2017) and the current study used the time estimation task. Compared to the simple gambling task (Kujawa et al., 2013; Osinsky et al., 2014), the time estimation task is learnable, and the outcomes are given according to participants' performances. Participants are more active in completing the task and care more about the outcomes. Thus, the good vs. bad classification could be found even at the local level in our study and in the study of Angus et al. (2017), but it disappeared in the studies of Kujawa et al. (2013) and Osinsky et al. (2014). The same conclusions have also been drawn by Heydari and Holroyd (2016), who found the valence effect over the prediction effect on FRN in an active task; these results failed to be found with the passive task (Talmi et al., 2013).
References Alexander, W.H., Brown, J.W., 2011. Medial prefrontal cortex as an actionoutcome predictor. Nat. Neurosci. 14, 1338–1344. https://doi.org/10.1038/nn.2921. Angus, D.J., Latham, A.J., Harmon-Jones, E., Deliano, M., Balleine, B., Braddon-Mitchell, D., 2017. Electrocortical components of anticipation and consumption in a monetary incentive delay task. Psychophysiology 54 (11), 1686–1705. https://doi.org/10. 1111/psyp.12913. Bismark, A.W., Hajcak, G., Whitworth, N.M., Allen, J.J., 2013. The role of outcome expectations in the generation of the feedback-related negativity. Psychophysiology 50, 125–133. https://doi.org/10.1111/j.1469-8986.2012.01490.x. Broyd, S.J., Richards, H.J., Helps, S.K., Chronaki, G., Bamford, S., Sonuga-Barke, E.J., 2012. An electrophysiological monetary incentive delay (e-MID) task: a way to decompose the different components of neural response to positive and negative monetary reinforcement. J. Neurosci. Methods 209 (1), 40–49. https://doi.org/10. 1016/j.jneumeth.2012.05.015. Faul, F., Erdfelder, E., Lang, A.G., Buchner, A., 2007. G* Power 3: a flexible statistical power analysis program for the social, behavioral, and biomedical sciences. Behav. Res. Methods 39 (2), 175–191. https://doi.org/10.3758/BF03193146. Gehring, W.J., Willoughby, A.R., 2002. The medial frontal cortex and the rapid processing of monetary gains and losses. Science 295, 2279–2282. https://doi.org/10. 1126/science.1066893. Goyer, J.P., Woldorff, M.G., Huettel, S.A., 2008. Rapid electrophysiological brain responses are influenced by both valence and magnitude of monetary rewards. J. Cogn. Neurosci. 20, 2058–2069. https://doi.org/10.1162/jocn.2008.20134. Gu, R., Lei, Z., Broster, L., Wu, T., Jiang, Y., Luo, Y., 2011. Beyond valence and magnitude: a flexible evaluative coding system in the brain. Neuropsychologia 49, 3891–3897. https://doi.org/10.1016/j.neuropsychologia.2011.10.006. Hajcak, G., Holroyd, C.B., Moser, J.S., Simons, R.F., 2005. Brain potentials associated with expected and unexpected favourable and unfavourable outcomes. Psychophysiology 42 (2), 161–170. https://doi.org/10.1111/j.1469-8986.2005. 00278.x. Hajcak, G., Moser, J.S., Holroyd, C.B., Simons, R.F., 2006. The feedback-related negativity reflects the binary evaluation of good versus bad outcomes. Biol. Psychol. 71, 148–154. https://doi.org/10.1016/j.biopsycho.2005.04.001. Hajcak, G., Moser, J.S., Holroyd, C.B., Simons, R.F., 2007. It's worse than you thought: the feedback negativity and violations of reward prediction in gambling tasks. Psychophysiology 44 (6), 905–912. https://doi.org/10.1111/j.1469-8986.2007. 00567.x. Heydari, S., Holroyd, C.B., 2016. Reward positivity: reward prediction error or salience prediction error? Psychophysiology 53 (8), 1185–1192. https://doi.org/10.1111/ psyp.12673. Holroyd, C.B., Coles, M.G.H., 2002. The neural basis of human error processing: reinforcement learning, dopamine, and the error-related negativity. Psychol. Rev. 109, 679–709. https://doi.org/10.1037/0033-295X.109.4.679. Holroyd, C.B., Krigolson, O.E., 2007. Reward prediction error signals associated with a modified time estimation task. Psychophysiology 44 (6), 913–917. https://doi.org/ 10.1111/j.1469-8986.2007.00561.x. Holroyd, C.B., Yeung, N., 2012. Motivation of extended behaviors by anterior cingulate cortex. Trends Cogn. Sci. 16 (2), 122–128. https://doi.org/10.1016/j.tics.2011.12. 008. Holroyd, C.B., Nieuwenhuis, S., Yeung, N., Cohen, J.D., 2003. Errors in reward prediction are reflected in the event-related brain potential. Neuroreport 14 (18), 2481–2484. https://doi.org/10.1097/01.wnr.0000099601.41403.a5. Holroyd, C.B., Larsen, J.T., Cohen, J.D., 2004. Context dependence of the event-related brain potential associated with reward and punishment. Psychophysiology 41 (2), 245–253. https://doi.org/10.1111/j.1469-8986.2004.00152.x. Holroyd, C.B., Pakzad-Vaezi, K.L., Krigolson, O.E., 2008. The feedback correct-related positivity: Sensitivity of the event-related brain potential to unexpected positive feedback. Psychophysiology 45 (5), 688–697. https://doi.org/10.1111/j.1469-8986. 2008.00668.x. Knutson, B., Greer, S.M., 2008. Anticipatory affect: neural correlates and consequences for choice. Philos. Trans. R. Soc. B 363 (1511), 3771–3786. https://doi.org/10.1098/ rstb.2008.0155. Knutson, B., Westdorp, A., Kaiser, E., Hommer, D., 2000. FMRI visualization of brain activity during a monetary incentive delay task. NeuroImage 12 (1), 20–27. https:// doi.org/10.1006/nimg.2000.0593. Kujawa, A., Smith, E., Luhmann, C., Hajcak, G., 2013. The feedback negativity reflects favorable compared to nonfavorable outcomes based on global, not local, alternatives. Psychophysiology 50 (2), 134. https://doi.org/10.1111/psyp.12002. Liu, Y., Nelson, L.D., Bernat, E.M., Gehring, W.J., 2014. Perceptual properties of feedback stimuli influence the feedback-related negativity in the flanker gambling task. Psychophysiology 51 (8), 782–788. https://doi.org/10.1111/psyp.12216. Mars, R.B., De Bruijn, E.R.A., Hulstijn, W., Miltner, W.H.R., Coles, M.G.H., 2004. What if I told you: ‘you were wrong’? Brain potentials and behavioral adjustments elicited by feedback in a time-estimation task. In: Ullsperger, M., Falkenstein, M. (Eds.), Errors, Conflicts, and the Brain. Current Opinions on Performance Monitoring. MPI of Cognitive Neuroscience, Leipzig, pp. 129–134. Meadows, C.C., Gable, P.A., Lohse, K.R., Miller, M.W., 2016. The effects of reward magnitude on reward processing: an averaged and single trial eventrelated potential study. Biol. Psychol. 118, 154–160. https://doi.org/10.1016/j.biopsycho.2016.06. 002. Miltner, W.H.R., Braun, C.H., Coles, M.G.H., 1997. Eventrelated brain potentials
4.4. Limitations Firstly, for the two types of context condition in this study, the main difference between the even and neutral contexts might be outcome frequency, and we also mainly used outcome frequency to explain the ERP results. Except valence evaluation, outcome frequency probably affects outcome prediction, which effect should be dissociated from valence processing. By using more convincing dada analyzing methods (e.g., the PCA and the time-frequency analysis) or more appropriate experimental design, we might resolve this issue in the future. Secondly, for the adaptive time window, one staircase procedure was applied for all incentive cues, which made behavioral adaptation occurred on the trial level, irrespective of the incentive cues. One adaptive algorithm for each cued condition would be more appropriate for this study. At last, except the even and neutral sections, including a control section, in which only the experimental conditions were conducted, would be better for revealing the context effect. 5. Conclusions The context of even or neutral conditions affects FRN amplitudes in experimental conditions (gain and loss). FRN in the even context was larger than that in the neutral context when outcomes were favorable. Although the unfavorable outcome evoked larger FRN than the favorable outcome in both even and neutral contexts, the amplitude difference between unfavorable and favorable outcomes was greater in neutral context than in even context. The discrimination between favorable and unfavorable outcome is better in neutral context, which derives from the better processing of favorable outcomes in neutral context. Acknowledgements This study was supported by the National Natural Science Foundation of China (31200784). The funding sources had no role in the study design, data collection, analysis, or interpretation of the 35
International Journal of Psychophysiology 141 (2019) 28–36
S. Zhu, et al. following incorrect feedback in a time-estimation task: evidence for a “generic” neural system for error detection. J. Cogn. Neurosci. 9, 788–798. https://doi.org/10. 1162/jocn.1997.9.6.788. Nieuwenhuis, S., Holroyd, C.B., Mol, N., Coles, M.G.H., 2004. Reinforcement-related brain potentials from medial frontal cortex: origins and functional significance. Neurosci. Behav. Rev. 28 (4), 441–448. https://doi.org/10.1016/j.neubiorev.2004. 05.003. Novak, K.D., Foti, D., 2015. Teasing apart the anticipatory and consummatory processing of monetary incentives: an eventrelated potential study of reward dynamics. Psychophysiology 52, 1470–1482. https://doi.org/10.1111/psyp.12504. Oliveira, F.T., Mcdonald, J.J., Goodman, D., 2007. Performance monitoring in the anterior cingulate is not all error related: expectancy deviation and the representation of action-outcome associations. J. Cogn. Neurosci. 19 (12), 1994–2004. https://doi.org/ 10.1162/jocn.2007.19.12.1994. Osinsky, R., Mussel, P., Hewig, J., 2012. Feedback-related potentials are sensitive to sequential order of decision outcomes in a gambling task. Psychophysiology 49, 1579–1589. https://doi.org/10.1111/j.1469-8986.2012.01473.x. Osinsky, R., Walter, H., Hewig, J., 2014. What is and what could have been: an ERP study on counterfactual comparisons. Psychophysiology 51 (8), 773–781. https://doi.org/ 10.1111/psyp.12221. Pfabigan, D.M., Alexopoulos, J., Bauer, H., Sailer, U., 2011. Manipulation of feedback expectancy and valence induces negative and positive reward prediction error signals manifest in event-related brain potentials. Psychophysiology 48, 656–664. https:// doi.org/10.1111/j.1469-8986.2010.01136.x. Pfabigan, D.M., Sailer, U., Lamm, C., 2015a. Size does matter! Perceptual stimulus
properties affect event-related potentials during feedback processing. Psychophysiology 52, 1238–1247. https://doi.org/10.1111/psyp.12458. Pfabigan, D.M., Seidel, E.M., Paul, K., Grahl, A., Sailer, U., Lanzenberger, R., Lamm, C., 2015b. Context-sensitivity of the feedback-related negativity for zero-value feedback outcomes. Biol. Psychol. 104, 184–192. https://doi.org/10.1016/j.biopsycho.2014. 12.007. Proudfit, G.H., 2015. The reward positivity: from basic research on reward to a biomarker for depression. Psychophysiology 52, 449–459. https://doi.org/10.1111/psyp. 12370. Sambrook, T.D., Goslin, J., 2015. A neural reward prediction error revealed by a metaanalysis of ERPs using great grand averages. Psychol. Bull. 141, 213–235. https://doi. org/10.1037/bul0000006. Santesso, D.L., Bogdan, R., Birk, J.L., Goetz, E.L., Holmes, A.J., Pizzagalli, D.A., 2012. Neural responses to negative feedback are related to negative emotionality in healthy adults. Soc. Cogn. Affect. Neurosci. 7 (7), 794–803. https://doi.org/10.1093/scan/ nsr054. Talmi, D., Atkinson, R., Elderedy, W., 2013. The feedback-related negativity signals salience prediction errors, not reward prediction errors. J. Neurosci. 33 (19), 8264–8269. https://doi.org/10.1523/jneurosci.5695-12.2013. Walsh, M.M., Anderson, J.R., 2012. Learning from experience: event-related potential correlates of reward processing, neural adaptation, and behavioral choice. Neurosci. Behav. Rev. 36 (8), 1870–1884. https://doi.org/10.1016/j.neubiorev.2012.05.008. Yeung, N., Sanfey, A.G., 2004. Independent coding of reward magnitude and valence in the human brain. J. Neurosci. 24, 6258–6264. https://doi.org/10.1523/jneurosci. 4537-03.2004.
36