(Re)Framing Resilience: A Trajectory-Based Study Involving Emerging Religious/Spiritual Leaders

Abstract

The COVID-19 pandemic has provided a unique circumstance for the study of resilience, and clergy resilience has garnered increased research attention due to greater recognition that religious/spiritual leaders are at risk for elevated levels of anxiety and burnout. We examined longitudinal patterns of change during the pandemic in a sample of emerging leaders (N = 751; M_age = 32.82; SD 11.37; 49.9% female; 59.8% White). In doing so, we offered a conceptual and methodological approach based on historical and critical evaluations of the study of resilience. Results revealed a subgroup that exhibited resilience over three waves of data. The labeling of this trajectory was based on established criteria for determining resilience: (a) significant adversity in the form of COVID-19 stress at time 1, which included the highest levels of the subjective appraisal of stress; (b) risk in the form of low religiousness/spirituality and greater likelihood of reporting marginalized identifications, relative to those who were flourishing; (c) a protective influence for transformative experiences to promote positive adaptation; and (d) interruption to the trajectory in the form of improvement in levels of symptoms and well-being. Practical implications center on the potential for transformative experiences to clarify emotional experience and construct new meaning.

1. Introduction

Clergy resilience is garnering increased research attention (e.g., Clarke 2022; Clarke et al. 2022), in large part due to greater recognition that religious/spiritual (R/S) leaders are an understudied, high-risk population (e.g., Terry and Cunningham 2020). As helping professionals, R/S leaders are susceptible to elevated levels of burnout (Clarke et al. 2022) and mental-health symptoms (Proeschold-Bell et al. 2015), including clinical levels of post-traumatic stress (Ruffing et al. 2021). R/S leaders face a variety of stressors, such as direct and indirect exposure to trauma (Wang et al. 2014), navigating multiple and potentially competing role demands, unrealistic expectations from some congregants, isolation, and limited remuneration. Studies on resilience are emerging within the larger literature on clergy well-being, and we agree that the study of resilience “may provide valuable intelligence to mitigate” the potential for burnout and mental-health symptoms (Clarke 2022, p. 2). A subset of clergy well-being research has focused on seminary students, during a time when the development of mitigation strategies is particularly pertinent (e.g., Lowe et al. 2022). However, conceptual and methodological problems with the study of resilience may limit the utility of findings to inform mitigation efforts.

1.1. Conceptual and Methodological Concerns

Raetze et al. (2022) suggested that the literature on resilience has a construct validity problem in the form of Kelley’s (1927) jingle fallacy, “where different meanings are attributed to a single construct label” (p. 868), while Williams and Kemp (2019) noted at least eight distinct definitions for resilience. Furthermore, the jingle fallacy calls attention to the misconception that all operationalizations of a similarly named construct are assessing the same construct (Hodson 2021; Lilienfeld and Strother 2020). Raetze et al. (2022) also suggested that the proliferation of research across disciplines has resulted in a conceptual “scattering in the wind” (p. 874). Possible conceptual drift can be seen in the move away from the origins of resilience as a developmental construct (Masten 2021; O’Connor et al. 2015; Zimmerman and Arunkumar 1994) to framing resilience as a self-perceived personality trait, defined as a capacity for or likelihood of “showing positive adaptation in the face of significant adversity” (Bonanno 2012; Britt et al. 2016, p. 380).

Infurna and Luthar (2018) challenged the conceptualization of resilience as a stable high-functioning trajectory (e.g., Bonanno 2012) and suggested that evidence for the prevalence of this trajectory was a methodological artifact. Historically, resilience was defined as “functional development against the odds” (Raetze et al. 2022, p. 884) and “doing better than expected, given the adversity” (Infurna and Luthar 2018, p. 51). Resilience seemed to be the exception rather than the norm. However, Masten (2001) also noted resilience to be “a common phenomenon” (p. 227). Yet, her use of common seemed less about prevalence and more about “ordinary human adaptive processes” (p. 234) and “ordinary, normative human resources” (p. 235), and particularly so relative to the view that resilience consisted of the “rare and special qualities” of individuals (p. 235). Masten contrasted a narrow trait perspective of resilience with that of a process view that highlighted greater potential for positive adaptation. In fact, Masten (2021) described multiple longitudinal pathways that depict resilience. Infurna and Luthar (2018) similarly broadened the definition of resilience to include other longitudinal patterns of change. Conceptualizations of resilience contextualize the term common, and perhaps over decades of research, there have been varying moves away from a view that resilience involves normative developmental and systemic factors that contribute to multiple positive adaptive processes to that of resilience as a single stable trajectory of healthy functioning.

From a construct validation perspective, evolution differs from drift. Research findings can contribute to the evolution of a theoretical construct (Hoyt et al. 2006), an aspect of which is clarification of its nomological network (Lilienfeld and Strother 2020). The construct of resilience has evolved such that systemic and process elements have become more pronounced (Masten 2021; Williams and Kemp 2019). Additionally, evolution is distinct from the aforementioned problem of meaning proliferation (i.e., jingle fallacy) and the problem of construct proliferation. Hodson (2021) applied Kelley’s (1927) jangle fallacy, defined as assuming “differently-named constructs are distinct from each other” (Hodson 2021, p. 577), to caution against redundant constructs. Brown et al. (2020) suggested that the constructs of resilience and growth may represent a jangle fallacy, and coping, flourishing, well-being, and thriving can be added to this list of conflated constructs (Ettinger et al. 2022; Gupta and McCarthy 2022). The conceptual overlap among these well-being-related constructs becomes even more challenging when R/S constructs are included in analyses, given noted concerns about the redundancy between various operationalizations of religiousness/spirituality and well-being (Jankowski et al. 2022c). Hodson (2021) called for “researchers to prioritize construct validity in their work” (p. 577), which at a minimum would seem to involve clarification of conceptual distinctions among constructs and an assessment of the discriminant validity evidence supporting those distinctions.

1.2. (Re)Framing Resilience

The “target stressor event” (Bonanno 2012, p. 755) is a key to punctuating resilience and has historically constituted “significant adversity,” typically delineated by events that can be considered traumatic or chronic (Britt et al. 2016, p. 381), whereas contemporary resilience scholarship also includes “acute life events” (Bonanno 2012, p. 754). Raetze et al. (2022) called for preciseness when defining significant adversity, and Britt et al. (2016) acknowledged that defining significant adversity involves the individual’s subjective appraisal of the exposure. Although an exposure or target stressor event can constitute significant adversity and therefore place individuals at risk for positive adaptation, it can be useful to consider adverse events as distinct from risk factors (e.g., Raetze et al. 2022). Risk factors include stressful life events (i.e., stressors) and other individual (e.g., experiential avoidance) and contextual (e.g., social inequality) variables that tend to have a negative influence on positive adaptation (Masten 2021).

Zimmerman and Arunkumar (1994) defined resilience as the “factors and processes that interrupt” a maladaptive trajectory to “adaptive outcomes even in the presence of adversity” (p. 4). They called for corresponding longitudinal data analytic strategies, and Bonanno (2012) emphasized the need for “repeated longitudinal [designs] …, with outcome measurements beginning as close as possible to the target stressor event” (p. 755). Nevertheless, Zimmerman and Arunkumar (1994) acknowledged a place for cross-sectional designs to provide “a snapshot” of resilience processes (p. 10), and yet, “the construct of interest is longitudinal,” despite the prevalence of cross-sectional designs (Raetze et al. 2022, p. 885). As such, designs are needed to map risk factors, protective factors, and outcomes over time (Zimmerman and Arunkumar 1994), each of which highlights additional methodological considerations. At least one clearly identifiable target stressor event is needed since resilience is a response, in the context of other individual and/or system-level risks or exposures that would typically predict maladaptation. Second, at least one protective factor is needed that serves to interrupt the predicted maladaptive trajectory (Zimmerman and Arunkumar 1994). Third, an outcome for conceptualizing positive adaptation is needed. O’Connor et al. (2015) defined resilience as “good outcomes in spite of serious threats to adaptation” (p. 602), and Masten (2021) suggested “observable ‘good adaptation’ in the context of adversity” (p. 117). The term observable connotes that resilience is inferred from “observed pathways of manifested resilience” (p. 117), and ideally with multiple indicators for determining good or positive adaptation.

In the current study, we examined self-reported COVID-19 stress as a risk factor, with the pandemic as a contextual factor serving as the target stressor event. Time 1 data were collected from seminary students during the 2021 spring semester, one year post-pandemic declaration, following a post-holiday surge in infections, the 2020 death toll in the United States (US) reaching 346,000 and 1,824,590 globally, public debates about vaccinations and stay-at-home orders, and political upheaval following the 2020 US election of Joe Biden as president (American Journal of Managed Care 2021). A January 2021 report by the American Psychological Association (2021) indicated that US adults reported their highest levels of stress since the pandemic began. Thus, the pandemic during the early months of 2021 connotes significant adversity. We also examined demographic risk factors for elevated mental-health symptoms during the pandemic (i.e., young adult, female gender, liberal ideological commitment, sexual minority, racial/ethnic minority; e.g., Filindassi et al. 2022; Fish et al. 2021; Na et al. 2022; Robillard et al. 2020; Thomeer et al. 2022).

We examined potential protective R/S factors, informed by the relational spirituality model (RSM; Sandage et al. 2020). The RSM focuses on ways individuals relate to whatever they consider sacred or ultimate, and the model has been widely utilized in research examining the well-being of seminary students and R/S leaders (e.g., Jankowski et al. 2019; Jankowski et al. 2022b, 2022d; Sandage et al. 2010, 2011). We used secure attachment to God as a protective R/S factor since it demonstrated a particular protective effect against elevated symptom levels in response to the pandemic in a diverse national US sample, which included various religious affiliations (Zhu and Upenieks 2022). Based on attachment theory, secure attachment to God typically facilitates emotion regulation in connection to a benevolent attachment figure (safe-haven function) and exploration of new R/S meaning (secure-base function; Sandage et al. 2020). As a developmental model, the RSM focuses on factors associated with change, and so we assessed transformative experiences as a protective factor (e.g., Manning et al. 2019). There is a long history of research on transformative experiences within the psychology of religion, and two prior studies with seminary students used the measure employed in this study (Sandage et al. 2010, 2011).

1.3. Transformative Experiences

Chirico et al. (2022) offered an integrative, interdisciplinary definition for transformative experiences as “brief experiences, perceived as extraordinary and unique, … involving epistemic expansion … heightened emotional complexity” (p. 14). Chirico et al. highlighted Miller and C’de Baca’s (2001) contribution of quantum change, defined as sudden rather than gradual change, although these experiences can be embedded in “a continuing growth process” (C’de Baca and Wilbourne 2004, p. 531). The construct of post-traumatic growth was also highlighted by Chirico et al. (2022), with emphasis on “stressful and traumatic events as key elicitors” of transformative experiences (p. 3). Post-traumatic growth (PTG) tends to emphasize change as “an ongoing process” (Tedeschi and Calhoun 2004, p. 1) and “not simply a return to baseline—it is an experience of improvement” (p. 4). Although quantum change and PTG connote positive growth, they can involve negative outcomes. C’de Baca and Wilbourne (2004), for example, observed that most descriptions of quantum change consisted of improvements in functioning, yet negative outcomes were also reported. Similarly, stressful events can result in negative outcomes (Tedeschi and Calhoun 2004), labeled “posttraumatic depreciation” (Tedeschi et al. 2017, p. 11).

Prior research by Sandage et al. (2010, 2011) found that a self-reported transformative experience, framed as quantum change, moderated curvilinear associations between R/S seeking (i.e., self-identity and meaning exploration) and R/S dwelling (i.e., felt security, perceived closeness), and R/S seeking and generativity, promoting greater seeking and dwelling, and protecting against a negative influence for seeking on generativity. From an RSM perspective, their results supported a dialectical–developmental association between secure R/S connections and the exploration of new R/S meaning, which might become integrated or reconciled for some individuals during transformative experiences (Sandage et al. 2020). Their results also highlight how transformative experiences often involve R/S themes (Miller and C’de Baca 2001; Skalski and Hardy 2013).

The construct of PTG provides a bridge between the resilience and the transformative experiences literatures. Specifically, growth is distinguished from return to baseline (Tedeschi and Calhoun 2004), and return to baseline is often used to further distinguish recovery from resilience (Infurna and Jayawickreme 2019). Recovery, resilience, and growth each focus on individuals’ responses to significant adversity; with longitudinal plots of outcomes for recovery showing the return-to-baseline trajectory, resilience depicted by a stable, plateau trajectory of positive adaptation, and a growth trajectory characterized by gradual improvement beyond baseline (Bonanno 2012; Infurna and Jayawickreme 2019; Masten 2021). However, interrupting the risk trajectory (Zimmerman and Arunkumar 1994) is central to historical and developmental definitions of resilience. Masten (2021) also depicted resilience as a plateau trajectory, with the qualifier that the trajectory differed from an anticipated pattern of maladaptation given the risk factor(s). As such, recovery and growth trajectories, which depict a change in the pattern of responses, may better represent resilience than a stable plateau of high functioning (Infurna and Luthar 2018).

Tedeschi and Calhoun (2004) also differentiated growth from effective coping, stating that growth “cannot easily be reduced to simply a coping mechanism” (p. 15). Elsewhere, Tedeschi and Kilmer (2005) suggested that resilience was defined by effective coping, implying that effective coping maintains stability rather than promotes growth. However, Infurna and Jayawickreme (2019) suggested that reporting self-perceived growth may itself be a coping strategy, rather than an indicator of improvement. In fact, self-perceived growth may be an indicator of positive appraisal style, which Schäfer et al. (2022) described as a convergence of protective factors that shape individuals’ perception of adversity. Self-perceived growth would therefore be akin to optimism, hope, and finding meaning as indicators of positive appraisal. We view effective coping strategies (e.g., positive appraisal; Schäfer et al. 2022) as potential protective factors contributing to positive adaptive responses to adversity and ineffective coping strategies (e.g., avoidance) as potential risk factors (Masten 2021). As such, coping strategies could be associated with any number of adversity-response trajectories.

1.4. The Current Study

In the current study, we used the RSM to conceptualize and model the R/S factors of secure God attachment and transformative experiences as protective, distinct from the outcomes of anxiety and burnout, using a sample of North American seminary students representing a diverse range of Christian traditions. We expected to identify distinct subgroups that differed by change processes on the outcomes of anxiety and work-related burnout over three time points, at least one of which would constitute a resilience trajectory. As such, our study is consistent with trajectory-based studies of resilience that examine post-adversity responses (Galatzer-Levy et al. 2018), and especially those studies that have examined responses to the pandemic (Schäfer et al. 2022). Furthermore, we expected that a resilience trajectory would differ from other trajectories on levels of risk (i.e., COVID-19 stress, demographic variables), protective factors (i.e., religiousness/spirituality), and subjective (i.e., positive emotion) and eudaimonic (i.e., life purpose) well-being. Last, we expected that a resilience trajectory could be further differentiated by transformative experiences, with these experiences promoting an adaptative response to adversity.

2. Method

2.1. Participants

Data were collected from graduate students at 18 North American seminaries. We used a sample of participants for whom we had three time points of data (N = 751; M_age = 32.82; SD 11.37; range = 21–72). Among those providing demographic data, 47.4% identified as male (49.9% female) and a majority identified as heterosexual (89.1%). Participants identified as 9.5% Black, 14.5% Asian, 59.8% White, 4.7% Hispanic, 2.7% other, and 5.3% multiracial. A majority indicated their religious affiliation as evangelical Protestant (46.9%), whereas others identified as 20.1% mainline Protestant, 15.6% Catholic, and 3.2% historically Black Protestant, and 2.9% identified as unaffiliated/none and 10.4% as “other” religion/affiliation. A majority (at wave 3; 75.3%) indicated vocational goals involving professional leadership in a church/parish, parachurch organization, and/or missions.

2.2. Procedure

As part of a larger study (e.g., Jankowski et al. 2022b, 2022c), seminary students consented and completed a self-report online survey during the spring of 2021. The survey included items on religious/spiritual beliefs, practices and experiences, symptoms, and well-being. The survey was constructed using select items from a variety of existing instruments to address the need for efficient measurement strategies within these educational contexts. In exchange for participating, students received a USD 25 gift card. This process was repeated at prior and subsequent time points. Data from waves 4–6 were used in the current study because data for burnout and COVID-19 stress were collected beginning at wave 4. Time 2 (wave 5) was approximately 6 months after time 1 (M = 6.78, SD = 0.79), during fall 2021, and time 3 (wave 6) was approximately 6 months later (M = 5.52, SD = 0.72), during spring 2022.

2.3. Measures

Anxiety. We used the 7-item General Anxiety Disorder-7 scale (ω = 0.91 at wave 4; e.g., “Worrying too much about different things;” Spitzer et al. 2006), with higher scores indicating greater levels of anxiety symptoms.

Burnout. We used the 7-item work-related burnout subscale from the Copenhagen Burnout Inventory (ω = 0.90 at wave 4; e.g., “Do you feel worn out at the end of the working day?” Kristensen et al. 2005), with higher scores indicating greater psychological and physical exhaustion related to work.

COVID-19 stress. We used the 6-item traumatic stress subscale from the COVID Stress Scales (ω = 0.92 at wave 4; e.g., “I had trouble sleeping because I worried about the virus;” Taylor et al. 2020), with higher scores indicating greater levels of perceived stress related to COVID-19.

Ideological commitment. We used a single item (Perry 2015) that assessed ideological commitment on religious/spiritual matters ranging from 1 (very conservative) to 7 (very liberal).

Attachment to God. We used the anxiety (ω = 0.89 at wave 4; five items, e.g., “I worry a lot about my relationship with God”) and avoidance subscales (ω = 0.80 at wave 4; five items, e.g., “I just don’t feel a deep need to be close to God;” Beck and McDonald 2004), with higher scores on each representing greater perceived insecurity in relating to God.

Gratitude. We used six items (ω = 0.76 at wave 4; e.g., “I have so much in life to be thankful for;” McCullough et al. 2002), with higher scores indicating greater levels of positive emotion, consistent with the conceptualization of gratitude as a positive emotion (e.g., Watkins et al. 2018). Prior factor analytic work found joy and gratitude to load on the same factor, consistent with prior findings of a strong correlation between joy and gratitude (Jankowski et al. 2022c). Joy and gratitude did show differential associations with external correlates suggesting that joy and gratitude are somewhat distinct positive emotions (Jankowski et al. 2022c).

Life purpose. We used the 4-item subscale from the Claremont Purpose Scale (ω = 0.92 at wave 4; e.g., “How clearly do you understand what it is that makes your life feel worthwhile?” Bronk et al. 2018), with higher scores indicating greater presence of life purpose.

Transformative experience. We used a single item to assess transformative experience, based on Miller and C’de Baca’s (2001) work. The item read: “Some people experience a highly memorable period of minutes or hours, through which they find themselves immediately, dramatically, and permanently changed (for better or for worse). These experiences usually take them by surprise, rather than being something the person chose or decided. Often this involves sudden significant shifts in spirituality, personality, self-identity, perceptions of reality, and emotional life. Over the past 6 months, have you ever had such an experience yourself?” Participants rated the item using “no/unsure/yes,” and then responded to an open-ended item asking them to briefly describe their experience.

Participants’ open-ended responses were coded using thematic analysis (Braun and Clarke 2006), with the aim of generating a multicategorical variable for the quantitative analysis. Transforming qualitative data for quantitative analysis can be consistent with thematic analysis (e.g., Robinson 2022), and it is also consistent with mixed-method research, and specifically, a concurrent nested design in which the integration of qualitative and quantitative data occurs during data analysis (Hanson et al. 2005). Thematic analysis moved from inductive initial coding based on the semantic content of responses (i.e., summary descriptions using participants’ words), to grouping initial codes into broader themes, and then to deductive, theory-informed coding to refine and name themes (e.g., insightful versus mystical transformative experiences; Miller and C’de Baca 2001; Skalski and Hardy 2013). This process resulted in four themes: (1) stressful life events (n = 68; or positive/negative life events; Skalski and Hardy 2013), including the subtheme of COVID-19 as a stressor (e.g., “I got married and moved into the married students apartments with my spouse;” “Realizing the toll the pandemic and isolation is taking on my family—my marriage, my child’s development, and myself was depressing and heart wrenching”); (2) self-identity development (n = 101; Skalski and Hardy 2013), including work-/vocation-/call-related subthemes (e.g., “I’ve learned a lot about myself during seminary, and have had several moments where I became aware of how I was impacting others (positively or negatively), and how this did or didn’t align with scripture. Seminary is helping me become more self-aware!” “Deep insights regarding my calling into Spiritual Direction, such that I heard several deep validations I was hearing a valid call”); (3) social justice/compassion (n = 19; C’de Baca and Wilbourne 2004; e.g., “Realization of how Christianity has been used as a weapon of oppression over and over. Also, the deep and lasting impact of colonialism, coming from a country that was colonized. How deeply entrenched is the white supremacist culture;” “Awakening to need to be a voice calling for justice within the church”); and (4) spiritual/mystical experiences (n = 62; Miller and C’de Baca 2001; Skalski and Hardy 2013; e.g., “While practicing corporate centering prayer, I had a powerful experience of God’s presence;” “During a time of private worship, I experienced a lift of heaviness that had settled into my ‘gut.’ It didn’t come back. I feel like in that time of worship I was set free of something dark, and assignment from the enemy”).

2.4. Data Analytic Plan

Data were analyzed using mixture modeling procedures in Mplus (version 8.4; Muthén and Muthén 1998–2019; i.e., type = mixture; estimation = maximum-likelihood estimation with robust standard errors; missing data were handled using full-information maximum-likelihood estimation (FIML). FIML estimation was appropriate given a nonsignificant Little’s MCAR test (χ² (38) = 40.82, p = 0.35). Data exhibited multivariate non-normality (Mardia’s kurtosis statistic = 67.29, p < 0.001). We used a parallel growth mixture model to examine simultaneous changes in anxiety and burnout, with each process sharing a single latent categorical variable (StatModel n.d.).

We used a two-step method to fix parameters of the latent growth mixture model before introducing external variables (Bakk and Kuha 2018; see also Asparouhov and Muthén 2021). We then examined the influence of covariates on subgroup membership using multinomial logistic regression for continuous variables and modeling categorical covariates as class indicators (Asparouhov and Muthén 2021; Muthén and Muthén 1998–2019). We used the Wald test of parameter constraints and the model constraint command to examine subgroup differences on the categorical covariates (Muthén and Muthén 1998–2019). For the multicategorical covariates we followed up a significant Wald test with tests of the difference between probability parameters (Muthén 2014). We then examined subgroup differences on the levels of continuous distal outcomes by modeling the variables as class indicators (Muthén 2013; Muthén and Muthén 1998–2019) and tested the difference within and between parameters at different time points (Muthén 2013). The complete data, which included covariates and distal outcomes, were missing completely at random based on a nonsignificant Little’s MCAR test (χ² (209) = 242.85, p = 0.054) and exhibited multivariate non-normality (Mardia’s kurtosis statistic = 664.19, p < 0.001).

Class enumeration was based on Bayesian and Akaike information criteria (BIC, AIC), with smaller values indicating better fit (Berlin et al. 2014). We also considered entropy, a measure of subgroup separation and classification accuracy, with values > 0.60 acceptable and > 0.80 good (Berlin et al. 2014), along with the average posterior class probability (AvePP), with values greater than 0.70 indicating that “the classes [are] well separated and the latent class assignment accuracy adequate” (Masyn 2013, p. 570). Last, we considered subgroup size, parsimony, and interpretability (Masyn 2013).

3. Results

We set the nonsignificant variance of the slope for anxiety to zero to aid convergence, thereby approximating latent trajectory analysis rather than remaining a growth mixture model (Jung and Wickrama 2008). A plot of information criterion values showed a “point of ‘diminishing returns’ in model fit,” (Nylund-Gibson and Choi 2018, p. 443; see Figure 1) at the three-class model. As such, we opted for interpreting the three-class model. Entropy for the three-class model was 0.85, suggesting good subgroup separation and classification accuracy, and the average posterior class probability (AvePP) values were adequate (≥0.69). Class 1 reported the highest levels of anxiety and work-related burnout symptoms and showed improvement over time (see Figure 2). Class 2 reported moderate levels of symptoms and showed deterioration over time. Class 3 reported the lowest levels of symptoms, a small decline in anxiety, and no change in burnout.

Class membership was associated with demographic variables. As age increased participants were less likely to belong to class 1 relative to class 3 (B = −0.07, SE = 0.02, p < 0.001), and identifying as more ideologically liberal was associated with a greater likelihood of belonging to class 1 (B = 0.48, SE = 0.10, p < 0.001). Similarly, as age increased participants were less likely to belong to class 2 relative to class 3 (B = −0.03, SE = 0.01, p = 0.002), and when participants identified as more ideologically liberal they were more likely to belong to class 2 relative to class 3 (B = 0.33, SE = 0.07, p < 0.001). In addition, participants who identified as heterosexual were less likely to belong to class 1 relative to class 3 (χ² = 4.52(1), p = 0.03) and class 2 relative to class 3 (χ² = 3.87(1), p = 0.049). Furthermore, participants who identified as female were more likely to belong to class 2 relative to class 3 (χ² = 6.36(1), p = 0.01). Participants who identified their religious affiliation as evangelical were more likely to belong to class 3 relative to classes 1 (χ² = 6.26(1), p = 0.01) and 2 (χ² = 5.22(1), p = 0.02).

There was also a significant difference between subgroups on the multicategorical covariate (i.e., “no,” “unsure,” “yes”) about a recent transformative experience (χ² = 11.77(4), p = 0.02). Comparisons of the multicategorical variable by class indicated that classes 2 and 3 differed (χ² = 7.42(2), p = 0.02). Pairwise comparisons by response revealed that class 3 was more likely to respond “no” relative to class 1 (ΔP = −0.16, SE = 0.09, percentile bootstrap (PC) 95% confidence interval (CI) [−0.36, −0.02]; 500 bootstrap samples) and class 2 (ΔP = −0.15, SE = 0.06, PC95%CI [−0.25, −0.03]). Alternatively, results based on dummy coding for the multicategorical variable revealed that class 3 was more likely to respond “no” relative to “yes” (χ² = 11.04(2), p = 0.004) relative to class 1 (ΔP = 0.16, SE = 0.09, PC95%CI [0.01, 0.36]) and class 2 (ΔP = 0.16, SE = 0.06, PC95%CI [0.03, 0.27]).

There was also a significant difference between subgroups on the multicategorical covariate (i.e., “stressful life events,” “self-identity development,” “social justice/compassion,” “spiritual/mystical experience”) involving participants’ descriptions of their transformative experience (χ² = 80.02(6), p < 0.001). Comparisons by class indicated that class 1 differed from classes 2 (χ² = 14.26(3), p = 0.003) and 3 (χ² = 59.10(3), p < 0.001). Pairwise comparisons by response revealed that class 1 was less likely to report a theme of “social justice/compassion” relative to class 2 (ΔP = −0.17, SE = 0.06, PC95%CI [−0.31, −0.06]) and class 3 (ΔP = −0.06, SE = 0.09, PC95%CI [−0.10, −0.03]), and class 1 was less likely to describe a theme of “spiritual/mystical experience” relative to class 3 (ΔP = −0.23, SE = 0.07, PC95%CI [−0.33, −0.08]). Alternatively, for the comparison based on dummy coding for the multicategorical variable, class 1 was more likely to report a theme of “stressful life events” relative to “social justice/compassion” than class 2 (χ² = 7.72(1), p = 0.005) and class 3 (χ² = 9.48(1), p = 0.002). Furthermore, there was a difference for reporting a theme of “stressful life events” relative to “spiritual/mystical experience” for class 1 relative to class 2 (χ² = 5.71(1), p = 0.02) and class 3 (χ² = 64.34(1), p < 0.001), with class 1 more likely to report a theme of “stressful life events.”

Class 1 reported the highest levels of COVID-19 stress at time 4 (M = 2.01, SE = 0.14), relative to both class 2 (M = 0.44, SE = 0.11; ΔM = 1.57, SE = 0.17, p < 0.001) and class 3 (M = 0.18, SE = 0.03; ΔM = 1.82, SE = 0.15, p < 0.001), with class 2 reporting higher levels of COVID-19 stress than class 3 (ΔM = 0.25, SE = 0.11, p = 0.02). Class 3 reported the highest levels of life purpose (M = 3.99, SE = 0.04) relative to class 1 (ΔM = −0.46, SE = 0.15, p = 0.003) and class 2 (ΔM = −0.86, SE = 0.11, p < 0.001), the highest levels of gratitude (M = 6.33, SE = 0.03) relative to class 1 (ΔM = −0.76, SE = 0.16, p < 0.001) and class 2 (ΔM = −0.61, SE = 0.12, p < 0.001), the lowest levels of anxious God attachment (M = 2.96, SE = 0.14) relative to class 1 (ΔM = 0.79, SE = 0.26, p = 0.002) and class 2 (ΔM = 1.31, SE = 0.19, p < 0.001), and the lowest levels of avoidant God attachment (M = 1.06, SE = 0.08) relative to class 1 (ΔM = 0.97, SE = 0.23, p < 0.001) and class 2 (ΔM = 0.96, SE = 0.18, p < 0.001). Classes 1 and 2 did not differ in levels of gratitude, anxious God attachment, or avoidant God attachment, but did differ in life purpose (ΔM = 0.40, SE = 0.19, p = 0.03) with class 2 reporting the lowest levels of life purpose (M = 3.13, SE = 0.10).

The lower symptom levels and greater levels of subjective (e.g., positive emotion) and eudaimonic (e.g., life purpose) well-being that characterized class 3 are consistent with conceptualizations of flourishing (e.g., Jankowski et al. 2020; Keyes 2002). In contrast, the moderate symptom levels and low levels of life purpose and gratitude that characterized class 2, along with moderate levels of the stressor and higher insecure God attachment and the deterioration over time, are consistent with formulations of languishing. Conceptualizations of languishing often emphasize cross-sectional assessments of low well-being, although the term languishing has been applied to declines in well-being over time (e.g., O’Donnell et al. 2022). We extended the latter notion to include the deterioration of symptoms over time. Last, the high symptom levels and low levels of life purpose and gratitude for class 1, along with highest levels of the stressor despite evidence of improvement on symptom levels, are consistent with multidimensional, developmental-process depictions of resilience (e.g., Masten 2021; Zimmerman and Arunkumar 1994).

In addition, the resilient class (i.e., growth trajectory) reported a significant decline in COVID-19 stress from time 4 to time 5 (ΔM = 0.76, SE = 0.37, p = 0.04; d = 1.77), whereas the flourishing (i.e., stable-plateau trajectory) and languishing (i.e., deterioration trajectory) classes reported no change in the level of COVID-19 stress from time 4 to time 5. Of note, the growth displayed by the resilient seemed tied to the decrease in levels of COVID-19 stress over time. In fact, among the resilient, reporting a transformative experience theme of “stressful life events” was associated with a greater rate of improvement in COVID-19 stress (B = −0.50, SE = 0.20, p = 0.01), as the dummy-coded multicategorical variable predicted the latent-change score for COVID-19 stress, which was modeled to be class specific in a follow-up analysis. The flourishing reported an increase in life purpose from times 4 to 6 (d = −0.17, SE = 0.05, p = 0.001), and the resilient reported an increase in gratitude from time 4 to time 6 (d = −0.42, SE = 0.20, p = 0.04). There were no changes in anxious and avoidant God-attachment dimensions over time. Last, wave 1 (fall 2019) anxiety did not differ from wave 4 anxiety levels among the resilient (ΔM = −0.64, SE = 0.43, p = 0.14), although the trend was toward increased anxiety. The flourishing subgroup reported an increase in anxiety symptoms from wave 1 to wave 4 (ΔM = −0.07, SE = 0.03, p = 0.02). Nevertheless, their levels of anxiety remained lower than the resilient and the languishing. The resilient (ΔM = 1.14, SE = 0.44, p = 0.01) and the languishing (ΔM = 1.00, SE = 0.32, p = 0.002) reported higher levels of anxiety than the flourishing at wave 1, whereas the resilient and the languishing did not differ.

Taken together, the mapping of “social justice/compassion” themes for the languishing seems consistent with their more marginalized demographic profile relative to the flourishing, with those in the languishing class more likely to identify as younger, more liberal, non-heterosexual, female, and non-evangelical. The mapping of “stressful life events” themes for the resilient seems consistent with their highest levels of anxiety, burnout, and COVID-19 stress, and their reported improvement may be indicative of the alleviation of stressors, including COVID-19-related stress and their transformative experience. For the flourishing, the mapping of “spiritual/mystical experiences” seems consistent with their predominantly evangelical identification and their highest levels of felt security and perceived closeness in their attachment relating to God.

Sensitivity Analyses

We conducted sensitivity analyses by “running alternative, justifiable analyses to see whether a reported result would still hold up” (Nuijten 2022, p. 392). For the first check, we examined a piecewise mixture model that included data for anxiety from waves 1–4 as a first slope segment and then the data for waves 4–6 as a second slope segment (Muthén and Muthén 1998–2019). We set the second slope segment for anxiety to zero to aid convergence. The results are depicted in Figure 3 and support our earlier contention and subsequent finding that a resilience trajectory need not be limited to a high-functioning plateau. Wave 1 was collected pre-pandemic declaration during fall 2019, with the declaration about a public-health emergency occurring on 31 January 2020 (U.S. Department of Health & Human Services 2020), just as the measurement window for wave 2 data collection was opening. As Figure 3 shows, the resilient and the languishing had comparable levels of anxiety at wave 1 (Δi = −0.26, SE = 0.40, p = 0.53), and although the languishing initially showed a plateau over times 1–4, their anxiety significantly increased over waves 4–6. In contrast, after showing an increase in anxiety over waves 1–4, the resilient reported a decline in anxiety during waves 4–6. The trajectory matches depictions of recovery (Infurna and Jayawickreme 2019), except that wave 6 anxiety was significantly lower than wave 1 anxiety (d = 0.48, SE = 0.10, p < 0.001), and therefore the trajectory is more consistent with growth, although this depends on conceptualizations of “gradual improvements to near-previous levels over time” (p. 156). However, as we describe below, descriptions of change over time are but one aspect to discerning resilience, and as we noted previously, resilient patterns of change can encompass growth and/or recovery.

Second, in another sensitivity analysis, we used gratitude and life purpose as additional indicators in the mixture model, along with anxiety and work-related burnout. Each process shared a single latent categorical variable. We set the nonsignificant variances of the slopes for anxiety and life purpose to zero to aid convergence. A plot of BIC values indicated three- and four-class solutions as viable. We opted for the three-class solution because two classes in the four-class solution seemed redundant. Three of the four intercepts exhibited a low degree of class separation (i.e., ≤ 1 standard deviation (SD) difference between classes, as indicated by standardized mean difference effect size; Grimm et al. 2021; for anxiety: Δi = −0.28, SE = 0.16, p = 0.07; d = 0.86; for burnout: Δi = −0.25, SE = 0.22, p = 0.25; d = 0.48; for life purpose: Δi = 0.67, SE = 0.31, p = 0.03; d = 0.99), whereas gratitude exhibited a high degree of separation (≥ 3 SD difference between classes; Grimm et al. 2021; Δi = 2.08, SE = 0.20, p < 0.001; d = 4.33). The slopes were in the same direction, except for gratitude, with one trajectory showing significant growth and the other a nonsignificant slope or plateau. Like Figure 2, Figure 4 shows the resilient trajectory with the highest levels of symptoms, and this time, also with the lowest levels of subjective and eudaimonic well-being, each of which showed improvement.

Next, we used the automated three-step procedures in Mplus to model auxiliary variables (Asparouhov and Muthén 2021). First, we compared subgroups on the dummy-coded multicategorical transformative-experiences variables using DCAT. Second, we examined demographic, COVID-19 stress, and attachment to God as risk and protective covariates predicting class membership using R3STEP. The flourishing subgroup was more likely than the languishing to respond “no” to a recent transformative experience relative to responding “yes” (χ² = 4.36(1), p = 0.037). The resilient subgroup was more likely to report a theme about “stressful life events” relative to the flourishing (χ² = 4.70(1), p = 0.03) and less likely to report a theme about “spiritual/mystical experience” than the flourishing (χ² = 121.47(1), p < 0.001) and the languishing (χ² = 9.55(1), p = 0.002). Relative to the flourishing, participants who identified as heterosexual were less likely to belong to the resilient subgroup (B = −1.13, SE = 0.57, p = 0.048). Relative to the flourishing, the languishing subgroup were more likely to identify as non-White (B = −0.87, SE = 0.42, p = 0.04), female (B = 0.89, SE = 0.40, p = 0.02), and younger (B = −0.05, SE = 0.02, p = 0.01), and less likely to identify as evangelical (B = −1.04, SE = 0.45, p = 0.02). The resilient were more likely to report higher levels of COVID-19 stress. Specifically, relative to the resilient, as levels of COVID-19 stress increased participants were less likely to belong to the flourishing (B = −1.55, SE = 0.27, p < 0.001) and the languishing (B = −1.11, SE = 0.62, p = 0.004). Last, as God-attachment anxiety increased participants were more likely to belong to the resilient (B = 0.57, SE = 0.18, p = 0.002), relative to the flourishing, and similarly, as God-attachment avoidance increased participants were more likely to belong to the resilient (B = 0.49, SE = 0.17, p = 0.004) relative to the flourishing.

Results supported the main analyses. The resilient reported transformative experience themes of “stressful life events” and the highest levels of anxiety, burnout, and COVID-19 stress, and their trajectory showed significant improvement over time on anxiety, burnout, life purpose, and gratitude.

4. Discussion

We found evidence of a subgroup of participants who exhibited resilience over three waves of data collection, approximately 12 months in elapsed time. In fact, we found evidence for each of the established criteria for determining resilience. First, we identified significant adversity in the form of self-reported levels of COVID-19 stress at time 1 (i.e., wave 4), one year post-pandemic declaration and during a time of rising infection rates and societal upheaval in North America, and in the US in particular. Second, we found additional evidence of risk, as the resilient reported lower religiousness/spirituality in terms of both secure attachment to God and spiritual/mystical transformative experiences. However, lower religiousness/spirituality need not necessarily be a risk factor, as this depends on socio-cultural context, the operationalization of religiousness/spirituality, and the outcome used to assess positive adaptation (Jankowski et al. 2022b). Nevertheless, low religiousness/spirituality can be a source of dissonance and stress for students in the context of seminary training. Members of the resilient subgroup were also more likely to identify as a sexual minority and younger relative to the flourishing. Members in the languishing subgroup also reported marginalized identities, with even more demographic identifications relative to the flourishing than the resilient, and yet there were no differences for covariate influence on membership between the resilient and the languishing, except that the resilient reported the highest levels of COVID-19 stress. Third, we found evidence of growth in the trajectory for the resilient in levels of symptoms and well-being, and fourth, we found evidence for transformative experiences to exert a protective influence and promote positive adaptation. Thus, not only did we find evidence of “positive outcomes in otherwise risky situations” (Zimmerman and Arunkumar 1994, p. 13), but we also found evidence for “stressful and traumatic events as key elicitors” of growth (Chirico et al. 2022, p. 3). As such, we subsumed growth under the broader construct of resilience given the interruption to a maladaptive trajectory (Zimmerman and Arunkumar 1994). Additionally, resilience seemed to correspond to a decline in COVID-19 stress among the resilient, and their transformative experiences related to “stressful life events” were elicitors of growth.

We did not conceptualize resilience as a healthy plateau (e.g., Bonanno 2012). In fact, we see a fuller depiction of the plateau trajectory we found as consistent with a conceptualization of flourishing, and specifically, low levels of reported COVID-19 stress, high levels of religiousness/spirituality, low symptom levels, and high well-being. Rather than resilience, Norris et al. (2009) suggested the term resistance to describe a stable low-symptom trajectory. They argued that resistance involves stability, whereas resilience involves adaptability. Masten (2021) also used the term resistance for a stable plateau trajectory; however, her depiction featured (a) midrange functioning, and (b) resistance as one type of resilience. Masten’s labeling thus differs from depictions of a single stable trajectory of high functioning as resistance (Norris et al. 2009) or resilience (Bonanno 2012). Norris et al. (2009) also suggested that resistance connotes the availability of resources to mitigate the effects of the stressor. As Willen (2022) suggested, “structural factors, ideological contexts, and relations of power … predispose some people to languish, and others to flourish” (p. 1). In our study, the flourishing were least likely to identify as marginalized on each of the demographic indicators, suggesting the availability of resources.

Furthermore, as Schäfer et al. (2022) stated, “the concept of resilience as a positive outcome despite stressor exposure (i.e., adversity) implies that it can only be assessed if individuals are exposed to stressors” (p. 1181). Schäfer et al. reviewed trajectory-based studies of resilience based on “the pandemic as a societal stressor” and noted that at the individual level, exposure “might have varied substantially” (p. 1184). Hence, there is a need to assess individual-level experience of the stressor, distinct from the outcome. Schäfer et al. also indicated that resilience as a high-functioning plateau trajectory may simply “reflect low levels of stress rather than better psychological adaptation” (p. 1184). The plateau trajectory we observed was associated with low levels of felt COVID-19 stress, lower likelihood of reporting a transformative experience, and lower likelihood of reporting themes related to “stressful life events,” suggesting that this was a low-exposure group. In fact, Lai et al. (2015) differentiated a low-symptom trajectory from a resilient trajectory on the basis that the low-symptom trajectory reported “the lowest exposure to disaster related stressors” (p. 519). Similarly, our flourishing subgroup represented low exposure to the COVID-19 stressor, consistent with Tedeschi and Calhoun’s (2004) description of flourishing as involving low perceived threat and/or distress related to adversity.

In a prior cross-sectional study involving a smaller subsample from the current study sample, Jankowski et al. (2022b) found that a flourishing subgroup was more likely to have an overly positive, exaggerated view of their level of religiousness/spirituality relative to other classes. An overly positive view may be an indicator of the general coping process of positive appraisal (Schäfer et al. 2022) or R/S coping in which greater perceived closeness with God shelters against felt adversity by regulating negative emotions, including perhaps through experiential avoidant processes that minimize threat and/or distress (Jankowski et al. 2022a). Flourishing as coping seems consistent with both Tedeschi and Kilmer’s (2005) suggestion that effective coping maintains stability, and Keyes’ (2002) notion that flourishing can function “as a stress buffer” (p. 219). Taken together, flourishing as a stress buffering coping process seems most consistent with Norris et al.’s (2009) depiction of resistance in which “coping resources have effectively blocked the stressor” (p. 2191). The flourishing subgroup in our study could therefore depict a low-exposure subgroup simply because the impact of the pandemic was minimal, or because coping resources effectively defended against negative impact from the pandemic.

The interruption of a maladaptive trajectory seems to be an oft-neglected yet essential aspect of the definition of resilience. Historically, resilience as a developmental construct had the connotation of an unexpected or contradictory outcome to that which would be predicted by the risk factor(s), including the target stressor event. As Masten (2001) stated, “risks are actuarially based predictors of undesirable outcomes drawn from evidence that this status or condition is statistically associated with higher probability of a ‘bad’ outcome” (p. 228). Our inclusion of waves 1–3 in the first sensitivity analysis permitted us to discern an interruption to the trajectory of the resilient that included pre-exposure (wave 1), one year post-declaration of the pandemic (waves 2 and 3) and heightened societal stress (wave 4), and finally one year post-peak societal stress levels (waves 5 and 6). In contrast, we did not see evidence that the flourishing experienced significant adversity, nor was there an interruption to their trajectory over time. Rather, the flourishing reported greater R/S protective factors in terms of secure God attachment and transformative-experience themes of “spiritual/mystical experiences,” and identity indicators of resource privilege that may have mitigated felt threat and/or distress related to the pandemic.

Last, the languishing trajectory we observed was consistent with a commonly reported trajectory in longitudinal studies of responses to adversity, that of delayed onset (Galatzer-Levy et al. 2018; Schäfer et al. 2022). As Schäfer et al. (2022) noted, “individuals with moderate-stable trajectories might be at particular risk for delayed responses to the pandemic” (p. 1181). Consistent with Schäfer et al.’s assertion, our sensitivity analysis involving waves 1–6 showed a stable plateau of moderate symptom levels over waves 1–4 and then an increase in symptoms over waves 4–6 for the languishing. Schäfer et al. further distinguished the delayed onset trajectory from a moderate-stable trajectory of mid-level functioning and a chronic trajectory with stable low levels of positive adaptation. Rather than delayed onset, we opted for the term languishing based on the results from the main analysis, which depicted deterioration over three time points, and because the term languishing has been used to describe a deterioration trajectory (O’Donnell et al. 2022), albeit the fullest descriptor is probably delayed-onset languishing. Finally, Keyes’ (2002) dual-continuum classification referred to types of languishers, which included a non-symptomatic and low-well-being group (i.e., pure languishers), however, those classifications were based on the operationalization of well-being and symptoms as categorical, using arbitrary cut points (Zhao and Tay 2022). Zhao and Tay (2022) pointed out that empirically, “an absolute zero point is often scarce” (p. 8), and as such, the notion that no symptoms should distinguish types of languishing seems questionable. In fact, Zhao and Tay, using continuous indicators to empirically generate latent classes, did not even find a languishing subgroup characterized by low symptoms and low well-being in one of their samples.

When symptoms and well-being are measured on a continuous scale, it seems better to conceptualize languishing as higher symptoms and lower well-being relative to the other classes. Furthermore, empirically generated latent classes should be interpreted within the research context, including data limitations (e.g., time-unstructured data, number of time points), indicator selection, psychometric evidence, analytic specifications of a particular model, and class separation (Masyn 2013; van der Nest et al. 2020). As such, it is important that latent classes not be reified (Masyn 2013). As Ram and Grimm (2009) cautioned, mixture modeling is an “exploratory … post-hoc analysis technique that seeks out the story the data are trying to tell—a story that is limited by the specific bounds imposed during model specification” (p. 572). Practically then, the meaning of any particular class is always relative to the others within a particular research context. It would seem, however, that the labels for dual-continuum classifications and post-adversity responses have become somewhat reified over time. Blaug (2015) defined reification as a cognitive bias that involves “the human tendency to invent knowledge, to forget that authorship and come to believe it real; to confront the socially constructed as natural; to make a ‘thing’ of an idea” (p. 3). Willen (2022) drew attention to the potential for research on flourishing and languishing to risk “reification of analytic categories” (p. 3), and Williams and Kemp (2019) noted the potential for descriptions of resilience to be reified. It seems perhaps that notions about languishing as an absence of symptoms despite low well-being and resilience as a single stable trajectory of healthy functioning have come to be seen as essential fixed properties of “literal entities” and the trajectories “literal depictions of reality” (Nagin and Tremblay 2005, p. 882). Rather, latent classes are but “useful statistical fiction” and “an approximation of a more complex … population distribution” (Nagin and Tremblay 2005, pp. 873, 888). Languishing and resilience are therefore descriptors, which appear to have use as signifiers for a variety of longitudinal processes. However, as important as it is to guard against reification, researchers must also be mindful to avoid jingle-jangle fallacies, or meaning and construct proliferation. Descriptors must be consistent with the existing nomological network for a phenomenon, with empirical evidence for convergent and discriminant conceptualizations. We chose our labels for the trajectories based on historical and critical evaluations of the research on patterns of change in response to adversity, and the empirical evidence supporting their use as distinct descriptors.

5. Conclusions

We suggest that promoting resilience among seminary students centers on the two core aspects of transformative experience: clarifying emotional experience and generating new meaning (Chirico et al. 2022). The notion that attending to novel and/or moving emotional experience can propel growth has a long history in dialectical constructivist theories of change (e.g., Greenberg and Pascual-Leone 2001; Mahoney 2002). Emotions indicate and facilitate the “disordering processes … of a complex system’s attempts to reorganize its life. New life patterns emerge out of the chaos and dysfunction that ensue when old patterns are no longer viable” (Mahoney 2002, p. 748). Mahoney (2002) highlighted the contrasts between self and other, and self and larger system, that can generate disruption, whereas Greenberg and Pascual-Leone (2001) highlighted the internal processes whereby an individual resolves the contrasts between bodily-felt experience and the narrative construction of meaning. They stated that the “emergence of new meaning is facilitated by vivid evocation … of emotionally laden experience, which brings emotions into contact with reflective processes” (p. 179). These dialectics of transformative experience fit the RSM framework in which stressful experiences can potentially challenge prior R/S understandings and necessitate emotional processing and new R/S meaning (Sandage et al. 2020).

Those who are training helping professionals within the seminary context could consider ways to provide opportunities for students to gain awareness of and reflection on felt, lived experience within safe, secure, and supportive relationships. As Mahoney (2002) stated, “the experience of who one is, what one is capable of, and personal worth—develop within human relationships. Changes … also develop within such relationships” (p. 748). In addition, one of the themes that emerged in Skalski and Hardy’s (2013) qualitative study of transformative experiences was the positive influence of a trusted, supportive other, which for some included “an intimate personal experience of God, however conceptualized” (p. 174). Transformative experiences typically emerge out of normative developmental processes as the evolving individual adapts to changes within the environment, in response to often unexpected stressful life events. However, there is also a sense in which transformative experiences can be intentionally elicited, for example, through R/S practices such as meditation, prayer, ritual, as well as aesthetic experiences and “social events hinging on connection with others” (Chirico et al. 2022, p. 9). In our experience, seminaries often offer resources for (a) cognitive theological reflection and (b) R/S practices, whereas it is more uncommon for students to find resources to help process and regulate complex emotions with the discovery of new meaning. Yet, the RSM suggests that positive development necessitates the integration of healthy relational holding environments, emotion regulation, and holistic reflection on meaning (Sandage et al. 2020).

Last, seminary leadership may want to consider ways to identify students who are languishing. The presence of a languishing subgroup in our findings draws attention to the influence of risk factors, including larger contextual factors. Willen’s (2022) critique about the scholarship on responses to adversity highlighted

troubling blind spots—including, above all, a worrying inattention to the ways in which structure, power, and inequity affect who gets to flourish, who is likely to languish, and who our social structures and institutions, as currently designed, are—and are not—designed to help recover from hardship.

(p. 6)

Willen’s critique is another reminder that resilience, languishing, and flourishing processes are embedded within a larger social context, the latter of which may not be conducive to positive adaptation for all. Willen’s critique also invites questions about ways seminaries can offer hospitable and supportive environments for the formation needs of diverse students, including those with marginalized identifications. Considering structure, power, and inequity led McCormick (2009) to recast resilience based on an Aboriginal “view of community and the philosophy of ‘all my relations’” (p. 4). Instead of resilience, McCormick offered the construct of response-able, that is, “the ability to respond to challenges” (p. 5). Response-ability is a relational construct, “combining spirituality, family strength, elders, ceremonial ritual, oral traditions, identity and support” (p. 4). Skalski and Hardy (2013) also referenced response-ability, and did so in the context of the potential for processing emotional experience within a safe, secure relationship to promote positive transformative experiences. Masten (2021) suggested that the social context may be such that “functioning deteriorates or remains poor until more favorable conditions occur, either naturally or through intervention” (p. 121). Thus, conditions may inhibit individuals from being able to respond in ways that promote positive adaptation. More specifically, in the context of our study, seminaries may not be meeting the unique developmental needs of those belonging to the languishing subgroup.

We examined longitudinal patterns of change in a sample of seminary students to offer a conceptual and methodological approach for studying resilience, based on historical and critical evaluations about the study of resilience. We found one trajectory that met select criteria for determining resilience. Nevertheless, we acknowledge limitations to our study. First, we began using the COVID-19 stress and burnout measures at wave 4 because the pandemic had not yet been declared at wave 1, the COVID-19 stress measure was developed after the start of the study, and burnout seemed of greater importance over the course of the pandemic (e.g., Filindassi et al. 2022; Lowe et al. 2022). Pre- and post-adversity data permit a fuller evaluation of interruption and change in the trajectories over time (Schäfer et al. 2022), although post-adversity data are frequently used to study adversity-response trajectories (Galatzer-Levy et al. 2018). We did conduct a sensitivity analysis to model pre- and post-adversity change for anxiety. Nevertheless, it would have been ideal to also have pre-adversity data for burnout. Relatedly, the pandemic is ongoing, as formal declaration about the end of the public-health emergency in the US has not yet been declared. Other change patterns may be identified were additional waves of data collection to occur. In addition, more frequent assessments closer together in time may capture greater heterogeneity in responses to adversity (Infurna and Luthar 2018). As Schäfer et al. (2022) commented, “the lessons learned from the pandemic indicate the need to frequently measure both multidimensional functioning and stressor exposure” (p. 1185). In addition, our sample consisted of students at Christian-affiliated seminaries, which may limit the generalizability of our findings. However, prior research using a subset of the current study sample at wave 1 found that the demographic profile by R/S classifications were comparable to those from a nationally representative US sample of religious- and non-religious-identifying persons (Jankowski et al. 2022b). Nevertheless, research is needed with students training for leadership in other religious traditions and cultural contexts, and with diverse R/S identifications. We also recognize that this research may not generalize to more experienced R/S leaders. The literature on burnout and well-being among practicing R/S leaders is weighted toward cross-sectional findings, and more research is needed using longitudinal designs. Last, future research should examine mechanisms of resilience (Infurna and Luthar 2018), and specifically, emotion-regulatory flexibility and positive appraisal, which seem to be emerging as particularly relevant (Schäfer et al. 2022).