Why compulsions persist: An ecological momentary assessment study of the reinforcement of compulsions

Abstract

Background:

The cognitive-behavioral model (CBM) for obsessive-compulsive disorder (OCD) posits that compulsions are negatively reinforced by providing temporary relief from distress, removal of unwanted thoughts, and reinforcement of obsessive beliefs. However, research has yet to test these assumptions using longitudinal and ecological methods.

Method:

We examined the CBM using an intensive-repeated measures ecological-momentary assessment design. Participants with OCD (n = 45) completed items assessing momentary compulsion severity, anxiety, unwanted thoughts, and perceived threat 7×/day for 7 days.

Results:

Greater anxiety, unwanted thoughts, and perceived threat immediately before compulsions were associated with greater compulsion severity. Greater compulsion severity was also associated with greater reduction from before to after compulsions of anxiety and was nonlinearly associated with unwanted thoughts, such that both large increases and decreases in unwanted thoughtswere associated with greater compulsion severity. All levels of anxiety reduction and no change or smaller (relative to larger) increases in anxiety from before to after a compulsion were associated with a greater likelihood of compulsions at the next timepoint.

Conclusions:

This is the first study to provide ecological validity to several aspects of the CBM for OCD. Anxiety reduction and avoidance of a negative emotional contrast (no change or small increase in anxiety from before to after the compulsion) were significant predictors of future compulsions, and were the only longitudinal reinforcers of compulsions. Findings underscore the importance of addressing anxiety tolerance in treatment and show the first evidence for a novel contrast avoidance effect in the reinforcement of compulsions.

Obsessive-compulsive disorder (OCD) is a chronic condition marked by repetitive, intrusive thoughts (i.e., obsessions) and ritualistic, compensatory behaviors (i.e., compulsions; American Psychiatric Association, 2022). Often leading to substantial psychosocial and functional impairment (Markarian et al., 2010), OCD rarely remits without treatment (Melkonian et al., 2022). Even with treatment, one review found that only 38% met recovery criteria (Fisher et al., 2020). Understanding causal and maintenance factors may inform a better understanding of OCD and its treatment.

One of the most widely accepted theories for the rise and maintenance of OCD is the cognitive-behavioral model (CBM). This model suggests that obsessions arise from negative interpretations or appraisals of intrusive thoughts. Compulsions are theorized to temporarily relieve anxiety, remove unwanted thoughts, and reinforce maladaptive beliefs (e.g., overestimation of threat), which negatively reinforces the perceived utility of compulsions (Rachman, 1998a, 1998b; Salkovskis, 1985, 1989, 1999). As a result of reinforcement, compulsions become persistent and excessive (Abramowitz et al., 2009; Salkovskis, 1985, 1989). For example, contamination fears may lead to excessive handwashing compulsions. Handwashing temporarily reduces contamination threat, unwanted thoughts, and reinforces the belief that excessive handwashing was warranted to avoid threat.

Despite its wide acceptance, research has revealed several contradictions to CBM assumptions. For example, in an event contingent ecological momentary assessment (EMA) study, Dean and Purdon (2021) found that only 5% of compulsive episodes were reported to have been terminated due to adequate reduction of distress or obsessional thoughts. Other daily dairy studies found that some types of compulsions were followed by more reduction in distress, guilt, and responsibility than others (Bouvard et al., 2020; Bucarelli and Purdon, 2015). Also ifn a daily diary study, a sense of “certainty”, which may be associated with one's threat perception, was only achieved in approximately half of compulsive episodes (Bucarelli and Purdon, 2015).

Thus, although the CBM emphasizes that compulsions serve to temporarily remove or reduce unwanted experiences (i.e., unwanted thoughts, anxiety), most existing data suggests that they do not always result in this relief (e.g., Bucarelli and Purdon, 2015; Dean and Purdon, 2021). Clinical anecdotes also support this case, as individuals with OCD often describe compulsive episodes as being filled with exhaustion, grief, and stress. Therefore, it is likely that compulsions are not solely reinforced under circumscribed conditions (e.g., by reducing anxiety), but rather through a variety of mechanisms.

The contrast avoidance model (CAM; Newman and Llera, 2011) may provide an alternative understanding of this reinforcement process. CAM theorizes that those with some mental health conditions, such as OCD, are more motivated than those without these conditions to engage in pathological behaviors such as compulsions because they increase the likelihood of experiencing a positive emotional contrast (PEC; a sharp decrease in negative emotions) and decrease the probability of experiencing a negative emotional contrast (NEC; a sharp increase in negative emotions). The theory further suggests that either the experience of a PEC or avoidance of an NEC reinforce and maintain the pathological behavior. For example, an individual engages in an obsession, experiences an increase in anxiety, and performs a compulsion to nullify this anxiety. If the compulsion is perceived to have been sufficient in nullifying the threat, individuals will experience a reduction in anxiety or a PEC from before to after the compulsion. This PEC reinforces the compulsion. This is the same reinforcement process as outlined in the CBM. However, the CAM extends this by also theorizing that compulsions can be reinforced by avoidance of a sharp increase in negative emotion (or avoidance of an NEC). That is, compulsions may sometimes lead to either no change or increased anxiety, especially if they are viewed as ineffective or only partially effective in nullifying the threat. It is therefore also possible that even an increase in anxiety (that might occur from a compulsion perceived as only partially nullifying the threat) may still be reinforcing via the assumption that one avoided a much larger increase in anxiety. Thus, CAM suggests that compulsions may be reinforcing either when they lead to anxiety reduction (i.e., a PEC) or were perceived to reduce the degree of anxiety increase that would have happened if the compulsions were entirely ineffective (i.e., avoiding an NEC). However, no known study has examined this as a reinforcing mechanism of compulsions in OCD.

Only one study to date examined the CAM in OCD, finding that individuals with probable OCD were more likely than a non-OCD control group to report being motivated to engage in these contrast avoidance behaviors (Swisher and Newman, 2025). However, no known study has examined how relative increases and decreases in unwanted experiences (e.g., anxiety) reinforce and maintain compulsions. Refining our understanding of the reinforcement of compulsions in terms of the CAM is important, as existing theoretical models have failed to elucidate how compulsions are reinforced despite not consistently leading to a decrease in unwanted experiences. Should both increases and decreases in unwanted experiences predict future compulsions, it would highlight important adaptations needed to improve current conceptualizations and treatment frameworks for OCD.

Lastly, despite the CBM outlining temporal relationships among hypothesized reinforcers and OCD symptoms (e.g., greater distress reduction reinforces compulsions), prior studies have relied heavily on retrospective reports and cross-sectional data (e.g., Foa et al., 2002; Wheaton et al., 2010). This is problematic, considering that OC symptoms have been found to fluctuate throughout the day (Nota et al., 2014) and were broadly context-dependent, exacerbated by factors such as sleep (Schubert et al., 2020) and stress (Adams et al., 2018). Additionally, most studies on OCD used total score severity measures of combined obsessions and compulsions, limiting the ability to determine how the theorized reinforcing mechanisms (e.g., maladaptive core beliefs) uniquely reinforced obsessional relative to compulsive symptoms within the CBM.

EMA functions to address these limitations by providing a robust assessment of temporal fluctuations in symptoms in one's natural context. This may be particularly relevant to OCD, wherein prior research showed impaired retrospective memory (Bucarelli and Purdon, 2015; MacLaren Kelly et al., 2019), and overestimated symptoms relative to EMA (Gloster et al., 2008; MacLaren Kelly et al., 2019; Tilley and Rees, 2014). Thus far, EMA studies on OCD have elucidated important findings on the characterization of compulsive behavior (Bouvard et al., 2020; Bucarelli and Purdon, 2015; Dean and Purdon, 2021), such as frequency of obsessions and thought suppression (Jaeger et al., 2024; Purdon et al., 2007), retrospective biases (Gloster et al., 2008; MacLaren Kelly et al., 2019; Tilley and Rees, 2014), and therapeutic change (e.g., Radomsky et al., 2020; for a full review see Braga et al., 2025). However, the current study is the first to examine predictors of momentary and subsequent compulsions using EMA.

The present study examined temporal associations and naturalistic reinforcers of compulsions using EMA to test the CBM and examine the possibility of a contrast avoidance effect. Specifically, in individuals diagnosed with OCD, we examined the following questions: (1) Does experiencing greater anxiety, unwanted thoughts, and perceived threat immediately prior to a compulsion predict greater compulsion severity?; (2) Is performing more severe compulsions associated with greater increases or decreases in anxiety, unwanted thoughts, and perceived threat from before to after compulsions? (3) Does experiencing greater increase or decreases in anxiety, unwanted thoughts, and perceived threat from before to after a compulsion predict engaging in a compulsion at the next time point? Based on the CBM, we hypothesized that greater anxiety, unwanted thoughts, and perceived threat immediately prior to a compulsion would be associated with more severe compulsions. We also hypothesized that reductions in these experiences from before to after a compulsion would be associated with greater severity of compulsions and predict engaging in compulsions at the next time point. However, based on prior research (e.g., Bucarelli and Purdon, 2015; Dean and Purdon, 2021), we expected not all compulsions to result in a decrease in unwanted experiences (e.g., anxiety). As such, in line with the CAM, we also hypothesized that smaller increases (relative to larger increases) in anxiety, unwanted thoughts, and perceived threat from before to after a compulsion would be associated with greater compulsion severity and future compulsive behavior.

1. Methods

1.1. Participants

Potential participants were recruited through three methods: (a) the undergraduate subject pool at a public University in the northeast United States, (b) a community-based study recruitment website; and (c) through the International OCD Foundation website. Individuals who scored above a 21 (i.e., the clinical-cutoff) on the Dimensional Obsessive Compulsive Scale (DOCS; Abramowitz et al., 2010) were invited to an initial screening visit (n = 65). During the visit, the DOCS was readministered along with the Diagnostic Interview for Anxiety, Mood, and OCD and Related Neuropsychiatric Disorders (DIAMOND; Tolin et al., 2018). Those who did not meet criteria for current OCD per the DIAMOND and the DOCS cutoff were excluded from the study (n = 10). Participants who met criteria for Bipolar I or II disorder (n = 2), a schizophrenic spectrum or other psychotic disorder (n = 2), or substance use disorder (within 6 months; n = 1) were excluded from the study. One eligible participant dropped out after the screening visit. As a minimum cluster size of greater than 5 is recommended for two level multilevel models, particularly with binary outcomes (McNeish and Stapleton, 2016), only participants with more than five prompts in which a compulsion was endorsed at time t were included (n = 45). In other words, participants had to endorse engaging in a compulsion greater than 0 (“None”) on at least 5 of the 49 prompts received in order to be included in the analysis. Therefore, the final sample consisted of 45 participants (n = 31 from a university subject pool, n = 6 from community-based website, n = 8 from IOCDF), aged 18 to 65 (M_age = 24.18 years; SD_age = 9.32 years). They were majority female (n = 44; 97.8%) and non-Hispanic White (n = 32; 71.1%). Based on the DIAMOND interview, the average duration of illness (calculated as current age minus age of onset) was 10.6 years (range = 2–35 years); however, it should be noted that onset ages were approximated in some cases, as several participants provided nonspecific onset ages (e.g., “early teens”), and two participants had missing data. See Table 1 for demographic information.

Table 1.

Demographic and clinical characteristics.

	M (SD)
Age	24.18 (9.32)
Female	44 (97.8%)
Race/Ethnicity
White Caucasian	32 (71.1%)
African American/Black	3 (6.7%)
Hispanic/Latino	3 (6.7%)
Arab/Middle Eastern	1 (2.2%)
Asian/Asian American	2 (4.4%)
More than one race	4 (8.9%)
DOCS total	36.02 (10.04)
Contamination	8.51 (3.53)
Harm	10.00 (3.85)
Unacceptable Thoughts	8.60 (3.91)
Order/Symmetry	8.91 (4.17)

Note. n = 45; DOCS = Dimensional Obsessive-Compulsive Scale.

1.2. Measures

1.2.1. Dimensional Obsessive-Compulsive Scale (DOCS; Abramowitz et al., 2010)

The DOCS is a 20-item questionnaire measuring OC symptom severity by assessing four symptom dimensions: contamination, responsibility for harm, unacceptable thoughts, and order and symmetry. Items assess avoidance, distress, interference, and difficulty controlling/resisting compulsions within each symptom dimension. Each item is rated on a 0 to 4 Likert scale, with a maximum score of 80 indicating highest symptom severity. A clinical-cutoff score of 21 was established as optimizing sensitivity and specificity in distinguishing OCD from other anxiety disorders (Abramowitz et al., 2010). The DOCS demonstrated good psychometric properties (Abramowitz et al., 2010), and good internal consistency in the present sample (α = 0.87).

1.2.2. Diagnostic Interview for Anxiety, Mood, and OCD and Related Neuropsychiatric Disorders (DIAMOND; Tolin et al., 2018)

Based on DSM-5 criteria, the DIAMOND is a semi-structured interview assessing psychiatric disorders. It demonstrates very good to excellent interrater reliability and good to excellent retest reliability for all disorders. Convergent validity, established via correspondence with self-report measures, was established for almost all disorders (Tolin et al., 2018). The DIAMOND was administered by undergraduate research assistants trained by a doctoral student. All interviews were audio and video recorded, and 50% were scored by a second blind rater. As marginal distributions were skewed due to pre-screening for OCD, Gwet's AC1 was used to calculate interrater reliability (Gwet, 2010; Wongpakaran et al., 2013). Gwet's AC1 was excellent 0.91 (95% CI [0.83, 0.98]) across all inclusion and exclusion diagnoses. For OCD specifically, interrater reliability was substantial to excellent (Gwet's AC1 = 0.84, 95% CI [0.60, 1]).

1.2.3. EMA measure

The EMA measure consisted of 24 items. Items relevant to the present study assessed compulsions and experiences (e.g., anxiety) before and after compulsions at each prompt.

1.2.3.1. Compulsions.

Participants rated two items per prompt on momentary compulsion severity. Items were adapted from the Yale-Brown Obsessive-Compulsive Scale (YBOCS; Goodman et al., 1989). Participants were first given the definition of a compulsion derived from the YBOCS at the start of each survey (“Compulsions are urges that people have to do something to lessen feelings of anxiety or other discomfort. Often they do repetitive, purposeful, intentional behaviors called rituals. The behavior itself may seem appropriate but it becomes a ritual when done to excess. Washing, checking, repeating, straightening, hoarding and many other behaviors can be rituals. Some rituals are mental. For example, thinking or saying things over and over under your breath”). Duration of compulsions (“In the last hour, how much of your time was occupied by compulsions?”) was assessed on a 5-point Likert scale (0 = None, 4 = Greater than 40 minutes or nearly constant performance of compulsions). Interference from compulsions (“In the last hour, how much did your compulsions interfere with your work, school, social, or other important role functioning? Was there anything that you did not do because of the compulsions?”) was assessed on a 5-point Likert scale (0 = none, 4 = incapacitating). Duration and interference of compulsions were summed to obtain a total momentary compulsion severity score with total scores ranging from 0 (none) to 8 (severe compulsions).

1.2.3.2. Experiences before and after compulsions.

Experiences before and after compulsions were assessed using eight items adapted for the present study. Participants were asked to think back to their most recent compulsive episode since the last prompt and rate how they felt before and after performing the compulsion in terms of their level of anxiety (“Think back to your most recent compulsion in the last hour: To what extent did you feel keyed up or on edge (see Newman et al., 2019 for evidence for this item as measuring anxious arousal) immediately [before/after] performing the compulsion?”), unwanted thoughts, (“Immediately [before/after] engaging in the behaviors, how much were you thinking about your intrusion?”), and perceived threat (“Immediately [before/after] engaging in the behaviors, how much did you feel in harm’s way, in danger, or threatened?”) on a 10-point Likert scale from 0 (Not at all) to 10 (Very much so).

1.3. Procedure

After screening, eligible participants met with a research assistant and were thoroughly trained on the EMA application and familiarized themselves with EMA question items. Specifically, participants were instructed to download the SEMA³ app (O'Brien et al., 2024) on their smartphone and were trained on the EMA questions using a PowerPoint presentation, where they were walked through each item and given a chance to ask questions. Participants were briefed on the definition of an obsession (“unwanted, distressing thoughts, urges, or images that repeatedly enter a person's mind, often seeming to happen against their will”) versus a compulsion (“Compulsions are urges that people have to do something to lessen feelings of anxiety or other discomfort. Often they do repetitive, purposeful, intentional behaviors called rituals. The behavior itself may seem appropriate but it becomes a ritual when done to excess. Washing, checking, repeating, straightening, hoarding and many other behaviors can be rituals. Some rituals are mental. For example, thinking or saying things over and over under your breath”) and were given examples of common obsessions and compulsions, including both behavioral (e.g., hand washing, checking, ordering) and mental (counting, reassurance seeking) compulsions. These examples also included discussion of “not just right experiences” and “incompleteness” as motivators for compulsions. Participants were also briefed on identifying experiences before and after compulsions. For example, to train participants on answering questions regarding threat perception before and after a compulsion, the following script was used: “We are interested in knowing how much you feel your compulsions helped you avoid potential threat or danger. For example, if you have contamination OCD, how much did you feel in danger/threatened by the germs/contaminants BEFORE you washed your hands? How much in danger/threatened did you feel by the germs/contaminants AFTER you washed?”

Participants were prompted on the SEMA³ app (O'Brien et al., 2024) seven times per day (once every 2 h) beginning at 10 AM and ending at 10 PM for seven days. They were instructed to respond to prompts as quickly as possible but were allotted a maximum of one-hour to complete the questions, which ensured a minimum of 1 h between prompts. They also received evening messages from a research assistant, informing them of their daily study compliance (e.g., “You completed 7 out of 9 prompts today.”). A research assistant was available each day to answer questions or resolve technical difficulties. Participants enrolled for monetary compensation ($50 Amazon gift card) received prorated amounts of compensation based on compliance. Participants enrolled for credits received 5 study credits.

1.4. Data analyses

A power analysis was conducted using 1000 Monte Carlo simulations with a standardized effect size of 0.25. There was 100% power to detect the association between experiences before and after compulsions with time t compulsions. There was 76% power to detect the association between each predictor at time t on presence or absence of a compulsion at time t + 1.

Two level multilevel models were conducted in R studio using the lme4 R package (Bates et al., 2015). Missingness was handled using full information maximum likelihood estimation (FIML; Graham, 2009). Time of day (i.e., minutes since the first prompt of the day), day in study, and participant age were included as covariates in all analyses. Predictors were person-mean centered. Additionally, the mean level of each participant's predictor were included as a covariate to account for individual differences in the predictor between participants.

Three separate two-level multi-level models for each predictor (i.e., anxiety, unwanted thoughts, and perceived threat; nine models total) were conducted to examine (1) the effect of each predictor immediately prior to performing a compulsion at time t on compulsion severity at time t; (2) the effect of increases or decreases in each predictor from before to after a compulsion at time t on compulsions at time t and (3) the effect of increases or decreases in the predictor from before to after a compulsion at time t on presence or absence of a compulsion at the next time point (time t + 1). For aims 1 and 2, as we were interested in examining experiences before and after a compulsion, only prompts where a compulsion was performed (1360 prompts; or 80.0% of completed prompts) were included in the models. For aim 3, only prompts in which a compulsion was endorsed at time t were included (1338 prompts; 78.7% of completed prompts).

To examine the effect of each predictor immediately prior to performing a compulsion at time $t$ on compulsion severity at time t, the following equations were used:

$$\begin{gathered} {\text{Compulsions}}_{\text{ti}}={\beta }_{0\mathrm{i}}+{\beta }_{1\mathrm{i}}\left({\text{Predictor}}_{\text{ti}}\right)+{\beta }_2\left({\text{Mean Predictor}}_{\mathrm{i}}\right)+ \\ {\beta }_3\left({\text{TimeofDay}}_{\text{ti}}\right)+{\beta }_4\left({\text{DayInStudy}}_{\text{ti}}\right)+{\beta }_5\left({\text{Age}}_{\mathrm{i}}\right)+{\mathrm{e}}_{\text{ti}} \end{gathered}$$ Level 1:

$$\begin{gathered} {\beta }_{0\mathrm{i}}={\tau }_{000}+u_{0\mathrm{i}} \\ {\beta }_{1\mathrm{i}}={\gamma }_{10}+u_{1\mathrm{i}} \end{gathered}$$ Level 2:

In these models, compulsions represent a continuous outcome variable at time point $t$, for person $i$. The term ${\beta }_{0\mathrm{i}}$ represents the intercept for person $i$ with a grand mean intercept ${\mathrm{\gamma }}_{00}$ and a random deviation, $u_{0\mathrm{i}}$ that varies across individuals. The term ${\beta }_{1\mathrm{i}}$ represents the slope of the momentary predictor (i.e., either anxiety, unwanted thoughts, and perceived threat), including a fixed component ${\mathrm{\gamma }}_{10}$ and a random component, $u_{1\mathrm{i}}$ that represents person-specific deviation from the average slope. ${\beta }_2$ represents the association between a person's mean level of the predictor and compulsions, ${\beta }_3$ represents the association between time since the first prompt and compulsions, ${\beta }_4$ represents the association between the day in the study and compulsions, and ${\beta }_5$ represents the association between participant's age and compulsions.

Next, to examine whether increases or decreases in each predictor from before to after a compulsion were associated with compulsion severity at time $t$, we calculated change scores, as models estimating slope of change with random slopes failed to converge. As an example, a change score was calculated for anxiety by subtracting responses to “To what extent did you feel keyed up or on edge immediately after performing the compulsion?” minus “To what extent did you feel keyed up or on edge immediately before performing the compulsion?” at each prompt. Thus, decreases in anxiety from before to after a compulsion were represented by negative change scores and increases in anxiety were represented by positive change scores.

To account for individual differences, scores were person-mean centered. Person-mean centering subtracts participants’ individual mean on a variable from their individual observations on that variable, so that each value represents their person-specific deviations from their own mean. Thus, increases and decreases from before to after a compulsion were relative to the participant (e.g., a positive value represents a greater-than-usual increase in anxiety relative to the participant's average anxiety across all compulsions).

After calculating change scores for all predictors, we used the same model equations as above, except change scores were individually added as predictors in separate models. As changes scores included both negative and positive values, which could represent both increases and decreases in the predictor, we probed for potential non-linear effects on all models using change scores, and compared model fit relative to the linear model based on Akaike Information Criterion (AIC) values. Nonlinear relationships were examined using natural cubic splines, set to 3 degrees of freedom (df) and using the default selection settings for knots (based on predictor quantiles) and boundaries (placed at minimum and maximum observed values) in the R package, spline. Lastly, to decompose the nonlinear effects, piecewise multilevel linear regression models were used to separately examine either the effect of (a) positive scores (increases in the predictor following a compulsion) or (b) zero (no change) or negative (decreases in the predictor following a compulsion) change scores on compulsions.

Finally, to examine whether increases or decreases in each predictor from before to after a compulsion predicted performing a compulsion at the next time point, the same steps as aim 2 were conducted, except lagged change scores were added as predictors (time $t$ change score), and the outcome variable was dichotomized to represent presence or absence of a compulsion at the next time point (time $t+1$). As the outcome variable was binary (performed compulsion at time $t+1$ was coded as 1, no compulsion at time $t+1$ was coded as 0), a binary generalized linear mixed model using the glmer function in the lme4 package was used. All predictors were rescaled to place variables on a common scale and improve model convergence. Random slopes were not included in these models as some models failed to converge, so these models were fitted without the random slope.

2. Results

2.1. EMA descriptives

Participants received a total of 2205 prompts (45 participants, 49 prompts each) and completed 1705 of those prompts. Four prompts were removed for failed attention checks (i.e., clicking other than the 3rd box when prompted to select the 3rd box). Thus, overall compliance rate (excluding failed attention checks) was 77.3% (i.e., total number of prompts responded / total number received). Anxiety and unwanted thoughts mostly decreased from before to after a compulsion (68.3% and 64.5% of compulsions, respectively). Anxiety remained unchanged or increased after 18.2% and 12.8% of compulsions, respectively. Unwanted thoughts remained unchanged after 18.2% and increased following 16.5% of compulsions. Perceived threat remained unchanged after 50.15%, decreased after 32.7%, and increased after 16.0% of compulsions. See Table 2 for descriptive statistics of EMA variables.

Table 2.

Descriptive statistics for EMA variables.

	M (SD)
Total prompts received (per person)	49
Average completed prompts (per person)	37.88 (7.83)
Compliance ratea	77.3%
Frequency of compulsionsb	79.8%
Compulsion severityc	2.62 (1.02)

Experiences before and after a compulsiond	Before	After
	% (n)
Anxietye	4.69 (1.81)	2.94 (1.45)
% of prompts anxiety decreased	68.31% (1057)
% of prompts anxiety did not change	18.16% (472)
% of prompts anxiety increased	12.79% (202)
Unwanted Thoughtsf	4.80 (2.00)	3.13 (1.48)
% of prompts thoughts decreased	64.49% (981)
% of prompts thoughts did not change	18.16% (489)
% of prompts thoughts increased	16.47% (259)
Perceived threatg	2.53 (2.29)	2.25 (2.20)
% of prompts perceived threat decreased	32.65% (484)
% of prompts perceived threat did not change	50.15% (987)
% of prompts perceived threat increased	15.96% (251)

Note. Data were first calculated within person and then averaged across individuals.

^aCompliance rate was calculated by dividing the total number of completed prompts by the total number of prompts received.

^bFrequency of compulsions was calculated by dividing the total number of prompts where compulsions at least “occasionally” by the total number of completed prompts.

^cCompulsion severity represents the average total compulsion score for prompts in which a compulsion was endorsed (range 1–8, with 8 indicating near constant, incapacitating obsessions/compulsions) averaged across individuals.

^dExperiences before and after a compulsion were rated on a 0 (not at all) to 10 (very much so) scale and means and percentages only include prompts in which a compulsion was endorsed.

^eAnxiety represents the mean response to “Immediately [before/after] engaging in the behaviors, to what extent did you feel keyed up or on edge?”.

^fUnwanted thoughts represents the mean response to “Immediately [before/after] engaging in the behaviors, how much were you thinking about your intrusion?”

^gPerceived threat represents the mean response to “Immediately [before/after] engaging in the behaviors, how much did you feel in harm's way, in danger, or threatened?”

2.2. Effect of anxiety on compulsions

Experiencing greater anxiety immediately prior to a compulsion was associated with greater compulsion severity (β = 0.28, SE = 0.02, t (40.43) = 10.62, p < .001, d = 3.34). The overall linear model showed a significant association between change in anxiety from before to after a compulsion (including both increases and decreases) and compulsion severity, β = −0.07, SE = 0.04, t(36.81) = −2.09, p = .042, d = −0.69. The natural cubic spline model suggested better fit for a nonlinear (vs. linear) association (p < .001; see Fig. 1). To examine the relative degree to which anxiety increases and decreases from before to after a compulsion were separately associated compulsion severity, two piecewise regression models were conducted. Greater degree of anxiety reduction from before to after a compulsion was associated with greater compulsion severity, β = −0.17, SE = 0.05, t (26.22) = −3.25, p = .003, d = −1.27. There was no association between degree of anxiety increase from before to after a compulsion and compulsion severity, p = .08.

Fig. 1.

Natural cubic splines of increases and decreases in anxiety from before to after a compulsion at time t on compulsion severity at time t.

Note. Anxiety change scores are person-mean centered and scaled. Negative values indicate instances where an individual experienced reduction in anxiety from before to after a compulsion and positive values indicate increases in anxiety. Knots and boundaries were placed on the default values (quantiles and minimum and maximum observed values, respectively, with degrees of freedom set to 3).

Lastly, the overall linear model revealed a negative association for anxiety increases and a positive association for anxiety decreases from before to after a compulsion and the likelihood of performing a compulsion at the next time point (OR = 0.84; β = −0.17, SE = 0.09, z = −2.04, p = .041). The natural cubic spline model suggested a marginally better fit (p = .051) for a nonlinear (vs. linear) association (see Fig. 2). Therefore, two piecewise regression models were conducted to examine the relative degree to which anxiety increases and decreases from before to after a compulsion were separately associated with the likelihood of performing a compulsion at the next time point. No change or smaller (relative to larger) increases in anxiety from before to after a compulsion were associated with a greater likelihood of performing a compulsion at the next time point, OR = 0.48, β = −0.72, SE = 0.28, z = −2.61, p = .009. There was no association between degree of anxiety reduction from before to after a compulsion and a subsequent compulsion, p = .801, though as shown in the nonlinear model (see Fig. 2), all levels of anxiety reduction were associated with a high probability of performing a compulsion at the next time point.

Fig. 2.

Natural cubic splines of increases and decreases in anxiety from before to after a compulsion at time t on probability of performing a compulsion at time t + 1.

Note. Anxiety change scores are person-mean centered and scaled. Negative values indicate instances where an individual experienced a reduction in anxiety from before to after a compulsion, and positive values indicate increases in anxiety. Scores are lagged, such that they predict performing a compulsion at the next time point (t + 1). Knots and boundaries were placed on the default values (quantiles and minimum and maximum observed values, respectively with degrees of freedom set to 3).

2.3. Effect of engagement with unwanted thoughts on compulsions

Higher unwanted thoughts immediately prior to a compulsion was associated with greater severity of compulsions, β = 0.27, SE = 0.03, t (40.82) = 10.34, p < .001, d = 3.24. The overall linear model showed no significant association between increases and decreases in unwanted thoughts from before to after a compulsion with compulsion severity, p = .921. However, the natural cubic spline showed a better fit for the nonlinear model (p < .001; Fig. 3). The piecewise regression revealed that both greater increases and greater decreases in unwanted thoughts from before to after a compulsion were separately associated with greater compulsion severity (increase: β = 0.27, SE = 0.06, t (12.89) = 3.58, p = .003, d = 1.99; decrease: β = −0.11, SE = 0.03, t(29.69) = −3.25, p = .003, d = −1.19). Lastly, neither increases nor decreases in unwanted thoughts from before to after a compulsion predicted performing a compulsion at the next time point, p = .251, nor was there a nonlinear relationship.

Fig. 3.

Natural cubic splines of increases and decreases in unwanted thoughts at time t on compulsion severity at time t.

Note. Unwanted thoughts are person-mean centered and scaled. Positive values indicate increases in unwanted thoughts and negative values indicate decreases in unwanted thoughts. Knots and boundaries were placed on the default values (quantiles and minimum and maximum observed values, respectively, with degrees of freedom set to 3).

2.4. Effect of perceived threat on compulsions

Higher perceived threat immediately prior to a compulsion was associated with greater severity of compulsions (β = 0.16, SE = 0.03, t (33.78) = 4.86, p < .001, d = 1.67). However, the overall model showed no significant associations between increases or decreases in perceived threat from before to after a compulsion and compulsion severity, p = .091. The natural cubic spline model had a significantly better model fit (p = .005) than the linear model, however, piecewise regression found no significant associations for both increases and decreases in perceived threat from before to after a compulsion with compulsion severity (ps = .093–.153). Lastly, neither increases nor decreases in perceived threat from before to after a compulsion predicted engaging in a compulsion at the next time point, p = .754, nor was there a nonlinear relationship.

3. Discussion

This is the first study to use an intensive repeated measures naturalistic design to examine the basic tenets of the CBM for the reinforcement of compulsions. In line with the CBM, greater anxiety, unwanted thoughts, and perceived threat immediately prior to a compulsion was associated with greater severity of compulsions. Greater compulsion severity was also associated with greater reduction of anxiety and was nonlinearly associated with unwanted thoughts, such that both large increases and decreases in unwanted thoughts from before to after the compulsion were associated with greater compulsion severity. Moreover, all levels of anxiety reduction from before to after a compulsion were associated with a greater likelihood of performing a compulsion at the next timepoint, though the relative degree of anxiety reduction did not affect their likelihood. In addition, no change or smaller (relative to larger) increases in anxiety from before to after a compulsion were associated with a greater likelihood of performing a compulsion at the next time point. This suggests that both anxiety reduction and prevention of a relatively sharper escalation in anxiety reinforced compulsions at the next time point, in line with the CAM. Findings highlight novel insights into the naturalistic reinforcement of compulsions and limitations in current conceptual models.

The CBM posits that temporary reduction of anxiety, unwanted thoughts, and perceived threat from before to after a compulsion reinforce and maintain compulsions in individuals with OCD (Rachman, 1998a, 1998b; Salkovskis, 1985, 1989, 1999). However, prior research on the reinforcement of compulsions was largely limited, with no known studies examining the temporal relationship between reduction in anxiety, unwanted thoughts, and perceived threat from before to after a compulsion with future compulsive behavior. Moreover, existing research suggested that reduction in these experiences did not fully explain compulsive behavior, as individuals did not consistently report a reduction in these experiences following a compulsion (Bouvard et al., 2020; Bucarelli and Purdon, 2015; Dean and Purdon, 2021). The present study provides the first evidence of this reinforcement process in a naturalistic setting in individuals with OCD and revealed a novel contrast avoidance effect for the reinforcement of compulsions.

As hypothesized, greater anxiety prior to performing a compulsion predicted engaging in more severe compulsions. Additionally, 68.3% of compulsions led to anxiety reduction and more severe compulsions were associated with greater anxiety reduction. This strongly supports the theorized mechanism of reinforcement for compulsions in the CBM (Salkovskis, 1985), highlighting anxiety avoidance as key to facilitating repetitive compulsive behavior within a discrete time period. Within a longer-time period, however, both anxiety reduction and prevention of its sharp escalation from before to after a compulsion appeared to collectively maintain future compulsive behaviors.

These findings bring forth novel insights into the reinforcement of compulsions. One possible explanation is that compulsions are reinforced via contrast avoidance. Namely, when individuals experienced a decrease in anxiety from before to after a compulsion (i.e., a PEC), it reinforced the compulsion. However, in line with previous work (Bouvard et al., 2020; Bucarelli and Purdon, 2015; Dean and Purdon, 2021), the present study found that not everyone experienced a reduction in anxiety from before to after a compulsion. Rather, 31.7% of compulsions resulted in no change or increased anxiety. No change and smaller (relative to larger) increases in anxiety from before to after a compulsion were also associated with a greater likelihood of performing a compulsion at the next time point. This is the first study to show that either no change or increases in anxiety from before to after a compulsion were reinforcing depending on the relative degree of the increase. In line with the CAM, this data suggested that individuals engaged in compulsions not only to reduce anxiety, but also to prevent large escalations in anxiety or avoid an NEC. These findings were consistent with Swisher and Newman (2025), which found that individuals with OCD (vs. those without OCD) were more likely to report engaging in behaviors to avoid an NEC. It is possible that either no change or smaller increases in anxiety were reinforcing because participants avoided the larger increase in anxiety they expected to experience if they had resisted the urge to perform a compulsion. It is also possible that a smaller increase in anxiety was associated with a compulsion perceived as only partially effective but still viewed as enabling avoidance of a relatively larger anxiety increase that would have occurred if the compulsion were completely ineffective. Future research should examine whether compulsions at the next timepoint are reinforced via how individuals rated prior compulsion contrast avoidance utility (e.g., “it was helpful to be emotionally prepared for the worst outcome”). This might help further clarify these findings.

Another possible explanation is that compulsions were reinforced by intermittent reinforcement. Intermittent reinforcement occurs when the reinforcement (i.e., anxiety reduction) only sometimes occurs following the behavior. In animal studies, intermittent reinforcement was associated with greater partial reinforcement acquisition, or performing the learned behavior more frequently or vigorously than if compulsions were rewarded every single time (continuous reinforcement; Anselme et al., 2013). The present study supports this notion, demonstrating that anxiety relief following compulsions only some of the time (intermittent reinforcement), led to compulsions remaining persistent.

In addition to anxiety, greater unwanted thoughts immediately prior to a compulsion were associated with more severe compulsions. Also, compulsions did not always lead to reduction of unwanted thoughts. In fact, 35.5% of the time compulsions led to either no change or increased unwanted thoughts. Nonetheless, both larger increases and larger decreases in unwanted thoughts from before to after compulsions were associated with greater compulsion severity. Interestingly, however, compulsion severity was highest at the highest increase in unwanted thoughts from before to after a compulsion (compared to highest decrease for which compulsion severity was lower; see Fig. 3). Thus, sharp increases in unwanted thoughts were linked to the most severe compulsive behavior. As the CBM has largely focused on compulsions as providing temporary removal in unwanted thoughts (e.g., Salkovskis, 1985), these findings suggest that it is also important to examine how increases in unwanted thoughts during a compulsive episode significantly drive compulsion severity. It is possible that when unwanted thoughts escalate during a compulsion, individuals try even harder to nullify the threat and intensify their compulsions. They may also keep engaging in a compulsion longer when it is effectively reducing unwanted thoughts. Contrary to expectations, neither increases nor decreases in unwanted thoughts from before to after a compulsion predicted compulsions at the next time point. Thus, in contrast to the CBM (Salkovskis, 1985), a decrease in unwanted thoughts did not appear to maintain future compulsive behavior.

Lastly, overestimation of threat has been posited to be a relevant bias predicting the maintenance of OCD symptoms (e.g., Cervin et al., 2022; Steketee et al., 1998; Taylor et al., 2006), though little research has examined the naturalistic reinforcement of this bias on OCD symptoms. In the present study, greater threat perception immediately prior to performing a compulsion was associated with greater compulsion severity. However, neither increases nor decreases in perceived threat from before to after performing a compulsion were associated with the severity of the compulsions. Perceived threat increases and/or decreases were also not predictive of performing a compulsion at the next time point. It is important to note that in the present study, general threat perception, rather than OC-specific threat (e.g., assessing threat specific to individuals' compulsion motivation) was assessed. Studies have found that threat estimation in OCD was elevated only when encountering the triggering stimulus relevant to the individual's OCD (e.g., estimated threat about a contamination), rather than general threat (Berman et al., 2019; Tata et al., 1996). Therefore, it is possible that only reduction in elevated threat specific to one's obsessional fears would have predicted future compulsive behavior.

It is also important to consider that OCD is a highly heterogenous condition, with many different motivators for compulsive behaviors. For example, individuals may feel driven to perform compulsions to prevent harm, to get rid of anxiety or unwanted sensations (e.g., “not just right experiences” or NJREs; Summerfeldt, 2004) or to placate feelings of incompleteness or asymmetry. The current study represented that heterogeneity given that all subscales of the DOCS were elevated within the sample, everyone met DSM-5 criteria for OCD, and we defined and assessed all types of compulsions, regardless of motivation. Importantly, no known study has examined whether different motivators for compulsions (e.g., harm avoidance, NJREs) would result in different reinforcers of compulsions (e.g., compulsions reinforced via distress reduction). It is possible that in the present study, the inconsistency in reduction in anxiety, unwanted thoughts, and perceived threat across prompts was driven in part by different reinforcers for performing a compulsion. As such, an interesting future study could use an EMA design to delineate whether different compulsion motivators are associated with different reinforcing mechanisms of compulsions.

Additionally, other theoretical models may offer alternative explanations for the persistence of compulsions not captured in the present study. Neurocognitive models for OCD, such as the habit formation model, posit that an imbalance between habit and goal-directed behavior in OCD results in an overreliance on habitual responses, leading to greater compulsive behavior (Gillan et al., 2011; Gillan and Robbins, 2014). A more recent theory integrated this model with the CBM, positing that both neuropsychological deficits (i.e., reliance on habit formation systems) and cognitive behavioral processes (i.e., anxiety reduction following compulsions) influence one another to maintain the persistence of compulsions (Kalanthroff and Wheaton, 2022). These models both highlight the potential importance of neurocognitive deficits in the persistence of compulsions, which were not captured in the present study. An interesting future study design might investigate how deficits in response inhibition interact with reductions in unwanted experiences following compulsions to maintain or exacerbate compulsive behavior.

Taken together, these findings have important implications for refining future research on the reinforcement and maintenance processes of OCD. First, the present findings expanded the CBM by showing that not all compulsions resulted in reduction of anxiety, unwanted thoughts or perceived threat. In addition, compulsions were reinforced via both anxiety reduction and facilitating avoidance of sharp increases in anxiety. That is, both no change in anxiety and a smaller increase in anxiety from before to after a compulsion were associated with greater likelihood of performing a compulsion relative to larger increases in anxiety. Future research extending the current findings would have important implications for refining treatment models to more precisely target the mechanisms reinforcing compulsive behavior. For example, including psychoeducation on how smaller (relative to larger) increases in anxiety perpetuate compulsions using the framework of the CAM may help OCD patients understand why compulsions persist even though they do not always make them feel better.

Additionally, these findings underscore the value of addressing anxiety tolerance in treatment settings. Importantly, contemporary treatment models for OCD have explored treatment options that excluded exposure to anxiety (via response prevention and tolerating the distress associated with not engaging in the compulsion), instead emphasizing obsessional doubt and inferential confusion (Julien et al., 2016). However, our findings suggest individuals with OCD specifically engage in compulsions to facilitate both anxiety reduction or prevention of its escalation, and that this is a key maintenance factor of the OCD cycle. As such, failing to include response prevention to address this contrast avoidance may result in suboptimal treatment outcomes or increase risk for relapse, as the core reinforcing mechanism would not be addressed. Although cognitive interventions showed efficacy in some trials (Aardema et al., 2022), the evidence base was smaller, non-inferiority has been inconclusive (Wolf et al., 2024), and longer-term outcomes and risk for relapse have not been assessed. The present findings support this caution toward cognitive-only treatment and as such, recommend future research to investigate ways to increase engagement in exposure, emphasize response prevention, and promote distress tolerance, rather than eliminate exposure and response prevention (ERP) as a treatment element.

Strengths of the present study include an ecological, naturalistic design with intensive-repeated sampling, limiting retrospective biases and allowing for a critical examination of the reinforcement of compulsions. However, our study had a relatively small sample size, restricted gender diversity (e.g., 97.8% female), low racial diversity (e.g.. 71.1% White non-Hispanic), and limited information on other relevant background variables, such as educational attainment of the non-university based sample or whether they have received treatment. Although we spent time thoroughly training participants with definitions of the constructs we assessed and reminded them of these definitions at each prompt, the level of awareness participants had about OCD prior to the study could have impacted how they answered and understood the EMA items. Indeed, EMA items required some level of abstraction to differentiate between experiences prior to their compulsion and experiences after their compulsion since the last prompt. As such, it is possible that participants' abilities and psychological mindedness may have influenced how they answered EMA items. Relatedly, retrospective biases were introduced by asking participants to reflect on their most recent compulsive episode in the last two hours. Event-contingent designs, in which participants self-initiate surveys immediately before or after completing a compulsion, may help mitigate these issues, though limitations to these designs should also be considered (e.g., forgetting to report immediately after every event). Additionally, not all EMA questions were derived from pre-existing psychometrically valid measures. EMA research for OCD is greatly needed, and future research should focus on validating EMA measures to move toward more precise and valid assessment. Another limitation was that although participants exhibited elevations in each subscale of the DOCS (see Table 1), the type of compulsions (e.g., hand washing, mental compulsions) and motivations for each compulsion (e.g., to placate feelings of incompleteness) were not assessed at each prompt. Moreover, as not all possible experiences occurring before and after compulsions could be captured, some relevant experiences (e.g., feelings of disgust) were omitted. This limited our understanding of how reinforcing mechanisms of compulsions differentially impacted different OCD presentations and if other potential reinforcers (e.g., changes in disgust) could have driven reinforcement. Finally, models predicting compulsion behavior at the next time point demonstrated moderate power (76%). Future research with larger samples is therefore necessary to replicate these findings.

In conclusion, this is the first study to provide ecological validity to several aspects of the CBM for the reinforcement of compulsions. It is also the first study to provide evidence for a contrast avoidance effect in the reinforcement of compulsions. Supporting the CBM, greater anxiety levels immediately prior to a compulsion were associated with greater compulsion severity. In contrast to CBM, both large increases and decreases in unwanted thoughts from before to after a compulsion were associated with greater compulsion severity. In line with the CAM, both anxiety reduction and prevention of its escalation were significant predictors of future compulsions. Change in unwanted thoughts and threat perception from before to after compulsions were not significant predictors of compulsions at the nexttime point. These results support several tenets of the CBM for OCD and call for further consideration of contrast avoidance as a relevant mechanism in OCD. Findings also emphasize the importance of addressing contrast avoidance within treatment, and caution third-wave movements emphasizing cognitive-only strategies, as these mechanisms appear to be less relevant to breaking the reinforcement cycle of compulsions. Overall, the present study offers ecological validity for the reinforcement of compulsions, and underscores the need for more robust ecological and longitudinal measurement to examine leading theoretical models for OCD.