Threats That Arise From Within: Changes in Error Processing and the Prospective Prediction of Everyday Avoidance

Abstract

Background

Errors are internally generated and can be harmful. Therefore, errors are considered endogenous threats. Like other forms of threat, errors receive increased processing in clinical anxiety, and enhanced error processing is predictive of anxiety onset in children. As is seen with other types of threat, avoidance of errors could also play a role in symptom worsening.

Methods

Seventy-four participants (56 female, 18 male) with internalizing psychopathology completed self-report measures and a flanker task during electroencephalography (EEG) recording at baseline (time 1) and 1 year later (time 2). Time 1 EEG event-related potentials, the error positivity (Pe) and the error-related negativity (ERN), as well as changes in error processing over 1 year, were examined as predictors of everyday avoidance (i.e., avoidance of anxiety-provoking stimuli in the real world) at time 2.

Results

Participants with greater elaborated processing of errors (Pe) at time 1 showed greater increases in everyday avoidance at time 2. This association was only evident for participants with reductions in early error processing (ERN) over the same time period, potentially indicating avoidance of errors in the laboratory. In addition, smaller time 2 ERNs were correlated with increased time 2 everyday avoidance.

Conclusions

Heightened elaborative error processing (indicative of sensitivity to endogenous threat) is predictive of increased everyday avoidance over 1 year. However, how patients respond to heightened elaborative error processing at baseline—i.e., via blunting of early error processing over the following year—is critical in determining worsening of everyday avoidance.

Plain Language Summary

Sensitivity to errors (a form of endogenous threat) may prompt avoidance in some individuals. In a mixed internalizing sample, we examined event-related potentials, the error positivity (Pe) and error-related negativity (ERN), as predictors of everyday avoidance over 1 year. Heightened Pe predicted increased avoidance, specifically for individuals who showed blunting of the ERN over 1 year. Therefore, heightened error (endogenous threat) processing prospectively predicts avoidance in individuals who disengage from error processing over the same time period.

Plain Language Summary

Sensitivity to errors (a form of endogenous threat) may prompt avoidance in some individuals. In a mixed internalizing sample, we examined event-related potentials, the error positivity (Pe) and error-related negativity (ERN), as predictors of everyday avoidance over 1 year. Heightened Pe predicted increased avoidance, specifically for individuals who showed blunting of the ERN over 1 year. Therefore, heightened error (endogenous threat) processing prospectively predicts avoidance in individuals who disengage from error processing over the same time period.

Errors can be considered endogenous threats because they are generated internally and can be harmful (1, 2, 3). Consistent with this, excessive response to errors may play a role in forms of internalizing psychopathology thought to be characterized by abnormal threat responding. For example, case-control and correlational studies indicate that clinical anxiety is associated with heightened error processing (4, 5, 6). Moreover, excessive error responding in children is a risk factor for anxiety disorder development (7,8). On the other hand, little is known about error processing as a transdiagnostic risk factor for the development or worsening of internalizing psychopathology in adults. Additionally, as is seen with other forms of threat, exaggerated error responding followed by avoidance may likely worsen internalizing psychopathology (9,10). That is, not only enhanced error processing, but also how individuals cope with heightened baseline sensitivity to errors may partially determine whether this initial risk factor serves as a liability for increased internalizing psychopathology over time.

Several lines of research suggest that errors, which have the potential to endanger individuals, can be considered endogenous threats [for a review, see (3)]. For example, like other forms of threat, errors prime defensive and fear systems. That is, errors potentiate startle, like other forms of threat (11). Errors also engage brain regions implicated in threat processing such as the amygdala and anterior cingulate cortex (ACC) (12, 13, 14, 15). Although exogenous threat processing is prominent in paradigms designed to model anxiety mechanisms, patients also avoid endogenous threats such as fear-eliciting memories/experiences (e.g., getting something wrong/erring) (16). Therefore, understanding individual differences in error processing may provide novel insight into avoidance across the internalizing spectrum.

Error processing can be measured via 2 electroencephalography (EEG) event-related potentials (ERPs): the error-related negativity (ERN) and the error positivity (Pe) (17,18). The ERN is a negative-going ERP that peaks 0 to 100 ms after error commission and likely originates in the ACC (19,20). The Pe is a positive-going ERP that emerges at approximately 200 to 400 ms post-error commission and likely originates in more rostral portions of the ACC (17,21). Whereas the ERN indicates relatively automatic, immediate error processing, the Pe likely reflects more conscious, elaborative error processing (17, 18, 19, 20). Compared with other neurobiological measures of error processing, these ERPs are cost-effective (22,23) and well tolerated by patients (24). Moreover, both ERPs are well suited to longitudinal investigations due to excellent test-retest reliability (25,26) and internal consistency (27).

Consistent with the notion that internalizing disorders may be characterized by exaggerated error (threat) processing (28,29), the ERN has been found to be increased (more negative) in multiple internalizing disorders, notably generalized anxiety disorder (GAD) and obsessive-compulsive disorder (5,7). However, evidence of altered ERNs in major depressive disorder, and the Pe’s cross-sectional relationship with internalizing psychopathology, has been mixed (30, 31, 32, 33, 34, 35, 36). In sum, despite fairly consistent evidence of larger ERNs in anxiety, conclusions regarding associations between the ERN and other internalizing psychopathology and those involving the Pe remain somewhat inconclusive.

Comparatively little work has focused on the prospective relationship between error processing and internalizing psychopathology, with most work having focused on children or adolescents. For example, previous work has found that larger ERNs at baseline predicted GAD onset or anxiety symptom increase in children, adolescents, and college students (8,37, 38, 39, 40, 41, 42). By contrast, the 3 studies that investigated the predictive utility of the ERN for depression in adults have shown varying results (43, 44, 45). For example, one study found that larger baseline ERNs predicted increased anhedonia 1 year later in patients with remitted depression (43). However, in undergraduate samples, ERN amplitudes did not predict depression, and during COVID-19, a blunted ERN predicted depression (44,45). Together, this work suggests that the ERN and Pe may also serve as a prognostic indicator of increased psychopathology for adults diagnosed with anxiety and possibly depression.

Examination of the ERN and Pe in relation to dimensional measures of internalizing psychopathology derived based on the same, more homogenous dimensions that cut across heterogenous disorders may map better onto underlying neurobiology (46). Avoidance is a transdiagnostic dimension thought to play a causal role in numerous internalizing disorders (9,10,47, 48, 49, 50, 51, 52). Evidence suggests that avoidance offers greater parsimony and noninferiority to diagnostic models in predicting critical aspects of internalizing psychopathology, such as impairment, comorbidity, and suicidality (53).

In the laboratory, enhanced error processing (the Pe and ERN) has been associated with increased avoidance learning [(54, 55, 56); but see (57)], possibly especially for individuals with clinical anxiety (58). However, only 2 studies have examined associations between habitual/everyday avoidance—avoidance of anxiety-provoking stimuli in the real world—and the ERN (3,59). One study found no association between traumatic avoidance and the ERN in a sample of adolescents with depression (3). The other study examined associations between the ERN, correct-related negativity (CRN), and emotion regulation strategies (including avoidance) in youths diagnosed with anxiety and found no relationship between the ERN and avoidance, but elevated CRNs positively correlated with avoidance (59). Therefore, while error processing may predispose individuals to greater avoidance in the laboratory, limited work has linked error processing and everyday avoidance. Other variables, such as how error processing changes over time, may distinguish individuals who develop everyday avoidance. For example, individuals who respond to initial exaggerations in conscious, elaborative error processing with reductions or blunting in automatic error processing, potentially indicative of avoidance in the laboratory, may be more likely to show broader evidence of everyday avoidance.

Here, we aimed to understand how baseline error processing might predict increased avoidance 1 year later in a mixed internalizing sample. We hypothesized that increased error processing at baseline (larger Pe and/or ERN) would be associated with greater increases in avoidance 1 year later. Furthermore, we investigated whether the relationship between baseline error processing and future avoidance would be moderated by changes in error processing over the same year.

Methods and Materials

Overall Study Design

Participants were individuals with mixed internalizing psychopathology who were part of a larger study. At study entry (time 1), participants completed a diagnostic interview [the Structured Clinical Interview for DSM-5 (60)], self-report questionnaires, and an EEG. Following their time 1 visits, participants returned approximately 1 year later (time 2) and repeated the same assessments in a counterbalanced order.

Participants

Inclusion/exclusion criteria for the parent study, and thus this sample, required that participants were 18 to 65 years of age and had at least 2 current (past 2 weeks) focal fear disorder—i.e., specific phobia or performance-only social anxiety—symptoms.a Participants could have any number of comorbid internalizing symptoms/diagnoses.

Of 79 participants, 5 were excluded for committing <6 errors (61). The final sample comprised 74 participants (56 female, 18 male; mean_age = 25.47 years, SD = 10.04). Participant demographic and clinical characteristics are presented in Table 1. Study procedures were consistent with the Helsinki Declaration of 1975 (as revised in 1983) and were approved by the Texas A&M University Institutional Review Board.

Table 1.

Demographic and Clinical Characteristics of Participants

	Time 1	Time 2
Race
American Indian or Alaska Native	3 (4.05%)	–
Asian	14 (18.92%)	–
Black	2 (2.70%)	–
More Than 1 Race	5 (6.76%)	–
Native Hawaiian or Pacific Islander	1 (1.35%)	–
White	49 (66.22%)	–
Ethnicity
Hispanic or Latino	25 (33.78%)	–
Not Hispanic or Latino	49 (66.22%)	–
Current Diagnoses
GAD	42 (56.76%)	49 (66.22%)
MDD	48 (64.86%)	35 (47.30%)
OCD	6 (8.11%)	6 (8.11%)
Panic disorder	4 (5.41%)	2 (2.70%)
PDD	7 (9.50%)	8 (10.81%)
PTSD	17 (22.97%)	13 (17.57%)
SAD, generalized	52 (70.27%)	51 (68.92%)
SAD, performance	53 (71.62%)	54 (72.97%)
Specific phobia	58 (78.38%)	52 (70.27%)

Values are presented as n (%).

GAD, generalized anxiety disorder; MDD, major depressive disorder, OCD, obsessive-compulsive disorder; PDD, persistent depressive disorder; PTSD, posttraumatic stress disorder; SAD, social anxiety disorder.

Psychiatric Measures

Using confirmatory factor analysis (see below), we combined the Contrast Avoidance Questionnaire-Worry (CAQ-W) (62) sum score, a self-report questionnaire that measures worry as an avoidance technique, with the agoraphobic and interoceptive avoidance items of the Panic Disorder Severity Scale (PDSS) (63) to create a measure of avoidance.b The PDSS is a self-report measure that assesses the frequency and symptoms of panic disorder. Higher scores on both CAQ-W and PDSS indicate greater symptoms.c

We selected only 2 avoidance items from the PDSS (PDSS-4 and PDSS-5) because the other items on the scale measure very different phenomena, such as frequency of panic attacks, anticipatory anxiety, distress, and social/occupational impairment (64). Notably, there is a precedent for the use of these 2 PDSS items as an avoidance measure (65). For example, previous work correlated each of the PDSS avoidance items with behavioral and interoceptive avoidance measures to independently validate these constructs on an item level (64). Other work has combined the avoidance items to make an avoidance scale (65). Therefore, these items have previously been identified as measures of avoidance, distinct from the rest of the PDSS items. In combination with the CAQ-W sum score, they provide a more well-rounded measure of avoidance than either measure alone (i.e., agoraphobic and interoceptive avoidance from the PDSS; cognitive/worry as avoidance from the CAQ-W).

Flanker Task

Participants completed a modified arrowhead version of the Eriksen flanker task (66). Presentation software displayed the task (Neurobehavioral Systems Inc.). On each trial, participants viewed 5 white arrows presented for 200 ms against a black background. Participants had up to 1800 ms to indicate the direction of the middle arrow by pressing the left or right mouse button. Additional details on the task are provided in the Supplement.

EEG Recording and Processing

Continuous EEG was recorded during the flanker task using an ActiCap and the ActiChamp amplifier system (Brain Products GmbH). Thirty-two electrode sites were used based on the 10/20 system. Additional data collection and processing details are available in the Supplement.

Correct and error trials were averaged separately, and baseline correction was performed using a 200-ms window from −500 to −300 ms before response onset. The ERN and the CRN were scored as the average activity on error and correct trials, respectively, from 0 to 100 ms after response at Fz (67). The Pe and correct positivity (Pc) were scored as the average activity on error and correct trials, respectively, from 200 to 400 ms after response at Cz (68).

To parse variance unique to the Pe versus the Pc and the ERN versus the CRN, unstandardized residuals were created by regressing each error ERP (e.g., time 1 Pe) onto the respective correct ERP (e.g., time 1 Pc) and saving the unstandardized residuals from each regression (69). In Results and Discussion, we refer to these unstandardized residuals simply as Pe and ERN. Notably, in figures and tables, we use Pe and ERN to refer to condition-specific (i.e., error and correct) values, not residuals.

Data Analyses

Descriptive statistics, correlations, and structural equation models (SEMs) were conducted in R using lavaan (version 4.3.1) (70,71). We used SEMs to test the hypothesized relationship between time 1 error processing (Pe, ERN) and time 2 avoidance and whether this relationship was moderated by a change in the ERN or Pe over the same time period.

At the first stage of the model, we specified and estimated baseline measurement models for transdiagnostic avoidance at each time point (time 1 and time 2) using the 2 avoidance items from the PDSS and the sum score from the CAQ-W scale as indicators using confirmatory factor analysis. The factor loading for 1 indicator was fixed to 1 to set the metric for the factors (time 1 avoidance and time 2 avoidance). The remaining factor loadings, factor variance, and residual errors were estimated freely.

Then, to determine whether the time 1 Pe or the time 1 ERN predicted increases in time 2 avoidance, we entered time 1 Pe, time 1 ERN, time 1 avoidance (latent variable), and the time passed in days between time 1 and time 2 visits (time passed) as simultaneous predictors of time 2 avoidance (latent variable). Covariances were allowed between indicators across time points (time 1 PDSS-4 and time 2 PDSS-4; time 1 PDSS-5 and time 2 PDSS-5; the sum score of the CAQ-W at time 1 and the sum score of the CAQ-W at time 2).

We were also interested in knowing whether a change in error processing from time 1 to time 2 would moderate prospective associations between time 1 error processing and time 2 avoidance. Therefore, we created an unstandardized residual reflecting time 2 ERN controlling for time 1 ERN and an unstandardized residual reflecting time 2 Pe controlling for time 1 Pe. These residuals, referred to hereafter as change in ERN and change in Pe, respectively, reflected the change in ERN from time 1 to time 2 and the change in Pe from time 1 to time 2.d

To assess change in ERN as a moderator of the association between time 1 Pe and time 2 avoidance (see below), we created an interaction term by multiplying time 1 Pe by change in ERN. We refer to this interaction term hereafter as time 1 Pe × change in ERN. The same steps were conducted to assess change in Pe as a moderator of the association between time 1 Pe and time 2 avoidance (time 1 Pe × change in Pe). The following predictors of time 2 avoidance were then added to the model: change in ERN, change in Pe, time 1 Pe × change in ERN, and time 1 Pe × change in Pe. The final model consisted of time 1 Pe, change in ERN, change in Pe, time 1 Pe × change in ERN, time 1 Pe × change in Pe, time 1 avoidance, and time passed as predictors of the latent variable time 2 avoidance. Significant interactions were followed up by simple slope analyses (72).

No missing data points were included in the models. We used a maximum likelihood estimator for all models. The models were evaluated using overall goodness-of-fit guidelines suggested by Hu and Bentler (73), i.e., root mean squared error of approximation (RMSEA) < 0.08 and comparative fit index (CFI) > 0.90 suggesting adequate model fit. Nonsignificant pathways were removed from the final model consistent with a trimming approach (74,75). Bivariate correlations were conducted between each of the Pe and ERN and avoidance factor scores, separately for time 1 and time 2, to test cross-sectional associations. Separate models testing the associations of accuracy, reaction time, and age with avoidance are reported in the Supplement.

Results

Task Effects

Table 2 presents means and SDs for ERPs and performance on the flanker task at time 1 and time 2 by trial type (incongruent and congruent). Reliabilities for the ERPs ranged from 0.620 to 0.979 (indicating moderate to excellent reliability) (76) and are reported in more detail in the Supplement together with t tests comparing correct and error ERP amplitudes.

Table 2.

Flanker Task Behavior

	Time 1		Time 2
Pe, μV	11.77 (5.96)		12.08 (5.96)
Pc, μV	3.07 (3.26)		3.88 (3.71)
ERN, μV	−2.15 (5.39)		−1.81 (5.28)
CRN, μV	4.04 (4.21)		4.44 (4.28)
	Congruent	Incongruent	Congruent	Incongruent
Error RT, ms	280.39 (126.06)	305.79 (54.62)	271.82 (145.61)	315.77 (56.73)
Correct RT, ms	358.49 (59.42)	432.39 (62.17)	351.51 (43.91)	422.38 (48.29)
Accuracy	95.49% (12.00%)	79.13% (11.99%)	95.73% (6.16%)	83.52% (10.26%)

Values are presented as mean (SD).

CRN, correct-related negativity; ERN, error-related negativity; Pc, correct positivity; Pe, error positivity; RT, reaction time.

Measurement Models

One-factor confirmatory models were fit to the avoidance indicators for time 1 avoidance and time 2 avoidance. Each factor only included 3 indicators making the models just identified; thus, both models were a perfect fit for the data (CFI = 1.00, RMSEA = 0.00, χ²₀ = 0.00). All indicators contributed significantly to each of the respective latent variables (i.e., ps < .05), and loadings were all >0.3 (77,78).e

Model Fit

Figure 1 depicts the final SEM with nonsignificant pathways removed by a trimming approach (74,75). The final model fit the data well (CFI = 0.99, RMSEA = 0.02, χ²₁₅ = 15.41, p = .422).

Figure 1.

Final, clarified structural equation model depicting only significant path coefficients consistent with a trimming approach. Betas depicted for relationships between observed and theorized variables. Latent variables are depicted as circles, and observed variables are depicted as rectangles. Only significant pathways are shown for brevity. Time 1 (T1) error positivity (Pe) significantly predicted time 2 (T2) avoidance. T1 Pe × change in error-related negativity (ERN) significantly predicted T2 avoidance. T1 avoidance significantly predicted T2 avoidance. CAQ-W, Contrast Avoidance Questionnaire-Worry; PDSS, Panic Disorder Severity Scale.

Time 1 Pe Predicts Time 2 Avoidance

Individuals with greater elaborative error processing at baseline (time 1 Pe) showed higher time 2 avoidance (controlling for time 1 avoidance; β = 0.20, p = .04), as depicted in Figure 2. There were no associations between earlier error processing (time 1 ERN) and time 2 avoidance (controlling for time 1 avoidance; β = 0.14, p = .16).

Figure 2.

Larger time 1 error positivity (Pe) predicts greater time 2 − time 1 avoidance. Grand-averaged time 1 waveforms at the pooling where the Pe and correct positivity were scored and time 1 error-correct voltage distributions shown separately for (A) individuals with more time 1 to time 2 increases in avoidance (high time 2 avoidance) and (B) individuals with little time 1 to time 2 increases in avoidance (low time 2 avoidance). Avoidance groups were created using a median split on time 2 minus time 1 avoidance for illustrative purposes only (all analyses were continuous). ERP, event-related potential.

Moderation by Change in the ERN

The association between time 1 Pe and time 2 avoidance was moderated by change in ERN (β = 0.35, p < .01), as depicted in Figures 3 and 4. Larger time 1 Pe amplitudes predicted greater time 2 avoidance for individuals with the greatest reduction in ERN amplitudes from time 1 to time 2 (t₇₀ = 0.23, p = .004) and individuals with medium-sized reductions in the ERN from time 1 to time 2 (t₇₀ = 0.09, p = .04). Time 1 Pe did not predict time 2 avoidance for individuals who failed to show a substantial reduction in the ERN from time 1 to time 2 (t₇₀ = 0.04, p = .49). The association between time 1 Pe and time 2 avoidance was not moderated by change in Pe (β = 0.16, p = .17). Change in Pe (β = −0.12, p = .24), change in ERN (β = 0.04, p = .67), and time passed (β = −0.08, p = .44) did not predict time 2 avoidance.

Figure 3.

Larger time 1 error positivity (Pe) only predicts greater time 2 avoidance for individuals with large and medium reductions in error-related negativity (ERN) amplitudes from time 1 to time 2. Grand-averaged time 1 error minus correct waveforms, shown separately for time 1 (red) and time 2 (black), individuals with low time 1 Pe (left column) and high time 1 Pe (right column), and individuals with low time 2 avoidance (top row); medium time 2 avoidance (middle row) and high time 2 avoidance (bottom row). Head maps depict time 2 (error − correct) minus time 1 (error − correct) voltage distributions; event-related potential (ERP) depictions are from representative single participants, for illustrative purposes only.

Figure 4.

Simple slopes depicting the relationship between time 1 error positivity (Pe) and time 2 avoidance, shown separately for individuals with the greatest reduction in the magnitude of the error-related negativity (ERN) from time 1 to time 2 (−1 SD below the mean; solid line), medium-sized reduction in the magnitude of the ERN from time 1 to time 2 (at the mean; dashed line), and little change in the size of the ERN from time 1 to time 2 (+1 SD above the mean; dotted line).

Cross-Sectional Associations Between ERPs and Avoidance

Time 2 ERN was positively correlated with time 2 avoidance (r = 0.317, p = .006), indicating that smaller (more positive) ERNs at time 2 were associated with greater time 2 avoidance, as depicted in Figure 5. By contrast, time 1 ERN was not correlated with time 1 avoidance (r = 0.175, p = .137). Likewise, the Pe was not correlated with avoidance cross-sectionally at time 1 or time 2 (0.096 < rs < 0.210, .072 < ps < .416).

Figure 5.

Smaller time 2 error-related negativities (ERNs) are associated with greater time 2 avoidance. Grand-averaged waveforms at the pooling where the ERN and correct-related negativity were scored and time 2 error-correct voltage distributions shown separately for (A) individuals with high avoidance at time 2 (high time 2 avoidance) and (B) individuals with low avoidance at time 2 (low time 2 avoidance). Avoidance groups were created using a median split on time 2 avoidance for illustrative purposes only (all analyses were continuous). ERP, event-related potential.

Discussion

Here, we aimed to test whether baseline heightened error processing (the Pe and ERN) would predict changes in transdiagnostic avoidance over 1 year in an adult mixed internalizing sample. Moreover, we sought to determine whether associations between baseline error processing and future avoidance would be moderated by changes in these ERPs over the same year. Results showed that individuals with greater baseline elaborated processing of errors (Pe) showed greater increases in self-reported everyday avoidance 1 year later. Furthermore, this association was strongest for individuals who also exhibited reductions in the ERN over the same year. A cross-sectional positive association between blunted ERNs and increased avoidance at time 2 was also observed.

Evidence of heightened error processing as a risk factor for transdiagnostic avoidance is consistent with previous work that has suggested that increased perception of threat/danger may be correlated with avoidance behaviors (16). For example, one study found that participants were more likely to avoid on high-threat trials (in a fearful avoidance task), but participants with increased threat sensitivity (indexed via skin conductance response and startle) were also more likely to avoid on safety trials (16). These findings, together with previous work showing that heightened error processing is correlated with better avoidance learning, suggest that increased threat sensitivity may be associated with greater avoidance (54, 55, 56). More specifically, our results suggest that heightened error processing (Pe) at baseline may indicate vulnerability toward greater future, everyday avoidance.

Critically, our results also showed that individuals most at risk for increased avoidance were characterized not only by heightened baseline Pe, but also by reductions in the ERN over the same year. One potential interpretation is that individuals with heightened baseline threat sensitivity (larger Pe) who also responded by avoiding (or disengaging from) threat (blunted ERN) 1 year later were more likely to also increase avoidance in everyday life. Along these lines, when participants are told to take their errors less seriously, the ERN is reduced (79) even though behavior is unchanged. Therefore, even though participants who showed a reduction in the ERN over time did not err less at time 2, they may have disengaged from error processing on a cognitive/emotional level, possibly to cope with initial error responsivity. Conversely, individuals who maintained intact levels of error processing over time, despite heightened baseline error sensitivity, appeared relatively protected from increases in everyday avoidance. Because we did not measure error avoidance per se, this potential explanation for our results requires confirmation in future studies.

Alternatively, reduced ERN amplitudes over time could indicate neuroplastic changes related to overstimulation of threat processing systems. From an evolutionary perspective, attenuated responses in threat processing neural circuitry could, in some circumstances (e.g., chronic adversity), help in evading danger/conserving resources (80, 81, 82). Supporting this, chronic stress and depression have been associated with threat reactivity deficits, as well as reductions in engagement with emotional stimuli (80, 81, 82, 83). Similarly, individuals with more severe/complex or comorbid depression symptoms show blunted fear responsivity over time potentially as an adaptation to chronic overengagement of defensive systems (80,84). In other words, participants with increased baseline error processing who go on to show increased avoidance may undergo changes in response to sustained threat sensitivity, contributing to reduced ERNs.

Notably, we did not observe a cross-sectional association at time 1 between ERN (or Pe) amplitudes and avoidance, although time 2 ERN amplitudes were observed to be blunted in individuals higher in avoidance. Although it is difficult to explain this null finding with certainty, one possibility is that participants responded differently to the task at time 2 than at time 1. That is, some participants might have engaged in avoidant error processing at time 2 only after they were familiar with the task. That is, if participants initially experienced making errors as aversive (11), some of them might have minimized error processing at time 2, when they returned to the laboratory and completed the same task again. This could explain why smaller time 2 ERNs were associated with greater avoidance at time 2 but not at time 1. Consistent with this notion, other work has suggested that participants retain memories of aversive experiences in laboratory tasks for 1 year or longer. For example, threat conditioning persists when participants are brought back to the laboratory 1 year later, especially for individuals with elevated anxiety (85,86).

Only the Pe at baseline predicted increased avoidance over 1 year, not the ERN. Although this was somewhat unexpected given previous work on the ERN in children (37, 38, 39, 40, 41, 42, 43), it suggests that individuals with more elaborated processing of errors trend toward future avoidance as a behavioral adaptation, whereas early error processing (ERN) is less relevant to cognitive and behavioral adaptation to threat (87). Furthermore, all our participants had some degree of psychopathology; thus, the Pe may be especially sensitive to predicting risk in an internalizing sample and across diagnoses, whereas the ERN has predominantly been found to predict anxiety symptomatology onset (37, 38, 39).

The existing literature is replete with mixed evidence regarding the role of error processing in internalizing disorders other than anxiety (88). For example, the ERN and Pe have been found to be elevated, blunted, or unchanged for individuals with depression (31, 32, 33). Our results suggest that more specific dimensional trajectories within internalizing psychopathology, such as avoidance, may map more cleanly onto error processing (89, 90, 91, 92). Moreover, although some forms of exogenous threat (e.g., shock) may arguably be more salient than errors, measuring individual variation in lower/more innocuous forms of threat (e.g., errors) may provide unique information regarding risk for pathophysiology and its correlates (e.g., truly threatening stimuli should scare everyone, but milder threats may be where individual differences in responding come to light) (93,94). Therefore, continued investigations of error processing as it is related to dimensional trajectories of psychopathology may clarify inconsistencies within diagnostic categories found in the current literature and may yield more fruitful understanding of trajectories of illness course than have been revealed using diagnostic categories (or dimensional measures inspired by these categories).

Limitations of our study include lack of assessment of additional variables during the year between participant visits, such as stressful life events. A better understanding of these potential moderators or mediators of changes in error processing may yield increased insight into why error processing becomes blunted over time for some individuals. In addition, more frequent assessment of error processing and avoidance throughout the year between visits could have further informed causal hypotheses (e.g., if error processing changed before, after, or concurrently with everyday avoidance). Finally, our sample composition was predominantly White women, making our results possibly less generalizable to minority groups and men.

Conclusions

Our results suggest that increased elaborative processing of errors is predictive of avoidance, especially in individuals who shift toward blunted early error processing over the same year (which may index neuroplastic changes in threat-processing regions over time). Avoidance predicts disorder course and persistence better than baseline anxiety (95,96) and impedes recovery (97). Therefore, continued work aimed at understanding the mechanisms that underlie avoidance should help advance more refined models of internalizing psychopathology and risk factors that contribute to illness course in these disorders. In the short term, continued progress in this line of work is likely to involve delineation of a mechanism that leads to the reduction in ERN amplitude over time or explains why intact error processing may serve as a protective factor against avoidance.