Components of Emotion Regulation Flexibility: Linking Latent Profiles to Depressive and Anxious Symptoms

Abstract

Emotion regulation flexibility has been conceptualized as a multicomponent construct that consists of context sensitivity, repertoire, and feedback responsiveness. Although individuals with greater abilities in each component show better psychological adjustment, the patterns of these components remain unknown. In two cross-sectional MTurk studies (Ns = 200 and 802), we identified four or five predominant latent profiles: high-flexibility regulators (HFR), medium-flexibility regulators (MFR), context-insensitive regulators (CIR), feedback-irresponsive regulators (FIR), and low-repertoire regulators (LRR; Study 2 only). Inflexible regulators (CIR, FIR, and LRR) exhibited greater depressive and anxious symptoms than MFR and then HFR. Although inflexible regulators did not differ from each other on depressive symptoms, CIR showed more anxious symptoms than FIR and LRR. These findings support the importance of all three flexibility components with a highlight on context sensitivity and, moreover, suggest one potential way in which future studies can integrate various flexibility components.

Keywords

emotion regulation flexibility context sensitivity repertoire depression anxiety

The field of emotion regulation (ER) has flourished in the past decade given the central role that ER plays in many forms of psychopathology (Aldao, Nolen-Hoeksema, & Schweizer, 2010; Webb, Miles, & Sheeran, 2012). Although traditionally the theories behind coping and ER highlighted the dynamic interplay between persons and situations (e.g., Gross, 1999; Lazarus & Folkman, 1984; Mischel, 1973), much of the research on ER has adopted a relatively static approach that emphasized the adaptiveness or maladaptiveness of specific strategies such as reappraisal, suppression, and rumination (e.g., Aldao et al., 2010; Gross & John, 2003; Webb et al., 2012). Recently, however, a growing number of investigators have revisited the interactionist approach and demonstrated that the efficacy of specific ER strategies varies markedly across situations and individuals (Birk & Bonanno, 2016; Bonanno, Papa, Lalande, Westphal, & Coifman, 2004; Bonanno, Pat-Horenczyk, & Noll, 2011; Sheppes et al., 2014; Tamir & Ford, 2012; Troy, Ford, McRae, Zarolia, & Mauss, 2017; Troy, Shallcross, & Mauss, 2013; Westphal, Seivert, & Bonanno, 2010). According to the flexibility literature, to manage the diverse demands across varying situations, people need to flexibly regulate their emotions (Aldao, Sheppes, & Gross, 2015; Bonanno & Burton, 2013; Cheng, 2001; Cheng, Lau, & Chan, 2014; Kashdan & Rottenberg, 2010).

Multicomponent Perspective on ER Flexibility

From a theoretical perspective, ER flexibility has been conceptualized not as a unitary phenomenon but as a broad multicomponent process involving a sequence of abilities in these Components (e.g., Bonanno & Burton, 2013; Cheng et al., 2014). For example, in their review of the literature on flexibility in coping and emotion regulation, Bonanno and Burton (2013) articulated three interrelated yet functionally distinct components. These components included (a) the ability to evaluate contextual demands, or context sensitivity; (b) the availability of a range of strategies that may be implemented to meet those demands, or repertoire; and (c) the capacity to monitor the efficacy of a chosen strategy and modify as needed, or feedback responsiveness. Each of these components was considered essential for successful self-regulation but also likely to vary measurably across individuals. Likewise, Cheng and colleagues (2014) proposed a flexibility model based on their review of more than 100 empirical studies. In their model, flexibility was conceptualized as ability displayed at three unique and interacted regulatory stages, respectively, assessing situational characteristics, adopting regulatory strategies, and monitoring outcomes, which resonates with the model suggested by Bonanno and Burton. We acknowledge that these models may likely still overlook the fact that flexibility may involve other processes, such as ER goals (Millgram, Joormann, Huppert, & Tamir, 2015), and situation selection (Gross, 2015). Nevertheless, given that these components are relatively established both theoretically and empirically in the flexibility literature, they were the focus of the current investigation.

Multicomponent models of ER flexibility have proposed that people who can display high flexibility across all three components would exhibit the best outcomes. However, to date, empirical research has focused almost exclusively on individual components (e.g., repertoire). One noteworthy exception (Southward & Cheavens, 2017), for example, examined the interplay of abilities in these components. Specifically, deficits in repertoire of expressive regulation predicted depression and anxiety only among participants who were low in context sensitivity. These findings were consistent with theories of flexibility that have underscored the primacy of context sensitivity and viewed deficits in this component as especially predictive of poor clinical outcomes (Aldao, 2013; Bonanno & Burton, 2013; Cheng et al., 2014). According to our knowledge, there is no empirical work to date that directly compares psychopathology symptoms among participants with different flexibility deficits. Adopting a person-centered approach has the potential not only to identify patterns of flexibility deficits but also to examine how these patterns are related to psychopathology symptoms. Although this is relatively unexplored in the flexibility literature, previous studies on ER strategy have adopted a person-centered approach and linked patterns of ER to psychopathology symptoms such as depression and anxiety (e.g., Dixon-Gordon, Aldao, & De Los Reyes, 2015).

In this study, we capitalized on recently validated measures matching the three components of Bonanno and Burton’s (2013) flexibility model: context sensitivity, repertoire, and feedback responsiveness. In addition, to understand the nature of flexibility deficits and how different flexibility profiles are linked to depression and anxiety, we conducted latent profile analysis on these three components and then linked the constructed latent profiles to depressive and anxious symptoms.

Context Sensitivity, Repertoire, and Response to Feedback

A growing body of work has suggested that the first and arguably the most crucial step in flexible self-regulation involves the ability to read, decode, and evaluate contextual cues that signal the impinging demands and opportunities inherent in the situation, commonly referred to as context sensitivity (Aldao, 2013; Bonanno & Burton, 2013; Bonanno, Maccallum, Malgaroli, & Hou, 2020). Note that this construct refers to the perception of context rather than the response to context. Research has shown that greater context sensitivity, in particular the ability to identify the absence of threatening cues, is associated with fewer psychopathology symptoms (Bonanno et al., 2020). Individual differences in context sensitivity move forward and exert an important impact on subsequent steps of flexible self-regulation because errors in initial evaluation or less sensitive appraisals made when decoding the contexts will make it more difficult and unlikely for one to select appropriate strategies and monitor feedback.

A subsequent step in flexible self-regulation recruits the ability to use a wide range of strategies, commonly referred to as repertoire (Bonanno & Burton, 2013; Dixon-Gordon et al., 2015). Whether one can successfully enact the chosen strategy depends on the repertoire of regulatory strategies available. Accumulating empirical findings have shown that better adjustment after stressful and potentially traumatic events is associated with a greater number of strategies (Orcutt, Bonanno, Hannan, & Miron, 2014), increased temporal variability (Cheng, 2001), and higher categorical variability (Bonanno et al., 2004; Burton et al., 2012). One particularly important indicator of repertoire is the ability to both up-regulate and down-regulate emotions, which has been consistently linked to greater psychological distress and better psychological adjustment (Bonanno et al., 2004; Rodin et al., 2017; Westphal et al., 2010).

Finally, after a strategy has been implemented, a third step in regulatory flexibility involves the postimplementation ability to monitor the effectiveness of the chosen strategy and either maintain or modify the strategy as needed, commonly referred to feedback responsiveness (Bonanno & Burton, 2013; Kato, 2012; Sheppes et al., 2014). Research has shown that the self-reported ability to discontinue maladaptive strategies and switch to adaptive strategies is associated with fewer depressive symptoms (Kato, 2015, 2017). Although experimental research on this component is relatively sparse, a few recent studies demonstrated that switching from reappraisal to distraction in the face of high-intensity negative images was associated with larger modulation in neural activity associated with emotional processing (Ilan, Shafir, Birk, Bonanno, & Sheppes, 2019) and that the switching frequency predicted better adjustment when it was in accord with internal physiological response (Birk & Bonanno, 2016). These findings provided support for the importance of monitoring the efficacy of an initial strategy and modifying or replacing ineffective strategies when necessary.

Toward an Integration of Flexibility Components

Although each of these components—context sensitivity, repertoire, and feedback responsiveness—has been hypothesized to be essential for successful self-regulation (Bonanno & Burton, 2013), they have been studied separately rather than jointly. Thus, whether and how these components work together in shaping one’s overall adjustment remains unknown. It has been hypothesized that strong abilities in all three components yield the most desirable outcomes (Bonanno & Burton, 2013). However, it is worth considering whether one component is more important than the others. For example, given that context sensitivity plays a crucial role in emotional experience and regulation (e.g., Aldao, 2013; Rottenberg, Gross, & Gotlib, 2005) and that it is sequentially the first step of self-regulation (Bonanno & Burton, 2013; Bonanno et al., 2020), it may be that deficit in this component is more detrimental for psychological health than deficit in the others. However, it is also likely that strong abilities in one or two components suffice to accommodate deficits in other components or that deficit in any flexibility component is universally associated with similar level of psychological distress. To develop a better understanding of the nature of flexibility deficits across individuals as well as how different patterns of flexibility components are associated with depressive and anxious symptoms, studies incorporating all three components of ER flexibility seem particularly beneficial and important.

The Current Investigation

To address these questions in the current investigation, we measured individual differences on each flexibility component using recently developed and well-validated measures. Two of the three measures have been validated against experimental paradigms (Bonanno et al., 2020; Burton & Bonanno, 2016). In addition, measures of all three flexibility components had low or no correlation with social desirability (Bonanno et al., 2020; Kato, 2012), which lowers the interference of social desirability bias with the interpretation of individual differences in flexibility (Crowne & Marlowe, 1960). We conducted latent profile analysis (LPA) to determine the predominant patterns of abilities across these flexibility components and compared concurrently measured levels of depression and anxiety across the constructed latent profiles.

The current investigation consisted of two studies. In Study 1, we used a relatively small sample to preliminarily evaluate the feasibility of applying LPA to identify predominant patterns of flexibility components. Specifically, we examined if the model provides interpretable latent profiles that were associated with clinical outcomes (i.e., depression and anxiety). Then, in Study 2, we recruited a considerably larger sample and adopted a more rigorous, double cross-validation LPA in an attempt to replicate and further our findings.

Study 1

From the perspective of sequential components of regulatory flexibility, ER flexibility can be viewed as context sensitivity at the evaluation stage, repertoire at the implementation stage, and feedback responsiveness at the modification stage (Bonanno & Burton, 2013). Questions then arise about whether there are individuals who are highly flexible across all regulatory components or alternatively deficient in at least some of flexibility components. If different flexible and inflexible profiles can be identified, a further question is whether and how these profiles are clinically relevant and meaningful. In other words, we were interested in learning whether highly flexible people are the least depressed and anxious and what inflexible profiles are most elevated in depression and anxiety. To answer these questions, we conducted Study 1, in which we used LPA to first classify potential subgroups and then relate outcomes with group memberships.

Method

Data and participants

Study 1 was conducted using Amazon’s Mechanical Turk (MTurk) service. MTurk facilitates high-quality data collection from a large pool of diverse participants. Recent studies have found that MTurk participants performed similarly to participants recruited offline (e.g., Paolacci, Chandler, & Ipeirotis, 2010) and showed high test–retest reliability (Casler, Bickel, & Hackett, 2013). The questionnaire session was advertised on MTurk as a “Life Events Survey” and consisted of demographic items, measures of different components of ER flexibility, and two attention-check items that directed participants to select certain items to ensure that they were paying attention to questions when responding.

Two hundred participants (115 men, 85 women; mean age = 34.19 years, SD = 9.54) completed the measures and were paid $2 for their participation. The sample was racially diverse: 83.6% were White, 9.0% were Black or African American, 6.5% were Asian American, 2.0% were American Indian, 0.5% were Native Hawaiian, and 0.5% of participants preferred not to answer. In the sample, 9.5% of the participants self-identified as Hispanic or Latino, 88.0% of the participants considered themselves non-Hispanic, and 2.5% preferred not to answer. The institutional review broad approved this study.

Measures

To establish latent profiles of the three flexibility components mentioned earlier and examine potential relationships between profile memberships and symptoms of depression and anxiety, we measured components of flexibility (i.e., context sensitivity, repertoire, and feedback responsiveness) and symptoms of depression and anxiety.

Context sensitivity was measured by the Context Sensitivity Index (CSI; Bonanno et al., 2020). The CSI is a scenario-based inventory that captures one’s sensitivity to the relative presence of contextual cues (Cue Presence Index, or CPI) and to the relative absence of cues (Cue Absence Index, or CAI). The CPI is composed of 10 appraisal ratings that are normed as relatively highly present in the scenarios; the CAI is composed of 10 items that are normed as relatively absent from the scenarios. The CAI items were reverse-coded so that high scores reflected sensitivity to the absence of a contextual cue. After reading each daily situation (e.g., “You take a medicine and it makes your nauseous. Your doctor tells you that it is not serious and that you just have to ‘wait it out.’”), participants were asked to rate their perception of cues including threat, self-control, other-control, urgency, and cooperation by circling the number that best corresponds with their response (1 = not at all, 7 = very much). The CSI is associated with flexible coping, ER, and psychopathology symptoms and is validated against behavioral measures of cue sensitivity such as the Picture Arrangement Test from the Wechsler Adult Intelligence Scale–Third Edition (Bonanno et al., 2020). In this study, we calculated the sum of CPI and CAI to reflect the context-sensitivity component of ER flexibility.

Repertoire was measured by the Flexible Regulation of Emotional Expression Scale (FREE; Burton & Bonanno, 2016). The FREE is a scenario-based questionnaire that measures participants’ perceived ability to enhance and suppress their emotional expressions to meet situational demands. The FREE includes two subscales that measure the ability to enhance and suppress emotional expression, from which a total score can be calculated by summing the enhancement and suppression scores. The higher score that the participants had, the more able they would be to use an adequate range of regulatory strategies that might accommodate divergent contextual commands and opportunities. Participants indicated how well they would be able to be even more expressive (e.g., “Your boss is complaining about a project you know little about and have no involvement with”) or conceal how they were feeling (e.g., “You are at a social event and the person you’re talking to frequently spits while they speak”) in each scenario (1 = unable, 6 = very able). The FREE is associated with measures of ER, psychopathology symptoms, and well-being (Burton & Bonanno, 2016; Chen, Chen, & Bonanno, 2018). Moreover, enhancement ability and suppression ability were previously found to be associated with participants’ corresponding regulatory abilities in a lab setting, which provides support for its validity (Burton & Bonanno, 2016). In this study, we used the total score of the FREE scale to reflect the repertoire component of ER flexibility.

Feedback responsiveness was measured by the Coping Flexibility Scale (CFS; Kato, 2012). The CFS is a 10-item self-report questionnaire that captures the ability to discontinue an ineffective strategy (e.g., “If I feel that I have failed to cope with stress, I change the way in which I deal with stress”) and produce and implement an alternative strategy (e.g., “When a stressful situation has not improved, I try to think of other ways to cope with it”). Participants were asked to indicate how the statements apply to them (1 = not applicable, 4 = very applicable). Higher CFS scores were associated with more adaptive outcomes (Kato, 2012). In this study, we calculated a sum score to reflect the feedback-responsiveness component of ER flexibility.

Depressive and anxious symptoms were measured by the depression and anxiety subscales from the 21-item version of the Depression, Anxiety, Stress Scale (DASS; Lovibond & Lovibond, 1995). The DASS is a self-report instrument designed to measure the three related negative emotional states of depression, anxiety, and stress. After reading each statement (e.g., “I felt that life was meaningless”), participants were asked to rate how much the statement applied to them over the past week (0 = did not apply to me at all, 3 = applied to me very much or most of the time). In this study, depression and anxiety scores were calculated by summing items within each of these two subscales. The αs for anxiety and depression subscales were .83 and .91, respectively. Scores of 9, 13, and 20 on the depression scale are considered cutoffs for mild, moderate, and severe depression, and scores of 7, 9, 14 on the anxiety scale are considered cutoffs for mild, moderate, and severe anxiety.

Data-analytic plan

All data were analyzed with Mplus (Version 7.4; Muthén & Muthén, 2012). A three-step analysis approach to latent variable modeling was used (Asparouhov & Muthén, 2013). This approach allows researchers to use the input variables to create latent profiles and then link constructed profiles to distal outcomes taking into account classification uncertainty (Asparouhov & Muthén, 2013; Wisco, Plate, May, & Aldao, 2018).

In the first step, LPA was performed, and the optimal number of profiles was identified. We did an LPA using the standardized scores of context sensitivity, repertoire, and feedback responsiveness to ease interpretation. To determine the most appropriate number of latent profiles, we used fit indices including Akaike information criterion (AIC), sample-size-adjusted Bayesian information criterion (BIC), entropy, adjusted Lo-Mendell-Rubin likelihood ratio test (LMR-LRT), and bootstrap likelihood ratio test (BLRT). Lower BIC or AIC values suggest better fit, and significant LMR-LRT or BLRT for n profiles (p < .05) demonstrates that n profiles fit better than n − 1 profiles. According to Nylund, Asparouhov, and Muthén (2007), when these two likelihood ratio tests yield different results, BLRT outperforms other information-theoretic and likelihood ratio statistical test methods. In addition, it has been shown through simulation studies (e.g., Morovati, 2014) that rare classes (lower than 5%) are generally difficult to replicate. Therefore, we opted not to include classes with a prevalence of less than 5% to avoid potentially unstable solutions (Nylund-Gibson & Choi, 2018). Finally, we took into consideration the interpretability of classes when making the final decision.

In the second step, the most likely profile memberships were obtained along with the classification uncertainty. In other words, not only were individuals assigned to the profile that best fit their abilities on three ER flexibility components but also the uncertainty of this assignment was considered and modeled.

In the third step, the most likely profile membership variables were analyzed as concurrent predictors of the distal outcome variables (i.e., depression and anxiety) when taking into consideration the classification uncertainty. Chi-square tests were conducted, which examine whether profiles differ significantly from each other on each distal outcome. In this process, we estimated overall chi-square likelihood ratio tests for each outcome in which all means of outcomes were constrained to be equal (i.e., the means in outcomes are the same) compared with a freely estimated model (i.e., the means in outcomes are different). Then, for each pairwise comparison of latent profiles, two models were estimated and compared: one in which the means for each outcome were freely estimated for each latent profile (i.e., the means in outcomes are different) and one in which the means were constrained to be equal between the two profiles (i.e., the means in outcomes are the same). We planned two separate models for anxiety and depression. The equality tests of means across latent profiles used posterior probability-based multiple imputations with 3 df for the overall test and 1 df for the pairwise tests.

To correct for the increased likelihood of Type I error as a result of multiple comparison, we adopted the Benjamini-Hochberg (B-H) procedure to adjust the critical value and decrease false discovery rate (Benjamini & Hochberg, 1995). Specifically, we set the false discovery rate (FDR) at .05 and calculated the B-H critical value for each pairwise comparison by first multiplying the individual p value’s rank (from low to high) by the FDR and then dividing the product by the total number of tests. Next, we identified the highest p value that was still smaller than the B-H critical value. Finally, all p values smaller than this identified highest p value were considered significant:

B - H Critical Value = \frac{rank of p value \times F D R}{total number of tests} .

To obtain variances accounted for (ω²) and effect sizes (Hedges’s g; Hedges, 1981), one-way analyses of variance (ANOVAs) were performed using most likely class membership from the LPAs as the independent variable and the two outcomes (i.e., depression and anxiety) as dependent variables. This approach is recommended and was adopted in previous studies (e.g., Merians, Baker, Frazier, & Lust, 2019) because the three-step method does not allow estimation of variances accounted for by profile memberships. It is important to keep in mind, however, that treating most likely class membership as an independent variable without considering the classification uncertainty will slightly underestimate the relation. Hedges’s g is a better and less biased estimate of effect size than Cohen’s d (Cohen, 1988) when samples are not equal in size. Rules of thumb for interpreting ω² are .01 = small, .06 = medium, and .14 = large (Albers & Lakens, 2018). Hedges’s g values are considered large, medium, and small at 0.80, 0.50, and 0.20, respectively (Hedges, 1981).

Results

Latent profile analysis

We report all fit indices in Table 1. The BLRTs were significant for the two-, three-, four-profile solutions (ps < .001) but not for the five-profile solution (p = .60), which suggests that the four-profile solution provided the best possible fit. Although LMR-LRTs were significant for the two- and three-profile solutions (ps < .01) but not the four-profile solution (p = .14), we did not choose the three-profile solution because the BLRT proved to be a consistent indicator of classes that outperformed the LMR-LRT (Nylund et al., 2007). The four-profile solution had lower AIC and BIC values than the two- and three-profile solutions, which suggested that it fit better. Although the five-profile solution had the highest entropy (.84) among all profile solutions, we did not choose it because it included a profile with few participants (2.2%, four participants). Previous research suggests that rare classes (lower than 5%) are potentially unstable and limited in replicability (Nylund-Gibson & Choi, 2018). In terms of interpretability, compared with the three-profile solution, the four-profile solution included a clearly interpretable profile that was low in feedback responsiveness and medium in context sensitivity and repertoire.

Table 1.

Model Fit Indices for Exploratory Latent Profile Analyses and Double Cross-Validation

	Exploratory latent profile analyses							Double cross-validation
N classes	Max LL	AIC	BIC	Entropy	LMR-LRT p	PB-LRT p	% Small	Max LL	BIC	LRTS (df), p	Entropy
Study 1: Full sample (N = 200)
1	−847.02	1,706.04	1,706.82	—	—	—	—	—	—	—	—
2	−832.85	1,685.69	1,686.99	.71	.004	< .001	17.20	—	—	—	—
3	−819.76	1,667.52	1,669.35	.74	.002	< .001	10.80	—	—	—	—
4	−812.49	1,660.98	1,663.36	.73	.46	< .001	11.20	—	—	—	—
5	−808.43	1,660.86	1,663.73	.84	.14	.60	2.20	—	—	—	—
6	−804.26	1,660.51	1,663.89	.74	.68	.43	2.10	—	—	—	—
Study 2: Sample 1 (N = 401)
1	−4,308.32	8,628.64	8,633.56	—	—	—	—	—	—	—	—
2	−4,267.57	8,555.13	8,563.34	.84	< .001	< .001	11.00	—	—	—	—
3	−4,244.46	8,516.92	8,528.42	.82	< .001	< .001	10.70	—	—	—	—
4	−4,214.80	8,465.60	8,480.37	.81	.03	< .001	11.80	−4,222.75	8,453.97	15.9 (18), .60	.81
5	−4,204.73	8,453.45	8,471.51	.77	.04	.01	9.00	−4,211.38	8,431.23	13.30 (22), .92	.78
6	−4,197.89	8,447.76	8,469.11	.79	.06	.08	2.31	−4,230.37	8,469.21	64.96 (26), < .001	.74
Study 2: Sample 2 (N = 401)
1	−4,308.51	8,629.01	8,633.94	—	—	—	—	—	—	—	—
2	−4,256.70	8,533.40	8,541.61	.84	< .001	< .001	12.98	—	—	—	—
3	−4,224.33	8,476.66	8,488.15	.83	< .001	< .001	13.26	—	—	—	—
4	−4,196.47	8,428.95	8,443.72	.85	< .001	< .001	13.38	−4,206.27	8,421.01	19.60 (18), .36	.81
5	−4,184.77	8,413.55	8,431.61	.78	.07	< .001	10.47	−4,192.93	8,394.33	16.32 (22), .80	.78
6	−4,166.20	8,384.41	8,405.75	.81	< .001	< .001	4.74	−4,191.14	8,390.73	49.88 (26), .003	.79

Note: Max LL = maximized log likelihood value; AIC = Akaike information criterion; BIC = sample-size-adjusted Bayesian information criterion; LMR-LRT = adjusted Lo-Mendell-Rubin likelihood ratio test; PB-LRT = parametric bootstrapped likelihood ratio test; % small = percentage of participants in the smallest group; LRTS = log likelihood ratio test statistic; — = not available. Bold lines indicate the best profile solution based on AIC, BIC, LMR-LRT, PB-LRT, % small, and interpretability of that solution.

Figure 1a shows the standardized group means for the four-profile solution. Participants were assigned to one of the following profiles with classification uncertainty: (a) high ability across three components (HFR, 13.6%), (b) average ability across three components (MFR, 55.7%), (c) average levels of context sensitivity and repertoire but low levels of feedback responsiveness (FIR, 19.5%), or (d) average levels of feedback responsiveness and repertoire but low levels of context sensitivity (CIR, 11.2%). Classification probabilities were high across the four profiles, with probabilities of .77 in the HFR, .91 in the MFR, .77 in the FIR, and .82 in the CIR.

Fig. 1.

Sample item means for the four latent profiles of emotion regulation (ER) flexibility (a) and comparisons of depressive and anxious symptoms among these profiles (b) in Study 1. DASS = Depression Anxiety Stress Scale; HFR = high-flexibility regulators; MFR = medium-flexibility regulators; CIR = context-insensitive regulators; FIR = feedback-irresponsive regulators.

Associations with depression and anxiety

We examined potential differences among these four latent profiles on depressive and anxious symptoms using each of the distal outcomes as an auxiliary variable in the model. Figure 1b shows the means and standard errors of depression and anxiety scores in each profile. The means along with standard deviations for depressive and anxious symptoms by groups are presented in Table 2. The p values and effect sizes are presented in Table 3. To adjust for multiple comparison and control false discovery rate at .05, the B-H procedure was adopted (Benjamini & Hochberg, 1995). A total of 12 pairwise comparison tests were conducted. The first 11 p values in rank were lower than their respective B-H critical values and were thus all considered significant. The only insignificant pairwise comparison was between FIR and CIR on depressive symptoms. We report detailed results of symptom comparison in the following paragraphs.

Table 2.

Differences in Depressive and Anxious Symptoms Among Emotion Regulation Flexibility Profiles

Profile	Study 1 (DASS; N = 200)		Study 2 (MASQ; N = 802)
Profile	Depression (0–21)	Anxiety (0–21)	Anhedonic depression (22–110)	Anxious arousal (22–110)
Full sample	8.84 (10.68)	6.37 (7.83)	60.88 (18.69)	25.97 (11.02)
CIR	15.71 (11.82)	17.01 (10.22)	66.62 (15.66)	39.39 (15.87)
FIR	16.11 (10.86)	8.79 (7.50)	68.12 (14.76)	27.88 (10.66)
LRR	—	—	64.47 (14.16)	25.32 (8.63)
MFR	6.42 (8.62)	4.46 (6.87)	58.70 (16.17)	22.75 (7.07)
HFR	0.75 (1.57)	2.14 (2.65)	44.13 (12.21)	21.11 (5.02)

Note: Values are means with standard deviations in parentheses. DASS = Depression Anxiety Stress Scale; MASQ = Mood and Anxiety Symptom Questionnaire; CIR = context-insensitive regulators; FIR = feedback-irresponsive regulators; LRR = low-repertoire regulators; MFR = medium-flexibility regulators; HFR = high-flexibility regulators.

Table 3.

Effect Sizes and Significant Tests for Pairwise Comparisons in Symptoms of Depression and Anxiety

Type	Study 1 (DASS; N = 200)				Study 2 (MASQ; N = 802)
	Depression		Anxiety		Anhedonic depression		Anxious arousal
	Hedges’s g	p	Hedges’s g	p	Hedges’s g	p	Hedges’s g	p
CIR > FIR	−0.04	.90	0.96	.001	−0.10	.46	0.86	< .001
CIR > LRR	—	—	—	—	0.14	.30	1.08	< .001
CIR > MFR	1.01	.001	1.68	< .001	0.49	< .001	1.73	< .001
CIR > HFR	1.84	< .001	2.06	< .001	1.57	< .001	1.45	< .001
FIR > LRR	—	—	—	—	0.25	.07	0.26	.06
FIR > MFR	1.06	< .001	0.62	.001	0.59	< .001	0.64	< .001
FIR > HFR	1.81	< .001	1.10	< .001	1.74	< .001	0.76	< .001
LRR > MFR	—	—	—	—	0.36	.001	0.35	.007
LRR > HFR	—	—	—	—	1.53	< .001	0.58	< .001
MFR > HFR	0.72	< .001	0.36	.005	0.93	< .001	0.24	.02

Note: Hedges’s values are considered large, medium, and small at 0.80, 0.50, and 0.20, respectively. Bold p values are considered significant after Benjamini-Hochberg procedure with false discovery rate set at .05. CIR = context-insensitive regulators; FIR = feedback-irresponsive regulators; LRR = low-repertoire regulators; MFR = medium-flexibility regulators; HFR = high-flexibility regulators.

For depression, both inflexible profiles (CIR and FIR) exhibited greater levels of symptoms than the flexible profiles (MFR and HFR), although they did not differ from each other on symptom severity. CIR showed higher symptom level than HFR (g = 1.84) and MFR (g = 1.01), χ²s > 11, ps < .01, but did not differ from FIR (g = −0.04), χ² = 0.016, p = .90. FIR exhibited more depressive symptoms than HFR (g = 1.81) and MFR (g = 1.06), χ² > 24, ps < .001. Between the two flexible profiles, MFR showed higher symptom level than HFR (g = 0.72), χ² = 44, p < .001. The variances of depression explained by the latent profile memberships were large, ω² = .15.

For anxiety, similarly, both profiles with deficit in one flexibility component (CIR and FIR) showed greater symptom severity than MFR and HFR. CIR showed higher symptom levels than FIR (g = 0.96), MFR (g = 1.68), and HFR (g = 2.06), χ²s > 7, ps < .01. FIR exhibited higher symptom levels than MFR (g = 0.62) and HFR (g = 1.10), χ²s > 10, ps < .01. MFR had more symptoms than HFR (g = 0.36), χ² = 7.83, p = .005. The variances of anxiety explained by the latent profile memberships were large, ω² = .20.

Discussion

In Study 1, we were able to subtype four different groups of individuals that clearly demonstrated unique patterns of flexibility components. Although more than half the sample were average across these components (i.e., MFR), there were clear subgroups that appeared either highly flexible (i.e., HFR) or deficient in one flexibility component (i.e., FIR and CIR). We did not see a profile characterized by low repertoire, but this may be attributed to both our relatively small sample size and sample characteristics. In fact, although a five-profile solution was not optimal based on the fit indices, a close scrutiny of this solution revealed that it did include a small profile that was low in repertoire (low-repertoire regulators [LRR], 2.2%). This would add theoretical interpretability because there was no profile showing deficits in repertoire in the four-profile solution. Given that a five-profile solution exhibited the highest entropy (.84) and included an interpretable profile, recruiting a larger sample may help further determine whether retaining this solution (i.e., including LRR) is reasonable.

When linking latent profiles to depression and anxiety, we found that HFR exhibited the lowest depressive and anxious symptoms and that participants with deficits in one flexibility components (i.e., FIR and CIR) showed greater symptom severity than participants who were relatively flexible (i.e., HFR and MFR). These findings were consistent with the flexibility literature that has emphasized the adaptive functions of these components (Aldao, 2013; Aldao et al., 2015; Bonanno & Burton, 2013; Bonanno et al., 2020; Burton & Bonanno, 2016; Cheng et al., 2014; Kato, 2015). Individuals with deficit in one flexibility component (i.e., CIR and FIR) showed relatively comparable elevation in depression, but participants with deficit in context sensitivity (i.e., CIR) reported more anxious symptoms than participants with deficit in feedback responsiveness (i.e., FIR). For one thing, this is partially in line with literature suggesting that context sensitivity is the most crucial and first step in regulatory process (Bonanno et al., 2020; Bonanno & Burton, 2013). For another, our findings seem to suggest that despite the role flexibility deficits play in both depression and anxiety, the type of flexibility deficit matters more for anxiety than for depression. That there is a high degree of overlap in the depression and anxiety measures in Study 1; thus, to better understand whether there are unique and shared flexibility deficits in relation to depression and anxiety, we decided that it might be helpful to attempt to replicate the current findings with measures that better differentiate depression and anxiety.

Study 2

In Study 1, we successfully identified four predominant patterns of the three flexibility components. Two of the latent profiles were characterized by deficit in one flexibility component (i.e., CIR and FIR), whereas the remaining two showed moderate to high ability across components (i.e., HFR and MFR). Moreover, the constructed latent profiles derived from three ER flexibility components were associated with depressive and anxious symptoms.

In Study 2, we aimed to further our findings in the following ways. First, we recruited a new and considerably larger sample, which may potentially help us determine whether to include a fifth profile of low repertoire. Second, unlike most prior LPA studies, in Study 2, we attempted to double cross-validate the profiles across two random halves of the sample to ensure replicability of the profile solutions. By constraining the parameters in one sample to the parameters obtained from the other sample, we could examine the replicability of a few candidate latent profile solutions and select one that can reliably cross-validate. Third, to gain a more nuanced understanding of the association between latent profiles and symptoms of depression and anxiety, we adopted a measure known to differentiate unique and shared aspects of depression and anxiety.

Method

Data and participants

Study 2 was conducted using MTurk. We recruited a total of 802 participants (361 men, 436 women, two others, and three who preferred not to answer) on average 36.41 years old (SD = 10.79), who completed the measures and were paid $2 for their participation. The sample was racially diverse; 75.9% were White, 10.3% were Black or African American, 9.0% were Asian American, 2.5% were American Indian, 0.5% were Native Hawaiian, and 1.7% of participants preferred not to answer. In the sample, 10.2% of the participants self-identified as Hispanic or Latino, 88.7% of the participants considered themselves non-Hispanic, and 1.1% preferred not to answer. The institutional review broad approved this study.

Measures

As in Study 1, we aimed to establish latent profiles that exhibit different patterns across the three flexibility components and investigate their potential relationships with depressive and anxious symptoms.

Context sensitivity, repertoire, and feedback responsiveness were measured using the same instruments reported in the Measures section of Study 1.

Depressive and anxious symptoms were measured by the 62-item Mood and Anxiety Symptom Questionnaire-Short Form (MASQ-SF; Watson et al., 1995). The MASQ-SF assesses symptoms that commonly occur in anxiety and mood disorders. Items were rated on a 5-point scale (1 = not at all, 5 = extremely); higher scores indicate greater levels of depression and anxiety. The MASQ-SF consists of four subscales: General Distress-Anxiety Symptoms (GDA) subscale, which includes 11 items that reflect nonspecific symptoms of anxiety (e.g., “was unable to relax”); General Distress-Depressive Symptoms (GDD) subscale, which includes 12 items reflecting nonspecific depressive symptoms (e.g., “felt sluggish or tired”); Anxious Arousal (AA) subscale, which includes 17 items assessing anxiety-specific symptoms (e.g., “startled easily”); and Anhedonic Depression (AD) subscale, which includes 22 items that assess symptoms specific to depression (e.g., “felt withdrawn from other people”). In the present investigation, we used only the AA and AD subscales because they have lower overlap and are therefore likely to produce stronger empirical differentiation between the highly overlapping constructs of anxiety and depression (in this sample, for AA and AD, r = .39, p < .01, whereas for GDA and GDA, r = .82, p < .01). It should be noted, however, that the AA subscale tends to capture physiological arousal of anxiety disorders (e.g., panic) but provides less information about anxious distress (e.g., social anxiety). Watson et al. (1995) reported high levels of internal consistency in multiple samples. In this study, the αs for AA and AD subscales were .90 and .92, respectively.

Data-analytic plan

All latent profile analyses were analyzed with Mplus (Version 7.4; Muthén & Muthén, 2012). In addition to the three-step approach mentioned in Study 1 (Asparouhov & Muthén, 2013; Wisco et al., 2018), we used a double cross-validation approach to assess the replicability of the latent profiles (Masyn, 2013). Specifically, we randomly halved the sample, performed exploratory LPAs on each half separately, selected the best fitting models in each half, and then conducted confirmatory LPAs to validate them in the other half of the sample. Because the two halves have different means and standard deviations, we used original scores instead of standardized scores to perform all LPAs. However, when visualizing the results, to ease interpretation, we standardized the item thresholds on the basis of the means and standard deviations of the scores on the three components of the joint sample.

In each exploratory LPA, we used fit indices including AIC, BIC, entropy, LMR-LRT, and BLRT to identify a few candidate models. To begin with, we looked at LMR-LRT and BLRT, and when those two yielded different results, we preferred BLRT because it proved more consistent than other criteria (Nylund et al., 2007). In addition, we avoided solutions that included rare classes because they were usually unstable and difficult to replicate (Morovati, 2014; Nylund-Gibson & Choi, 2018). After deciding on a few candidate models, we performed double cross-validation to select solutions that could be reliably replicated in the other half of the sample. When more than one candidate model was successfully replicated, we chose the one that had higher interpretability. In terms of relationships between constructed latent profiles and outcomes, we set latent profile means and item thresholds to the values that resulted from the exploratory analyses. This ensured that the outcome variables (i.e., anhedonic depression and anxious arousal) were compared across latent profiles that were decided in previous steps of analysis, again enhancing replicability (Masyn, 2013).

Finally, to obtain variances accounted for (ω²) and effect sizes (Hedges’s g), two one-way ANOVAs were performed separately in each half and then in the combined sample using most likely class membership from the LPAs as the independent variable and the two outcomes (i.e., anhedonic depression and anxious arousal) as dependent variables. This will slightly underestimate the relationship, but it is recommended given that the three-step approach does not allow estimation of variances accounted for by profile memberships. Rules of thumb for interpreting ω² are .01 = small, .06 = medium, and .14 = large (Albers & Lakens, 2018). Hedges’s g values are considered large, medium, and small at 0.80, 0.50, and 0.20, respectively (Hedges, 1981).

Results

Exploratory latent profile analysis

We report all fit indices in Table 1. In the first half of the sample, the BLRTs were significant for the two-, three-, four-, and five-profile solutions (ps < .05) but not for the six-profile solution (p = .08), which suggests that the five-profile solution provided the best possible fit. The LMR-LRTs were significant for the two-, three-, four-, and five-profile solutions (ps < .05) but not for the six-profile solution (p = .06), again supporting the five-profile solution. In addition, the six-profile solution yielded a low-frequency profile (2.31%), which suggested that six-profile solution was not ideal. The AIC and BIC of the five-profile solution were lower than those of the four-, three-, and two-profile solutions, supporting the five-profile solution.

In the second half of the sample, the BLRTs were significant for the two-, three-, four-, five-, and six-profile solutions, supporting the six-profile solution. The LMR-LRTs were significant for the two-, three-, and four-profile solutions (ps < .001) but not for the five-profile solution (p = .07), supporting the four-profile solution. The six-profile solution had low-frequency profile (4.74%) and was thus not considered. Between the five- and four-profile solutions, the BIC and AIC values were lower for the five-profile solution, suggesting that it provided better fit.

Double cross-validation

The results of exploratory LPA in two random halves were mixed regarding whether a four- or five-profile solution should be adopted. In addition, given the BLRT statistics, the six-profile solution fit better than either the four- or five-profile solution for the second half of the sample. Therefore, we further performed double cross-validation for the four-, five-, and six-profile solutions to evaluate their respective replicability. Specifically, we examined whether the parameters for the four-, five-, and six-profile solutions estimated in one half of the sample (i.e., item thresholds and latent profile means) displayed good fit when applied to the other half of the sample and vice versa.

For both the four- and five-profile solutions, the likelihood ratio tests did not find significant differences in both halves between the freely estimated models and the models constrained by parameters estimated in the other half of the sample, χ²s < 19.70, ps > .50, indicating that these constraints did not reduce fit and that the models successfully cross-validated across samples. For the six-profile solution, however, the likelihood ratio tests revealed significant differences between the freely estimated and the constrained models, χ²s > 49, ps < .01, suggesting that the six-profile solution did not cross-validate for either half of the sample. In other words, the six-profile solution was not reliable across random halves and thus should not be adopted. Given that both the four- and five- profile solutions cross-validated successfully, we chose the five-profile solution as the optimal solution because it included a new profile that was clearly interpretable. The fifth profile was low in repertoire but medium in context sensitivity and feedback responsiveness.

Because the patterns of the five-profile solution were similar in both halves, we report the standardized group means and symptom comparison using the full sample for brevity (Fig. 2). Figure 2a presents the standardized group means for the five-profile solution. Participants were assigned to one of the following profiles: (a) high ability across three components (HFR, 9.8%), (b) average ability across three components (MFR, 47.5%), (c) average levels of context sensitivity and repertoire but low levels of feedback responsiveness (FIR, 15.3%), (d) average levels of feedback responsiveness and repertoire but low levels of context sensitivity (CIR, 14.0%), or (e) average levels of feedback responsiveness and context sensitivity but low levels of repertoire (LRR, 13.3%). Classification probabilities were high across the five profiles, with probabilities of .92 in HFR, .85 in MFR, .89 in FIR, .90 in CIR, and .81 in the LRR.

Fig. 2.

Sample item means for the five latent profiles of emotion regulation (ER) flexibility (a) and comparisons of depressive and anxious symptoms among these profiles (b) in Study 2. Here we show findings of only the full sample for brevity. MASQ = Mood and Anxiety Symptom Questionnaire; HFR = high-flexibility regulators; MFR = medium-flexibility regulators; CIR = context-insensitive regulators; FIR = feedback-irresponsive regulators; LRR = low-repertoire regulators.

Associations with depression and anxiety

We examined potential differences between the five profiles on anhedonic depression and anxious arousal using each of the distal outcomes as auxiliary variables in the model. Because results were similar in both halves, we report results using the full sample for brevity (Fig. 2b). The p values for each pairwise comparison and corresponding effect size are presented in Table 3. To adjust for multiple comparison and control false discovery rate at .05, the B-H procedure was adopted (Benjamini & Hochberg, 1995). A total of 20 pairwise comparison tests were conducted. We ranked the p values for pairwise tests and compared them with their respective B-H critical value. All p values below .02 were lower than their respective B-H critical values and thus considered significant. We report detailed results of symptom comparison in the following paragraphs.

For anhedonic depression, the inflexible profiles (CIR, LRR, and FIR) showed higher symptom levels than that of the flexible profiles (HFR and MFR), χ²s > 3.6, ps < .01. HFR had the lowest symptom level, χ²s > 12, ps < .001. CIR did not differ from LRR (g = 0.14) or FIR (g = −0.10) in symptom severity, χ²s < 1.1, ps > .10, but exhibited greater symptom severity than MFR (g = 0.49) and HFR (g = 1.57). FIR and LRR did not differ in symptom severity (g = 0.25), χ² = 3.34, p = .07, which was consistent with the findings from Study 1 that the type of deficit was not associated with different elevation in depressive symptoms. The variances explained by the latent profile memberships were medium, ω² = .09.

For anxious arousal, consistent with findings on anhedonic depression, the inflexible profiles (CIR, LRR, and FIR) exhibited greater symptom severity than the flexible profiles (HFR and MFR), χ²s > 15, ps < .01. CIR exhibited higher symptom levels than any of the other groups (gs > 0.85), χ²s > 40, ps < .001. FIR showed higher symptom levels than MFR (g = 0.64) and HFR (g = 0.76), χ²s > 23, ps < .001, but did not differ from the LRR (g = 0.26), χ² = 3.68, p = .06. LRR had more symptoms than MFR (g = 0.35), χ² = 7.16, p = .007, and showed greater symptom severity than HFR (g = 0.58), χ² = 15.73, p < .001. In sum, participants who showed high ability across three ER flexibility components exhibited the least anxious symptoms, whereas participants showing one deficit, especially context insensitivity, exhibited greater symptom severity. The variances explained by the latent profile memberships were large, ω² = .21.

Discussion

In Study 2, we used a rigorous double cross-validation approach to identify subtypes of regulators that differ in components of ER flexibility. We adopted the five-profile solution on the basis of fit indices, replicability across samples, and interpretability. This profile solution suggests that there were clearly distinguishable subpopulations that are highly flexible (i.e., HFR), moderately flexible (i.e., MFR), or inflexible in one of three regulatory components (i.e., CIR, LRR, and FIR). These findings appeared inconsistent with the results of Study 1, in which the four-profile solution fit the best. However, this could be partially attributed to the inability to always identify rare profile with small samples. Moreover, in Study 2, we found that both the four- and five-profile solutions could reliably cross-validate, which suggests that both the four- and five-profile solutions could fit the data relatively well. We preferred the inclusion of a fifth profile when identifiable because it clearly added theoretical interpretability and allowed comparison of symptoms between groups with different flexibility deficits.

Furthermore, these groups differed in depressive and anxious symptoms. Participants who were highly flexible across regulatory components exhibited the least depressive and anxious symptoms. On the contrary, participants with flexibility deficits showed worse outcomes, but results were divergent for anxious and depressive symptoms. CIR exhibited the highest level of anxiety, followed by FIR and LRR. For depressive symptoms, the three inflexible profiles with deficits in one regulatory component did not differ with each other. These findings were consistent with those from Study 1, which suggests that flexibility deficits, regardless of which component, are associated with elevated depressive symptoms, whereas impairment in context sensitivity is clearly related to increased anxious symptoms beyond other flexibility deficits.

General Discussion

Sequential, multicomponent models of flexibility have been proposed in review and meta-analyses (Bonanno & Burton, 2013; Cheng et al., 2014), and individual components of ER flexibility have consistently shown links to health and psychopathology. However, no empirical study has yet examined all components of the flexibility model within the same study and examined their hypothesized combined benefits to mental health. In two studies, we addressed this gap by applying LPA to the three components of ER flexibility proposed by Bonanno and Burton (2013) and by connecting profile memberships with the severity of depressive and anxious symptoms.

Our findings demonstrated unique profiles of ER flexibility. Specifically, the most common profile included around half of the participants, who showed an average ability across all three flexibility components (i.e., MFR). By contrast, superior flexibility, manifesting as high scores across all three regulatory components, was evident in just about one tenth of the sample (i.e., HFR). Approximately 30% of the participants belonged to any of the three relatively inflexible profiles, manifesting as impairment in one regulatory component: Some exhibited difficulty accurately perceiving the cues in stressful situations (i.e., CIR), some experienced challenges in enhancing and suppressing emotions (i.e., LRR), and others could not effectively monitor and adjust strategy use via feedback (i.e., FIR). Note that inflexible regulators with deficit in one component generally did not show high abilities in the other two domains comparably to HFR, which suggests that failure at one flexibility component may make it unlikely to succeed at the others (Bonanno & Burton, 2013).

The number of profiles that emerged was similar but not identical across the two studies. In Study 1, we did not observe a latent profile that captured a deficit in repertoire (i.e., LRR). In Study 2, however, we were able to identify LRR in both halves of the sample (N = 401 in each half). Whereas Study 1 supported a four-profile solution, Study 2 showed that both the four- and five-profile solutions successfully cross-validated across random halves. That the first study did not support the five-profile solution may result from both the small sample size and sample variability. This occurs frequently for small classes, as previously demonstrated in simulation studies (Morovati, 2014). In other words, given the relatively low frequency of LRR and sample variability, it could not always be identified in studies with small sample size, such as Study 1 (N = 200). As mentioned in Study 2, the inclusion of a fifth profile increases theoretical interpretability by capturing deficits in all components of flexibility. Moreover, the inclusion of LRR allows comparison of symptoms among participants with different types of flexibility deficits (CIR, FIR, and LRR).

Despite the apparent inconsistency, profile solutions in both studies shared great similarity. First, the order in which the latent profiles emerged was consistent in both studies: MFR, HFR, CIR, and FIR. Second, although the fit indices reported in Study 1 failed to support the five-profile solution, we did observe a fifth profile characterized by low repertoire (LRR, 2.2%) when setting the number of profiles to five. Although likely unstable because of its small size, this profile was consistent with the fifth profile identified in Study 2. Third, the pattern of relations between profile memberships and psychopathology symptoms was similar across the two studies. For instance, inflexible profiles (CIR, LRR, and FIR) had greater levels of depressive and anxious symptoms than the two flexible profiles (MFR and HFR). In addition, inflexible groups did not differ in symptoms of depression, but CIR exhibited significantly greater severity of anxiety than groups with other deficits.

From a clinical science perspective, our studies contribute to a growing body of literature that emphasize the effect of flexibility on psychological health. Crucially and consistent with prevailing assumptions in the flexibility literature, the latent profiles were associated with depressive and anxious symptoms. As previously hypothesized (e.g., Bonanno & Burton, 2013), HFR had the lowest levels of symptoms, which again highlights the importance of all three flexibility components. However, varied combinations of the three components also showed differential relations to adjustment. Also as previously hypothesized (Bonanno & Burton, 2013; Bonanno et al., 2020), CIR had the poorest overall adjustment. CIR showed greater elevation of depression and anxiety than that of the HRF and MFR and more anxiety symptoms than all other profiles. These findings suggest that context insensitivity may confer particular risk for anxiety, or anxiety may make people less sensitive to context. Although there has been relatively little research on the feedback component of ER flexibility, FIR also showed poor adjustment as evidenced by high levels of anxiety and depression. In fact, FIR did not differ statistically from CIR on depression in both studies. These findings suggest that anxiety is particularly high for individuals with deficits in context sensitivity, whereas elevation of depression seems similar across individuals with different types of flexibility impairments. Future studies adopting longitudinal designs and recruiting clinical population with depression, anxiety, and comorbid condition may elucidate shared and unique flexibility impairments in depression and anxiety and establish temporal priority of flexibility and psychopathology, which could potentially improve the understanding of the highly comorbid nature of these two forms of disorders.

The advances suggested by our findings should be understood in the context of several noteworthy methodological limitations. First of all, the cross-sectional nature of our study precludes us from establishing causal relationships or evaluating the stability of latent profiles over time. Although previous studies have demonstrated the longitudinal predictability of flexibility (e.g., Bonanno et al., 2004; Westphal et al., 2010), no study to date has examined whether deficits in flexibility components precede depression and anxiety or whether the onset of these conditions decreases flexibility over time. Although our measures were previously found to have very low social desirability, it is still likely that the presence of depression and anxiety may lead participants to down-rate their perceived ability in these flexibility components. Longitudinal studies could potentially further our findings by establishing temporal order of flexibility and psychopathology. This may help shed light on the etiology of depression and anxiety as well as differential diagnosis. In addition, our assessment of the ER flexibility components was limited to self-report measures. Although two of the three measures were validated against experimental and behavioral indexes, it is necessary for future research to adopt other flexibility assessments, including self-report as well as experimental measures.

Moreover, our measure of repertoire was limited in that it assessed only up-regulation and down-regulation of emotional expression, which prohibits the understanding of repertoire in ER flexibility. Although this was frequently examined in the flexibility literature (e.g., Bonanno et al., 2004; Burton & Bonanno, 2016; Westphal et al., 2010), other studies have conceptualized repertoire as the use of a wider range of strategies (e.g., Dixon-Gordon et al., 2015; Southward, Altenburger, Moss, Cregg, & Cheavens, 2018). This limitation also precluded us from examining whether individuals with high ability across flexibility components are overlapping or populated by individuals who use putatively adaptive strategies. One study, for example, showed that individuals relying heavily on adaptive strategies exhibited fewer depressive and anxious symptoms (Dixon-Gordon et al., 2015). To examine additive and unique effects of flexibility, comparing flexibility and trait emotion regulation in the same study could be potentially informative and beneficial. Future studies may benefit from considering various operationalizations of regulatory repertoire and their potential impact on findings. Finally, our analyses were bound to one theoretical perspective on ER flexibility. It may be worth examining other models (Aldao et al., 2015; Pruessner, Barnow, Holt, Joormann, & Schulze, 2020) as well as alternative flexibility components, such as preexisting goals (e.g., Millgram et al., 2015).

In conclusion, five subtypes of regulators were identified on the basis of the flexibility model proposed by Bonanno and Burton (2013). Individuals with great abilities in all regulatory components exhibited the least depressive and anxious symptoms. On the contrary, inflexibility, as demonstrated by deficits in context sensitivity, repertoire, or feedback responsiveness, was associated with depressive and anxious symptoms. Individuals who were relatively insensitive to context exhibited the highest level of anxiety. These findings are consistent with theory highlighting the importance of flexibility (Aldao et al., 2015; Bonanno & Burton, 2013; Cheng et al., 2014; Kashdan & Rottenberg, 2010). Future research may extend these findings by incorporating longitudinal designs, adopting multimethod measures of different flexibility components, and examining the relationship between flexibility and psychopathology in various samples.

Footnotes

Acknowledgements

Data and materials (except for third-party materials) associated with this article are available at .

Transparency

Action Editor: Christopher G. Beevers

Editor: Kenneth J. Sher

Author Contributions

S. Chen developed the study concept, designed the study, and performed the data analysis and interpretation under the supervision of G. A. Bonanno. S. Chen drafted the manuscript, and G. A. Bonanno provided critical revisions. Both authors approved the final manuscript for submission.

ORCID iD

Shuquan Chen

References

Albers

Lakens

(2018). When power analyses based on pilot data are biased: Inaccurate effect size estimators and follow-up bias. Journal of Experimental Social Psychology, 74, 187–195. doi:10.1016/j.jesp.2017.09.004

Aldao

(2013). The future of emotion regulation research: Capturing context. Perspectives on Psychological Science, 8, 155–172. doi:10.1177/1745691612459518

Aldao

Nolen-Hoeksema

Schweizer

(2010). Emotion-regulation strategies across psychopathology: A meta-analytic review. Clinical Psychology Review, 30, 217–237. doi:10.1016/j.cpr.2009.11.004

Aldao

Sheppes

Gross

J. J.

(2015). Emotion regulation flexibility. Cognitive Therapy and Research, 39, 263–278. doi:10.1007/s10608-014-9662-4

Asparouhov

Muthén

(2013). Auxiliary variables in mixture modeling: A 3-step approach using Mplus (Mplus Web Notes: No. 15, Version 6). Retrieved from https://www.statmodel.com/examples/webnotes/AuxMixture_submitted_corrected_webnote.pdf

Benjamini

Hochberg

(1995). Controlling the false discovery rate: A practical and powerful approach to multiple testing. Journal of the Royal Statistical Society: Series B: Methodological, 57, 289–300. doi:10.1111/j.2517-6161.1995.tb02031.x

Birk

J. L.

Bonanno

G. A.

(2016). When to throw the switch: The adaptiveness of modifying emotion regulation strategies based on affective and physiological feedback. Emotion (Washington, D.C.), 16, 657–670. doi:10.1037/emo0000157

Bonanno

G. A.

Burton

C. L.

(2013). Regulatory flexibility: An individual differences perspective on coping and emotion regulation. Perspectives on Psychological Science, 8, 591–612. doi:10.1177/1745691613504116

Bonanno

G. A.

Maccallum

Malgaroli

Hou

W. K.

(2020). The Context Sensitivity Index (CSI): Measuring the ability to identify the presence and absence of stressor context cues. Assessment, 27, 261–273. doi:10.1177/1073191118820131

10.

Bonanno

G. A.

Papa

Lalande

Westphal

Coifman

(2004). The importance of being flexible: The ability to both enhance and suppress emotional expression predicts long-term adjustment. Psychological Science, 15, 482–487. doi:10.1111/j.0956-7976.2004.00705.x

11.

Bonanno

G. A.

Pat-Horenczyk

Noll

(2011). Coping flexibility and trauma: The Perceived Ability to Cope with Trauma (PACT) scale. Psychological Trauma: Theory, Research, Practice, and Policy, 3, 117–129 doi:10.1037/a0020921

12.

Burton

C. L.

Bonanno

G. A.

(2016). Measuring ability to enhance and suppress emotional expression: The Flexible Regulation of Emotional Expression (FREE) Scale. Psychological Assessment, 28, 929–941. doi:10.1037/pas0000231

13.

Burton

C. L.

Yan

O. H.

Pat-Horenczyk

Chan

I. S. F.

Bonanno

G. A.

(2012). Coping flexibility and complicated grief: A comparison of American and Chinese samples. Depression and Anxiety, 29, 16–22. doi:10.1002/da.20888

14.

Casler

Bickel

Hackett

(2013). Separate but equal? A comparison of participants and data gathered via Amazon’s MTurk, social media, and face-to-face behavioral testing. Computers in Human Behavior, 29, 2156–2160. doi:10.1016/j.chb.2013.05.009

15.

Chen

Bonanno

G. A.

(2018). Expressive flexibility: Enhancement and suppression abilities differentially predict life satisfaction and psychopathology symptoms. Personality and Individual Differences, 126, 78–84. doi:10.1016/j.paid.2018.01.010

16.

Cheng

(2001). Assessing coping flexibility in real-life and laboratory settings: A multimethod approach. Journal of Personality and Social Psychology, 80, 814–833.

17.

Cheng

Lau

H.-P. B.

Chan

M.-P. S.

(2014). Coping flexibility and psychological adjustment to stressful life changes: A meta-analytic review. Psychological Bulletin, 140, 1582–1607. doi:10.1037/a0037913

18.

Cohen

(1988). Statistical power analysis for the behavioral sciences (2nd ed.). Mahwah, NJ: Lawrence Erlbaum.

19.

Crowne

D. P.

Marlowe

(1960). A new scale of social desirability independent of psychopathology. Journal of Consulting Psychology, 24, 349–354. doi:10.1037/h0047358

20.

Dixon-Gordon

K. L.

Aldao

De Los Reyes

(2015). Repertoires of emotion regulation: A person-centered approach to assessing emotion regulation strategies and links to psychopathology. Cognition and Emotion, 29, 1314–1325. doi:10.1080/02699931.2014.983046

21.

Gross

J. J.

(1999). Emotion regulation: Past, present, future. Cognition and Emotion, 13, 551–573. doi:10.1080/026999399379186

22.

Gross

J. J.

(2015). Emotion regulation: Current status and future prospects. Psychological Inquiry, 26, 1–26. doi:10.1080/1047840X.2014.940781

23.

Gross

J. J.

John

O. P.

(2003). Individual differences in two emotion regulation processes: Implications for affect, relationships, and well-being. Journal of Personality and Social Psychology, 85, 348–362. doi:10.1037/0022-3514.85.2.348

24.

Hedges

L. V.

(1981). Distribution theory for Glass’s estimator of effect size and related estimators. Journal of Educational Statistics, 6, 107–128. doi:10.2307/1164588

25.

Ilan

Shafir

Birk

J. L.

Bonanno

G. A.

Sheppes

(2019). Monitoring in emotion regulation: Behavioral decisions and neural consequences. Social Cognitive and Affective Neuroscience, 14, 1273–1283. doi:10.1093/scan/nsaa001

26.

Kashdan

T. B.

Rottenberg

(2010). Psychological flexibility as a fundamental aspect of health. Clinical Psychology Review, 30, 865–878. doi:10.1016/j.cpr.2010.03.001

27.

Kato

(2012). Development of the Coping Flexibility Scale: Evidence for the coping flexibility hypothesis. Journal of Counseling Psychology, 59, 262–273. doi:10.1037/a0027770

28.

Kato

(2015). The impact of coping flexibility on the risk of depressive symptoms. PLOS ONE, 10(5), Article e0128307. doi:10.1371/journal.pone.0128307

29.

Kato

(2017). Effects of flexibility in coping with menstrual pain on depressive symptoms. Pain Practice, 17, 70–77. doi:10.1111/papr.12412

30.

Lazarus

R. S.

Folkman

(1984). Stress, appraisal, and coping. New York, NY: Springer.

31.

Lovibond

S. H.

Lovibond

P. F.

(1995). Manual for the Depression Anxiety Stress Scales (DASS) (2nd ed., Psychology Foundation Monograph). Kensington, Australia: University of New South Wales.

32.

Masyn

K. E.

(2013). Latent class analysis and finite mixture modeling. In Little

T. D.

(Ed.), The Oxford handbook of quantitative methods in psychology (Vol. 2, pp. 551–611). New York, NY: Oxford University Press. doi:10.1093/oxfordhb/9780199934898.013.0025

33.

Merians

A. N.

Baker

M. R.

Frazier

Lust

(2019). Outcomes related to adverse childhood experiences in college students: Comparing latent class analysis and cumulative risk. Child Abuse & Neglect, 87, 51–64. doi:10.1016/j.chiabu.2018.07.020

34.

Millgram

Joormann

Huppert

J. D.

Tamir

(2015). Sad as a matter of choice? Emotion-regulation goals in depression. Psychological Science, 26, 1216–1228. doi:10.1177/0956797615583295

35.

Mischel

(1973). Toward a cognitive social learning reconceptualization of personality. Psychological Review, 80, 252–283. doi:10.1037/h0035002

36.

Morovati

(2014). The intersection of sample size, number of indicators, and class enumeration in LCA: A Monte Carlo study (Unpublished doctoral dissertation). University of California, Santa Barbara.

37.

Muthén

L. K.

Muthén

B. O.

(2012). Mplus users’ guide (7th ed.). Los Angeles, CA: Author.

38.

Nylund

K. L.

Asparouhov

Muthén

B. O.

(2007). Deciding on the number of classes in latent class analysis and growth mixture modeling: A Monte Carlo simulation study. Structural Equation Modeling: A Multidisciplinary Journal, 14, 535–569. doi:10.1080/10705510701575396

39.

Nylund-Gibson

Choi

A. Y.

(2018). Ten frequently asked questions about latent class analysis. Translational Issues in Psychological Science, 4, 440–461. doi:10.1037/tps0000176

40.

Orcutt

H. K.

Bonanno

G. A.

Hannan

S. M.

Miron

L. R.

(2014). Prospective trajectories of posttraumatic stress in college women following a campus mass shooting. Journal of Traumatic Stress, 27, 249–256. doi:10.1002/jts.21914

41.

Paolacci

Chandler

Ipeirotis

P. G.

(2010). Running experiments on Amazon Mechanical Turk. Judgment and Decision Making, 5, 411–419.

42.

Pruessner

Barnow

Holt

D. V.

Joormann

Schulze

(2020). A cognitive control framework for understanding emotion regulation flexibility. Emotion, 20, 21–29. doi:10.1037/emo0000658

43.

Rodin

Bonanno

G. A.

Rahman

Kouri

N. A.

Bryant

R. A.

Marmar

C. R.

Brown

A. D.

(2017). Expressive flexibility in combat veterans with posttraumatic stress disorder and depression. Journal of Affective Disorders, 207, 236–241. doi:10.1016/j.jad.2016.09.027

44.

Rottenberg

Gross

J. J.

Gotlib

I. H.

(2005). Emotion context insensitivity in major depressive disorder. Journal of Abnormal Psychology, 114, 627–639. doi:10.1037/0021-843X.114.4.627

45.

Sheppes

Scheibe

Suri

Radu

Blechert

Gross

J. J.

(2014). Emotion regulation choice: A conceptual framework and supporting evidence. Journal of Experimental Psychology: General, 143, 163–181. doi:10.1037/a0030831

46.

Southward

M. W.

Cheavens

J. S.

(2017). Assessing the relation between flexibility in emotional expression and symptoms of anxiety and depression: The roles of context sensitivity and feedback sensitivity. Journal of Social and Clinical Psychology, 36, 142–157. doi:10.1521/jscp.2017.36.2.142

47.

Southward

M. W.

Altenburger

E. M.

Moss

S. A.

Cregg

D. R.

Cheavens

J. S.

(2018). Flexible, yet firm: A model of healthy emotion regulation. Journal of Social and Clinical Psychology, 37, 231–251. doi:10.1521/jscp.2018.37.4.231

48.

Tamir

Ford

B. Q.

(2012). When feeling bad is expected to be good: Emotion regulation and outcome expectancies in social conflicts. Emotion, 12, 807–816. doi:10.1037/a0024443

49.

Troy

A. S.

Ford

B. Q.

McRae

Zarolia

Mauss

I. B.

(2017). Change the things you can: Emotion regulation is more beneficial for people from lower than from higher socioeconomic status. Emotion, 17, 141–154. doi:10.1037/emo0000210

50.

Troy

A. S.

Shallcross

A. J.

Mauss

I. B.

(2013). A person-by-situation approach to emotion regulation: Cognitive reappraisal can either help or hurt, depending on the context. Psychological Science, 24, 2505–2514. doi:10.1177/0956797613496434

51.

Watson

Weber

Assenheimer

J. S.

Clark

L. A.

Strauss

M. E.

McCormick

R. A.

(1995). Testing a tripartite model: I. Evaluating the convergent and discriminant validity of anxiety and depression symptom scales. Journal of Abnormal Psychology, 104, 3–14. doi:10.1037/0021-843X.104.1.3

52.

Webb

T. L.

Miles

Sheeran

(2012). Dealing with feeling: A meta-analysis of the effectiveness of strategies derived from the process model of emotion regulation. Psychological Bulletin, 138, 775–808. doi:10.1037/a0027600

53.

Westphal

Seivert

N. H.

Bonanno

G. A.

(2010). Expressive flexibility. Emotion, 10, 92. doi:10.1037/a0018420

54.

Wisco

B. E.

Plate

A. J.

May

C. L.

Aldao

(2018). A latent profile analysis of repetitive negative thinking: Distinguishing ruminators from worriers. Journal of Experimental Psychopathology. doi:10.1177/2043808718811428