The effects of induced state negative affect on performance monitoring processes

Abstract

Previous research suggests that performance-monitoring processes are related to personality traits; relationships with affective states, however, remain unclear. The purpose of this study was to replicate and extend previous findings that induced state negative affect alters electrophysiological reflections of performance monitoring. High-density event-related potentials (ERPs) were obtained from 69 healthy individuals (41 female, 28 male) who completed an Eriksen flanker task and received either encouraging or derogatory feedback based on mean reaction times (RTs) for 30-trial sub-blocks. Affective state, behavioral measures (i.e. error rates, RTs) and ERP measures [i.e. error-related negativity (ERN), post-error positivity (Pe) and N2] were assessed. Reaction times did not differ between feedback groups. Participants who received derogatory feedback committed more errors over time. Despite changes in affect, no significant group differences were demonstrated for behavioral or ERP measures of performance monitoring. Increases in vigilance were associated with more negative N2 amplitudes; no other changes in affective state were associated with changes in ERP measures. Results are consistent with findings suggesting performance-monitoring processes are only slightly affected by changes in affective state and fail to replicate previous studies suggesting the ERN is related to state changes in affect—supporting the possibility of the ERN as an endophenotype.

Research on neuroelectric indices of performance monitoring often focuses on three related processes—early error detection, error awareness and conflict monitoring. The error-related negativity (ERN) is a response-locked negative deflection in the scalp-recorded event-related potential (ERP) with fronto-central scalp distribution that putatively reflects early error-detection processes and peaks within 100 ms after an erroneous response (Falkenstein et al., 1991; Gehring et al., 1993; Nieuwenhuis et al., 2001). The post-error positivity (Pe) is a positive deflection in the ERP with more posterior scalp distribution than the ERN that occurs within 200–500 ms after conscious erroneous responses (Falkenstein et al., 1991, 2000; Nieuwenhuis et al., 2001). Various theories implicate the Pe as a reflection of conscious error processing (Nieuwenhuis et al., 2001; Endrass et al., 2007; Larson and Perlstein, 2009; Shalgi et al., 2009; Hughes and Yeung, 2011) or as an affective response to conscious errors (Falkenstein et al., 2000; Overbeek et al., 2005). The conflict N2 is a negative deflection in the stimulus-locked ERP with a fronto-central scalp distribution that peaks ∼250–350 ms after stimulus presentation and represents conflict detection (Nieuwenhuis et al., 2003; Yeung et al., 2004; Yeung and Cohen, 2006; Folstein and Van Petten, 2008). This role in conflict detection is supported by studies demonstrating that N2 amplitude is more negative (larger) on incongruent trials (i.e. when the flankers and target stimulus queue opposite responses) relative to congruent trials (i.e. when the flankers and target stimulus prime identical responses; Danielmeier et al., 2009; Forster et al., 2011; Clayson and Larson, 2011). Source localization studies have implicated the anterior cingulate cortex in ERN (van Veen and Carter, 2002; Stemmer et al., 2004; Brazdil et al., 2005), Pe (Herrmann et al., 2004; Overbeek et al., 2005) and N2 generation (Ridderinkhof et al., 2004; Yeung et al., 2004; Ladouceur et al., 2007).

The preponderance of current research indicates that performance-monitoring processes are closely associated with trait measures of personality. For example, individuals with high levels of trait anxiety generally exhibit increased ERN and N2 amplitudes when presented with competing response options relative to controls (e.g. Dennis and Chen, 2009; Aarts and Pourtois, 2010). Similarly, individuals with obsessive–compulsive disorder generally exhibit enhanced ERN amplitudes, despite normal Pe amplitudes (Ruchsow et al., 2005; Endrass et al., 2008) and ERN amplitudes remain unchanged after treatment (Ladouceur et al., 2006; Hajcak et al., 2008). The trait–anxiety relationship with ERN amplitudes is further bolstered by a recent study indicating increased ERN amplitudes in individuals with generalized anxiety disorder, despite similar Pe amplitudes (Weinberg et al., 2010). Moreover, individuals scoring high on trait measures of punishment sensitivity have demonstrated increased ERN amplitudes, and individuals scoring high on trait measures of reward sensitivity have shown increased Pe amplitudes (Boksem et al., 2006, 2008). High levels of negative affect have also been associated with enhanced ERN amplitudes in consonance with decreased Pe (e.g. Luu et al., 2000; Hajcak et al., 2004; Ladouceur et al., 2010) and larger N2 amplitudes (Ladouceur et al., 2010). Recently, Tops and Boksem (2010) suggested that persistence may underlie the relationship between ERN amplitude and personality traits such as sensitivity to reward or punishment, neuroticism and agreeableness. Other studies have shown that increased ERN amplitudes are associated with higher levels of trait empathy (Santesso and Segalowitz, 2008; Larson et al., 2010). In sum, trait measures of personality, particularly highly anxious and perfectionistic personality traits seem closely related with neural indices of performance monitoring.

In contrast, research on the relationship between affective state and neural indices of performance monitoring is less clear and somewhat contradictory. Moser et al. (2005) demonstrated that after inducing fear by presenting a feared spider stimulus, resulting in increased affective distress, the ERN remained unaltered and the Pe became diminished, suggesting that individuals oriented less to errors. Tops et al. (2006) evidenced that the ERN is related to trait measures, such as behavioral shame and agreeableness, but not affective states, such as depression, anger and tension. Trial-by-trial performance feedback is also not associated with changes in performance-monitoring indices (Olvet and Hajcak, 2009a). However, in contrast to these studies, changes in affect using emotionally-valent pictures have been shown to alter performance-monitoring processes (Larson et al., 2006; Wiswede et al., 2009a). Larson et al. (2006) found larger ERN amplitudes when individuals completed a flanker task superimposed on pleasant pictures compared to neutral and unpleasant pictures. Wiswede et al. (2009a) found that when presenting pictures between flanker trials, the ERN is enhanced following negative pictures compared to positive and neutral pictures. Neither of these studies, however, directly assessed changes in affective state. Wiswede et al. (2009b) used derogatory feedback to induce negative affect and found that increased levels of negative affect may be reliably associated with enhanced ERN amplitudes. Taken together, research to date is contradictory regarding the effects of affective state on neural substrates of performance monitoring.

Building upon previous findings indicating that performance-contingent feedback may reliably induce changes in affective states (see Westermann et al., 1996; Gottesman and Gould, 2003), Wiswede et al. (2009b) induced affective states using encouraging or derogatory feedback based upon reaction times in a small sample of 24 female participants. During the derogatory feedback condition, error detection became enhanced, as indicated by increased ERN amplitudes at the end of the task compared to the beginning of the task. Amplitude of the ERN was also correlated with changes in levels of endorsed negative affect (post-task minus pre-task). These results suggest that higher induced state negative affect may be related to increased early error-detection processes. Individuals in the encouraging feedback group demonstrated no changes in ERN amplitudes throughout the task. No effects were demonstrated for the Pe. These findings challenge the state-independent characteristics of early error-detection processes.

In order to extend our understanding of the influence of induced affective state on neural indices of performance monitoring, the present study used an identical paradigm to Wiswede et al. (2009b). A potential weakness to Wiswede et al.’s study is the small sample of only 12 female participants in each feedback group. Given that their study contradicts research suggesting that affective state does not influence early error-detection processes, we sought to verify their findings in a large sample. Another potential limitation to the generalizability of Wiswede et al.’s findings is the exclusion of male participants. Recent research indicates that females exhibit lower levels of error-related (Larson et al., 2011) and conflict-related (Clayson et al., 2011) neural activation compared to males. Furthermore, females may be more sensitive to derogatory feedback, particularly associated with errors (Moeller and Robinson, 2010). Considering the greater susceptibility of females to derogatory feedback, we expected to find enhanced performance-monitoring processes for females in the derogatory group compared to males in the derogatory group. We also hypothesized that there would be an enhancement in early error-detection processes associated with increases in negative affect. The present study will provide the possibility for replication of the Wiswede et al. (2009b) findings, elucidation of how changes in affective state alter behavioral and electrophysiological performance-monitoring processes, and reconciliation of contradictory research suggesting that changes in affective state may (Larson et al., 2006; Wiswede et al., 2009a, b) or may not alter indices of performance monitoring (e.g. Moser et al., 2005; Tops et al., 2006).

In addition, an important implication of the influence of state and trait negative affect on neural substrates of performance monitoring is in the determination of cognitive endophenotypes. In the past decade, a considerable amount of research has investigated the utility of indices of neural processing in identifying individuals at risk for developing psychopathology. In order for a neural-processing abnormality to be characterized as an endophenotype, it must be associated with a psychopathology, be heritable, manifest regardless of changes in state characteristics or whether the illness is activated, co-segregate with illness in families, and occur at higher rates in non-affected family members compared to the general population (see Gottesman and Gould, 2003). Recently, Olvet and Hajcak (2008) suggested that irregularities in neural activation associated with early error-detection processes, the ERN, may provide insight regarding information-processing abnormalities in psychiatric disorders and serve as an endophenotype. The functionality of the neural substrates of performance monitoring as endophenotypes relies on the state-independent characteristics of the ERN.

METHODS

Participants

All participants provided written informed consent as approved by the Brigham Young University Institutional Review Board. Participants were recruited from undergraduate psychology courses. Exclusion criteria, assessed via participant self report, included current or previous diagnosis of a psychiatric disorder, psychoactive medication use, substance use or dependence, neurological disorders, head injury, left-handedness or uncorrected visual impairment. Study enrollment included 69 individuals (41 female, 28 male) between the ages of 18 and 43 (M = 20.95, s.d. = 3.64). There were 35 participants (14 males and 21 females) in the encouraging feedback group and 34 participants (14 males and 20 females) in the derogatory feedback group. Groups did not differ in male-to-female ratio, χ²(1) = 0.01, P = 0.92. Descriptive statistics and mean summary data as a function of group are presented in Table 1.

Table 1.

Demographic and mean summary data as a function of feedback group

	Encouraging feedback (n = 35)	Derogatory feedback (n = 34)
	Mean (s.d.)	Mean (s.d.)
Age (years)	21.2 (4.7)	20.6 (2.3)
Congruent trial RT (ms)	458 (71)	450 (50)
Incongruent trial RT (ms)	485 (71)	476 (62)
Congruent trial error rates (%)	4 (3)	6 (5)
Incongruent trial error rates (%)	10 (6)	11 (5)
BIS	12 (2)	12 (3)
BAS	25 (5)	24 (5)
BAS-R	7 (1)	7 (2)
BAS-D	9 (3)	9 (2)
Agreeableness	36 (5)	36 (5)
Neuroticism	21 (5)	22 (6)

BAS-R = BAS reward responsiveness; BAS-D = BAS reward drive.

In order to assess changes in affective state, participants were administered the Profile of Mood State Scale (POMS) before and after task completion (McNair et al., 1992). The POMS is a 65-item self-report inventory designed to measure transient or fluctuating affective states on six different scales: anxiety, depression, anger, vigilance, fatigue and confusion. Individuals rate adjectives that describe moods or feelings on a scale ranging from 1 (not at all) to 5 (extremely). Wiswede et al. (2009b) used the ‘Aktuelle Stimmungsskala’ (Dalbert, 1992), a test written in German that is similar to the POMS and possesses similar subscales. Mean summary pre- and post-test data for the POMS are presented in Table 2.

Table 2.

Mean summary data for POMS subscales as a function of feedback group

Time	Subscale	Encouraging feedback (n = 35)	Derogatory feedback (n = 34)	Analysis
		Mean (s.d.)	Mean (s.d.)	t	P
Before task	Anxiety	15.2 (3.1)	15.3 (3.3)	0.12	0.90
	Depression	17.9 (3.6)	17.8 (3.9)	− 0.04	0.97
	Anger	13.5 (2.6)	13.2 (2.3)	−0.62	0.54
	Vigilance	22.8 (5.4)	23.4 (5.6)	0.44	0.66
	Fatigue	13.1 (4.3)	13.9 (5.2)	0.62	0.54
	Confusion	11.0 (3.4)	11.0 (2.9)	−0.08	0.94
After task	Anxiety	16.4 (5.2)	17.6 (5.3)	0.92	0.36
	Depression	18.5 (5.3)	19.3 (4.7)	0.69	0.49
	Anger	16.5 (7.7)	18.3 (6.8)	1.05	0.30
	Vigilance	18.8 (6.5)	17.6 (5.2)	−0.83	0.41
	Fatigue	16.2 (5.1)	19.0 (6.8)	1.92	0.06
	Confusion	11.3 (3.6)	12.1 (3.8)	0.81	0.42

The Behavioral Inhibition System (BIS) scale was administered to assess trait sensitivity to punishment (Carver and White, 1994). The Behavioral Activation System (BAS) scale was used to assess reward-related sensitivities (Carver and White, 1994). The BAS contains three subscales: reward responsiveness, drive and fun seeking. The Big-Five Inventory (BFI) was administered to assess openness, conscientiousness, extraversion, agreeableness and neuroticism (John and Srivastava, 1999). Based on previous research (Tops and Boksem, 2010), composite persistence-related trait scores were calculated using principle component analysis. Scores were transformed into z-scores and summated based on factor loadings. The reward drive/responsiveness component consisted of the combined scores of the BAS subscales of drive for reward and reward responsiveness. Neuroticism and BIS z-scores were combined to create a punishment-related component. The third component consisted only of agreeableness z-scores. These measures were administered before the experimental task.

Experimental task

Participants completed a task identical to the paradigm used by Wiswede et al. (2009b), although adapted for an English-speaking sample. Each trial consisted of either congruent (e.g. HHHHH) or incongruent (e.g. HHSHH) letter arrays presented in black on a gray background of a 17 inch computer monitor ∼20 inch from the participant’s head. Participants were instructed to respond as quickly and accurately as possible to the central letter of the five-letter string. Participants responded with a right-hand, index-finger key press to the H and a left-hand, index-finger key press to the S. Each flanker trial was preceded by a fixation cross that remained on screen for between 600 and 800 ms (mean 700 ms); flanker stimuli remained on screen until button press. Trials were presented randomly and consisted of 60% congruent and 40% incongruent letter arrays. There were 10 blocks of 210 trials each. A feedback screen was presented after every 30 trials (i.e. sub-block) that remained on the screen until terminated by a button press. Each participant received a total of 70 feedback screens.

Feedback was contingent upon the mean reaction time (RT) of the previous sub-block (i.e. 30 trials). Feedback ranged from mild to moderate to strong (Table 3) and was provided in a sentence format, e.g. ‘Your reaction time in the last sub-block was good’. If the mean RT of the current (N) sub-block was faster than the last-performed (N − 1) sub-block, participants in the encouraging feedback group received positive feedback (Table 3) and participants in the derogatory group received no feedback (‘press the mouse button to continue’). If the mean RT of N sub-block was slower than N − 1 sub-block, participants in the encouraging feedback group received no feedback, whereas participants in the derogatory group received negative feedback (Table 3). Thus, feedback was individually tailored to the participant’s actual performance. A button press terminated feedback screens, allowing participants to rest after each sub-block and view the feedback as long as desired.

Table 3.

Feedback algorithm as a function of feedback group

	Condition	Feedback group
		Encouraging	Derogatory
Level 1	RTN ≤ 20 ms faster than RTN−1	Good	No feedback
Level 2	RTN between 20 and 40 ms faster than RTN−1	Very good	No feedback
Level 3	RTN > 40 ms faster than RTN−1	Brilliant	No feedback
Level 1	RTN ≤ 20 ms slower than RTN−1	No feedback	Not good
Level 2	RTN between 20 and 40 ms slower than RTN−1	No feedback	Bad
Level 3	RTN > 40 ms slower than RTN−1	No feedback	Very bad

RT_N = mean reaction time for current sub-block (30 trials); RT_N−1 = mean reaction time for previous sub-block; table adapted from Wiswede et al. (2009b).

Electrophysiological data recording and reduction

Electroencephalogram (EEG) was recorded from 128 scalp sites using a geodesic sensor net and Electrical Geodesics, Inc. (EGI; Eugene, OR) amplifier system (20 K nominal gain, bandpass = 0.05–30 Hz). EEG was initially referenced to the vertex electrode and digitized continuously at 250 Hz with a 24-bit analog-to-digital converter. Impedances were maintained below 50 kΩ. Data were average-rereferenced off-line. Eye movement and blink artifacts were corrected using the Gratton et al., (1983) algorithm.

For the ERN and Pe, individual-subject response-locked averages were calculated using a window from 400 ms prior to participant response to 600 ms following participant response. We used a 200 ms time window from 400 to 200 ms before the response for baseline correction. Trials containing errors of omission were excluded from averages. Individual-subject correct-trial N2 data were segmented spanning 250 ms prior to stimulus presentation to 500 ms after stimulus presentation. Epochs were baseline corrected using a 150 ms window from 250 to 100 ms before presentation of the target stimulus. Electrode sites for analysis were chosen based on the scalp distribution of the ERP components of interest (Figure 1; e.g. Gehring et al., 1993; Falkenstein et al., 2000; Nieuwenhuis et al., 2003; Overbeek et al., 2005). Epochs were chosen based on examination of the grand-averaged waveforms and previous literature on the components of interest. ERN and N2 amplitudes were averaged across four fronto-central electrode sites [numbers 6 (FCz), 7, 106 and Ref (Cz); Figure 1]. Correct-trial and error-trial ERN amplitudes were extracted as the average of 15 ms pre-peak to 15 ms post-peak negative amplitude within 100 ms of the response. Correct-trial congruent and incongruent amplitudes for the N2 were extracted as the average of 15 ms pre-peak to 15 ms post-peak negative amplitude between 270 and 380 ms. Error-trial and correct-trial Pe amplitudes were extracted as the mean amplitude from 200 to 400 ms post-response across seven centro-parietal electrode sites [Cz, 31, 54, 55 (PCz), 62 (Pz), 79 and 80; Figure 1]. In order to examine the effects of changes in affect associated with feedback over time, behavioral indices were divided into 10 blocks (210 trials each). In order to maintain an adequate signal-to-noise ratio, ERPs were divided into four blocks (525 trials each).

Fig. 1.

Sensor layout of the 128-channel geodesic sensor net. Solid-line circle indicates fronto-central recording sites averaged for the ERN and N2; dashed-line circle indicates centro-parietal recording sites averaged for measurement of the Pe. Voltage maps are displayed for the (A) ERN, (B) Pe and (C) N2. Note the different scale for N2.

Data analysis

To ensure the state-related affective manipulations were effective and to replicate Wiswede et al. (2009b), data from the POMS were analyzed first to assess fluctuations in mood associated with task-relevant feedback. A 2-Group (encouraging feedback condition, derogatory feedback condition) × 6-Mood (anxiety, depression, anger, vigilance, fatigue, confusion) × 2-Time (before task, after task) mixed-model analysis of variance (ANOVA) was conducted on the POMS subscales. Partial-eta² (η²) is reported for all ANOVA effect sizes and the Huynh–Feldt epsilon adjustment was applied to correct for possible violations of sphericity for factors with more than two levels. Paired-samples t-tests were used to decompose significant interactions and orthogonal trend analyses were used to decompose main effects and interactions involving time. To determine the effects of changes in affect and time-in-task, A 2-Group (encouraging feedback condition, derogatory feedback condition) × 10-Time (10 blocks consisting of 210 trials each) × 2-Congruency (congruent, incongruent) ANOVA was conducted on RTs and error rates. A 2-Group × 4-Time (four blocks consisting of 525 trials each) × 2-Accuracy was conducted for ERP amplitudes. Pearson’s correlations were used to investigate the relationship between mood-state scores and ERP indices of performance monitoring. For correlation analyses as a function of group, POMS subscale difference scores (post-task minus pre-task) were correlated with difference waveform amplitudes for ERN (error minus correct), N2 (incongruent minus congruent) and Pe (error minus correct; see Wiswede et al., 2009b). We additionally looked at the relationship between changes in ERP amplitude across time (block 4 amplitude minus block 1 amplitude) and POMS subscale difference scores as well as trait scores related to persistence (see Tops and Boksem, 2010).

In order to assess possible sex differences in performance monitoring following feedback, we conducted a 2-Sex (male, female) × 2-Group × 4-Time × 2-Accuracy ANOVA on ERN and Pe amplitudes and a 2-Sex × 2-Group × 4-Time × 2-Congruency ANOVA on N2 amplitudes.

RESULTS

State changes in mood

A Group × Mood × Time ANOVA on mood states revealed significant main effects of mood and time, F(5,335) = 48.35, P < 0.001, η²= 0.42; F(1,67) = 10.99, P = 0.001, η²= 0.14, respectively. The Mood × Time interaction was significant, F(5,335) = 37.35, P < 0.001, η²= 0.36. Individuals endorsed higher levels of anxiety, t(68) = −3.28, P = 0.002, anger, t(68) = −5.03, P < 0.001 and fatigue, t(68) = −6.16, P < 0.001 and decreased vigilance, t(68) = −8.41, P < 0.001, post-task relative to pre-task (Table 2). No significant differences were demonstrated between times for POMS-rated depression, t(68) = −1.79, P = 0.08 or confusion, t(68) = −1.89, P = 0.06. No other main effects or interactions approached significance, including those between groups (F’s < 1.9, P’s > 0.15). Considering that both groups demonstrated increased levels of negative affect from pre-task to post-task, a sub-group analysis for ERP amplitudes was conducted for individuals in the derogatory feedback group that increased in levels of negative affect and individuals in the positive feedback group that did not increase in levels of negative affect (see below for further details). See Table 4 for summary feedback screen data as a function of group for the overall analyses as well as the subgroup analyses.

Table 4.

Mean (SD) feedback screens received as a function of feedback group

Analysis	Feedback group		Analysis
	Encouraging	Derogatory	t	P
Overall analysis
No Feedback	35 (2)	35 (3)	0.14	0.89
Level 1	16 (6)	15 (6)	0.54	0.59
Level 2	9 (3)	9 (3)	−0.59	0.56
Level 3	7 (6)	10 (6)	1.85	0.07
Affective subgroup analysis
No Feedback	35 (2)	35 (2)	0.17	0.87
Level 1	15 (5)	15 (6)	0.81	0.42
Level 2	9 (3)	9 (3)	−0.67	0.51
Level 3	8 (6)	10 (7)	1.33	0.19
Outcome-based subgroup analysis
No Feedback	34 (2)	35 (2)	0.96	0.35
Level 1	16 (6)	15 (6)	−0.79	0.44
Level 2	10 (3)	8 (3)	−1.48	0.15
Level 3	6 (5)	12 (6)	3.07	0.004

See Table 3 for explanation of feedback for each level as a function of group.

Behavioral data

Reaction times

A Group × Time × Congruency ANOVA on correct-trial RTs indicated that the main effect of congruency was significant, F(1,67) = 208.49, P < 0.001, η²= 0.76. Individuals responded more quickly to congruent than incongruent trials. No other main effects or interactions were significant (F’s < 1.6, P’s > 0.15).

Error rates

The Group × Time × Congruency ANOVA on error rates showed a significant main effect of congruency, with more errors committed on incongruent compared to congruent trials, F(1,67) = 176.78, P < 0.001, η²= 0.73. Individuals committed more errors through the course of the task, as supported by a main effect of time, F(9,603) = 20.62, P < 0.001, η²= 0.24 (Figure 2). This pattern was confirmed by a significant linear trend over time, F(1,67) = 35.67, P < 0.001, η²= 0.35. The main effect of group was not significant, F(1,67) = 2.60, P = 0.21, η²= 0.02. The Time × Group interaction, however, was significant, F(9,603) = 3.82, P = 0.02, η²= 0.05 (Figure 2). For the derogatory feedback group, participants committed more errors over time, as supported by a significant linear trend, F(1,33) = 19.35, P < 0.001, η²= 0.37. Participants in the encouraging feedback group also committed more errors over time, as supported by a significant linear trend, F(1,34) = 27.96, P < 0.001, η²= 0.45; although, the quadratic trend was also significant, F(1,34) = 11.76, P = 0.002, η²= 0.26. Independent samples t-tests revealed no differences between groups for blocks one through nine, t’s < 1.9, P’s > 0.06. Individuals in the derogatory feedback group committed more errors for block 10 relative to individuals in the encouraging feedback group, t(67) = 2.30, P = 0.02. The Time × Congruency interaction was also significant, F(9,603) = 4.36, P < 0.001, η²= 0.06, with congruent-trial error rates demonstrating a significant linear trend across blocks, F(1,68) = 16.24, P < 0.001, η²= 0.19. Incongruent-trial error rates demonstrated significant linear, F(1,68) = 50.69, P < 0.001, η²= 0.43 and quadratic trends, F(1,68) = 7.41, P = 0.008, η²= 0.10, over time. Neither the Group × Congruency interaction nor the Group × Time × Congruency interaction were significant (F’s < 0.7, P’s > 0.70).

Fig. 2.

Error rates as a function of feedback group for each 210-trial block.

ERPs

Grand averaged ERP difference waves (error minus correct or incongruent minus congruent) for the ERN, Pe and N2 waveforms as a function of sex and feedback group are presented in Figure 3. For response-locked ERPs, seven individuals with fewer than eight errors per block were excluded in order to maintain an adequate signal-to-noise ratio, resulting in 62 individuals retained for response-locked analyses (Olvet and Hajcak, 2009b).

Fig. 3.

Grand averaged difference waveforms for ERN and N2 over fronto-central recording sites and Pe over centro-parietal recording sites as a function of feedback group and sex.

ERN

A Group × Time × Accuracy ANOVA on ERN amplitudes revealed a significant main effect of accuracy with larger ERN amplitudes on error trials relative to correct trials, F(1,60) = 173.62, P < 0.001, η²= 0.74. The main effect of time was also significant, F(3,180) = 14.22, P < 0.001, η²= 0.19. Amplitudes became less negative over time, as supported by a significant linear trend, F(1,60) = 29.77, P < 0.001, η²= 0.33. No other main effects or interactions, including those associated with group, were significant (F’s < 0.5, P’s > 0.43).

Wiswede et al. (2009b) used a mean amplitude from 0 to 80 ms post-response to extract CRN and ERN amplitudes. Using this mean amplitude approach does not alter any of the findings. The main effect of group and group interactions were not significant (F’s < 0.74; P’s > 0.49).

Pe

A Group × Time × Accuracy ANOVA for Pe amplitudes showed a significant main effect of accuracy with larger Pe amplitudes for error trials compared to correct trials, F(1,60) = 94.66, P < 0.001, η²= 0.61. The main effect of time was significant, with Pe amplitudes generally becoming more positive over time, F(3,180) = 3.86, P = 0.01, η²= 0.06. No other main effects or interactions were significant (F’s < 2.0, P’s > 0.16).

N2

The Group × Time × Congruency ANOVA on N2 amplitudes revealed a significant main effect of congruency with more negative N2 amplitudes on incongruent compared to congruent trials, F(1,67) = 13.83, P < 0.001, η²= 0.17. The main effect of time was significant; N2 amplitudes became more negative over time, F(3,201) = 5.08, P = 0.01, η²= 0.07. The main effect of group was not significant, F(1,67) = 2.44, P = 0.12, η²= 0.04. The Time × Congruency interaction was significant, F(3,201) = 3.61, P = 0.02, η²= 0.05. For congruent and incongruent trials, N2 amplitudes became more negative over time as supported by significant linear trends, F(1,67) = 4.93, P = 0.03, η²= 0.07; F(1,67) = 7.40, P = 0.008, η²= 0.10, respectively. All other interactions were non-significant (F’s < 0.9, P’s > 0.40).

Sub-group analyses

Considering that both feedback groups demonstrated increased negative affect after task completion, separate analyses on the ERP components were conducted for individuals in the derogatory feedback group that increased in levels of negative affect and individuals in the encouraging feedback group that demonstrated no changes in levels of negative affect. A significant change from pre-task to post-task was based on the reliable change index that was calculated using procedures outlined by Temkin et al. (1999). After excluding seven individuals who demonstrated a reliable increase in depression levels from the encouraging feedback group, 28 individuals remained who did not demonstrate increases in negative affect (anxiety, anger, confusion, ts < 0.1, P’s > 0.34; with these individuals excluded a significant decrease in depression levels was shown from pre-task to post-task, t[27] = 2.38, P = 0.03). Seven individuals in the derogatory feedback group who showed reliable decreases in depression from pre-task to post-task were excluded from analysis, resulting in 27 remaining individuals who demonstrated increases in negative affect across the time of the task (anxiety, depression, anger, confusion, ts > 6.0, P’s < 0.01). The main effects of group and interactions involving group for ERN and Pe remained non-significant in this subset of participants for both Group × Time × Accuracy ANOVAs (F’s < 0.7, P’s > 0.52), and the main effect of group and interactions with group were similarly were non-significant in the Group × Time × Congruency ANOVA on N2 amplitudes (F’s < 1.6, P’s > 0.22). Using a mean amplitude for the ERN, identical to Wiswede et al. (2009b), does not alter the findings; the main effect of group and the group-related interactions remained non-significant (F’s < 0.82; P’s > 0.44).

A subgroup based on BIS scores was created. Considering that the BIS is a measure of punishment sensitivity, which has been shown to be associated with ERN amplitude in conditions with punishment (Boksem et al., 2008), those individuals with higher trait sensitivity to punishment may show larger increases in negative affect and subsequent increases in ERN amplitude. A median split of BIS scores from the derogatory feedback group was conducted and 15 individuals that showed the highest BIS scores were compared to 15 individuals with the lowest BIS scores. Both high and low BIS groups demonstrated increased anger, decreased vigilance and increased fatigue (|t’s| > 3.3, P’s < 0.02). The group with low BIS scores also showed greater anxiety after task completion, t(14) = −2.80, P = 0.01 (group with high BIS scores, t(14) = −1.85, P = 0.09). No differences were shown for depression or confusion levels, |ts| < 1.2, P’s > 0.26. The main effect of time remained significant in the Group × Time × Accuracy ANOVA on ERN amplitude, F (3,84) = 7.26 P < 0.001; ERN amplitudes decreased over time. The main effects of group and interactions involving group for ERN remained non-significant (F’s < 0.90, P’s > 0.35). In sum, results from both subgroup analyses remained consistent with the overall analysis.

Sex differences

When including sex as a factor in the Group × Time × Accuracy ANOVA on ERN and Pe amplitudes, no alterations were demonstrated in the pattern of results. The main effects of sex and interactions involving sex were not significant (F’s < 2.2, P’s > 0.12); however, the main effect of sex for Pe amplitudes approached significance, F(1,39) = 3.44, P = 0.07, η²= 0.05. Separate subsequent Group × Time × Accuracy ANOVAs on ERN and Pe amplitudes for females, identical to the analyses conducted by Wiswede et al. (2009b), revealed non-significant main effects of group and non-significant interactions involving group (F’s < 2.5, P’s > 0.12). Thus, we do not replicate the Wiswede et al. findings in any analyses.

The Group × Sex × Time × Congruency on N2 amplitudes revealed a main effect of sex approached significance, F(1,65) = 3.44, P = 0.07, η²= 0.05. No interactions with sex were significant (F’s < 2.2, P > 0.12). The Group × Time × Congruency ANOVA on N2 amplitudes for females indicated a non-significant main effect of feedback group and non-significant interactions involving feedback group (F’s < 1.2, P > 0.30).

ERP and mood correlations

No significant correlations between ERP difference waves and mood difference scores were found in either feedback group (derogatory feedback group: ERN difference wave, |r’s| < 0.17, P’s > 0.32; Pe difference wave, |r’s| < 0.22, P’s > 0.20; N2 difference wave, |r’s| < 0.21, P’s > 0.21; encouraging feedback group: ERN difference wave, |rs| < 0.24, P’s > 0.17; Pe difference wave, |r’s| < 0.22, P’s > 0.20; N2 difference wave, |r’s| < 0.27, P’s > 0.11).

For mood differences scores and ERP block difference scores, N2 amplitude enhancements across the task were associated with higher vigilance difference scores in the encouraging feedback group, r(34) = −0.44, P = 0.008. No other mood difference score or ERP block difference score correlations were significant for either group (|r’s| < 0.28, P’s > 0.11).

For correlations between composite trait scores and ERP block difference scores, more negative ERN block difference scores were associated with higher reward drive/responsiveness scores in the encouraging feedback condition, r(30) = −0.35, P = 0.04. No other correlations were significant for the encouraging feedback group (|r’s| < 0.12, P’s > 0.07). No significant relationships were shown for the derogatory feedback group (agreeableness and N2 block difference scores, r[36] = −0.31, P = 0.06; remaining correlations, |rs| < 0.20, P’s > 0.24).

DISCUSSION

The purpose of the present study was to reconcile contradictory findings indicating that affective state may (e.g. Larson et al., 2006; Wiswede et al., 2009a,b) or may not be associated with changes in neural substrates of performance monitoring (e.g. Moser et al., 2005; Tops et al., 2006). To this end, we replicated a paradigm that has previously evidenced alterations in performance monitoring associated with induced negative affect in a larger sample that included both males and females (Wiswede et al., 2009b). Contrary to previous findings (cf. Wiswede et al., 2009b), in the present sample reductions in state levels of negative affect were present for both encouraging and derogatory feedback groups. For behavioral data, RTs did not differentiate feedback groups. Although there was a significant interaction of time and feedback group for error rates, both feedback groups demonstrated reliable linear trends with more errors committed in only the last block in the derogatory feedback group compared to the encouraging feedback group.

Changes in negative affective state did not alter early error-detection, error-awareness or conflict-monitoring processes as indexed by the ERN, Pe and N2 components of the ERP. Notably, both groups showed increases in levels of negative affect and attenuation in ERN amplitude; thus, contrary to Wiswede et al. (2009b) the association between increases in negative affect and increases in ERN amplitude were not shown for either group, despite both groups showing increases in negative affect. Even after excluding individuals from the derogatory feedback group that demonstrated decreases of negative affect and individuals from the encouraging feedback group that demonstrated increases of negative affect, results indicate that affective state did not affect ERP components associated with performance monitoring. Moreover, in a subgroup of individuals from the derogatory feedback group high in punishment sensitivity scores and low in punishment sensitivity both groups showed increases in negative affect; however, both groups showed ERN amplitude attenuation, indicating that results from the overall analysis remained consistent in both subgroup analyses. No sex differences were demonstrated in the present analyses. Taken together, results seem to support the notion that changes in affective state measured in the present sample (i.e. changes in anxiety, depression, anger, vigilance, fatigue and confusion measured by the POMS) did not affect electrophysiological indices of performance monitoring.

Current results contradict previous findings by Wiswede et al. (2009b). Wiswede et al. in a sample of 24 female participants demonstrated significant differences in ERN amplitude between encouraging and derogatory feedback conditions as well as larger ERN amplitudes at the end of the task compared to the beginning for individuals receiving derogatory feedback. In the current study, no significant differences between feedback groups were shown for ERN, Pe or N2 amplitudes, suggesting that performance monitoring is not affected by changes in negative affective state. Even in a subgroup analysis of individuals that showed increases in negative affect from the derogatory group and individuals that demonstrated no change in negative affect from the encouraging feedback group as well as a subgroup analysis of individuals from the derogatory feedback group that were either high in trait punishment sensitivity or low in trait punishment sensitivity, negative affect was not associated with increases in ERN amplitude in any analyses. Although the present sample included males and females, analyses examining only females were consistent with our overall findings. Thus, we fail to replicate any portion of the Wiswede et al. findings.

In regards to previous findings of changes in performance monitoring associated with presentation of emotionally valent pictures (Larson et al., 2006; Wiswede et al., 2009a), differential performance monitoring may be unrelated to affective state and more associated with changes in conflict-associated processing. According to the conflict monitoring theory of cognitive control, ERN amplitude is contingent upon continued target-stimulus processing after erroneous responses (Yeung and Cohen, 2006; Yeung et al., 2007). Thus, larger ERN amplitudes for pleasant pictures relative to unpleasant or neutral pictures may primarily be the result of facilitated stimulus processing (see Larson et al., 2006). Whereas in the task used by Wiswede et al. (2009a), they noted that unpleasant pictures may be associated with increased allocation of attentional resources following the picture, which would result in greater focus on target-stimulus processing and subsequent larger ERN amplitudes. Considering that neither study measured changes in affective state, ERN differences may simply be accounted for using the conflict monitoring theory.

Other studies have found relationships between ERN amplitude and changes in happiness (West and Travers, 2008), levels of pre-task boredom (West and Travers, 2008), alcohol intake (Ridderinkhof et al., 2002), sleep deprivation (Tsai et al., 2005) and competition (van Meel and van Heijningen, 2010). Changes in these domains may be coupled with changes in attention, which is supported by findings indicating that decreased attention is associated with increased boredom (see Vodanovich, 2003 for review), alcohol intake (Steele and Josephs, 1990; Curtin and Fairchild, 2003) and sleep deprivation (Beebe et al., 2010; Chee and Tan, 2010). For example, early studies of the ERN found increased ERN amplitudes when accuracy was emphasized relative to when speed was emphasized (Gehring et al., 1993; Falkenstein et al., 2000). According to the conflict monitoring theory, greater attention to task-relevant information should result in enhanced ERN amplitude; whereas decreased attention should be associated with reductions in ERN amplitude (Yeung and Cohen, 2006; Yeung et al., 2007; Danielmeier et al., 2009). Indeed larger ERN amplitudes have been related to improved performance on neuropsychological tests of attention and executive function (Larson and Clayson, 2011). Thus, attention to the task likely subsumes the relationships between ERN amplitudes and changes in affective state, such as those mentioned above, and may account for the general decrease in ERN amplitude across the task in the current data. This seems consistent with recent research indicating a relationship between trait persistence and ERN amplitude (Tops and Boksem, 2010); individuals with higher persistence scores are more likely to remain engaged and proactively focus on the task. Future studies of the effects of induced affective state should take into account changes in attention to the task.

Findings by Olvet and Hajcak (2009a) further support the notion that performance monitoring is not affected by feedback. Olvet and Hajcak used two different tasks (one with trial-by-trial feedback, one without trial-by-trial feedback) to investigate the relationship between trait measures of depression, anxiety and stress and error-related processing. Trait anxiety and error detection were only related in the condition without feedback, as supported by a significant correlation between trait anxiety and ERN amplitude; no ERN or Pe differences were demonstrated between the feedback condition and condition without feedback. Although changes in affect were not measured, findings by Olvet and Hajcak (2009a) in concert with the current results, suggest that affectively valenced feedback may not affect early error-detection or error-awareness processes.

Reductions of ERN amplitude across the course of the task in both encouraging and derogatory feedback groups support previous findings indicating that error detection may be affected by task engagement. Tops et al. (2006) demonstrated a linear relationship between ERN difference wave amplitudes and pre-task cortisol levels and ERN difference wave amplitudes and cortisol decrease during the task. Tops et al. concluded that higher pre-task levels of cortisol were associated with a larger decrease of cortisol during the task indicating an allocation of resources for task completion (Ennis et al., 2001; Lewis and Ramsay, 2002). More recently, Tops and Boksem (2010) found that persistence-related traits were reliably associated with ERN amplitude changes across time. The present study found that reward drive/responsiveness, a persistent-related component measured in the Tops and Boksem (2010) study, was reliably associated with changes in ERN amplitude for the encouraging feedback group but not the derogatory feedback group; these results likely indicate that individuals in the encouraging feedback group with high reward drive/responsiveness were most likely to persist through the task and remain engaged. In sum, these findings seem to support the conclusion that traits associated with task engagement, such as persistence, may account for variations in ERN amplitude across tasks. This conclusion is further supported by studies demonstrating a relationship between the motivational significance of errors and ERN amplitudes that show enhanced ERN amplitudes correlating with increased error significance (e.g. Gehring et al., 1993; Hajcak et al., 2005; Chiu and Deldin, 2007).

Other explanations of the present outcomes should be noted. First, both feedback groups demonstrated increases in negative affect in the overall sample. This may be the result of the encouraging feedback group in the present sample experiencing the ‘no-feedback’ screen as a negative event. In the study by Wiswede et al. (2009b), only the derogatory feedback group showed increases in negative affect and decreases in positive affect, changes which were coupled with ERN enhancements across the task. Considering that in the present study both groups showed increased negative affect, it would then be expected that both feedback groups would show increased ERN amplitudes; however, neither feedback group showed ERN enhancements across time, despite increases in state negative affect. Although the present study attempted to address differences in personality traits related to punishment sensitivity and reward drive/responsiveness, other personality traits related to motivation and persistence may better account for the group similarities on negative affect (see Tops and Boksem, 2010). Another potential explanation of the current findings could be differences between the number of feedback screens presented to the respective feedback groups compared to the groups from the study by Wiswede et al. (2009b); however, it is unclear as Wiswede et al. did not note the number of feedback screens for each respective level of feedback.

By demonstrating that error detection was not susceptible to changes in affective states using a feedback-based paradigm in the present sample, current findings support the role of the ERN as a possible endophenotype for psychopathology by evidencing its state-affect-independent properties. Previous research suggests that error-detection processing may be closely related to affective traits, such as anxiety (Aarts and Pourtois, 2010; Weinberg et al., 2010), negative affect (Luu et al., 2000; Hajcak et al., 2004; Ladouceur et al., 2010) and persistence/task engagement (Tops and Boksem, 2010); trait reward drive/responsiveness, a component underlying persistence, was associated with enhancements in ERN amplitude for individuals that received encouraging feedback. Studies of the effects of affective state are less unified. The present results suggest that affect-related changes do not alter performance-monitoring processes, such as early error detection, error awareness or conflict monitoring. Future studies replicating the picture-related findings of Larson et al. (2006) and Wiswede et al. (2009a) and assessing state-related changes in affect are needed to further understand the influence of state affective changes on electrophysiological indices of performance monitoring.

Conflict of Interest

None declared.