Applying person-centered techniques in organizational behavior research: challenging assumptions and providing new directions

Introduction

The vast majority of organizational behavior (OB) research has been variable-centered (VC). That is, the focus has been on describing, understanding, predicting, and controlling important variables (e.g., leadership style, employee commitment, work motivation, engagement, turnover, and job performance). Although this approach has yielded a wealth of information, the analytic techniques used in these investigations involve the assumption that the population is homogenous and that variables relate similarly for all individuals within that population. Of course, there are times when group differences are expected and/or detected (e.g., gender, age, or culture differences). However, these groups are defined a priori with clear classification metrics rather than latent (i.e., unobserved), so the focus remains variable centered.

OB researchers have increasingly started to use an alternative person-centered (PC) approach that relaxes the assumption of population homogeneity (Morin et al., 2020; Woo et al., 2024). Many early PC studies involved the use of non-model-based (algorithmic) clustering approaches (e.g., k-means cluster analysis), and variations of these approaches continue to be used. However, over the last decade, researchers have increasingly relied on advanced methods anchored in mixture modeling techniques (see Woo et al., 2018; for a summary of the differences). The objective of the family of mixture modeling techniques (e.g., latent class analysis; latent profile analysis; latent transition analysis) is to identify unobserved (i.e., latent) subpopulations (i.e., profiles) characterized by different configurations of variables (e.g., commitment mindsets; leader behaviors; indicators of wellbeing) or relations among those variables. That is, in this research, the focus shifts from the variables being measured to the unit of analysis (individual, team, organization, etc.), with the objective of identifying subpopulations for whom the variables, the relations among them, or both, demonstrate different configurations (Howard and Hoffman, 2018; Woo et al., 2024).

Our objective is to review PC research pertaining to OB, with emphasis on research published within the last decade using a mixture modeling approach. Although we acknowledge earlier studies where relevant, we focus on more recent research reflecting important advances in analytic techniques, allowing for greater sophistication in both the questions addressed and the answers provided. Unlike earlier reviews of PC research in OB (Morin et al., 2020; Woo et al., 2024) and related domains (Hofmans et al., 2020; Spurk et al., 2020; Woo et al., 2018), we focus primarily on substantive issues and how they have been addressed using PC methods rather than on the methods themselves. Our objective is to illustrate when and how PC analytic techniques can be applied to advance our understanding of OB-relevant issues and to encourage broader application where appropriate. We begin by providing a user-friendly introduction to the data analytic techniques made available within the overarching Generalized Structural Equation Modeling (GSEM) framework to conduct PC research both within [e.g., latent profile analysis (LPA); multilevel LPA (MLPA); mixture regression analyses (MRA)] and across time [e.g., latent transition analysis (LTA); growth mixture modeling (GMM)]. Our aim in this section is simply to provide the background necessary for the review of substantive research to follow. We refer interested readers to existing sources (e.g., Hofmans et al., 2020; Morin et al., 2020; Spurk et al., 2020; Woo et al., 2018, 2024) for more detail.

Despite the relatively recent adoption of PC methodology in OB research, applications are already too extensive and varied to review in their entirety. Thus, we did not seek to conduct a systematic review within the broad field of OB, nor within any specific sub-domain. Rather, we began with a search of primary databases (Google, Scopus, Web of Science) using key words intended to identify PC studies (e.g., person-centered, mixture modeling, and latent profile analysis) on major topics within OB (e.g., burnout, commitment, engagement, leadership, motivation, and wellbeing). The selection of substantive keywords was admittedly guided by earlier reviews of the methodology (e.g., Morin et al., 2020; Woo et al., 2024) and our personal familiarity with the literature, and may have influenced the selection of topics highlighted in our review. Given the increasing and wide-ranging applications of PC methods in OB research, for purposes of this review, we sought to strike a balance between breadth and depth. We sought to achieve breadth by sampling from research within several broad categories of theory/research. To provide depth, we sampled from research within these categories in such a way as to illustrate the cumulative contribution of multiple studies addressing selected issues from a PC perspective. Within each mini-review we discuss (a) how taking a PC perspective allowed for the testing of unique hypotheses to complement or extend those tested in VC research, and (b) the implications of the findings for theory development/refinement, and/or practical application.

Following our review across OB domains, we identify and discuss a series of issues that have arisen, and/or been addressed, within selected domains such that they might inform research in other domains. That is, our objective is to stimulate cross-fertilization by suggesting ways in which techniques applied in one domain might be applied in others included within our review, as well as in domains where a PC approach has yet to be, or is only beginning to be, applied. Moreover, we identify issues arising from the existing research that could have important implications for interpretation of findings and their application, and therefore warrant more systematic investigation.

Review of person-centered analytic strategies

In this section, we provide a user-friendly introduction to data analytic techniques within the GSEM framework. We focus here on the analytic techniques used to identify the profiles of interest (e.g., variables, change trajectories, and relations); we describe related procedures used for purposes of validation and/or theory testing as they become relevant in our substantive review. It is important to note that the input for the analyses discussed here can be raw scores for the measures involved, fully latent constructs obtained from preliminary factor analyses, or factor scores obtained from these same analyses. In practice, models involving fully latent constructs often fail to converge due to complexity, in which case factor scores can be used as a partial control for measurement error. We address the implications of this decision where relevant in our review and again in our concluding section on lessons learned and future directions.

Cross-sectional data analytic techniques

The most basic analytic technique within the GSEM framework is latent class analysis (LCA), also referred to as LPA when applied with continuous variables. Because most research in OB is conducted with continuous variables, we focus here on LPA. LPA is used to identify prototypical configurations of variables (e.g., Big 5 personality traits). Rather than assuming that there is a fixed pattern of correlations among these variables within the population, LPA allows for the detection of subpopulations for whom the variables combine differently. If it is found that these variables relate differently across profiles, both with one another and with variables not included in the LPA, it challenges the assumption of population homogeneity and raises questions regarding the interpretation of statistical parameters (e.g., correlation or regression coefficients) obtained in VC research.

It is important to note that LPA is generally viewed as an exploratory analysis. That is, like exploratory factor analysis in VC research, the results are dictated by the data and the optimal solution is typically guided by a set of established statistical indicators (see Morin et al., 2020, for more detail), as well as interpretability. The resulting profiles are also prototypical and probabilistic. That is, the profiles reflect the different ways that the variables commonly combine in the sample and each individual will have a probability of belonging to each of the prototypical profiles. However, despite the exploratory nature of the analysis, it is important to consider whether observed profiles are valid and meaningful. This can be achieved using procedures to evaluate the stability of profiles across comparable samples (Morin et al., 2016) or over time (Morin and Litalien, 2017), and by evaluating compatibility of the findings with existing theory (Marsh et al., 2009).

Longitudinal data analytic techniques

As a complement to the cross-sectional techniques described above, there are PC analytic techniques that can be applied to investigate the stability of profiles over time. As noted above, consistency in profile structure is one potential indicator of validity, particularly for the same sample over time. However, under conditions of change, it is possible for the profile structure to vary over time. Indeed, it might be possible to develop and test hypotheses regarding the expected change. Alternatively, the profile structure might remain the same but the individuals involved might move from one profile to another. Both of these possibilities can be investigated using latent transition analysis (LTA: for details, see Morin et al., 2020). The analysis typically starts with an LPA conducted for each time-point followed by comparison of the profile solution across occasions. The profile structures at different time-points need not be identical to conduct LTA, but if a comparison of the profiles provides evidence of configural (number of profiles), structural (profile shape), dispersion (within-profiles variability), and distributional (profile size) similarity, a common profile structure based on the model with the highest level of similarity can be used for the transition analysis. The LTA then provides an estimate of the proportion of the sample that maintains the same profile over time and the proportions that shift across profiles (see Morin and Litalien, 2017; Morin et al., 2016). Covariates can then be introduced into the model to examine their associations with these shifts. Note that, although we discuss LTA here as a longitudinal analysis, as it is commonly used in OB research, it is a highly versatile technique that can be applied with cross-sectional or longitudinal data and varying combinations of mixture models (see Nylund-Gibson et al., 2014, 2023). We address other potential uses in our discussion of future directions.

Another PC analytic technique that can be applied to longitudinal data is growth mixture modeling. GMM is the PC variant of latent curve analysis (LCA) used to identify initial levels and trajectories of change in variables over time. Unlike LCA, which identifies single population-level parameter estimates, GMM allows for the identification of subpopulations with distinct patterns (profiles) of initial levels (e.g., low, moderate, or high) of a variable combined with differing trajectories of change over time (e.g., stable, increasing, and decreasing). In more advanced forms of GMM, researchers can relax restrictions on additional parameters (e.g., intercept and slope variances and covariances, time-specific residuals) and allow for trajectories characterized by distinct functional forms (i.e., linear, quadratic). Supplementary analyses can be conducted to investigate potential predictors and outcomes of both profile membership and within-profile (state-like) fluctuations in the construct of interest. Thus, GMM is well-suited to investigating dynamic and time-structured organizational phenomena, including those occurring naturally or resulting from specific interventions (see Morin et al., 2020, for more detail).

Other analytic techniques

Although the cross-sectional and longitudinal techniques described above are the most widely used, additional techniques have been developed for use with multilevel data, and to identify subpopulation differences in the ways that variables relate to one another. Multilevel LPA (MLPA) makes it possible to identify profiles (defined in terms of variables, relations, or trajectories) across different levels of analyses. Mixture regression analyses (MRA), with their focus on variable relations, might be used to identify subpopulations differing in the extent to which predictor variables relate to criteria of interest. In MRA, profiles are defined by allowing all components of the regression equation (slopes, intercepts, and residuals) to differ across profiles, and can be combined with LPA when one also allows predictor means (and potentially variances) to differ across profiles in hybrid MRA. Finally, there are other techniques that involve the combination of PC and VC analyses. For example, latent profiles (PC) and factors (VC) can be used in the same models in various ways (e.g., factor mixture analyses, random intercept LTA: see Lubke and Muthén, 2005; and Muthén and Asparouhov, 2022; respectively). Profiles can also be used as moderators for a specific VC analysis (see McLarnon and O'Neill, 2018; for details). We provide additional details on these analytic techniques where relevant as we describe example applications in the section to follow.

Applications of PC techniques in OB research

In this section, we provide a selected summary of PC research to illustrate how PC techniques have been applied to test, challenge, extend, and integrate theory within major domains in the field of OB. These domains were selected based on both the extent and breadth of application of PC techniques. We also discuss studies within less well-represented domains in cases where the techniques are novel and have broader applicability in OB research.

Workplace commitments

Although applications of PC methods did not originate in the investigation of workplace commitments, it is arguably one of the OB domains in which it has been applied most frequently and using the greatest variety of analytic techniques (for reviews, see Kabins et al., 2016; Meyer and Morin, 2016; Meyer et al., 2013). The two most common applications to date focus on identifying profiles of affective (AC), normative (NC), and continuance (CC) commitment from Meyer and Allen's (1991) three component model (TCM), as well as profiles of commitments to multiple work-relevant foci (e.g., organization, occupation, supervisor, and team).

Applications of LPA to the TCM measures have confirmed the hypothesis that employees can experience organizational AC, NC and CC to varying degrees, but failed to support the prediction that the optimal profile would be characterized by strong AC combined with weak NC and CC (Meyer and Herscovitch, 2001). Indeed, it is commonly found that NC and CC combine synergistically with AC to improve employee retention, work engagement, and wellbeing (e.g., Oh, 2019; Pinho et al., 2024; Stark et al., 2025). This is particularly the case when CC is operationalized as perceived sacrifice as opposed to a lack of alternatives (e.g., Meyer et al., 2018; Stanley et al., 2013). Although VC investigations of the TCM allow for detection of interactions among the components in the prediction of outcomes (e.g., Gellatly et al., 2006; Jaros, 1997), PC studies have the added advantages of (a) identifying potentially meaningful constructs reflected within the patterns of indicators, and (b) allowing for the investigation of personal and/or situational antecedents of these constructs. That is, evidence suggests that the commitment mindsets associated with specific profiles (e.g., moral imperative; indebted obligation; Meyer and Parfyonova, 2010) may be distinct from those typically attributed to the individual components. Moreover, membership in profiles reflecting these mindsets have been linked to work conditions (e.g., HRM practices: Meyer et al., 2018; Pinho et al., 2024) that can be targeted for intervention.

Studies investigating TCM profiles have also been conducted for occupational commitment (e.g., Houle et al., 2020; Somers et al., 2019) as well as dual commitments to the organization and occupation (e.g., Meyer et al., 2021; Morin et al., 2015) and the organization and supervisor (Meyer et al., 2015). Studies involving dual commitments provided some support for the previously untested hypothesis that dependencies can exist among “nested commitments” (Meyer and Allen, 1997). For example, some evidence suggests that, in the absence of strong organizational AC, teachers with strong commitment to their occupation may experience a sense of obligation (NC) to the organization where they work (Meyer et al., 2019; Morin et al., 2015). These studies also revealed varying combinations of AC, NC, and CC to the dual foci, suggesting the potential for complementarity, synergy, or competition regarding outcomes (Johnson et al., 2009). However, computational demands and interpretive complexity limit the extent to which these interactive effects can be investigated across more than two targets while still retaining the three components. Consequently, multiple commitment profile studies have generally been conducted using only AC or related measures (e.g., Cooper et al., 2016; Meyer et al., 2021, 2025; van Ravenstein and Sentse, 2026). The latter studies typically yield profiles reflecting varying combinations of commitments, often in accord with differing theoretical perspectives (e.g., Meyer et al., 2021, 2025). This is an important contrast with VC research where theories are typically tested individually or in competition.

One of the challenges identified in multi-foci research is that commitments are often positively correlated, thereby increasing the likelihood of identifying quantitatively rather than qualitatively distinct profiles (i.e., profiles differing in level rather than shape). This problem was identified by Morin et al. (2011) and addressed using factor mixture modeling (FMA) to control for (remove) the shared variance. When conducted with unique (independent) indicators, the LPA yielded more qualitatively distinct profiles. More recently, researchers have used bifactor analysis (see Reise, 2012) to extract global (G) and specific (S) factors reflecting shared and unique variance, respectively, and including both in subsequent LPA (e.g., Meyer et al., 2021, 2025; Morin et al., 2025). This approach allows for variation across subpopulations in both levels of the G factor (i.e., what is common among the individual commitments) and in configurations of the S factors (i.e., what is unique to each target commitment), in accord with Perreira et al.'s (2018) hierarchical model of multiple foci. These studies commonly find that profiles with stronger levels of G are associated with more positive outcomes, and that the outcomes associated with different configurations of the S factors provide evidence for complementarity, synergy, and competition rather than one or the other.

In addition to identifying static profiles of commitment mindsets and foci, researchers have conducted longitudinal studies using LTA to investigate stability in profile structure and membership over time (e.g., Anderson et al., 2022; Houle et al., 2020; Kam et al., 2016; Xu and Payne, 2018), or GMM to identify profiles of change trajectory (Houle et al., 2022, 2024). For example, Kam et al. (2016) found the same profile structure for organizational commitment over an 8-month period among employees experiencing a large-scale organizational change. Houle et al. (2020) found evidence for the stability of occupational commitment profiles over a 2-year period. Xu and Payne (2018) found a stable profile structure involving AC and CC among military personnel, and also found that individual transitions across profiles aided in the prediction of turnover.

Evidence for stability of the profile structure, despite movement of individuals across profiles, as observed in the foregoing studies, is testament to the meaningfulness of the profiles (Marsh et al., 2009). Incidentally, it was in the context of commitment profiles that Morin et al. (2016) developed procedures to compare profile similarity across samples and/or time. These procedures have become standard practice in profile comparison across research domains. Their own study provided evidence for cross-cultural stability of TCM profiles, and a subsequent study by Meyer et al. (2018) demonstrated stability over time prior to and following an economic crisis.

In other longitudinal studies, GMM has been used to identify commitment trajectory profiles (e.g., Houle et al., 2022, 2024, 2025a). For example, Houle et al. (2024) used GMM to identify trajectories of nurses' commitment to both the organization and occupation over the first 5 years of their careers. They identified four trajectory profiles that were similar across the two targets of commitment: High, Moderate, Low and Increasing, and Average/Low and Decreasing. Subsequent profile comparisons within and across targets revealed similarities and differences in associations with antecedents (e.g., socialization) and outcomes (e.g., need satisfaction). More recently, Houle et al. (2025a) investigated and found multiple trajectories of organizational commitment among military recruits, again observing within-profile differences in variability and cross-profile associations with antecedents and outcomes. Thus, in addition to demonstrating that there are multiple patterns in the development of commitment, studies of this type help to identify where and how interventions might best be targeted to achieve optimal outcomes.

In sum, applications of PC methods in commitment research have contributed to the discovery of more nuanced commitment mindsets and their implications, illustrated the multiple ways that workplace commitments can combine and interact to influence important organizational- and employee-relevant outcomes, and identified the nature, antecedents, and implications of changes in commitments over time.

Leadership

Leadership is another domain in which there has been a steady increase in the use of PC methods. Not surprisingly, given its predominance over the last several decades (Gardner et al., 2020), many PC studies have focused on leadership styles and, in particular, variations of Bass and Avolio's (1997) full-range leadership model (e.g., Arnold et al., 2017; Doucet et al., 2015; Hancock et al., 2023; Poetz and Volmer, 2024; Stadnick et al., 2019; Willis et al., 2021). Findings regarding profile structure vary across studies, but a common theme is the observation that the leadership styles included in the full-range model combine in different ways across subpopulations of leaders. It is common to find profiles combining transformational and contingent reward leadership (e.g., Arnold et al., 2017; Hancock et al., 2023; Stadnick et al., 2019; Willis et al., 2021), but some studies also identified profiles combining transformational leadership with management by exception and/or passive leadership in what have been labeled comprehensive (Arnold et al., 2017; Poetz and Volmer, 2024), moderate (Doucet et al., 2015), or undifferentiated (Stadnick et al., 2019) profiles. Together, these studies raise questions about, or at least the need to qualify, the findings of research addressing these leadership styles in isolation.

Looking beyond the full-range leadership models, studies have investigated profiles pertaining to multidimensional leadership constructs such as leader-member exchange (LMX: Gillet et al., 2022a), transformational leadership (Tian et al., 2023), authentic leadership (Cougot et al., 2025), paternalistic leadership (Chou et al., 2015), leader behavior (Howard et al., 2025; Laborie et al., 2026; Lornudd et al., 2016), and need-supportive leadership (Huyghebaert-Zouaghi et al., 2023, 2025; Paquette et al., 2025). In a recent study, Bartz and Felfe (2026) identified four profiles reflecting varying combinations of transformational leadership and abusive supervision from two sources: upper- and lower-level leaders. Again, across all studies, a key finding is that distinct configurations of the dimensions emerge and relate differently with important outcomes, raising questions regarding the generalizability and meaningfulness of findings pertaining to omnibus measures of these leadership constructs.

Importantly, whether dealing with multiple styles or multidimensional constructs, the indicator variables of interest are often correlated (Howard et al., 2025). As noted previously, this can increase the likelihood of identifying quantitatively rather than qualitatively distinct profiles. To address this issue, some researchers conducted FMA to control for shared variance (e.g., Tian et al., 2023) and others conducted bifactor analyses to disaggregate the common and unique variance (e.g., Gillet et al., 2022a). Indeed, this issue is becoming increasingly common in PC research and we address it in more detail in our concluding section.

PC research involving the full-range model and multidimensional leadership constructs typically include a set of theory-specific indicators. As a result, the findings have direct relevance for the underlying theory. However, in a few of the studies identified above, the authors introduced additional indicators for purposes of clarifying or elaborating on the meaning of the profiles. For example, Tian et al. (2023) tested Christie et al.'s (2011) theory that true- and pseudo-transformational leaders could be distinguished by their configuration of the four facets of transformational leadership (inspirational motivation, idealized influence, individualized consideration, and intellectual stimulation), with inspirational motivation dominating for pseudo leaders. However, to be consistent with other treatments of the true vs. pseudo distinction, they also included a separate measure of leader self-interest that ultimately helped in clarifying the distinction between the two types of leaders. Howard et al. (2025) included a measure of behavioral integrity to augment the leader behavior measures from Yukl's (2012) taxonomy. They found that the integrity measure helped to distinguish among leaders with comparable levels of task-oriented behavior, as well as the outcomes associated with the profiles. Finally, Hancock et al. (2023) included a measure of abusive supervision along with measures of full-range leadership. Across four studies, they found that abusive supervision was most likely to accompany passive leadership, but that not all passive leaders were abusive. Instead, they found distinct passive and passive-abusive profiles. Interestingly, the passive-abusive profile was associated with lower physical wellbeing and perceived organizational support, but the passive profile was associated with lower psychological wellbeing. Thus, the detrimental effects of passive leadership may be more nuanced than observed in previous VC research.

A PC approach has also been used to identify profiles of various individual differences, including leader personality (Parr et al., 2016; Stackhouse and Rickley, 2026), leadership skills (Dai and De Meuse, 2013; Müller and Turner, 2010), and motivation to lead (Auvinen et al., 2020). Again, these studies revealed different configurations of the relevant indicators, suggesting the existence of more nuanced “gestalt” constructs. For example, Parr et al. (2016) identified six distinct trait profiles using data obtained by administering the Hogan Personality Inventory (Hogan and Hogan, 2007) to executives participating in an assessment center. Whereas, one profile with elevated scores on all seven HPI traits (i.e., Power Player) scored highest across all performance measures (e.g., defining and executing strategy), the remaining profiles demonstrated different patterns of performance across measures.

Although most PC leadership studies have been conducted to identify distinct configurations of theory-relevant indicators, a few studies have applied other PC techniques. For example, Chénard-Poirier et al. (2017, 2022) used a hybrid MRA approach to identify profiles of leader behavior and investigate between- and within-profile relations with various outcomes. The analytic procedures used in these studies combine PC (LPA) and VC (MRA) techniques thereby capitalizing on the strengths of both. To illustrate, using LPA, Chénard-Poirier et al. (2017) found that profiles of leader empowerment behaviors (delegation, coaching, and recognition) primarily reflected differences in level of the three behavior types, and that followers' experienced psychological empowerment was greater in the higher empowering behavior profiles. This is consistent with Lawler's (1992) theoretical argument that empowering behaviors should be “coherent” (i.e., at matching levels). However, in analyses conducted to predict experienced empowerment from the facets of leader behavior within profiles, Chénard-Poirier et al. found some potentially beneficial effects of “incoherence.” For example, they found that, within profiles, experienced empowerment was greater when leaders emphasized delegation over coaching and recognition. Thus, in addition to finding some support for Lawler's coherence hypothesis, the findings suggest that, under some conditions, incoherence might also affect empowerment perceptions.

Taking a similar approach, Chénard-Poirier et al. (2022) used LPA to identify profiles of constructive and destructive leadership and found three profiles, one reflecting predominantly constructive, another predominantly destructive, and a third inconsistent leader behaviors. Between-profile comparisons revealed that the inconsistent profile was associated with the most negative outcomes. Within-profile analyses involving MRA to predict outcomes revealed, among other things, that within the inconsistent profile, increases in both transformational leadership and petty tyranny were both associated with improved outcomes, suggesting that movement toward greater consistency in either direction is viewed positively.

Finally, Afota et al. (2026), applied GMM to measures of leader-member exchange (LMX) obtained on three occasions over a 6-month period from employees engaged in hybrid work. In addition to identifying five distinct profiles (trait-like distinctions), they also identified within-profile (state-like) fluctuations in LMX. Within-profile fluctuations were greater in profiles with decreasing trajectories, and both profile membership and within profile fluctuations were associated with telework intensity (time spent) and monitoring practices (observational vs. interactional) as predictors, and with performance, turnover intentions, and emotional exhaustion as outcome. Thus, by using GMM, the researchers were able to demonstrate the dynamic nature of LMX among hybrid workers and identify predictors of change trajectory and within-profile fluctuations that might be targeted for intervention.

In sum, PC research on leadership demonstrates that leadership styles, personality traits, and behaviors typically investigated individually in VC research, can combine and relate to other variables in different ways. Differences in relations with important outcomes serve to qualify relations observed in VC research. Interestingly, most PC leadership research focuses on outcomes, with little attention to the prediction of profile membership. One of the advantages of examining the interplay among profile indicators (e.g., leadership behaviors) using a PC approach is the ability to link membership to potential antecedents. Therefore, this could be an important direction for future research with implications for leader selection and development. There is also a need for more longitudinal PC research to address the dynamic nature of leadership processes.

Work motivation

Much of the PC research on work motivation has been conducted using a self-determination theory (SDT: Ryan and Deci, 2017) framework. According to SDT, individuals can experience different types of motivation varying in degree of autonomy (self-determination) vs. external control. The most controlled type is external regulation (external reward/punishment), followed by introjected regulation (sense of pride/guilt). The autonomous types include identified regulation (value-driven), integrated regulation (assimilation into self), and intrinsic motivation (task enjoyment). SDT also acknowledges “amotivation” (absence of goal direction) as another distinct type. VC studies of work motivation typically investigate the antecedents (e.g., supportive leadership) and/or outcomes (e.g., performance, wellbeing) of the motivation types or the underlying autonomy/control dimension. Relative autonomy scores are commonly computed by subtracting scores for the controlled motivation types from scores on the autonomous types (e.g., Ryan and Connell, 1989; Sheldon et al., 2017). Results generally support SDT predictions that autonomous motivation, and the more autonomous types, are associated with the most desirable outcomes (see Van den Broeck et al., 2021; for meta-analytic evidence).

More recently, researchers have taken a PC approach to test SDT. Among the advantages of this approach is that it allows for the possibility that (a) employees can experience varying combinations of the motivation types, and (b) the types will relate differently to other variables in combination than when considered individually. Numerous studies have now revealed that this is indeed the case (e.g., Howard et al., 2016; Levesque-Côté et al., 2021; Meyer et al., 2022; Parker et al., 2021; Zhou S. et al., 2025: see Blechman et al., 2025; for a review). These studies typically yield four to six distinct profiles, the most common of which have been labeled as amotivation, external regulation, autonomous, and mixed. The mixed profiles include moderate to strong scores on a combination of “controlled” and “autonomous” motivation types. This is an important finding because it acknowledges that employees can simultaneously recognize both external and internal sources of motivation for their actions. Moreover, mixed profiles are often found to be associated with the same antecedents and positive outcomes as the autonomous profiles. To investigate why this might be the case, Meyer et al. (2022) administered direct measures experienced autonomy and control along with the type measures and found that individuals with mixed profiles experienced greater autonomy than external control, despite having moderate to strong scores on external regulation and/or introjected regulation. Thus, these PC studies raise questions about the interpretation of correlations involving the individual types, especially those involving extrinsic and introjected regulation, as well as the practice of using type measures to create an index of relative autonomy.

A more recent development in the investigation of SDT motivation profiles has been the use of bifactor analysis to disaggregate the variance shared among the motivation types (i.e., degree of autonomy) from that which is unique to each type. In addition to increasing the likelihood of identifying qualitatively distinct type profiles, this allows for independent investigation of the implications of the autonomy continuum and the distinct motivation types. Studies using this approach have indeed found unique predictive effects largely in accord with SDT predictions (e.g., Blechman et al., 2025; Fernet et al., 2020; Gillet et al., 2020; Tóth-Király et al., 2021). In another study, Gillet et al. (2018a) used the “uncontaminated” G factor to investigate profiles of trajectories of self-determined motivation using GMM. One caveat with the bifactor approach, however, is uncertainty regarding the meaning of the global (G) and type-specific (S) factors. Recent research investigating the meaning of the G factor provide some support for its interpretation as global self-determination (e.g., Espinoza and Meyer, 2025), but the meaning of the S factors still requires investigation. We address this issue more broadly in the concluding section of this article.

PC techniques have also been used to investigate Vallerand's (2015) dualistic passion model (e.g., Astakhova et al., 2024; Gillet et al., 2023b; Li et al., 2020). This model distinguishes harmonious passion (HP: psychological investment reflecting free choice) from obsessive passion (OP: psychological investment reflecting external or internal contingencies). These PC studies provide evidence that HP and OP can each dominate for some individuals, but can co-exist (mixed profile) for others. In addition, Astakhova et al. (2024) observed profiles with different combinations of HP and OP for both work and non-work activities. Again, as in the case of the SDT motivation types, these studies reveal that profiles reflecting OP in combination with HP are associated with more positive outcomes than profiles dominated by OP.

PC investigations involving other motivational frameworks have been less systematic, but nevertheless provide additional evidence that distinct motives, often studied in isolation in VC research, can combine in varying ways. For example, Jang et al. (2024) found evidence for four distinct profiles of motives for engaging in organizational citizenship behavior (OCB). Lohbeck and Frenzel (2022) found four profiles reflecting varying motives for pursuing a teaching career. Zhou S. et al. (2025) also found four distinct profiles reflecting autonomous and controlled motives for teaching but, in an interesting twist, also included measures of coping behaviors. Profiles reflecting motivation and coping behavior in combination predicted the psychological wellbeing and occupational commitment of teachers during preservice training. Kilinç et al. (2025) found three teacher-level motivation profiles but also conducted MLPA to investigate how these profiles were distributed across schools characterized by two distinct climate profiles (high-performance and high mastery). They found that the school-level profiles moderated the effectiveness of transformational leadership on teachers' commitment and sense of alienation. Although it is more difficult to synthesize the results of these studies because of their unique objectives, measures, and analytic techniques, they are consistent in demonstrating the existence of multiple subpopulations along with differences in how the indicator variables relate to other variables across profiles.

In sum, like the components of commitment in the TCM, PC studies involving types of motivation and passion reveal that they can combine in different ways and that the experienced motivational state associated with the combinations might be different from that for individual types. Moreover, the relations observed between the type profiles and other variables, both antecedents and outcomes, suggest the need to qualify the findings of VC research where the types are investigated individually. Recent findings regarding other motivation taxonomies have similar implications.

Employee wellbeing

Employee wellbeing has become an issue of increasing interest in the OB literature, both for its own sake and its implication for organizational effectiveness (Kelloway et al., 2023; Martela, 2025). Perhaps not surprisingly given the breadth of the topic, it has been investigated from a variety of perspectives with wellbeing being variously defined and measured. Among the most common application of PC methods in this literature has been to identify distinct configurations of components within multidimensional constructs. This includes measures of wellbeing in general (e.g., Bujacz et al., 2020; Houle et al., 2025b; Rantanen et al., 2024), as well as work engagement (e.g., Gillet et al., 2019a; Gillet et al., 2020), need satisfaction (e.g., Huyghebaert-Zouaghi et al., 2022; Tóth-Király et al., 2024; Gillet et al., 2024a), coping and recovery (e.g., Gabriel et al., 2019; Kleshinski et al., 2024), stress (e.g., Somers et al., 2019), workaholism (e.g., Gillet et al., 2025), work fatigue (e.g., Blais et al., 2020; Zhai et al., 2025), and burnout (e.g., Pyhältö et al., 2021; Sandrin et al., 2022).

An important distinguishing characteristic of PC wellbeing studies is the selection of the indicators to be included in the analysis. In some cases, these indicators reflect the dimensions of well-established theoretical constructs (e.g., engagement, need satisfaction, coping, and burnout). In other cases, a variety of indicators are included with no well-established underlying connection. As we discuss in our concluding section, the former studies are generally simpler to interpret and evaluate (e.g., generalizability, meaningfulness), and tend to have more direct implications for theory and application. For this reason, we give priority here to studies involving well-established constructs and/or measures. We begin by considering studies addressing single constructs (e.g., engagement, burnout) using both cross-sectional and longitudinal designs, and then address studies similarly investigating a combination of constructs (e.g., engagement and burnout). We conclude by providing examples of research involving multilevel analyses.

As was the case with commitment, leadership, and motivation, the components within multidimensional well-being constructs can reflect both common and unique construct-relevant variance. Thus, as in these other domains, researchers have begun to conduct bifactor analyses to disaggregate the common (G) from unique (S) variance in the components of engagement (e.g., Gillet et al., 2019a; Perera et al., 2021), need satisfaction (e.g., Gillet et al., 2019a; Huyghebaert-Zouaghi et al., 2022, 2023; Tóth-Király et al., 2023), coping (e.g., Gillet et al., 2025), and burnout (e.g., Gillet et al., 2023c; Sandrin et al., 2022). G is presumed to reflect the unifying construct, with S reflecting what is unique to each component.

A common finding in these studies is that the G factor has the strongest association with antecedent and outcome variables. This is consistent with what is typically found in VC studies using composite measures. However, in PC studies, these antecedents and outcomes are often found to differ as well across profiles with varying patterns of the S factors. For example, Sandrin et al. (2022) found that, in some burnout profiles, the cynicism factor had a counterintuitive association with job satisfaction (i.e., high cynicism was associated with high job satisfaction and low cynicism was associated with low job satisfaction). Thus, cynicism in these profiles had exactly the opposite relation from what is typically expected and found in VC studies.

The different patterns of indicator scores in profiles observed in LPA are sometimes interpreted in terms of their relevance to developmental theory. For example, in burnout studies, the patterns of emotional exhaustion, cynicism, and professional efficacy have been linked to theories regarding the burnout process (e.g., Duan et al., 2023; Mäkikangas et al., 2021). Importantly, as was the case for multiple commitment research, a PC approach can provide evidence consistent with competing theories. For example, in a sample of nursing home care aides, Duan et al. identified four burnout profiles. Whereas, emotional exhaustion and cynicism covaried together across profiles, for some individuals, professional efficacy (a contra indicator of burnout) remained high even when exhaustion and cynicism were high. Importantly, the authors noted that, although some of the observed patterns were consistent with dominant developmental theories, none of these theories could account for all of the distinct patterns observed.

Of course, cross-sectional PC studies do not allow for direct tests of the developmental process and/or the effectiveness of interventions. For this, alternative PC strategies are required, including the application of LPA with multi-wave data (e.g., Mäkikangas et al., 2021, 2022) and GMM (e.g., Hätinen et al., 2013; Mäkikangas et al., 2012). For example, Mäkikangas et al. (2022), conducted LPA on a composite measure of work engagement obtained on four occasions from workers forced to work remotely during COVID. In this use of LPA, the measures obtained on the four occasions served as the indicator variables. They identified four distinct profiles varying primarily in level of engagement, with one profile involving low initial engagement as well as a decline over time. Moreover, they found that perceptions of organizational support, the functionality of the home work environment, job-related self-efficacy, and job crafting were highest among those employees who remained engaged over the course of the study.

In another adaptation of LPA for use with multi-wave data, Mäkikangas et al. (2021) conducted their analyses using the three facets of burnout measured on five occasions over 8 years as indicator variables. In addition to a stable low burnout profile, they found two burnout patterns reflecting different developmental trajectories. Moreover, they found that job demands and resources had different relations with these trajectory profiles. Specifically, job demands were associated with a profile reflecting increasing burnout instigated by early levels of exhaustion, whereas job resources were associated with an inverted U-shaped pattern of burnout. They noted that these different patterns align with distinct developmental theories and suggest the need for tailored intervention strategies.

Hätinen et al. (2013) used GMM to identify burnout trajectories among clients undergoing a 1-year rehabilitation intervention. They identified three profiles: low burnout, high-burnout—benefitted, and high-burnout—not benefitted. Moreover, they demonstrated that recovery from burnout was associated with increased job resources and decreased job demands (antecedents) and increased job satisfaction and decreased depression (outcomes).

PC studies have also been conducted with combinations of wellbeing measures, including engagement and burnout (Gillet et al., 2024b; Salmela-Aro et al., 2019; Villanen et al., 2026) and engagement and workaholism (Cook and Gilin, 2023). These studies allow for tests of theory pertaining to relations among these constructs and, again, have the advantage over a VC approach of finding evidence consistent with competing theories across subpopulations. For example, Gillet et al. (2024b) identified five profiles reflecting facets of both burnout and engagement: burned-out disengaged, burned-out involved, engaged, engaged exhausted, and normative (all average). Of note in this study, bifactor analyses were used to extract separate G factors for burnout and engagement along with S factors for the facets. In four of the five profiles, some of the S factor scores deviated from G. No profile had matching high/low levels of G and S—but there was more alignment for the engagement factors than the burnout factors. Turnover intention was greater in the engaged exhausted than the normative profile despite higher “global” engagement in the former. Thus, by taking a PC approach, it was possible to identify profiles combining comparable levels of engagement and burnout facets, and to illustrate how these facets might be experienced differently, and relate differently with other variables, across profiles. Cook and Gilin (2023) likewise demonstrated that engagement combined with both low and high workaholism, and that employees with an engaged profile reported less work to family conflict than and those with an engaged workaholic profile.

Complementing these cross-sectional studies is a longitudinal study by Mäkikangas et al. (2012) who used GMM to identify trajectory profiles for two facets of engagement (vigor and dedication) and two facets of burnout (exhaustion and cynicism). They identified four profiles for exhaustion and cynicism, five profiles for vigor, and six profiles for dedication. Importantly, in addition to identifying multiple trajectory patterns, they observed that trajectories of cynicism and dedication were opposite to one another with a strong negative relation, whereas trajectories of exhaustion and vigor were not connected and seemed independent. Thus, their findings are consistent with the supposition that cynicism and dedication may be polar opposites (antipodes), whereas this does not appear to be the case for vigor and exhaustion. Consequently, the latter may need to be addressed separately in interventions.

Finally, a few multilevel PC studies have been conducted to complement the individual-level studies described above (e.g., Benitez et al., 2019; Gillet et al., 2024b). In these studies, LPA is conducted to identify individual-level profiles followed by MLPA to identify the prevalence of the profiles within higher-level units. For example, Benitez et al. identified four distinct profiles of job satisfaction, emotional exhaustion, and cynicism, and found that individuals with these profiles were unevenly dispersed among work groups. Importantly, groups with greater representation of the contented profile (high job satisfaction and low exhaustion and cynicism) had the highest levels of performance. Similarly, in a study discussed above, Gillet et al. (2024b) followed their LPA of engagement and burnout with an MLPA to investigate the distribution of profiles across work units. This also allowed them to investigate associations with job demands and resources at both an individual and group level.

In sum, PC studies again complement VC studies of wellbeing by demonstrating that the facets of multidimensional constructs (e.g., burnout, engagement), can combine in different ways with unique relations with theoretical antecedents and outcomes. The same applies when these constructs are investigated together and in studies of change over time. By identifying subpopulations with varying facet combinations and change trajectories, PC studies can qualify the results of VC studies and illustrate how even competing theoretical predictions can hold for different subpopulations. They also allow for the investigation of predictors of wellbeing profiles and change trajectories. This can have important implications for intervention.

Other examples

There are many other substantive domains within the field of OB where researchers are beginning to apply PC methods. Indeed, there are far too many studies at this point to do justice to in this review. Thus, we simply provide a sampling of domains and studies to illustrate the diversity of topics that are well-suited to a PC approach.

As in the foregoing review, most PC studies in the broader OB literature have been conducted to address the interplay among the components or dimensions of focal constructs. These include investigations of (a) individual differences such as personality (Haag et al., 2024; Min and Su, 2020; Perera et al., 2018; She et al., 2024; Smith et al., 2021), emotional intelligence (Zhou J. et al., 2025), empathy (Anderson and Gilin, 2025), affect (Gabriel et al., 2022), psychological capital (Bouckenooghe et al., 2019; Lipshits-Braziler et al., 2026), self-efficacy (Perera et al., 2019), work values (Busque-Carrier et al., 2022; Drouin-Rousseau et al., 2024), work ethic (Woehr et al., 2024), goal orientation (Kunst et al., 2018), psychological empowerment (Gillet et al., 2023a), change readiness (Köhler et al., 2024), mindfulness (Lee et al., 2024), and social identities (Manzi et al., 2024); (b) situational variables including organizational climate (Marinova et al., 2019; Wang et al., 2026), job demands and resources (Marzocchi et al., 2023; Müller et al., 2025; Vangrieken et al., 2026), perceived support (Caesens et al., 2020), organizational justice (Eib et al., 2025; Ju and Ning, 2022), diversity and inclusion (Qu and Wang, 2022; Sommovigo et al., 2025), job security (Urbanaviciute et al., 2021), safety climate/culture (Lefsrud et al., 2021), feedback environment (Dahling et al., 2017), team conflict (O'Neill et al., 2018; Schoen et al., 2026), and structural empowerment (Cougot et al., 2024); and (c) work behavior, including job performance (Peeters et al., 2021), creativity (Zhang et al., 2021), impression management (Chawla et al., 2021; Debus et al., 2024), emotional labor (Cheung and Lun, 2015; Clarke et al., 2026; Gabriel et al., 2015; Hu et al., 2023; Nguyen and Stinglhamber, 2020; Zeng et al., 2025), voice (He et al., 2026), job crafting (Ho et al., 2024; Mäkikangas and Schaufeli, 2021; Van Aerden et al., 2015; Zhang et al., 2024), responses to diversity initiatives (Kanitz et al., 2024), bystander sexual-harassment interventions (Liang and Park, 2025), deviance (Bigelow et al., 2024), and mistreatment (Nixon et al., 2021). They have also been used in research involving the novel construct of death awareness during the COVID pandemic (Zhong et al., 2021).

A smaller body of research has been conducted using longitudinal data to assess the stability of the profile structure and investigate transitions of individuals across profiles. Notable examples include longitudinal investigations of workplace supports (Caesens et al., 2020), goal-orientation (Kunst et al., 2018), job crafting (Mäkikangas, 2018), job demands and resources (Beeler et al., 2026; Thomas et al., 2026) psychological capital (Zyberaj et al., 2022), and emotions (Slaughter et al., 2021). Other studies used multilevel analyses with longitudinal measures to control for the non-independence (nesting) of time-lagged measures (e.g., Lu et al., 2025 for job crafting and Shockley et al., 2022, for attributions for work-family conflict). Finally, longitudinal data have also been used in studies to identify distinct patterns (profiles) in trajectories of change, and to investigate both potential predictors and outcomes of these trajectories. These include trajectories of change in perceptions of justice (Eib et al., 2025), performance of expatriate managers (Takeuchi et al., 2019), and perceived organizational support (Caesens et al., 2020).

Again, the foregoing is simply a sampling of domains and studies intended to illustrate the diversity of applications, provide illustrative examples, and serve as an impetus for future research in these and related domains. We turn now to discussion of the lessons to be learned from the existing body of research that we hope might serve as a guide for research going forward.

Lessons learned and future directions

Our objective in this section is to encourage cross-fertilization by discussing how the analytic techniques, problems, and solutions from PC research in one domain within the OB literature, might inform and guide research in others. We begin with an overview of the rationale and justification for taking a PC approach and then address a series of more specific issues that arose in our review and should be considered by researchers going forward.

Why and when to take a PC approach

The distinctions between, and complementary nature of, the VC and PC approaches have been well-articulated elsewhere (e.g., Morin et al., 2020; Woo et al., 2024). We focus here on the specific reasons why one might consider using a PC approach and, importantly, the conditions required to support such a decision. For present purposes, we focus our attention on the basic decision to treat the sample under investigation as reflecting subpopulations rather than a single homogeneous population.

As we conducted our review, we were struck by the number of studies we encountered in which the profiles identified with LPA were quantitatively (level) rather than qualitatively (shape) distinct. Although, as we discuss later, this can be important information, in many of these cases, authors continued to interpret the findings as evidence of distinct subpopulations reflecting unique variable interrelations and perhaps requiring tailored treatments. In these cases, the alternative, and more parsimonious, interpretation of a single population varying along a single continuum amenable to a VC approach cannot be ruled out. The primary rationale for taking a PC approach in many such studies was that most previous research had been VC. In contrast, the studies making the strongest contributions were those that (a) provided a strong rationale for taking a PC approach, (b) demonstrated that their data were appropriate for such an investigation, and (c) interpreted findings regarding profiles, whether quantitatively and/or qualitatively distinct, appropriately and in accord with the intended objectives.

There are many reasons for taking a PC approach. One is to test theory regarding how distinct facets of a construct (e.g., commitment components; motivation types; dimensions of engagement and burnout), or multiple variables (e.g., commitment foci; leader behaviors; wellbeing indicators), can combine, and how the combinations relate to other variables. As we noted in our review, such tests can provide support for an overarching theory, but might also suggest need for refinement. Importantly, by acknowledging multiple subpopulations, PC studies can sometimes provide support for multiple, even competing, theories. Second, even in the absence of theory, there may be reasons to expect a potentially complex interplay among a set of variables (e.g., personality traits; facets of satisfaction; cultural or work values). Detecting interactions and non-linear relations becomes exceedingly difficult in VC research as the number of variables increases (Aguinis et al., 2005). Moreover, even if detected, they may be extremely difficult to interpret and communicate (e.g., visualizing beyond three-way interactions). In PC research, these complex interrelations can be captured in profiles that are arguably easier to interpret (but see below for more detailed discussion).

Regardless of the specific rationale for taking a PC approach, an important first step is to demonstrate that the variables involved are sufficiently distinct to combine in different ways. Variables that are highly correlated will at best yield profiles varying in level (e.g., all high, moderate, and low). Therefore, an important preliminary step is to investigate the factor structure of the variables involved using confirmatory factor analyses or exploratory structural equation modeling (Marsh et al., 2014) and considering various model specifications (e.g., bifactor structures). Interested readers should consult Morin et al. (2017) for a step-by-step guide to sequential model evaluation. Subsequent LPA is warranted only if there is sufficient evidence for distinct factors. If a bifactor analysis indicates a hierarchical structure, then extracting one or more general factors (G) along with multiple specific factors (S) is most likely to lead to detection of qualitatively distinct profiles (see below for more detail). Assuming that a PC approach is indeed justified, the following are additional issues to be considered.

Open vs. closed indicator sets

One notable difference across studies, even within a substantive domain, is the selection of indicator variables. We refer here to a set of indicators as “closed,” when they are specifically identified and differentiated within a theory, construct, or measure. For example, the Big 5 personality traits (Digman, 1997), the three components of organizational commitment (Meyer and Allen, 1991), the motivation types in SDT (Ryan and Deci, 2017), the leadership styles included within the full-range leadership theory (Avolio and Bass, 2004), and the dimensions of engagement (Schaufeli et al., 2002; Rich et al., 2010) and burnout (Maslach et al., 1996), all constitute closed sets. By contrast, variables selected because they represent work demands/resources, work characteristics, or indicators of stress/wellbeing, but that are not unique to a particular theory or part of a standard multidimensional construct, are considered an “open” set. Whereas, use of a closed set is useful for theory testing/refinement, an open set can be useful for theory development. For example, a researcher might expect that employees can have multiple reactions to hybrid work (e.g., affective, attitudinal, and/or behavioral), but have no a priori theory regarding how the reactions are related, within or over time, and whether these relations might differ across subpopulations. A PC approach provides a suite of analytic techniques this researcher might use to explore the relations among an open set of indicators (i.e., reactions) with the objective of developing theory regarding the nature, antecedents, and outcomes of reaction patterns within time and trajectories over time.

A third approach, involving what we describe as an “augmented closed set,” combines the advantages of the other two. Examples of an augmented approach are the inclusion of self-interest along with facets of transformational leadership to distinguish between true and pseudo-transformational leadership (Tian et al., 2023), the inclusion of abusive supervision within the full-range model to identify distinct types of passive leadership (Hancock et al., 2023), inclusion of integrity with measures of leader behavior to distinguish among leaders with a task-orientation (Howard et al., 2025), and inclusion of coping behaviors along with SDT-based measures of autonomous and controlled motivation (Zhou S. et al., 2025) to understand pre-service teachers' psychological wellbeing and occupational commitment.

Finally, a related issue to consider is the combination of variable sets. Interestingly, a variable treated as auxiliary (e.g., antecedent or outcome) in one PC study might itself be the focal construct in another PC study. For example, burnout might be considered an outcome variable in a study of multiple commitment profiles whereas, in other studies, commitment could be treated as a predictor of burnout profiles. As we noted in our summary of PC analytic techniques, LTA can be applied for many purposes beyond investigation of transitions in profile membership over time (see Nylund-Gibson et al., 2014, 2023). For example, it could be used in the present example to investigate how membership in commitment profiles relates to membership in burnout profiles or, with longitudinal data, to burnout trajectory profiles. Thus, with due consideration to the caveat regarding complexity below, we encourage more research involving the combinations of variable sets within PC investigations. Although the analysis would be identical to LTA, the approach might be better described as an investigation of latent mixture associations given that no actual transitions are involved.

Generalizability and cross-study comparison

The ultimate goal of applied OB research is to generate useful information to guide practice. In some cases, the findings might have direct implications for practice. For example, studies identifying patterns of burnout, or burnout trajectories, within a defined population could include predictor variables to aid in identifying strategies for targeted intervention (e.g., Mäkikangas et al., 2022; Pyhältö et al., 2021), or to evaluate the effectiveness of interventions (e.g. Hätinen et al., 2013). Such findings might have most immediate practical implications in investigations involving well-defined occupations or settings (e.g., nurses, teachers, and military personnel). In other cases, the findings might serve to develop, test, or revise theory from which general principles can be derived and applied as a guide to practice. For example, studies involving the SDT motivation types provide evidence that the potential negative implications of extrinsic regulations can be mitigated when accompanied by intrinsic motives (e.g., Howard et al., 2016; Meyer et al., 2022). In either case, it is important that findings are replicable and accumulate meaningfully across studies; rarely will the findings of a single study be definitive.

As noted in our review, there are domains within OB where substantial bodies of research already exist and efforts have been made to synthesize the results (e.g., organizational commitment: Kabins et al., 2016; Meyer and Morin, 2016; SDT motivation types: Blechman et al., 2025; burnout: Mäkikangas and Kinnunen, 2016; FFM personality traits: Yin et al., 2021). Even here, there are factors that make cross-study comparison/synthesis difficult, and these factors are exacerbated in research without guiding theory and/or standard measures. We focus here on three factors that can contribute to difficulties in comparison/synthesis—the exploratory nature of PC methods, small sample sizes, and scaling strategy—and suggest ways to address them.

Mixture modeling, as it is typically applied in OB research, is exploratory (data-driven) and this can contribute to variability in findings across studies. The differences may in fact be meaningful as, for example, when studies are conducted with unique populations or distinct contexts. These differences warrant investigation in their own right. However, in other cases, differences may reflect noise in the data and/or differences in judgment calls (e.g., number of profiles to extract) made by the researchers. In the absence of guiding theory (see below), researchers might rely on evidence of profile consistency (stability) across samples or within samples over time to attest to the meaningfulness of the profiles (see Marsh et al., 2009; Meyer and Morin, 2016). Strategies have been developed to evaluate cross-sample (Morin et al., 2016) and cross-time (Morin and Litalien, 2017) stability in LPA results, and these have since been extended to address stability in findings for GMM, MRA, and MLPA (see Collie et al., 2020, 2021; Morin et al., 2020). To make effective use of these strategies, researchers attempting to achieve meaningful results from their exploratory analyses should collect data from multiple samples and/or over multiple occasions.

When there is theory and/or a large body of existing research to guide expectations, Schmiege et al. (2018) proposed the use of a “confirmatory approach” that might also be useful in achieving greater consistency across studies. In this approach, parameters are fixed to align with the theory and/or previous research. Importantly, however, in the absence of absolute fit indices for mixture models, it remains necessary to contrast models with fixed (confirmatory) and free (exploratory) parameters in evaluations of fit with theory and/or previous research. Recent applications of this approach include Stark et al.'s (2025) confirmation of the TCM commitment profiles and their relations with wellbeing, and Schoen et al.'s (2026) confirmation of team conflict profiles, including demonstration of the dual nature and implications of task conflict.

Another potential reason for the variability in findings is the use of small samples. For example, Herzberg and Roth (2006) argued that sample size was a contributing factor to the variability in personality profile structure observed in their review. To address this problem, Espinoza et al. (2020) and Daljeet et al. (2026) demonstrated how parameter estimates obtained from very large (population-like) samples can be applied in small-sample studies to reproduce the normative profiles. This approach could allow for greater cross-sample consistency in profile structure and facilitate the accumulation of findings. However, it requires high confidence in the robustness and generalizability of the “normative” profiles (see Daljeet et al., 2026, for more detail). In this case, normative profiles were identified using data obtained from over 370,000 respondents to an on-line version of the HEXACO personality traits (Ashton and Lee, 2007). Another approach to obtaining such a large data set might be to solicit raw data from studies using other widely used measures (e.g., see Kabins et al., 2016).

Finally, comparison across studies can be made difficult by differences in how authors conduct their analyses and report profiles. Most investigators generate and interpret profiles using either raw or standardized (e.g., factor) scores. Both have advantages and disadvantages. For example, the use of latent factor scores as input provides partial control for measurement error (Meyer and Morin, 2016) and provides a common metric to facilitate interpretation and comparison of profiles within the sample. However, one downside to this approach is that scores reflect deviations from the sample means, and these have the potential to vary widely across studies.

One approach to addressing the sample-specificity problem has been to use normative data to create standard scores. For example, Kam et al. (2016) used normative data from a large-scale meta-analysis to produce standard scores reflecting deviations from the population rather than the sample means. However, even this approach fails to address the possibility that the various indicators can differ meaningfully in strength and/or dispersion. For example, in an investigation of personal aspirations, Bradshaw et al. (2021) noted that profiles reflecting the relative strength of intrinsic and extrinsic aspirations in standard score form did not allow for definitive conclusions regarding the actual strength of these aspirations. That is, despite having a lower standard score within a profile, actual extrinsic aspirations could be higher than intrinsic aspirations on average. To address this issue, they computed weighted estimates for the raw scores within profiles and verified that differences in actual scores corresponded to those for the standard scores. This procedure allows researchers to take advantage of the benefits of using latent factor scores, while also providing raw scores for use in interpretation. The raw scores might also be more meaningful in practice where emphasis is often placed on absolute values.

Thus, although there is no single strategy for ensuring comparability/integration of findings across studies, it is important for researchers be aware of steps they can take to enhance the meaningfulness of their own findings so that, over time, the potential for, and boundaries around, generalizability can become apparent.

Dealing with correlated indicators

As noted above, an important preliminary step in PC studies is to demonstrate that the potential indicator variables are meaningfully distinct. In some cases, although distinguishable, these variables will be correlated, sometimes quite strongly. Indeed, this has been found to be the case for measures of multiple commitments (e.g., Meyer et al., 2021; 2025; Morin et al., 2011), motivation types (e.g., Bureau et al., 2023; Espinoza and Meyer, 2025; Howard et al., 2016), engagement (e.g., Perera et al., 2021; Tóth-Király et al., 2023), need satisfaction (e.g., Gillet et al., 2019b), burnout (Gillet et al., 2023c), and leadership styles (e.g., Tian et al., 2023). Again, this can undermine the ability to detect qualitatively distinct profiles. One approach to addressing this issue has been to use FMA to “control” for the high correlations by partialing out the common variance (e.g., Morin et al., 2011; Tian et al., 2023). Although this approach increases the likelihood of identifying qualitatively distinct profiles, it has the disadvantage of ignoring a large portion of variance (Morin et al., 2017). An alternative has been to conduct a bifactor analysis to extract one or more G factors along with orthogonal specific (S) factors. The S factors presumably reflect the unique variance in the indicators and can be combined with the G factor to identify profiles that are typically qualitatively distinct (e.g., Gillet et al., 2018a; Gillet et al., 2024b; Meyer et al., 2021, 2025; Tóth-Király et al., 2021). Although this approach has the advantage of retaining all relevant variance, it raises questions regarding the meaning of the G and S factors.

The G factor is often interpreted as a reflection of the umbrella construct, be it self-determination (e.g., Tóth-Király et al., 2021), engagement (e.g., Gillet et al., 2020), burnout (e.g., Sandrin et al., 2022), or commitment to work life (e.g., Houle et al., 2024). However, it cannot be interpreted unambiguously without independent confirmation and ruling out alternative interpretations. One strategy employed recently to address the meaning of the G factor underlying the SDT motivation types involved the inclusion of a direct measure of self-determination along with the type measures to anchor the G factor (see Espinoza and Meyer, 2025, for details). A similar approach might be taken for the investigation of other multidimensional constructs, perhaps including different measures for purposes of comparison when there are competing interpretations.

Less attention has been paid to the meaning of the S factors. Although some authors caution readers that S factors must be interpreted differently than the raw measurement scores (e.g., Gillet et al., 2024b), this may be difficult to do. Indeed, it is often not immediately obvious how much unique variance is left in each of the S factors, how much this varies across factors, and what implications this might have for profile structure and interpretation. Interestingly, in their study of need satisfaction profiles, Huyghebaert-Zouaghi et al. (2022) found that there was not sufficient variability remaining in satisfaction of the need for competence to be included as a separate indicator in the LPA. Clearly, therefore, the variance remaining in the S factors varies, but what are the implications for profile interpretation when such indicators are included in the LPA but do not capture sufficient variance? Alternatively, what are the implications of excluding these indicators from the analyses when their constituent items are embedded in the factor scores extracted for the G factor, and potentially other factors (in the case of ESEM)? At a minimum, researchers should report the results of their bifactor analyses to allow readers to determine how much unique variance remains for each indicator. Ideally, more systematic research should be conducted to establish the meaning of the S factors and inform interpretation of profiles.

Finally, it is important to keep in mind that finding evidence for only quantitatively distinct profiles can be useful in some cases. For example, in a 10-year investigation of union commitment, Morin et al. (2021) found that, across time, LPA yielded only quantitatively distinct profiles involving three dimensions: union loyalty, responsibility to the union, and willingness to work for the union. These findings provided further evidence that union commitment, as it is commonly measured, is best considered to be a unidimensional construct. Moreover, the results of an LTA revealed that, over time, there was greater movement from the extreme (low and high) profiles to the more moderate profiles than vice versa, and that individuals within profiles became more similar. While the reasons for these changes were not immediately apparent, the findings raise interesting new directions for research.

It is also possible that the construct under investigation will have non-linear relations with its antecedents or outcomes. Although non-linear relations with outcome variables can be investigated using VC analyses, this is not the case for antecedents. That is, one cannot enter predictors of non-linear relations in a multiple regression analysis. In contrast, one can conduct multinomial logistic regression analyses to predict profile membership with the possibility of discovering non-linearity even with quantitatively distinct profiles (e.g., membership can be predicted only for profiles with low or high scores on the indicators). Such findings could have practical implications such as identifying cut scores for the effectiveness of interventions. Thus, finding only quantitatively distinct profiles can be informative, but such findings must be interpreted appropriately.

Complexity vs. utility

Among the arguments for the value added by PC techniques is that profiles better capture potential complexities (e.g., interactions; non-linear associations) among the indicator variables than is possible with VC techniques (e.g., moderated multiple regression; response surface analyses). As noted previously, in PC research, these complex interrelations are presumably captured in the distinct configurations observed across subpopulations. Importantly, unlike the additive, interactive, or non-linear relations detected in VC research, the profiles obtained in PC research can be included in analyses to investigate their origins (i.e., predict profile membership). The results of such investigation can inform practice by targeting interventions to subpopulations. And this is just the beginning. Other PC analyses can be applied to identify profiles involving trajectories of change, relations among variables, or relations among variables within trajectory profiles, all of which can yield information not available in VC research.

That said, the utility of the PC approach can itself be undermined by the inclusion of too many indicators and/or analyses that identify highly nuanced connections among variables. Moreover, as useful as the disaggregation of common and unique variance can be in identifying qualitatively distinct profiles, this advantage can be lost under conditions where the meaning of the G and S factors are not clearly established as discussed above. Therefore, in the interests of the ultimate utility of PC research for theory testing/development and practice, we call for a more programmatic approach to research. Variables, whether in open, closed, or augmented sets, should be chosen carefully to be in line with the objectives of the research, but not so numerous that profiles and their relations with other variables become impossible to interpret.

Again, as noted above, preliminary analyses regarding the structure of the indicator variables should be conducted to guide whether and how they should be included in the PC analyses. Similarly, the use of innovative and sophisticated analyses to detect profiles of change trajectory, variable relationships, or variable relationships within profiles, will be most useful when it can be demonstrated that the findings have meaningful implications for theory and/or practice, and when these implications can be clearly articulated. Noting that the results of a PC analysis suggest the need for more targeted interventions can be useful in early exploratory studies, but the true value will only be realized with replication and more detail on the nature of the targeting. Ideally, as PC research progresses within the various OB domains, there will be a move away from the predominant exploratory approach to more confirmatory studies (see Schmiege et al., 2018; for more detail and examples).

Sweating the small stuff

Despite common advice against “sweating the small stuff,” we believe there may be good reason to address the implications of some of the mundane decisions that are made, either explicitly or implicitly, in PC studies. It is true that these same decisions have implications for VC research, but that makes them no less important. In fact, it may be at the preliminary stages of conducting a PC investigation—the selection/development of measures and evaluation of their factor structure—that many of these issues will need to be addressed. In some cases, there may already be sufficient VC research available to make informed decisions going into a PC investigation. These might include studies pertaining to the degree to which particular measures are contaminated by social desirability or other response biases. Other studies might inform decisions regarding the treatment of variables as bi-polar vs. multidimensional (e.g., Kam and Meyer, 2015) and/or whether to include both regular and reverse-keyed items (e.g., Kam and Sun, 2022; Weijters et al., 2013). To the extent that decisions such as this affect the correlations among the variables, and perhaps the interpretation of G, they are also likely to affect the number, nature, and interpretation of profiles.

Another decision involves the time lag used in longitudinal studies. Given the difficulty in obtaining longitudinal data, these lags are often determined by logistics rather than being theory-driven. This is a limitation that commonly is, and should be, reported. However, to the extent that researchers have control over the time lag, it is important to tailor it to the constructs of interest. For example, daily diary studies may be well-suited to measures of motivation, task engagement, leader behavior, or stress and coping, but they are less well-suited to identifying profiles for constructs such as burnout, commitment, or work values which, by their nature, evolve over longer periods of time.

Yet, another decision involves the selection of samples. Not surprisingly, many PC studies are conducted with convenience samples from one, several, or many (e.g., through crowdsourcing) organizations. In contrast, research within some domains (e.g., stress and burnout) are particularly prevalent within certain occupational groups (e.g., teachers, healthcare professionals, military personnel, and first responders). This has implications for the selection of variables as well as the interpretation and generalizability of results. Convenience samples are arguably appropriate for exploratory purposes and, when guided by theory and involving comparable measures, the cumulative use of convenience samples can aid in the detection of common patterns and underlying principles. Examples include investigations of organizational commitment and motivation type profiles. Both involve a relatively standard set of measures and findings have been largely consistent. On the other hand, there are situations where the sample and context can play an important role in the selection of measures and the implications of the results. For example, because job demands and resources vary widely across contexts, the findings of PC studies are most likely to be meaningful and practically useful when measures are tailored to the population and context under investigation.

Finally, once profiles are identified, researchers are challenged with the task of labeling them. We were struck by the variability we observed even for studies involving the same constructs. Assigning labels is not an easy task, but it is important because it conveys the essence of the profiles. One approach has been to use descriptive labels reflecting the pattern of indicators. This is arguably a useful approach particularly in early exploratory studies. However, this must be done carefully so as to reflect both the high and low scores (see Meyer and Morin, 2016, for an example involving the TCM commitment profiles). Labels that reflect a more holistic description of the profile [e.g., Parr et al.'s (2016) labels of leader profiles as “Power Players,” “Protocol Followers,” etc.] might ultimately be most useful (and memorable), but these must be assigned carefully with a strong theoretical rationale and/or systematic validation (see above). Failure to do so will only contribute further to the jingle-jangle problem within OB and related fields (i.e., common labels for distinct constructs and different labels for the same construct: Kelley, 1927).

Importantly, rather than contributing to the jingle-jangle problem, PC research has the potential to address it directly by combining relevant constructs within a single investigation and exploring their similarities and differences. To take just one example, multiple measures of leadership style (e.g., transformational, authentic, and ethical) might be included in an LPA. As demonstrated recently by Eva et al. (2025), preliminary analyses are likely to reveal a bifactor structure reflecting what is common (G) and unique (S) among the styles. Subsequent PC analyses might complement this analysis by identifying subpopulations of leaders with unique style profiles, along with potential antecedents and outcomes. Longitudinal analyses might also be conducted to identify similarities and/or differences in the way these styles evolve over time. Such an investigation must of course be conducted systematically with careful attention to the labeling issue raised above.

The issues we addressed here are by no means exhaustive. Moreover, with the exception of some of the issues noted at the beginning of this section, there may not be ready answers to guide decisions. Rather, they may reflect “empirical questions” to be addressed as relevant to a particular line of PC investigation. Indeed, these should be part of the systematic investigation we called for above.

Conclusion

OB researchers are beginning to use PC analytic techniques with increasing frequency across many domains. PC methods provide an important complement to VC research by allowing for the possibility of multiple subpopulations with unique interrelations among variables. Consequently, in addition to addressing and/or introducing more nuance into existing theory, PC methods can provide supporting evidence for different, even competing, theories across subpopulations. PC analytic techniques are also becoming increasingly sophisticated, allowing for tests of a much wider range of questions, including the existence of distinct profiles of change trajectory, relations among variables, and even relations among variables within trajectory profiles. Indeed, the potential for further applications of PC methods within the field of OB is extremely promising. However, to have meaningful and lasting impact for theory and practice, they must be applied judiciously. There is always the risk that “when one has a hammer, everything looks like a nail.” Hence our call for a systematic approach to the use of PC methods, beginning with preliminary investigation to determine whether the variables of interest are well-suited to a PC analysis, and whether there are reasonable grounds, theoretical or otherwise, to expect differential relations across multiple subpopulations. Finally, in our view, the ultimate value of PC research in OB will be its implications for practice, either directly or through the confirmation of applicable theoretical principles. Thus, we call for researchers to move beyond acknowledging the need for more “targeted interventions” by elaborating, where possible, on the targeting process (i.e., how to identify subpopulations), the nature of the distinct interventions, and the impact that can be expected from targeting relative to a more homogeneous approach. Admittedly, no single study is likely to be sufficient for these objectives, which is again why we call for more systematic research that builds toward a solid basis for evidence-based practice.

Generative AI statement

The author(s) declared that generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher's note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Footnotes

Edited by: Alexandre J. S. Morin, Concordia University, Canada

Reviewed by: Joeri Hofmans, Vrije University Brussels, Belgium

Simon Houle, Concordia University, Canada