Trial-by-trial ERP-behavior relationships in psychosis: Between- and within-person variability in performance monitoring adjustments

Abstract

Cognitive impairment in schizophrenia, characterized by deficits in performance monitoring, predicts clinical and functional outcomes. The error-related negativity (ERN), a neurophysiological index of error detection, is reduced in psychosis, but it is unclear why this impaired error detection is not closely linked to behavioral adjustments. A possibility is that research has overrelied on examining between-person relationships of average ERN and behavior, rather than focusing on within-person, trial-by-trial changes. This study aimed to determine whether neurophysiological indices of error detection (ERN, error positivity [Pe]) predict within-person post-error behavioral adjustments in psychotic disorders and whether these relationships are weaker in people with psychosis than in controls. ERN and Pe were assessed during a flanker task in 72 participants with psychosis (PwP) and 82 healthy comparison participants. Multilevel location-scale models examined trial-by-trial changes in the relationships between event-related potentials (ERPs) and behavior (response [RTs], accuracy). Results showed that ERP-RT relationships were similar across PwP and controls. In both groups, greater within-person increases in ERN predicted longer and more variable RTs following correct trials. Larger within-person increases in Pe predicted shorter and more variable RTs following correct trials, but less variable RTs following error trials. Exploratory analyses in a subset of participants with schizophrenia showed similar effects. ERP-accuracy relationships were neither observed nor moderated by diagnostic group. Within-person ERP-behavior relationships were preserved in psychosis, indicating intact performance monitoring at the individual level. This supports performance monitoring as a transdiagnostic construct and underscores the importance of examining intra-individual variability to understand performance monitoring in psychotic disorders.

1. Introduction

Cognitive impairment is a core feature of schizophrenia and robustly predicts clinical and functional outcomes (Green, 2016; Koshiyama et al., 2021; Lepage et al., 2014). An aspect of this impairment is performance monitoring—the process of evaluating and adjusting behavior to achieve goals (Foti et al., 2012; Kirschner and Klein, 2022; Perez et al., 2012). For example, neural indices of error detection are consistently and robustly impaired in psychotic disorders (Martin et al., 2018). In healthy participants, these indices predict post-error behavioral adjustments (Dutilh et al., 2012; Li et al., 2022; Rabbitt and Vyas, 1970; Schroder and Moser, 2014), referring to changes in behavior following an error to improve performance. However, the relationship between error detection and post-error behavioral adjustment appears weak in psychotic disorders (Becerril et al., 2011; Morris et al., 2006). It remains unclear why impaired error detection in psychotic disorders does not consistently lead to maladaptive behavioral adjustments, particularly since intact post-error adjustments are sometimes observed. A possible explanation is the emphasis on between-person differences in average scores rather than on between-person differences in trial-by-trial adjustments (i.e., within-person changes) in performance monitoring indices (Kirschner and Klein, 2022; Ullsperger, 2024), where meaningful variations are collapsed. This study aims to determine whether neural indices of error detection predict within-person post-error behavioral adjustments in psychotic disorders.

Error-related negativity (ERN) and error positivity (Pe) are two event-related brain potentials (ERPs) associated with performance monitoring. ERN appears within 100 ms after an error at fronto-central scalp sites (Falkenstein et al., 1991; Gehring et al., 1993; Larson et al., 2014). It is followed by Pe, occurring 200–400 ms post-error at centro-parietal sites (Nieuwenhuis et al., 2001; Overbeek et al., 2005). Prevailing theoretical explanations suggest that ERN reflects early error detection and Pe reflects later error awareness, with both ERPs supporting post-error behavioral adjustments (Danielmeier and Ullsperger, 2011; Overbeek et al., 2005). Although findings on Pe group differences are mixed (Foti et al., 2016; Kirschner and Klein, 2022), reduced ERN is consistently observed in psychosis (Martin et al., 2018).

Effective performance monitoring involves adapting behaviors to achieve goals, which is evident in trial-by-trial behavioral adjustments during speeded forced-choice response tasks. Response times (RTs) are typically longer following errors—known as post-error slowing (PES)—reflecting a shift toward prioritizing accuracy (Dutilh et al., 2012; Rabbitt and Vyas, 1970). In healthy participants, larger average ERN and Pe predict longer PES and improved post-error accuracy (PEA; Carp and Compton, 2009; Hajcak et al., 2003; Schroder et al., 2020; Steinhauser and Andersen, 2019), referring to the increased likelihood of a correct response following an error. Considering ERN deficits in psychosis (Martin et al., 2018), abnormal PES and PEA would be expected; however, some studies show PES appears intact (Bates et al., 2002; Kirschner and Klein, 2022; Perez et al., 2012; Polli et al., 2006), while others show impairment (Alain et al., 2002; Becerril et al., 2011). To our knowledge, studies to date have not examined PEA in psychosis (Kirschner and Klein, 2022).

The analysis of within-person, trial-by-trial changes in ERN might uncover relationships between ERN and PES/PEA that are obscured in between-person analyses. Studies have typically emphasized between-group/–person differences in average scores, which might mask intact within-person changes in performance monitoring in psychosis. In other words, performance monitoring might be preserved at the individual level in psychotic disorders, despite abnormal responses observed at the group level.

The first aim of the present study was to determine whether diagnostic status moderates ERP-behavior relationships—the associations between neural responses to errors and behavioral adjustments—in psychotic disorders. We expected that the participants with psychotic disorders (PwP) would exhibit weaker between-person ERP-behavior relationships than healthy comparison participants, consistent with blunted neural response to errors (Martin et al., 2018). The second aim was to determine whether within-person ERP-behavior relationships are preserved in psychosis. We predicted these relationships would be intact, indicating preserved performance monitoring at the individual level (Morris et al., 2006). Exploratory analyses examined a subset of participants with schizophrenia.

2. Methods and materials

The present study adheres to open science practices (Garrett-Ruffin et al., 2021; Paul et al., 2021). A project repository on Open Science Framework (OSF) serves as a hub with links to data processing and analysis code (https://osf.io/qyjs5/). Present analyses were performed on publicly available data that are posted to the NIMH Data Archive (https://nda.nih.gov/edit_collection.html?id=2477) for the “Trajectories of Aging in Psychotic Disorders Over 27 Years” NIMH grant awarded to Roman Kotov, PhD (1R01MH110434–01). Between- and within-person relationships between previous-trial ERN/Pe and subsequent behavioral adjustments have not been examined in this dataset.

2.1. Participants

Study participants were drawn from the Suffolk County Mental Health Project (Bromet et al., 2011; Foti et al., 2021; Tramazzo et al., 2024), an epidemiologic longitudinal study of first-admission psychosis. Participants were initially recruited between 1989 and 1995 from inpatient psychiatric facilities in Suffolk County, New York, USA. Participant data from the 25-year follow-up wave were used for this study. Inclusion criteria for present analysis included 1) diagnostic information verified by clinical interview (see Section 2.2), 2) at least five post-correct trials and five post-error trials (see supplement for justification), and 3) above chance accuracy during the flanker task (see Section 2.4). Therefore, final sample sizes included 72 PwP (schizophrenia-spectrum disorders, mood disorders with psychosis, and other psychotic disorders) and 82 without history of psychosis (similar to cases on zip code and demographics).

2.2. Clinical assessments

Psychiatric diagnoses were assessed using the Structured Clinical Interview for DSM-IV Axis I Disorders – Patient Edition (SCID-I/P; First et al., 1996). Symptom levels were assessed using the Scale for the Assessment of Positive Symptoms (SAPS; Andreasen, 1984a) and the Scale for the Assessment of Negative Symptoms (SANS; Andreasen, 1984b). Composite scores for Disorganization, Reality Distortion, Apathy/Asociality, and Inexpressivity were used for analysis (Kotov et al., 2016). The Social and Occupational Functioning Assessment Scale (SOFAS) was used as a measure of daily functioning (American Psychiatric Association, 2000). Clinical information is presented in Table 1 to characterize the sample, but scores were not used in further analysis.

Table 1.

Demographic, diagnostic, BPRS, and SAPS/SANS summary statistics by group.

	PwP (n = 72)		Control group (n = 82)
	M	SD	M	SD
Age (years)	52	9	57	10
Sex	n	%	n	%
Female	32	44.44 %	26	31.71 %
Male	40	55.56 %	56	68.29 %
Race
White	63	87.5 %	73	89.02 %
Black/African American	7	9.72 %	7	8.54 %
Asian	1	1.39 %	1	1.22 %
Unknown or more than one race	1	1.39 %	1	1.22 %
Diagnosis
Schizophrenia	18
Schizoaffective Disorder	10
Bipolar I	25
Other Bipolar	2
Substance Induced Psychotic Disorder	5
Psychotic Disorder NOS	1
Major Depressive Disorder	7
Depressive Disorder NOS	3
SOFAS	49	16.3	71.7	13.7
BPRS
Affect	8	4.0	6	2.9
Positive Symptoms	7	4.2	4	0.9
Negative Symptoms	6	3.0	5	1.5
Resistance	6	2.9	4	1.2
Activation	4	1.1	4	1.0
SAPS/SANS
Inexpressivity	8	9.1	2	3.8
Reality Distortion	3	5.3	0	0.3
Apathy/Asociality	13	7.7	5	6.2
Thought Disorder	5	5.4	2	2.6

Note: PwP = Participants with psychosis; NOS = Not otherwise specified; SOFAS = Social and Occupational Functioning Assessment Scale; BPRS = Brief Psychiatric Rating Scale; SAPS/SANS = Scale for the Assessment of Positive and Negative Symptoms.

2.3. Experimental task

A modified Eriksen flanker task (Eriksen and Eriksen, 1974), as described in Foti et al. (2012) and in the supplement, was used. Briefly, each trial presented five horizontally aligned white arrowheads. In congruent trials, all arrows pointed in the same direction; in incongruent trials, the central arrow pointed opposite to the others. Participants responded by pressing the mouse button matching the central arrow’s direction. They completed 11 blocks of 330 trials and received performance-based feedback after each block.

2.4. Electrophysiological data recording and reduction

EEG data collection and processing are briefly described here, with complete details provided in the supplement. Continuous EEG was recorded at a sampling rate of 1024 Hz from 34 scalp electrodes, including FCz and Iz. Two additional electrodes were placed on the left and right mastoids. Electrooculogram was recorded from additional sensors placed above and below the eyes and near the outer canthi.

Data were rereferenced to averaged mastoids, filtered (0.01 to 30 Hz), and segmented around participant responses (−400 to 800 ms). Ocular artifacts were removed using independent components analysis. Bad channels were then identified and interpolated. The first 200 ms was used for baseline adjustment. Single-trial ERN (0 to 100 ms at FCz) and single-trial Pe scores (200 to 400 ms at Pz) time-window mean amplitudes were extracted. Participants with fewer than five post-correct/–error trials or below-chance accuracy were excluded (see supplement for justification).

Subject-level internal consistency estimates were examined to characterize data quality relative to between-person differences (Clayson, 2024a; Clayson and Miller, 2017b). Coefficients of dependability were estimated using generalizability theory equations to characterize internal consistency (Baldwin et al., 2015; Brennan, 1992; Clayson et al., 2021a; Clayson et al., 2021b; Shavelson and Webb, 1991). Dependability as a function of the number of trials needed for a stable average ERN and Pe score was separately estimated for each event type and group using the ERP Reliability Analysis (ERA) Toolbox (Clayson et al., 2021c; Clayson and Miller, 2017a). For PwP, subject-level internal consistency estimates were on average 0.98 (SD = 0.01) for correct-related negativity (CRN), 0.78 (SD = 0.10) for ERN, 0.97 (SD = 0.02) for correct positivity (Pc), and 0.83 (SD = 0.09) for Pe. For controls, subject-level internal consistency estimates were on average 0.99 (SD = 0.01) for CRN, 0.85 (SD = 0.08) for ERN, 0.97 (SD = 0.01) for Pc, and 0.78 (SD = 0.11) for Pe.

2.5. Data analysis

Multilevel location-scale models (MLSMs) were used to examine trial-level changes in neural and behavioral indices within individuals, accommodating the nested structure of trials within persons. Unlike traditional multilevel models that estimate only conditional means (location), MLSMs simultaneously estimate fixed and random effects for both means and variances (scale) (Walters et al., 2018). The approach facilitates analysis of within-person variance influenced by categorical and dimensional covariates (e.g., individual differences, event type, psychiatric symptoms). MLSMs have been successfully applied to studies of ERPs (Clayson et al., 2022, 2024; Holbrook et al., 2025a, 2025b; Park et al., 2024). In the present study, MLSMs were used to model RTs, and traditional multilevel location-only models with a Bernoulli distribution and logit link were used to model accuracy, accommodating the binary nature of the dependent variable (0, incorrect; 1, correct).

The first analysis examined ERP-behavior relationships irrespective of diagnostic status. Base models predicted current-trial RTs or accuracy from previous-trial accuracy (correct, error), person-specific mean previous-trial ERP (person-specific mean ERN [psmERN]/ person-specific mean Pe [psmPe]), person-specific difference previous-trial ERP (person-specific difference in ERN [psdERN]/ person-specific difference in Pe [psdPe]), and the three-way interaction (see supplement for full description of models with notation). Person-specific mean ERP scores (psmERN, psmPe) represent a subject’s own average amplitude (constant within person), while person-specific differences (psdERN, psdPe) reflect deviations from this mean on each trial. Therefore, psmERN and psmPe characterize between-person relationships (e.g., do participants with larger average ERN exhibit shorter RTs?), whereas psdERN and psdPe characterize within-person relationships (e.g., do trials with larger ERN relate to shorter RTs within a person?). Random intercepts were included for participants, and random slopes were included for the Previous-Trial Accuracy × Person-Specific Difference Previous-Trial ERP interactions.

The second analysis examined whether diagnostic status moderated ERP-behavior relationships. Diagnostic status was added to the base models, along with four-way interactions involving each ERP component (i.e., psmERP, psdERP) and previous-trial accuracy. Three models were used to test the hypotheses. Model 1 was the base ERP model that included previous-trial accuracy, person-specific mean previous-trial ERPs, and person-specific difference previous-trial ERPs. Model 2 included all predictors from Model 1 and the main effect of diagnostic status and interaction effects with diagnostic status. Model 3 included only Previous-Trial Accuracy and diagnostic status, and Model 3 was used as a comparison model to test if inclusion of ERP components improved the prediction of behavioral changes. The same predictors were included on location and scale portions of the models predicting RT. The three models used in the main analysis were also used in the exploratory analyses for a subset of participants with schizophrenia.

2.6. Model estimation

Bayesian multilevel models were fit using R v4.3.1 (R Development Core Team, 2023) with the package brms v2.20.4 (Bürkner, 2017), a front-end wrapper for Stan (Stan Development Team, 2021). Details of the priors are provided in the supplementary material. Model fit was compared using leave-one-out cross-validation via Pareto smoothed importance sampling (PSIS-LOO; Vehtari et al., 2017) using the loo package (Vehtari et al., 2017). A better-fitting model was identified if the difference in expected log predictive density (${\widehat{\mathit{\text{elpd}}}}_{loo}$) exceeded 4 and was greater than twice the standard error of ${\widehat{\mathit{\text{elpd}}}}_{loo}$. If these criteria did not single out a model, the principle of parsimony was applied among models satisfying the criteria.

3. Results

PwP were younger than controls,¹ t(151.62) = 2.94, p < .01, and gender distribution was similar across groups (χ²(1) = 2.13, p = .14). Summary demographics for each group are presented in Table 1, and ERP and behavioral data for each group are presented in Table 2. Grand average waveforms and topographical maps are shown in Fig. 1.

Table 2.

Event-related potential amplitude (μV), response times (RTs; ms), and accuracy data for each group.

		PwP		Controls
		M	SD	M	SD
ERN	Correct	2.3	4.7	7.2	6.7
	Error	1.1	4.5	2.1	6.7
Pe	Correct	0.2	3.6	2.9	3.7
	Error	3.5	5.3	7.6	5.7
RT	Post-correct	560	111	505	93
	Post-error	610	173	540	122
Accuracy	Post-correct	0.87	0.10	0.91	0.05
	Post-error	0.82	0.19	0.92	0.09

Note. PwP = Participants with psychosis; Correct-trial ERN and Pe refer to correct-trial counterpart of these ERP components: the correct-related negativity (CRN) and correct positivity (Pc), respectively.

Fig. 1.

Grand average response-locked waveforms at FCz for error-related negativity (ERN; Panels A & B) and for error positivity (Pe; Panels C & D). Difference waveforms represent error minus correct activity (Panels B and D). Shaded regions represent time-windows used for scoring. PwP = participants with psychosis.

3.1. Response times

3.1.1. Model comparison

Model comparisons indicated that the base ERP model without diagnostic status (Model 1) showed better fit for RT data than the model with diagnostic status (Model 2) or the model without previous-trial ERPs (Model 3). Therefore, modeling clinical group differences² did not improve the predictive accuracy of the models, but the inclusion of previous-trial ERPs did. These findings provide evidence of similar ERP-RT relationships across people with psychotic disorders and healthy comparison participants.

A brief note on the abbreviations: “psm” stands for “person-specific mean”, representing each individual’s average ERP amplitude across all trials (e.g., psmERN for the average ERN amplitude, psmPe for the average Pe amplitude). “Psd” stands for “person-specific difference,” which reflects the deviation of the ERP amplitude on a trial from the individual’s mean (e.g., psdERN for ERN deviations, psdPe for Pe deviations).

3.1.2. Location portion

Parameter estimates for the model predicting RTs are shown in Table 3, and those parameters that predicted the score averages are interpreted below. Previous-trial accuracy predicted current-trial RTs, with longer RTs following error trials than correct trials (95 % credible interval [CrI]: −43.22, −20.67 ms). Larger psmERN (i.e., more negative) predicted longer RTs on the following trial (95 % CrI: −9.35, −3.41), and larger psdERN also predicted longer RTs (95 % CrI: −1.09, −0.37). A psmPe × Previous-Trial Accuracy interaction was observed (95 % CrI: 0.46, 5.47). Larger psmPe predicted shorter RTs on the following trial, and this relationship was stronger for post-error trials (95 % CrI: −14.33, −3.65) than for post-correct trials (95 % CrI: −10.39, −1.61).

Table 3.

Estimates from location-scale multilevel model predicting RTs.

	Estimate	SE	95 % CrI
Location portion
Intercept	605.12	11.02	583.38, 626.87
Previous-Trial Accuracy	−31.77	5.74	−43.22, −20.67
Subject-Mean Previous-Trial ERN (psmERN)	−6.08	1.70	−9.35, −2.74
Subject-Mean Difference Previous-Trial ERN (psdERN)	−0.55	0.33	−1.20, 0.09
Subject-Mean Previous-Trial Pe (psmPe)	−8.93	2.70	−14.33, −3.65
Subject-Mean Difference Previous-Trial Pe (psdPe)	−0.33	0.30	−0.91, 0.25
Previous-Trial Accuracy * psmERN	−0.60	0.80	−2.16, 0.97
Previous-Trial Accuracy * psdERN	−0.35	0.34	−1.01, 0.31
psmERN * psdERN	0.05	0.03	−0.01, 0.11
Previous-Trial Accuracy * psmPe	2.92	1.26	0.46, 5.47
Previous-Trial Accuracy * psdPe	0.27	0.31	−0.35, 0.88
psmPe * psdPe	−0.003	0.05	−0.11, 0.10
Previous-Trial Accuracy * psmERN * psdERN	−0.02	0.03	−0.08, 0.05
Previous-Trial Accuracy * psmPe * psdPe	0.004	0.06	−0.11, 0.11
Scale portion (SD)
Intercept	5.17	0.05	5.07, 5.27
Previous-Trial Accuracy	−0.17	0.04	−0.25, −0.1
Subject-Mean Previous-Trial ERN (psmERN)	−0.03	0.009	−0.04, −0.01
Subject-Mean Difference Previous-Trial ERN (psdERN)	0.002	0.002	−0.002, 0.006
Subject-Mean Previous-Trial Pe (psmPe)	−0.04	0.013	−0.07, −0.02
Subject-Mean Difference Previous-Trial Pe (psdPe)	−0.005	0.002	−0.01, −0.001
Previous-Trial Accuracy * psmERN	−0.001	0.007	−0.014, 0.01
Previous-Trial Accuracy * psdERN	−0.006	0.002	−0.01, −0.002
psmERN * psdERN	−0.0001	0.0002	−0.001, 0.0003
Previous-Trial Accuracy * psmPe	0.02	0.01	0.004, 0.043
Previous-Trial Accuracy * psdPe	0.007	0.002	0.003, 0.01
psmPe * psdPe	0.0005	0.0004	−0.0003, 0.001
Previous-Trial Accuracy * psmERN * psdERN	0.0003	0.0002	−0.0001, 0.0008
Previous-Trial Accuracy * psmPe * psdPe	−0.0008	0.0004	−0.002, 0.0001

Note: Estimates of parameters represent the median, and parameters in standard deviations (SD) units are shown on a log scale. SE = standard error; 95 % CrI = 95 % credible interval; ERN = error related negativity; Pe = error positivity.

In summary, a PES effect was observed. Larger psmERN and psdERN predicted longer RTs, while larger subject-average Pe predicted shorter RTs on the next trial. Overall, between-person differences predicted trial-by-trial changes in average RTs.

3.1.3. Scale portion

The scale portion of the model describes relationships between predictors and within-person variability of RTs. Previous-trial accuracy predicted greater RT variability, with post-error RTs being more variable than post-correct RTs (95 % CrI: −0.25, −0.10). For ERN, larger psmERN predicted greater within-person variability of RTs (95 % CrI: −0.04, −0.01). A psdERN × Previous-Trial Accuracy interaction was observed (95 % CrI: −0.01, −0.002). Greater psdERN predicted greater post-correct trial variability of RTs (95 % CrI: −0.006, −0.002), but not post-error trial variability (95 % CrI: −0.002, 0.006).

For Pe, there was a psmPe × Previous-Trial Accuracy interaction (95 % CrI: 0.004, 0.04), such that larger psmPe predicted smaller within-person variability of the post-error RTs (95 % CrI: −0.07, −0.02) than of correct-trial RTs (95 % CrI: −0.04, −0.001). There was also a psdPe × Previous-Trial Accuracy interaction (95 % CrI: 0.003, 0.01). Larger psdPe predicted greater post-correct trial variability of RTs (95 % CrI: 0.00002, 0.003), but an opposite pattern was observed for post-error trials—larger psdPe predicted smaller post-error variability of RTs (95 % CrI: −0.009, −0.001).

In summary, previous-trial accuracy predicted greater RT variability, with more variability following error trials than correct trials. Larger psmERN was associated with increased overall within-person RT variability, and larger psdERN predicted higher RT variability following correct trials. For Pe, larger psmPe generally predicted smaller RT variability; however, whereas larger psdPe on a previous correct trial predicted greater RT variability, larger psdPe following an error trial predicted smaller RT variability. Therefore, between-person and within-person differences in ERP components amplitudes predicted changes in RT variability.

3.1.4. Interim summary

Considering that Model 1 showed the best fit for the data and ERP-RT relationships were observed, ERP-RT relationships were not weaker in people with psychotic disorders than healthy comparison subjects. Therefore, hypothesis 1 was not supported. However, within-person differences in RT variability were successfully predicted by within-person changes in ERP component amplitudes. Therefore, these findings support hypothesis 2.

3.2. Accuracy

3.2.1. Model comparisons

When accuracy was the dependent variable, Model 3 showed the best fit, suggesting that the inclusion of previous-trial ERPs in the model did not improve prediction of accuracy above and beyond diagnostic status. Model 3 is reported below, and Model 1 is reported in the supplementary material for the sake of comparability to RT data.

3.2.2. Location portion

Parameter estimates for the accuracy model are presented in Table 4. There was a Group × Previous-Trial Accuracy interaction (95 % CrI: 0.21, 0.90). Only PwP showed a post-error change in accuracy (95 % CrI: −0.58, −0.07), with PwP showing worse accuracy following error trials. Accuracy was similar across post-correct and post-error trials for controls (95 % CrI: −0.04, 0.50). Controls showed higher accuracy than PwP following correct trials (95 % CrI: 0.07, 0.50) and error trials (95 % CrI: 0.42, 1.24).

Table 4.

Estimates from location-only multilevel model predicting accuracy.

Predictor	Estimate	SE	95 % CrI
Location portion
Intercept	2.61	0.16	2.31, 2.93
Previous-trial accuracy	−0.22	0.14	−0.50, 0.04
Diagnostic group	−0.84	0.21	−1.24, −0.42

Note: Estimates of parameters represent the median. SE = standard error; 95 % CrI = 95 % credible interval.

3.2.3. Interim summary

Model 3 showed the best fit for the data, indicating that diagnostic group was a better predictor of current-trial accuracy than previous-trial ERP amplitudes. Therefore, our hypotheses were not supported for accuracy data, as ERP-accuracy relationships were neither observed nor moderated by clinical status.

4. Discussion

The present study investigated trial-by-trial changes in ERP-behavior relationships in psychosis to determine whether diagnostic status moderates ERP-behavior relationships and whether performance monitoring is intact at the individual level. The intact link between error-monitoring ERPs and subsequent behavioral adjustments complements previous findings on group-level differences observed at the neural level (Foti et al., 2021; Holbrook et al., 2025a, 2025b). Furthermore, deviations from an individual’s average ERP predicted their RTs on the following trial similarly across groups. This finding supports the notion of intact within-person changes in ERP-behavior relationships in psychosis, indicating preserved performance monitoring at the individual level (Morris et al., 2006)—a person’s own relative ERP size predicted a person’s own relative change in RTs. Exploratory analyses focusing only on participants with schizophrenia revealed similar findings (see supplement).

The relationships between error-monitoring ERPs and subsequent behavior align with findings in healthy participants, where larger ERN predicted longer and more variable RTs and larger Pe predicted shorter and less variable RTs (Mattes et al., 2022; Park et al., 2024). Generally speaking, between-person differences in ERP component scores (i.e., psmERN, psmPe) showed numerically stronger relationships with subsequent behavior than within-person changes in amplitudes (i.e., psdERN, psdPe). Considering that variable neural activity is often considered maladaptive (Clayson et al., 2022; Dinstein et al., 2015; Kebets et al., 2021), the association between a larger ERN and longer, more variable RTs—without improved accuracy—potentially supports the bottleneck account of post-error slowing (Lavro et al., 2018). This contrasts with the idea of a strategic adjustment in the speed-accuracy trade-off (Dutilh et al., 2012; Wickelgren, 1977), which proposes that reactivity to an error, or the potential for one, may tax early attentional processes during the subsequent trial, leading to longer RTs without improving post-error accuracy (Notebaert et al., 2009; Wessel, 2018).

Conversely, the relationship between larger Pe and shorter, less variable RTs on the subsequent trial may reflect the exertion of cognitive control following conscious error processing (Mattes et al., 2022; Park et al., 2024). The similar pattern of ERP-RT relationships observed across PwP and controls suggests performance monitoring is a transdiagnostic construct that is more sensitive to individual differences than to traditional diagnostic group differences. Although speculative, this within-person approach might provide a more proximal index of the actual neural processes or constructs of interest than constituent average RT or ERN scores alone. Taken together, these findings suggest that examining within-person response dynamics could provide a direct measure of the neural processes that support performance monitoring, making this approach particularly useful for relating performance-monitoring measures to other clinical, cognitive, or functional outcomes and for evaluating intervention responses.

Consistent with our second hypothesis, previous-trial ERPs predicted current-trial RTs, supporting preserved trial-by-trial adjustments of ERP-behavior relationships. While studies reported inconsistent findings on post-error behavioral adjustments in psychosis (Alain et al., 2002; Araki et al., 2013; Bates et al., 2002; Becerril et al., 2011; Perez et al., 2012), the present trial-level (i.e., within-person) analysis indicated that PwP showed comparable levels of behavioral adjustment across trials, and these changes were similarly predicted by previous-trial ERPs across groups. Thus, within-person analysis demonstrated intact trial-by-trial adjustment of ERP-RT relations in psychosis, despite the diminished neural response to errors (Foti et al., 2016; Holbrook et al., 2025b). Notably, when a traditional repeated measures analysis of variance was used, clinical group differences in average RTs were observed (see footnote ¹). However, the corresponding MLSM showed group differences only in variability of RTs, not averages (see supplement). This discrepancy underscores the importance of using analytical approaches that model intraindividual variability to examine performance monitoring in psychosis. It highlights how focusing solely on average scores can miss the nuances of performance monitoring, such as intraindividual differences in variability of responses. Present findings also have an important methodological implication: failing to account for between-person differences in intraindividual variability of behavioral responses can lead to biased statistical estimates and erroneous inferences (see also Clayson et al., 2022; Leckie et al., 2014; Walters et al., 2018).

PwP showed worse accuracy on post-error trials than post-correct trials. Although speculative, PwP may have been more strongly influenced than controls by the unexpected and infrequent event of an error, impairing adjustments in attention. However, controls did not show post-error accuracy adjustments, possibly indicating that attentional changes from the previous-error trial to the current trial remained unaffected in controls. The use of computational approaches, such as drift-diffusion modeling (DDM), could test this notion by decomposing current-trial RT into distinct processes involved in decision-making (Ratcliff, 1978; Ratcliff and McKoon, 2008). DDM is a computational approach that models decision-making, helping to distinguish cognitive processes involved in speeded forced-choice tasks. DDM could elucidate between-person differences in cautiousness and efficiency of evidence accumulation, helping to understand how previous-trial characteristics and processes impact performance.

The study had limitations. First, the PwP group was chronically ill, making it unclear whether the present findings would generalize to earlier stages of psychotic disorders. Second, some PwP were on anti-psychotic drugs or other medications, which could have influenced ERP findings (Joshi et al., 2023). Third, the present study examined ERP-behavior relationships recorded during one task, but ERP-behavior relationships vary across different performance-monitoring tasks—present findings may not generalize to data from other tasks (e.g., Go/Nogo and Stroop; Park et al., 2024). To clarify how task features influence ERP-behavior relationships, future work could incorporate multiverse analyses that systematically compare task-specific effects (Clayson, 2024b). Nonetheless, the present study had many strengths, including successful identification of trial-by-trial changes in ERP-behavior relationships in psychosis and analysis of a large sample for an ERP study (Clayson et al., 2019). Furthermore, the study’s methodology is likely widely applicable to other commonly studied measures, providing a novel approach that might change interpretations of widely studied effects at the average score level.

Present findings indicate that performance-monitoring ERP-RT relationships are preserved in psychosis and in schizophrenia. Importantly, the use of a transdiagnostic sample of participants with broadly defined psychotic disorders enhances the generalizability of these findings across the spectrum of psychosis. This transdiagnostic approach is a strength and highlights performance monitoring as a transdiagnostic dimensional construct. Modeling within- and between-person variances identified an intact functional link between ERP component amplitudes and subsequent RT adjustments. This preserved coupling suggests that performance-monitoring processes can support behavioral adjustments in psychosis, despite blunted ERN in this cohort (Foti et al., 2021). Although not yet clinically actionable, these findings can help refine pathophysiological models by identifying preserved circuitry. Additionally, exploratory analyses indicated that diagnostic group differences in intraindividual variability of RTs erroneously appear as group differences in average RTs when intraindividual variability is not modeled. Therefore, failing to consider intraindividual variability may lead to misleading inferences. Future research might consider using MLSMs to examine the variability of ERP-behavior relationships in other forms of psychopathology using different tasks and employing DDM to elucidate differences in the relationship between neural responses and decision-making processes.

Appendix A. Supplementary data

Supplementary data to this article can be found online at https://doi.org/10.1016/j.schres.2025.06.010.