Good,Bad,and Ugly Leadership Patterns: Implications for Followers’ Work-Related and Context-Free Outcomes

Patterns Where Destructive Styles are Dominant

Based on prior theoretical and empirical research on negative leadership styles, we hypothesize the natural clustering of mistake-oriented transactional styles (MBE-A), passive styles (MBE-P and LF/A), and AS. As alluded to earlier, of particular interest is whether followers associate AS with these passive forms of leadership.

We propose that followers will experience MBE-A, an active behavioral style that portrays a leader who intentionally seeks out and focuses on correcting follower mistakes, and AS together. The notion that followers might see MBE-A and AS occurring together echoes the arguments of Kark and Van Dijk (2007), who suggest that MBE-A behaviors likely result from a desire for stability and a need to control the workplace. Some abusive behaviors, such as invading subordinates’ privacy and reminding followers of their past mistakes and failures, are similarly restrictive in focus and may be aimed at controlling follower behavior. MBE-A behaviors, such as focusing on followers’ past mistakes and failures, could also fall within the range of abusive behaviors described by Tepper (2000). The implied threats that could often accompany criticism may be interpreted by followers as both MBE-A and abuse (Landay, Harms, & Credé, 2019), suggesting compatibility between these styles.

That said, a case can also be made that AS will be perceived to co-occur with passive, disengaged, and avoidant forms of leadership (i.e., MBE-P and LF/A). Mitchell and Ambrose (2007: 1162) reported that behaviors such as my boss “doesn't give me credit for jobs requiring a lot of effort” could be experienced as a combination of neglect and deliberate abuse. Both AS and LF/A are styles, for instance, that encompass acts of ignoring and, in turn, not fully engaging with followers (Tepper, 2000). Abusive supervisors are known to not give followers credit for hard work, which again is passive in nature (Tepper, 2000). However, passive leadership need not be interpreted as abuse if followers attribute the neglectful/avoidant behavior to time constraints and/or the number of direct reports (Bass & Riggio, 2006). Thus, we propose:

Patterns Where Constructive Styles are Dominant

In contrast to patterns that feature destructive leadership styles, much more is known about the co-occurrence of TFL and CR. TFL theory informs us that leaders are most effective when they enact TFL behaviors along with CR and low levels of the passive styles (Bass, 1999). As expected, studies that employ a pattern-oriented analytic strategy consistently find in samples of followers (ratings of leadership styles: Doucet et al., 2015) and leaders (self-reported leadership styles: Arnold et al., 2017) a pattern where the scores on both TFL and CR are high relative to other styles in the pattern. This combination, often referred to as an optimal configuration, leverages the advantages of both transformational and transactional leadership (e.g., Bass, 1999; Bass & Riggio, 2006). Here followers see their leader emphasizing behaviors that inspire and motivate followers, build commitment toward the goals, and facilitate the execution of strategy (e.g., performance management)—behaviors that, collectively, are inconsistent with acts of passive leadership or AS (e.g., putting them down in front of others). Given the ubiquitous nature of this pattern, we expect to find a subgroup of followers who will perceive their leaders as displaying both transformational and transactional behaviors with greater frequency, relatively speaking, than other styles in the pattern.

Predicting Cohort Membership

We considered the extent to which followers’ demographic (gender, age, tenure) and personal (negative affectivity) characteristics predicted the likelihood of being classified into one of the pattern groups. Research shows that people who exhibit high levels of negative affectivity are predisposed to hold negative views of themselves and others and generally perceive the world as a hostile and threatening place (Thoresen et al., 2003). According to social exchange theory and the norm of reciprocity (Blau, 1964), people holding these negative views of others will be motivated to act accordingly. Therefore, relative to pattern(s) where positive styles are dominant, the probability of being classified into one of the negative leadership patterns—especially passive-abusive—should be greater for followers who report higher rather than lower levels of negative affectivity.

Moreover, it has been shown that people who feel particularly vulnerable are more likely to be a target of, or at the very least more sensitive to, negative leader behavior (Mackey et al., 2017; Tepper, 2007; Wang et al., 2019). Thus, relative to patterns where positive styles are dominant, we expect that the likelihood of classification into one of the negative leadership patterns should be greater for females as opposed to males, young as opposed to old, and newly hired as opposed to longer term employees (see Tepper, 2007). Wang et al. (2019) have shown that female followers tend to perceive their leaders in more extreme terms when rating them than do males (e.g., females are more likely to rate transformational leaders in more favorable terms and abusive leaders in more negative terms than do males). Similarly, we propose that younger and newly hired workers are more likely to be unsure of themselves and their status within the organization and as such more sensitive to destructive behavioral displays that increase their anxiety and negative self-perceptions. In contrast, older and more experienced workers should have access to extensive social supports and coping skills, which, in turn, makes them less dependent on their leaders for positive affirmations (Zhang & Bednall, 2016). In short, these demographic characteristics can make followers more vulnerable and thus more, rather than less, likely to describe their leader's behavior in negative terms than will those who are more confident in their abilities and status within the organization.

Outcomes Associated With Destructive and Constructive Patterns

COR theory (Hobfoll, 1989) recognizes that people strive to build and maintain a personal reservoir of resources that they value. This theory defines resources broadly—from objects (e.g., cars), conditions (e.g., supportive work relationships), personal skills and traits (e.g., self-efficacy), to energy-based qualities (e.g., knowledge) (Hobfoll, 1989). For some time, management scholars have recognized that leaders serve as poignant sources of resource gain or drain by contributing to or depleting their followers’ resource reservoirs and related outcomes (e.g., Bono & Judge, 2004; Halbesleben et al., 2014; Hobfoll et al., 2018). As such, a COR perspective offers a theoretical basis for hypotheses about follower reactions to leadership patterns. Unfortunately, shortcomings in the literature have been found to limit our understanding of how leaders impact followers, especially with respect to feelings of well-being (Hancock, Daher Moreno, & Arnold, 2021). First, as alluded to earlier, there appears to be an absence of studies that examine different combinations of negative and positive leadership styles on follower well-being outcomes despite reports of these individual styles co-occuring (for exceptions, see Breevaart & Zacher, 2019; Fiset et al., 2019). Second, management research has tended to conflate one's experience of well-being with a myriad of specific work-related concepts (e.g., job satisfaction, commitment) while overlooking direct measures of physical and psychological health (Inceoglu et al., 2017). To address these shortcomings and to emphasize well-being as an outcome that is distinct from employee attitudes and job perceptions, we assess physical and psychological well-being directly. Thus, it is of theoretical relevance to assess outcomes that are specific to and embedded within the work context as well as those that are more generally inclusive of the many facets of life experience (i.e., context free).

As alluded to earlier, in this article we examine the implications of membership within a particular leadership-pattern on a set of context-free (physical health and psychological well-being) and work-related criteria. Work-related outcomes with particular relevance for a COR perspective include (a) work-related burnout, defined as followers’ feelings of emotional exhaustion, diminished personal accomplishment, and depersonalization arising from role demands (Maslach et al., 2012); (b) work-related vigor, defined as followers’ feelings that they possess the necessary physical strength, emotional energy, and cognitive liveliness to fulfill role demands (Shirom, 2003); (c) affective commitment, defined as the extent to which followers’ personally identify with and experience an emotional bond with their organization (Meyer, 2016); and (d) perceived organizational support, defined as followers’ belief that the organization and its agents value their contributions and care about their well-being, in turn evoking within these individuals an obligation to reciprocate this good will (Eisenberger et al., 1986).

To discern the potential implications of membership in patterns where negative styles are dominant, we offer the following predictions for context-free and work-related criteria. The basis for our predictions draws upon what is known about the individual styles and the potential implications of these styles experienced together as a whole. We know, for instance, that passive behaviors such as MBE-P and LF/A have consistently been associated with higher absenteeism and presenteeism (Frooman, Mendelson, & Murphy, 2012) and have negatively predicted both organizational safety climate and safety consciousness (Kelloway et al., 2006). LF/A has been consistently shown to be associated with the lowest levels of group and organizational performance ratings and the highest levels of de-motivation (Judge & Piccolo, 2004). Other studies have determined additional unfavorable outcomes of LF/A, such as increased coworker conflict, role conflict and ambiguity, and decreased job satisfaction (Skogstad et al., 2014). Likewise, AS has been associated with a range of outcomes (e.g., Mackey et al., 2017), including reduced psychological and physical wellness (e.g., Liang et al., 2018; Tepper et al., 2017), increased burnout (Mackey et al., 2017), lower affective commitment (Meyer, 2016), and lower perceived organizational support (Kurtessis et al., 2015). Moreover, the link between the passive leadership styles (e.g., LF/A) and employee outcomes, such as higher sick leave and injury rates (Frooman et al., 2012; Skogstad et al., 2014) and reduced organizational performance/motivation (Judge & Piccolo, 2004), suggests that elevated passiveness will be a source of resource drain (loss). It is noteworthy that Doucet et al. (2015) found that cohorts of followers who experienced different forms of passiveness experienced worse outcomes than did cohorts characterized by high levels of TFL (superleaders) and high levels of CR (transactors) relative to the other styles in the pattern. Similarly, Arnold et al. (2017), using a sample of leaders’ self-reports, found that individuals who reported a pattern with elevated levels of MBE and LF/A relative to the levels of TFL and CR exhibited signs of resource drain as expressed through elevated burnout.

In contrast, we expect a different set of associations with patterns where positive styles dominate. It is well known that transformational leaders play an integral role in their followers’ social support systems and can be a significant source of resource gain for those reporting to them (e.g., Bakker & Demerouti, 2007). When we consider an integration of the TFL and CR styles as a perceived behavioral configuration, we surmise that this pattern should be associated with lower burnout (Harms et al., 2017) and higher employee well-being (e.g., Barling et al., 2011), work-related vigor (Gözükara & Şimşek, 2015), and more affirmative employee-organization relationships such as affective commitment (e.g., Bono & Judge, 2004; Meyer, 2016) and perceived organizational support (Kurtessis et al., 2015). Whether assessed as leader self-reports or follower perceptions, preliminary evidence suggests that when the TFL and CR styles are prominent relative to other leadership styles within a pattern, members of these cohorts will experience lower levels of burnout and role demands (Arnold et al., 2017) and higher employee job satisfaction and commitment (Doucet et al., 2015). As such, we predict a positive leadership-style pattern will be associated with the most favorable personal and work-related relationships and outcomes for followers.

To the extent that negative leader behaviors drain followers’ resources (e.g., Hobfoll, 1989; Hobfoll et al., 2018), COR theory would suggest that resource loss will be even more pronounced for individuals who perceive a negative pattern that involves overt abuse. Acts of mistreatment should not only be more salient to followers but should a have a greater detrimental effect than simple neglect. Neglect can have many interpretations (e.g., my leader is too busy) and may not always be experienced as a resource-draining event, but passive leaders who also exhibit acts of abuse may be more difficult to ignore, which would trigger resource loss (Barling & Frone, 2017). Thus, followers who perceive their leader as more abusive than passive should experience additional resource loss. This prediction draws on previous work that observed differential outcomes for followers whose leaders were unethical rather than merely passive (Barling et al., 2008).

Because pattern-oriented research has not yet considered how AS fits with these leadership style combinations, our expectations for resource gains or losses rely on the extant literature. We know, for instance, that followers who report having an abusive supervisor are more likely to have negative attitudes towards their leader and resist their direction to a greater extent than followers who do not describe their leader as abusive (Schyns & Schilling, 2013; Tepper et al., 2017). From the perspective of organizational support theory, employees who encounter abuse are more likely to feel undervalued, threatened, and, thus, more likely to engage in negative social exchange (e.g., Kurtessis et al., 2015). Employees with abusive supervisors are also more likely to display negative well-being on context-specific indicators such as work-related burnout, vigor, and workplace negative affect (Palmer et al., 2017). In addition, AS is likely to result in lower levels of organizational commitment, job satisfaction, citizenship behavior, and performance in employees compared to those who have a less hostile supervisor (Kurtessis et al., 2015; Mackey et al., 2017). With regards to context-free outcomes, the literature suggests a substantive relationship between AS and employee physical illness and psychological harm (Liang et al., 2018). Furthermore, since AS involves actively mistreating followers, patterns in which abuse is salient should represent an additional source of resource loss. Thus, combinations involving high levels of abuse should adversely affect employee outcomes more than would be the case if AS were not as prominent in the pattern.

Assessment of Leadership Styles

Six leadership styles were assessed in all three studies. Our measure of AS consisted of five items drawn from Tepper’s (2000) scale that best represented visible (active and salient) acts of abuse (cf. Mitchell & Ambrose, 2007). Followers reported how often they observed their supervisor engaging in behaviors, such as “My supervisor ridicules me” and “My supervisor tells me my thoughts or feelings are stupid.” Ratings were provided on a 5-point frequency scale (1 = I cannot remember him/her ever using this behavior with me; 5 = He/she uses this behavior very often with me). When respondents were leaders (one of the samples in Study 1), the five behaviors were rated on a 5-point scale (1 = I don't ever use the behavior with them; 5 = I use this behavior very often with them). Our measures of TFL, CR, MBE-A, MBE-P, and LF/A were drawn from the Multifactor Leadership Questionnaire (MLQ: Bass & Avolio, 2014) with permission from the copyright holder. Respondents in our follower samples were asked to rate how often they observed their immediate supervisor exhibiting a range of behaviors in their day-to-day interactions. Ratings for each of the specific behaviors were provided on a 5-point scale (e.g., 1 = not at all; 5 = frequently, if not always). Respondents in our leader sample (Study 1) used the same scale for MLQ items but were asked to rate how often they exhibited a range of behaviors in their day-to-day interactions with followers. Responses for items belonging together were averaged into a single score for AS, TFL, CR, MBE-A, MBE-P, and LF/A.

Data Sources: Planning, Implementation, and Reporting

We tested our study hypotheses using online panel data (OPD) acquired from the most frequently used online panel platform: Amazon's Mechanical Turk (MTurk) (Porter et al., 2018). Using OPD allowed us to sample across social and organizational contexts, capturing follower perceptions of different leaders working in a myriad of situations. Sampling broadly also allowed us to overcome any natural restriction of leader behavior that might have been evident if followers shared a common organizational context (i.e., if all participants were members of the same company). Furthermore, given that followers were asked to rate the frequency of positive and negative leader behaviors, the anonymity afforded by OPD encourages more honest responses and, in turn, allows us to optimize the measurement of leader behaviors with historically low base rates such as AS (Tepper, 2000). A recent meta-analysis concluded that OPD exhibited psychometric properties and criterion validities that approached those reported in research using conventional data (Walter et al., 2019). However, to address concerns about the trustworthiness of online convenience samples (e.g., Zack, Kennedy, & Long, 2019), we observed best practice recommendations for OPD (Porter et al., 2018) and MTurk data collection recently articulated by Aguinis et al. (2021) in the planning, implementation, and reporting stages.

In terms of planning, our goal for the follower samples was to recruit participants who were employed full-time (i.e., 35+ hours per week) and who held an approval rating on previous MTurk Human Intelligence Tasks (HITs) of 95% or higher. In the leader sample, we recruited participants who held a similar approval rating and occupied a full-time supervisory position with direct reports. The recruitment text communicated a detailed description of the study requirements including estimated time commitment, description of task, and compensation rules. The HITs were intentionally given a vague title to avoid cues that might provide signals about the study's aims or purpose, which could encourage self-misrepresentation or social desirability bias. Compensation was set to match U.S. hourly rates of minimum wage (since we were drawing from a U.S. sample) based on estimated survey completion time. Pursuant to institutional ethical guidelines, participants were informed that they would receive full payment for partially completed responses, however careless or inattentive responses would prevent them from being eligible to return to future studies and likely result in the exclusion of their data from the current analysis. To verify the participant had completed the study, the online survey hosting software Qualtrics assigned an auto-generated five-digit code when participants reached the end of the survey. Workers’ unique MTurk identification numbers were collected and matched with their completion codes to identify previous cases of careless responding or misrepresentation. With respect to planning the survey itself, reverse-item coding and a variety of question formats (e.g., matrix tables with Likert scales, dropdown menus, slider scales) were used (Podsakoff et al., 2012). In addition, rather than assuming similarity with earlier MTurk samples, we have collected and reported detailed sample characteristics including age, sex, organizational tenure, and industry per best practice recommendations (Aguinis et al., 2021). Finally, we anticipated having to overcollect participants by 25% to compensate for participant attrition in the time-lagged survey and failure to pass attention checks.

With respect to implementation, a pilot HIT was launched to test the survey instrument prior to the main launch. During data collection, the first author monitored queries coming from participants as data collection was ongoing to gauge MTurkers’ reactions to the study, resolve issues, or respond promptly to any questions or concerns raised by participants in a timely manner (Aguinis et al., 2021; Porter et al., 2018). Data were screened using multiple quality control checks such as attention checks (e.g., please select “strongly agree” for this question). All scale points were labeled and thus we did not engage in the practice of only labeling the “end” points (Aguinis et al., 2021). Furthermore, we checked for rushed responding by deleting responses that were submitted in less than 40% of the median completion time (McGonagle et al., 2016). Consistent with best practices, we used the “prevent ballot box stuffing” feature in survey hosting software Qualtrics® to prevent surveys from being completed more than once by the same IP address. In addition, a database of workers was maintained and assigned qualifications based on response quality (Aguinis et al., 2021). We also addressed the concern of high attrition rates and perceived researcher unfairness by paying U.S. minimum wage relative to the average completion time and followed ethics protocol of providing payments within 24 hours to all participants who completed the survey.

Finally, with respect to reporting, the current manuscript provides information regarding all procedures followed, decisions made, and results obtained during each stage of the study (Aguinis et al., 2021). As recommended, to facilitate secondary analyses we have provided demographic information as well as descriptive statistics for the measures.

Analytic Strategy

Given the tendency for measures of leadership styles to correlate with one another (e.g., Hoch et al., 2018), we felt it prudent to examine the discriminant validity of the six leadership measures prior to conducting LPA (Morin et al., 2016). When possible, the fits of several plausible measurement models were estimated using the Mplus (Muthén & Muthén, 1998-2021) robust weighted least square estimator (WLSMV). WLSMV estimation has been found to be well-suited to the ordered-categorical nature of the Likert scales used to assess leadership styles in the present study and is more robust to nonnormality assumptions than traditional maximum likelihood estimation (cf. Morin et al., 2017).

The aim of mixture modeling techniques, like LPA, is to test the extent to which one or more classes or subgroups coexist within a population based on their distinct pattern or profile of scores on a set of input variables (e.g., various leadership styles). We followed typical procedures (see Morin et al., 2020; Morin et al., 2011; Pastor et al., 2007) to first assess fit of a two-pattern model and in subsequent models specified an additional cohort until it was clear that increasing the number of patterns could not be justified (i.e., because of statistical errors [e.g., nonpositive definite solutions]). When determining the optimal solution for each study, we were guided by the following. An optimal LPA solution should have higher log likelihood (LL) values and lower Akaike information criterion (AIC), Bayesian information criterion (BIC), and sample-size-adjusted BIC (SSA–BIC) values in comparison to other pattern solutions, and entropy values should be larger in comparison to other solutions. Additionally, p values <.05 associated with the Lo-Mendell-Rubin likelihood ratio test (LMR; Lo, Mendell, & Rubin, 2001) and the bootstrap likelihood ratio test (BLRT: Nylund et al., 2007), which assesses fit of a k-pattern model over a k-1 model, can be used to support an optimal model (Lo et al., 2001; Nylund et al., 2007). Finally, researchers should ensure that all classes or groups are sufficiently populated (i.e., do not contain less than 5% of the sample), and the optimal solution should be parsimonious and, perhaps most importantly, interpretable from a theoretical perspective (Morin et al., 2011). Thus, a solution that balanced empirical fit, parsimony, and theoretical consistency was emphasized.

For Studies 2 and 3, once an optimal LPA solution had been determined, relations between the patterns and various auxiliary variables (i.e., outcomes not used to determine profile membership; Asparouhov & Muthén, 2014) were considered, taking into account the most likely class membership and classification error rate (Wang & Hanges, 2011). To model predictors, we used the R3STEP command in MPlus (Asparouhov & Muthén, 2014), which conducts a series of multinomial logistic regressions that assess whether an increase in a predictor variable would result in a higher probability that a respondent belongs to one pattern group over another class. To model the outcome relations, we utilized the DU3STEP command (Asparouhov & Muthén, 2014), which provides a test of mean equality of each outcome across the pattern groups.

Sample Characteristics, Measures, and Descriptive Statistics

The data for Study 1 involved independent samples of followers and leaders. The followers sample consisted initially of 215 adults who worked full-time and reported to a manager or supervisor. Fourteen followers were removed from the sample, 10 because they failed attention checks (including invalid completion codes) and 4 due to speedy responding, thus reducing our sample to 201. In terms of follower characteristics, over half of the respondents (54%) identified as male, and the average age was 34 years (M = 34.29, SD = 8.89). On average, respondents had been a member of their organization for approximately 6 years (M = 6.07, SD = 5.06). A variety of industrial contexts were represented in the sample, with most followers employed in telecommunications, technology, internet and electronics industry, health care and pharmaceuticals industries, and education. Measures of follower-rated leadership styles were as follows: AS (M = 1.33, SD = 0.59, α = 0.90), TFL (M = 3.39, SD = 0.88, α = 0.95), CR (M = 3.55, SD = 0.90, α = 0.85), MBE-A (M = 2.83, SD = 0.85, α = 0.73), MBE-P (M = 2.51, SD = 0.97, α = 0.79), and LF/A (M = 2.26, SD = 1.06, α = 0.87).

The leader sample consisted initially of 536 full-time employees who managed or supervised the activities of other people in a work setting. Thirty-one participants were removed from the sample based on the data-quality criteria described earlier, thus reducing our sample to 505. In terms of leader characteristics, 60% identified as male and the average age was 37 years (M = 37.30, SD = 9.89). On average, the leaders had been in their current supervisory position for an average of 6 years (M = 5.97, SD = 4.90) and had been with their organization approximately 8 years (M = 8.10, SD = 6.00) years. A variety of industrial contexts were represented in the leader sample, with most respondents employed in information technology, education, sales/retail, health care, and manufacturing. Measures of leader-rated leadership styles were as follows: AS (M = 1.36, SD = 0.63, α = 0.88), TFL (M = 3.84, SD = 0.59, α = 0.89), CR (M = 3.87, SD = 0.64, α = 0.68), MBE-A (M = 3.13, SD = 0.80, α = 0.68), MBE-P (M = 2.20, SD = 0.78, α = 0.70), and LF/A (M = 1.75, SD = 0.75, α = 0.74). For each data source, correlations among the demographic and leadership-style measures are displayed (Table 1).

OPEN IN VIEWER

Table 1

Study 1: Correlations Among the Study Measures (Follower and Leader Samples).

Measures	1	2	3	4	5	6	7	8	9
1. Respondent Sex (1 = male; 2 = female)		0.07	0.06	−0.01	−0.06	−0.16	−0.04	0.03	−0.09
2. Respondent Age	0.05		0.50	0.11	0.14	−0.12	−0.15	−0.23	−0.14
3. Organizational Tenure (Years)	0.00	0.61		0.04	0.05	−0.09	−0.02	−0.10	−0.03
4. Transformational Leadership	−0.07	−0.17	−0.05		0.77	0.27	−0.31	−0.42	−0.25
5. Contingent Reward	−0.04	−0.07	0.03	0.90		0.28	−0.24	−0.39	−0.14
6. Management By Exceptions–Active	−0.16	−0.21	−0.07	0.25	0.20		0.19	0.10	0.18
7. Management By Exceptions–Passive	0.11	−0.06	0.04	−0.46	−0.49	0.17		0.62	0.43
8. Laissez Faire/Avoidant	0.10	−0.08	0.01	−0.49	−0.50	0.15	0.81		0.50
9. Abusive Supervision	−0.03	−0.20	−0.07	−0.35	−0.35	0.25	0.37	0.42

Note: Correlations among follower-rated measures below the diagonal (N = 201). Correlations among leader-rated measures above the diagonal (N = 505). Bolded correlations are statistically significant, p < 0.05 (two-tailed test).

Testing the Measurement Model: Confirmatory Factor Analyses

The validity of our proposed measurement model was assessed by testing the fit of three competing models. We began with the sample of followers. Model 1 was specified as a one-factor model whereby scale items (indicator variables) were specified to load on a common latent factor. In light of the correlations reported in Table 1, we tested the viability of an alternative measurement model that involved four rather than six correlated factors. As such, Model 2 was specified as follows: the indicators of TFL and CR were combined to load on one factor; the indicators of MBE-P and LF/A were combined to load on one factor; the indicators of AS and MBE-A, respectively, were retained as separate factors. Model 3 was specified so that the measures were retained as six distinct yet correlated factors. Using the recommended WLSMV estimator within Mplus, we tested the fit of the three measurement models: Model 1 (χ² = 1,612.55, df = 275; RMSEA = 0.16; SRMR = 0.16), Model 2 (χ² = 592.06, df = 269; RMSEA = 0.08; SRMR = 0.08), and Model 3 (χ² = 575.03, df = 260; RMSEA = 0.08; SRMR = 0.08). Of the three models, the fit of Models 2 and 3 was found to be superior to Model 1. We then used the DIFFTEST command within Mplus to perform a chi-square test to assess the difference in fit between Models 2 and 3, χ² = 25.22, df = 9, p < 0.005; thus, six latent factors (Model 3) provided a better explanation of the observed variance-covariance matrix than did four latent factors.

Next, we replicated this series of CFAs with the sample of leaders. The following fit indices were found for the three measurement models: Model 1 (χ² = 3,082.55, df = 275; RMSEA = 0.14; SRMR = 0.14), Model 2 (χ² = 924.73, df = 269; RMSEA = 0.07; SRMR = 0.07), and Model 3 (χ² = 874.00, df = 260; RMSEA = 0.07; SRMR = 0.06). Once again, the application of the DIFFTEST command within Mplus revealed that the fit of Model 3 was superior to that of Model 2 (χ² = 56.48, df = 9, p < 0.001). Thus, regardless of whether leadership behaviors are viewed through the eyes of followers or leaders, a six-factor measurement model provided a superior fit than did a four-factor measurement model.

LPA

Next, a series of LPAs were estimated, respectively, on the follower and the leader data. The input variables for all analyses were measures of the six leadership styles. To determine the number and nature of cohorts in each of the two populations, we considered the fit indices associated with each of the proposed models (Table 2). For both samples, as the number of patterns increased, we observed increasingly higher LL values and increasingly lower values of AIC, BIC, and SSA-BIC. To help determine the point at which further extraction produced a nontrivial improvement of fit, we graphed these three fit values as an elbow plot (Morin & Marsh, 2015). The “elbow” in both samples occurs clearly at the three-pattern solution. For both samples, the three-pattern model represented the best choice given that it exhibited lower AIC, BIC, and sample-size adjusted BIC (SSA–BIC) values than did the two-pattern model. Although the four-pattern model had lower AIC, BIC, and SSA-BIC and a higher entropy value than did the three-pattern cohort, the more parsimonious solution represented three theoretically interpretable, qualitatively distinct patterns that differed in shape rather than level (Morin & Marsh, 2015). We also observed that the three-pattern model demonstrated a high probability that respondents were correctly classified (i.e., classification errors were minimal, with the average latent class probabilities for most likely pattern membership ranging from 0.96 to 0.97 for followers and 0.87 to 0.98 for leaders). With the exception of the four-pattern group, all of the models tested exhibited significant likelihood ratios as determined by the LMR and BLRT values. Thus, when we consider the decision criteria needed to determine the optional solution, for both followers and leaders the three-pattern model seems most interpretable relative to the other models tested.

OPEN IN VIEWER

Table 2

Studies 1-3: Model Fit Statistics for Pattern Cohorts.

Models	LL	FP	AIC	BIC	SSA-BIC	LMR (p)	BLRT (p)	Entropy
Study 1 (Followers Sample; N = 201)
2-Pattern	−1,358.96	19	2,755.93	2,818.69	2,758.49	0.00	0.00	0.89
3-Pattern	−1,273.64	26	2,599.27	2,685.16	2,602.78	0.03	0.00	0.92
4-Pattern	−1,211.86	33	2,489.71	2,598.72	2,494.17	0.08	0.00	0.95
Study 1 (Leaders Sample; N = 505)
2-Pattern	−2,882.01	19	5,802.01	5,882.28	5,821.97	0.00	0.00	0.98
3-Pattern	−2,703.97	26	5,459.95	5,569.78	5,487.26	0.00	0.00	0.84
4-Pattern	−2,630.82	33	5,327.64	5,467.05	5,362.30	0.30	0.00	0.86
Study 2 (N = 408)
2-Pattern	−2,829.44	19	5,696.88	5,773.10	5,712.81	0.00	0.00	0.88
3-Pattern	−2,645.57	26	5,343.13	5,447.43	5,364.92	0.11	0.00	0.93
4-Pattern	−2,540.94	33	5,147.89	5,280.26	5,175.54	0.13	0.00	0.89
Study 3 (N = 246)
2-Pattern	−1,719.30	19	3,476.60	3,543.20	3,482.97	0.00	0.00	0.89
3-Pattern	−1,641.32	26	3,334.63	3,425.77	3,343.35	0.07	0.00	0.85
4-Pattern	−1,567.49	33	3,200.99	3,316.66	3,212.06	0.10	0.00	0.91

Note. LL = log-likelihood; FP = free parameters; AIC = Akaike information criteria; BIC = Bayesian information criteria; SSA-BIC = sample-size-adjusted BIC; LMR = Lo, Mendell, and Rubin (2001) test; BLRT = bootstrapped log-likelihood ratio tests.

The nature of the patterns that characterize the three follower cohorts are evident in Table 3 (Followers Sample; N = 201) and Figure 1a. Table 3 portrays the relative status of the six behavioral-style components within each pattern. To keep with the traditions in the literature, we labeled Pattern 1 (N = 121 or 60.20% of the sample) as optimal in that followers perceived their leaders using a combination of both transformational and transactional styles more often than the other behaviors under consideration (Bass & Riggio, 2006). Finding evidence that followers recognize a pattern dominated by positive styles provides support for Hypothesis 1 and is consistent with work that has used a similar analytic strategy with leaders themselves (e.g., Arnold et al., 2017). Pattern 2 (N = 49 or 24.38% of the sample) was labeled passive as this group of followers perceived their leaders’ pronounced tendency to engage in passive behavior characterized by MBE-A, MBE-P, and LF/A but not TFL, CR, or AS (see Figure 1a), supporting Hypothesis 2. Finally, we found evidence supporting Hypothesis 3. Pattern 3 (N = 31 or 15.42% of the sample) was labeled passive-abusive because followers in this cohort viewed their leaders as high on the passive styles and on AS relative to their perceptions of TFL or CR behaviors.

OPEN IN VIEWER

Figure 1

Three-Pattern solution across four samples. 1a. Study 1, Followers, N = 201, 1b. Study 1, Leaders, N = 505, 1c. Study 2, Followers, N = 408, 1d. Study 3, Followers, N = 246.

OPEN IN VIEWER

Table 3.

Studies 1-3: Mean Differences Across the Three Pattern Cohorts.

Leadership Patterns	% of Sample	TFL		CR		MBEA		MBEP		LF/A		AS
		M	95% CI	M	95% CI	M	95% CI	M	95% CI	M	95% CI	M	95% CI
Study 1 (Followers Sample; N = 201)
Optimal	60.20%	3.79	[3.66, 3.92]	3.96	[3.82, 4.09]	2.71	[2.55, 2.88]	1.92	[1.77, 2.06]	1.56	[1.44, 1.68]	1.06	[1.02, 1.09]
Passive	24.38%	2.88	[2.55, 3.21]	3.00	[2.69, 3.32]	2.85	[2.55, 3.14]	3.49	[3.23, 3.75]	3.36	[2.99, 3.73]	1.21	[1.11, 1.32]
Passive-Abusive	15.42%	2.71	[2.37, 3.05]	2.87	[2.52, 3.22]	3.23	[2.99, 3.47]	3.21	[2.93, 3.49]	3.11	[2.75, 3.47]	2.56	[2.32, 2.79]
Study 1 (Leaders Sample; N = 505)
Optimal	58.22%	4.16	[4.07, 4.24]	4.18	[4.09, 4.27]	3.17	[3.06, 3.29]	1.84	[1.72, 1.95]	1.36	[1.28, 1.44]	1.14	[1.11, 1.17]
Passive	32.67%	3.36	[3.22, 3.50]	3.36	[3.18, 3.53]	2.94	[2.77, 3.10]	2.59	[2.41, 2.77]	2.16	[1.96, 2.36]	1.30	[1.21, 1.38]
Passive-Abusive	9.11%	3.50	[3.33, 3.68]	3.63	[3.42, 3.84]	3.53	[3.34, 3.72]	3.14	[2.94, 3.35]	2.85	[2.59, 3.10]	3.03	[2.76, 3.29]
Study 2 (N = 408)
Optimal	69.12%	3.85	[3.74, 3.96]	4.03	[3.91, 4.15]	2.79	[2.68, 2.90]	2.10	[1.97, 2.22]	1.68	[1.57, 1.79]	1.06	[1.04, 1.08]
Passive	24.76%	2.27	[2.04, 2.50]	2.28	[2.04, 2.52]	2.81	[2.62, 2.99]	3.43	[3.13, 3.74]	3.08	[2.75, 3.41]	1.27	[1.19, 1.36]
Passive-Abusive	6.13%	3.11	[2.77, 3.45]	3.11	[2.76, 3.46]	3.04	[2.73, 3.34]	2.86	[2.51, 3.22]	2.87	[2.54, 3.21]	3.20	[2.80, 3.59]
Study 3 (N = 246)
Optimal	45.12%	4.01	[3.82, 4.20]	4.07	[3.87, 4.27]	3.02	[2.84, 3.21]	1.92	[1.74, 2.11]	1.37	[1.24, 1.50]	1.07	[1.02, 1.12]
Passive	36.18%	2.76	[2.49, 3.04]	2.79	[2.52, 3.06]	2.87	[2.66, 3.07]	2.63	[2.37, 2.89]	2.22	[1.94, 2.50]	1.28	[1.18, 1.37]
Passive-Abusive	18.70%	2.65	[2.37, 2.93]	2.67	[2.37, 2.97]	3.21	[3.01, 3.40]	3.47	[3.25, 3.70]	3.35	[3.05, 3.65]	2.86	[2.60, 3.11]

Note: CI = Confidence Interval. Leadership Styles: TFL = Transformational Leadership; CR = Contingent Reward; MBEA = Management by Exception Active; MBEP = Management by Exception Passive; LF/A = Lassize Faire/Avoidant; AS = Abusive Supervision.

To offer context for follower perceptions, we examined the patterns rendered from a sample of leaders (see Table 3 and Figure 1b). Here we observed an optimal pattern (N = 294; 58.22% of leaders) where the perceived frequency of TFL and CR behaviors were higher than the other styles under consideration. We also observed a passive cohort (N = 165; 32.67% of leaders) whereby the pattern involved elevated scores on MBE-P and LF/A relative to the other styles. Finally, we observed a passive-abusive cohort (N = 46; 9.11% of leaders) whereby elevated scores were observed on MBE-A, MBE-P, LF/A, and AS relative to TFL and CR.

To summarize the findings of Study 1, we found three clusters of followers held distinct views of their leaders—patterns that were also confirmed within an independent sample of leaders. A constructive pattern emerged (which we labeled as optimal), demonstrated by elevated scores on TFL and CR (relative to the strength of the other styles), replicating the findings of other pattern-oriented studies with follower samples (e.g., Doucet et al., 2015) and leader samples (Arnold et al., 2017). Two destructive patterns emerged: one where passive behaviors were elevated relative to the other styles and one where passive behaviors co-occurred with abuse in a context where TFL and CR were low. To test whether these patterns replicate in independent samples of followers and to incorporate outcomes into our analyses, we turn to our next two studies.

Sample Characteristics, Measures, and Descriptive Statistics

We repeated our data collection procedure to generate a second sample of followers who provided ratings on both leader behaviors and personal outcomes. The data for Study 2 consisted of 437 adults who met the eligibility criteria, worked full-time, and reported to a manager or supervisor. Twenty-nine respondents were removed from the sample because they failed attention checks and/or were speedy responding, thus reducing our final sample to 408. In terms of sample characteristics, over half of the followers (53%) identified as male, the average reported age was 38 years (M = 37.85, SD = 9.98), and they had been a member of their organization for approximately 7 years (M = 6.80, SD = 6.73). A variety of industries were represented in our sample with most participants employed in education, telecommunications, technology, internet and electronics industry, manufacturing, and health care.

In addition to describing their leader's behaviors, we asked followers to provide information about themselves. We assessed the physical health of respondents using the Physical Health Questionnaire, a 14-item measure validated for self-report of somatic symptoms (Schat et al., 2005). Sample items included “How often have you experienced headaches?” and “How often have you woken up during the night?” and were rated on 7-point scales that ranged from never to always. We assessed the psychological well-being of respondents using four items from the WHO Well-Being Index (Bech, 2012). For this measure, each of the followers were asked to reflect on the past 4 weeks and indicate how frequently (on a 5-point scale) they experienced a particular indicator of well-being, such as “I have felt cheerful and in good spirits” and “My daily life has been filled with things that interest me.” We assessed affective commitment using the six items from Meyer et al. (1993). Using a 7-point scale, each respondent expressed their level of agreement with statements that reflected one's emotional commitment to the organization. We assessed the level of work-related burnout experienced by followers using seven items from the Copenhagen Burnout Inventory (Kristensen et al., 2005). On a 5-point scale (always to never), participants were asked to indicate how often they experienced each burnout indicator. Finally, we assessed the trait negative affectivity using a 10-item subscale described by Watson et al. (1988). In this measure, respondents were presented with one-word descriptions of 10 negative states (e.g., nervous, upset, afraid, irritable). For each of these items, followers indicated on a 5-point scale how often they felt this way. Composite scores were computed by averaging item responses for each of the above-mentioned outcomes. Means, standard deviations, scale reliabilities, and correlations among the study measures are displayed in Table 4.

OPEN IN VIEWER

Table 4

Study 2: Descriptive Statistics and Correlations.

Measures	M	SD	1	2	3	4	5	6	7	8	9	10	11	12	13	14
1. Respondent Sex (1 = male; 2 = female)	n/a	n/a	n/a
2. Respondent Age	37.85	9.98	0.07	n/a
3. Organizational Tenure (Years)	6.80	5.73	0.06	0.41	n/a
4. Negative Affectivity	2.04	0.87	0.05	−0.12	−0.05	0.92
5. Transformational Leadership	3.41	0.91	−0.01	−0.08	0.04	−0.18	0.96
6. Contingent Reward	3.54	1.00	−0.01	−0.05	0.06	−0.21	0.89	0.87
7. Management by Exceptions–Active	2.81	0.89	−0.14	−0.09	−0.05	0.12	0.09	0.06	0.73
8. Management by Exceptions–Passive	2.48	1.01	−0.04	0.02	−0.03	0.24	−0.51	−0.48	0.23	0.77
9. Laissez Faire/Avoidant	2.10	0.99	−0.02	−0.04	−0.04	0.25	−0.54	−0.54	0.17	0.79	0.84
10. Abusive Supervision	1.24	0.59	−0.11	−0.10	−0.07	0.20	−0.20	−0.22	0.13	0.23	0.30	0.92
11. Physical Health	5.47	0.83	−0.13	0.04	0.05	−0.49	0.25	0.27	−0.05	−0.21	−0.28	−0.21	0.86
12. Psychological Well-Being	3.50	0.88	−0.09	0.04	0.05	−0.48	0.49	0.49	0.07	−0.26	−0.29	−0.06	0.46	0.87
13. Affective Commitment	4.69	1.75	0.00	0.01	0.16	−0.26	0.63	0.55	0.12	−0.34	−0.29	−0.10	0.20	0.53	0.96
14. Work-Related Burnout	2.49	0.88	0.10	−0.03	−0.06	0.48	−0.47	−0.46	0.09	0.37	0.40	0.12	−0.55	−0.64	−0.50	0.90

Note: N = 408. Bolded correlations are statistically significant, p < 0.05 (two-tailed test). Where appropriate, scale reliabilities are presented in the diagonal.

Testing the Measurement Model: Confirmatory Factor Analyses

Following the same logic and procedures described in Study 1, the fit of three models of the leadership-style measures were evaluated and compared: Model 1 (χ² = 3,284.53, df = 275; RMSEA = 0.16; SRMR = 0.16), Model 2 (χ² = 1,225.50, df = 269; RMSEA = 0.09; SRMR = 0.08), and Model 3 (χ² = 1,195.44, df = 260; RMSEA = 0.09; SRMR = 0.08). As was observed in Study 1, a chi-square test confirmed that the six-factor measurement model was superior to the four-factor measurement model (χ² = 47.72, df = 9, p < 0.001). Beyond the leadership-style measures, we then tested several measurement models that involved the five auxiliary variables (physical health, psychological health, affective commitment, work-related burnout, and negative affectivity). The first model specified a single (common) latent factor: χ² = 2,565.73, df = 119; RMSEA = 0.22; SRMR = 0.14. The second model specified three latent factors that included negative affectivity, context-free outcomes (combining physical and psychological health items), and work-related outcomes (combining affective commitment and burnout items): χ² = 1,802.45, df = 116; RMSEA = 0.19; SRMR = 0.15. The final model in this series involved five latent factors whereby each item loaded on its intended latent factor: χ² = 248.24, df = 109; RMSEA = 0.06; SRMR = 0.04. The fit of the five-factor model was superior, which confirmed the discriminant validity of these measures.

LPA

Table 2 (Study 2) provides the LPA fit statistics associated with different models tested. As with Study 1, the three-pattern model represented the best choice based on our inspection of elbow plots of AIC, BIC, and SSA-BIC values; entropy values; and the significance levels of the LMR values. The three-pattern model demonstrated a high probability that respondents were correctly classified (0.94 to 0.98), was theoretically interpretable, and replicated the three patterns observed in Study 1. The mean values of the six leadership styles within each of the three patterns are evident in Table 3 (Study 2). Figure 1c presents a graph of the three latent patterns of follower perceptions: Pattern 1 (N = 282 or 69.12% of the sample) optimal, Pattern 2 (N = 101 or 24.76% of the sample) passive, and Pattern 3 (N = 25 or 6.13% of the sample) passive-abusive.

The next step in our analytic strategy involved regressing class membership on a set of follower characteristics—namely, sex, age, organizational tenure, and negative affectivity. The reference category used for all of the multinomial logistical regression coefficients was the optimal pattern. Inspection of the regression coefficients revealed that both age and negative affectivity had implications for class membership relative to the reference category (see Table 5, Study 2). To aid interpretation, an odds ratio (OR) was associated with each regression coefficient. An OR greater than 1 means that for every unit change in the predictor variable, the likelihood of being classified in the target profile increases relative to the reference group with all of the other variables in the model held constant. Thus, the ORs associated with negative affectivity revealed that respondents who reported higher rather than lower scores were 1.64 and 2.81 times, respectively, more likely to belong to the passive and passive-abuse patterns than they were to the referent profile. When OR values are less than 1, it is helpful to consider how much the OR deviates from 1. From Table 5, it is evident that a one-unit change in age reduces the likelihood of membership in the passive-abusive pattern by 6% (1–0.94) relative to the reference group. Expressed another way, when age is high rather than low, membership in the passive-abusive pattern is 1.06 times (1.06 as the reciprocal of 0.94) less likely than is membership in the reference profile with other variables in the model held constant. Thus, Hypothesis 4 was supported for age and negative affectivity but not for gender or tenure.

OPEN IN VIEWER

Table 5

Studies 2-3: Predicting Pattern Membership From Personal Characteristics.

Predictor Variables	Passive Pattern			Passive-Abusive Pattern
	Coefficient	SE	OR	Coefficient	SE	OR
Study 2 (N = 408)
Respondent's Sex (1 = male; 2 = female)	0.17	0.25	1.19	−0.77	0.50	0.46
Respondent's Age	0.02	0.01	1.02	−0.06	0.03	0.94
Organizational Tenure	0.00	0.00	1.00	0.00	0.00	1.00
Negative Affectivity	0.49	0.15	1.64	1.03	0.24	2.81
Study 3 (N = 246)
Respondent's Sex (1 = male; 2 = female)	−0.44	0.36	0.64	−1.01	0.45	0.37
Respondent's Age	0.03	0.02	1.03	0.01	0.03	1.01
Organizational Tenure	−0.06	0.06	0.94	−0.03	0.05	0.97
Negative Affectivity	1.00	0.42	2.72	1.96	0.39	7.12

Note. For both studies the Optimal Pattern is the reference category. Coefficient = multinomial logistic regression coefficient; SE = standard error of the coefficient; OR = odds ratio. Bolded coefficients are statistically significant, p < 0.05.

Table 6 (Study 2) displays the results of the outcome analyses. To determine if the three pattern groups differed with respect to context free (physical health, psychological well-being) and work-related outcomes (affective commitment and work-related burnout), we conducted Wald chi-square (χ²) tests of mean equality. For all four criteria, significant group differences were observed for physical health (χ² = 37.56, p < 0.01), well-being (χ² = 55.89, p < 0.01), affective commitment (χ² = 70.90, p < 0.01), and work-related burnout (χ² = 70.89, p < 0.01). Inspection of the means associated with each pattern provided clear support for Hypothesis 5. Followers who perceived their leader to exhibit an optimal pattern reported the highest levels of physical health, psychological well-being, affective commitment, and the lowest level of work-related burnout relative to those classified in the two passive patterns. We observed that the passive-abusive pattern was particularly detrimental on physical health, providing partial support for Hypothesis 6. Followers who perceived their leader as exhibiting the passive pattern reported the lowest level of psychological well-being and affective commitment. Followers exposed to either passive or passive-abusive leaders reported the most work-related burnout, although these groups were not significantly different from one another. Thus, the findings of Study 2 provided support for Hypotheses 1-5 and offered support for Hypothesis 6 with respect to physical health.

OPEN IN VIEWER

Table 6

Studies 2-3: Evaluating the Effects of Pattern Membership on Distal Outcomes.

	Optimal	Passive	Passive-Abusive	Overall	Summary of Tests
Outcomes	Pattern	Pattern	Pattern	Chi-Square	of Pattern Means
Study 2 (N = 408)
Physical Health	5.64	5.23	4.49	χ2(2) = 37.56, p < 0.01	1 > 2 > 3
Psychological Well-Being	3.73	2.89	3.34	χ2(2) = 55.89, p < 0.01	1 > 3 > 2
Affective Commitment	5.27	3.20	4.37	χ2(2) = 70.90, p < 0.01	1 > 3 > 2
Work-Related Burnout	2.22	3.15	2.86	χ2(2) = 70.89, p < 0.01	1 < (2 = 3)
Study 3 (N = 246)
Physical Health	5.73	5.25	4.43	χ2(2) = 56.13, p < 0.01	1 > 2 > 3
Psychological Well-Being	3.36	2.87	2.78	χ2(2) = 25.19, p < 0.01	1 > (2 = 3)
Perceived Organizational Support	5.94	4.19	3.65	χ2(2) = 142.05, p < 0.01	1 > 2 > 3
Work-Related Vigor	5.49	4.20	4.59	χ2(2) = 81.68, p < 0.01	1 > 3 > 2

Note: All analyses were run utilizing the DU3STEP procedure within MPlus. Means for the four outcomes are displayed for each pattern group a test of mean equality across the pattern groups was performed using the three-step procedure with 2 degrees of freedom for the overall chi-square (χ²).

Sample Characteristics, Measures, and Descriptive Statistics

To confirm and extend our understanding of follower perceptions and experiences, we collected a third sample of data. To provide temporal separation between the variables used to create the pattern profiles and those used to assess context-free and work-related outcomes, followers provided data at two points in time. Initially (Time 1) followers reported their status on several personal characteristics and rated the behavior of their leaders. Approximately 2 months later (Time 2), the outcome measures were collected. At Time 1, 486 eligible participants completed study measures. Of these followers, 71 were excluded for speedy responses and another 30 failed attention check questions, reducing the N from 486 to 385. At Time 2, 269 of the 385 followers provided data. Of the 269, 23 respondents were removed (16 because of speeding and 6 due to failed attention checks). In total, 246 followers provided data at both temporal periods. In terms of sample characteristics, over half of 246 followers (59%) identified as male, and the mean reported age was 35 years (M = 34.77, SD = 10.04). On average, followers had been a member of their organization for approximately 7 years (M = 7.01, SD = 5.65). As with the previous two studies, a variety of industries were represented with most followers indicating they worked in retail/sales, internet and electronics industries, finance, manufacturing, and health care contexts.

The measures of the six leadership styles, physical health, and negative affectivity were the same as described in Study 2. Although the focal constructs were the same, several new outcome measures were featured to test the conceptual rigor of the findings. In this study, followers’ psychological well-being was captured using the 12-item General Health Questionnaire (GHQ: Goldberg & Williams, 1988). Statements that reflected various indicators of psychological health were presented to respondents (e.g., been able to concentrate on whatever you are doing; felt capable of making decisions about things; been able to enjoy your normal day-to-day activities). In turn, followers were asked to express how often they felt this way in the past 2 months using a 4-point scale (1 = never to 4 = all the time). Perceived organizational support was measured using eight items from the longer scale described by Eisenberger et al. (1986). Here, followers were asked to express their agreement or disagreement on 7-point scales with statements such as “The organization really cares about my well-being.” Work-related vigor was assessed using the 14-item measure developed by Shirom (2003). Followers were presented with a series of statements that reflected physical strength, cognitive liveliness, and affective energy (e.g., “I feel vigorous,” “I feel able to be creative,” “I feel I am capable of investing emotionally in coworkers and customers”) and then asked to express on a 7-point scale how often they felt this way at work (1 = never or almost never to 7 = always or almost always). Composite scores were computed by averaging item responses. Means, standard deviations, scale reliabilities, and correlations among the study measures are displayed in Table 7.

OPEN IN VIEWER

Table 7

Study 3: Descriptive Statistics and Correlations.

Measures	M	SD	1	2	3	4	5	6	7	8	9	10	11	12	13	14
1. Respondent Sex (1 = male; 2 = female)	n/a	n/a	n/a
2. Respondent Age	34.77	10.04	0.10	n/a
3. Organizational Tenure (Years)	7.01	5.65	−0.07	0.47	n/a
4. Negative Affectivity	1.62	0.73	0.02	−0.22	−0.10	0.91
5. Transformational Leadership	3.31	0.90	0.09	0.00	0.11	−0.26	0.95
6. Contingent Reward	3.35	0.93	0.09	0.02	0.14	−0.31	0.91	0.82
7. Management by Exceptions–Active	3.00	0.82	−0.01	−0.07	0.01	0.15	0.12	0.13	0.62
8. Management by Exceptions–Passive	2.47	0.94	−0.02	0.02	−0.05	0.33	−0.43	−0.42	0.17	0.74
9. Laissez Faire/Avoidant	2.05	1.00	−0.01	−0.13	−0.07	0.44	−0.55	−0.56	0.11	0.75	0.87
10. Abusive Supervision	1.48	0.78	−0.11	−0.16	0.01	0.42	−0.36	−0.34	0.17	0.45	0.59	0.92
11. Physical Health	5.31	1.03	0.01	0.12	0.06	−0.50	0.19	0.21	−0.16	−0.33	−0.34	−0.43	0.91
12. Psychological Well-Being	3.07	0.57	0.04	0.18	0.12	−0.60	0.37	0.37	−0.07	−0.28	−0.33	−0.29	0.63	0.90
13. Perceived Organizational Support	4.86	1.46	−0.03	0.06	0.14	−0.40	0.65	0.61	−0.04	−0.50	−0.53	−0.44	0.39	0.58	0.94
14. Work-Related Vigor	4.84	1.15	0.01	0.09	0.07	−0.28	0.47	0.46	0.11	−0.19	−0.24	−0.22	0.40	0.57	0.51	0.94

Note: N = 246. Bolded correlations are statistically significant, p < 0.05 (two-tailed test). Where appropriate, scale reliabilities (Cronbach’s alpha) are displayed in the diagonal.

Testing the Measurement Model: Confirmatory Factor Analyses

For the reasons described above, the fit of three models was evaluated and compared: Model 1 (χ² = 1,849.62, df = 275; RMSEA = 0.15; SRMR = 0.14), Model 2 (χ² = 902.65, df = 269; RMSEA = 0.10; SRMR = 0.09), and Model 3 (χ² = 884.78, df = 260; RMSEA = 0.10; SRMR = 0.08). As observed in prior studies, a chi-square test confirmed that the six-factor measurement model was superior to the four-factor measurement model, χ² = 31.80, df = 9, p < 0.001. We then tested several measurement models that involved the five auxiliary variables (physical health, psychological health, perceived organizational support, work-related vigor, and negative affectivity). The first model specified a single (common) latent factor: χ² = 1,230.79, df = 119; RMSEA = 0.20; SRMR = 0.12. The second model specified three latent factors that included negative affectivity, context-free outcomes (combining physical and psychological health items), and work-related outcomes (combining perceived organizational support and vigor): χ² = 555.73, df = 116; RMSEA = 0.12; SRMR = 0.10. The final model in this series involved five latent factors whereby each item loaded on its intended latent factor: χ² = 204.43, df = 109; RMSEA = 0.06; SRMR = 0.05. Once again, the five-factor model fit was superior, which confirmed the discriminant validity of these measures.

LPA

Table 2 (Study 3) provides the LPA fit statistics associated with different pattern models tested. As with Studies 1 and 2, the three-pattern model demonstrated a high probability that respondents were correctly classified (0.90 to 0.97), was theoretically interpretable, and replicated the three cohorts in previous follower samples. The means of the six leadership styles within the three cohorts are evident in Table 3 (Study 3). Figure 1d presents a graph of the three distinct latent patterns of follower perceptions observed in the second sample: Pattern 1 (N = 111 or 45.12% of the sample) optimal, Pattern 2 (N = 89 or 36.18% of the sample) passive, and Pattern 3 (N = 46 or 18.70% of the sample) passive-abusive. Once again, the configuration of the six behavioral styles within the three patterns replicated across follower (and leader) samples.

Following the analytic strategy used in the previous study, we regressed class membership on a set of follower characteristics. The reference category used for all of the multinomial logistical regression coefficients was the optimal pattern (see Table 5, Study 3). Inspection of the multinomial regression coefficients revealed that both sex and negative affectivity had implications for class membership relative to the reference category (optimal pattern). Not unlike what was observed in Study 2, the ORs associated with negative affectivity revealed that followers who reported higher rather than lower scores were 2.72 and 7.12 times, respectively, more likely to belong to the passive and passive-abuse patterns than they were to the referent profile. In this sample, gender played a role. From Table 5 (Study 3) we see that identifying as a female reduces the likelihood of membership in the passive-abusive pattern by 63% (1–0.37) relative to the optimal pattern. Expressed another way, when followers are female rather than male the likelihood of membership in the passive-abusive pattern is 2.70 times (2.70 as the reciprocal of 0.37) less likely than is membership in the reference profile with all of the other variables in the model held constant. In summary, the finding for negative affectivity confirms Hypothesis 4, whereas the findings for sex, age and tenure do not.

Table 6 (Study 3) displays the results of the outcome analyses. To determine if the three pattern groups differed with respect to well-being (physical health, psychological well-being) and work (perceived organizational support, work-related vigor) outcomes, we conducted Wald chi-square (χ²) tests of mean equality. For all four criteria, significant group differences were observed for physical health (χ² = 56.13, p < 0.01), well-being (χ² = 25.19, p < 0.01), perceived organizational support (χ² = 142.05, p < 0.01), and work-related vigor (χ² = 81.68, p < 0.01). Inspection of the means associated with each pattern provided clear support for Hypothesis 5. Followers who perceived their leader to exhibit an optimal pattern reported the highest levels of physical health, psychological well-being, perceived organizational support, and work-related vigor relative to those classified in the passive groups. For the two patterns where negative styles were dominant, we observed that the passive-abusive pattern was particularly detrimental for context-free and work-related outcomes, as indicated by followers’ reported physical health and perceived organizational support. It is noteworthy that followers exposed to the passive-abusive pattern had the lowest level of psychological well-being (another context-free outcome), yet this was not significantly different from the level reported in the passive pattern. Followers who perceived their leader as exhibiting the passive pattern reported the lowest level of work-related vigor. To summarize, the findings of Study 3 provided support for Hypotheses 1-5 and offered support for Hypothesis 6 with respect to physical health and perceived organizational support.

Theoretical Implications

Across the three studies, we were struck by how followers’ reactions to the passive leadership styles changed depending on the context of the pattern. For instance, in a context where leaders are not perceived as acting with malice (AS is low), passive leadership is generally experienced as benign neglect. But in a context where AS is high, followers might interpret passive leadership as a consistent, deliberate, and intentional means of undermining them (e.g., avoiding interactions, failure to provide helpful feedback). The role of follower attributions has recently been introduced as a potential moderator of how negative leader behaviors impact followers (Harvey et al., 2014; Schyns et al., 2018). In the present case, we speculate that when followers attribute the cause of their leader's disengaged and avoidant style to the situation (e.g., too many subordinates; industry or organizational norms) the negative reaction should be muted (i.e., forgivable) given that there are no other cues to suggest otherwise (cf. Schyns et al., 2018). This view is consistent with previous research showing that recipients of mistreatment may engage in different sense-making processes to understand their experience (Fiset et al., 2019; Herschovis & Barling, 2010; Mackey et al., 2017). In particular, more severe and active forms of mistreatment heighten the salience of group identification, which allows employees to minimize psychological harm to themselves by viewing the supervisors’ actions as targeting the group. When passiveness occurs in a context when leaders are seen to also engage in targeted acts of abuse, the leader's neglect may appear to be intentional, deliberate, and vindictive. Furthermore, the destructive effects of the passive-abusive pattern should be exacerbated in situations where employees feel trapped in their jobs. It was not surprising in the present research that the response to ugly leadership (i.e., passive-abusive) was felt on a visceral level involving elevated reports of sleep disturbances, headaches, gastrointestinal problems, and respiratory illness (Schat et al., 2005).

From a broader perspective, our findings shed light on the behavioral contexts in which individual styles occur. As previously discussed, we see that passive leadership has its most harmful effect on followers when it is conflated with AS. In a context of leader abuse (e.g., public ridicule, broken promises, deliberate lying), passivity is likely experienced as targeted hostility and aggression, which, in turn, erodes physical health. When passive leadership is disassociated from abuse, followers may be more likely to experience this as benign neglect—an experience associated with higher levels of burnout and lower levels of well-being, vigor, organizational commitment, and perceived organizational support. In other words, passive leadership in a context of low abuse feels differently than it does when abuse is high. Finally, this work shows that the relationship between any one style and an outcome of interest depends on the relative strength of the other styles in the leader's overall pattern (cf. van Knippenberg & Sitkin, 2013). This observation might be relevant for middle managers who, because of their unique position in the organization, must employ a variety of different leadership styles depending on the situation (referred to as code switching: Anicich & Hirsh, 2017).

From a resources perspective, these results provide further evidence to support the conceptualization of transformational-transactional leaders as a source of resource gain for followers, while passive and passive-abusive patterns are likely to trigger resource loss. Through processes of interpersonal resource exchange, both passive and passive-abusive patterns are less likely to elicit desirable follower attitudes and outcomes than constructive leadership. In an employment context, resource loss experienced under destructive leadership can decrease the likelihood that followers will achieve their work goals, increase job demands and the associated physiological and psychological costs, and inhibit personal development (Bakker & Demerouti, 2007). Additionally, by measuring physical health, psychological well-being, affective commitment, vigor, and perceived organizational support, we have responded to calls to extend COR literature beyond stress and emotional exhaustion (Hobfoll et al., 2018).

Practical Implications

From a pragmatic perspective, the current study offers insights for leadership development. Rather than designing leadership training around individual styles, a more effective pedagogy would stem from an acknowledgement that different leadership styles co-occur. Increasing the incidence of a particular leadership style may not provide the desired effect if the other styles within the overall leadership pattern remain constant. More importantly, the nature of these combinations of leadership styles matters when it comes to understanding follower responses and outcomes. Managers should be aware that the adverse effects of AS with respect to follower well-being might also occur under passive leadership. Although leadership development typically focuses on increasing transformational behaviors, our findings show, as have other pattern-based studies (e.g., Arnold et al., 2017) and research on destructive leadership (Skogstad et al., 2014), that organizational training initiatives should encompass and mitigate other leader behaviors, such as AS. Training should also focus on reducing abusive-passive behaviors and equipping managers and employees alike with conflict management skills. Leaders who avoid followers may be perceived as passive and should consider the implications of their absence on followers to determine whether followers need or want more guidance.

By demonstrating that passive leadership can occur on its own or together with abusive leadership, we can make several inferences to provide incremental clarity to anecdotal cases of high-profile leaders who have a reputation for abusive behaviors some of the time. Contrary to the suggestion that even the most abusive supervisors act supportively sometimes (cf. Tepper et al., 2017), we did not observe a subpopulation of leaders who exhibited this pattern. While some leaders might be described in popular culture as both transformational and abusive, our data suggest this might be an exceptional circumstance or an outlier scenario rather than a dominant behavioral pattern. What we did find is a subpopulation of passive leaders who are also abusive. Overall, these results are relevant for both academics and practitioners, with the takeaway message that a leader who is perceived by followers as a disengaged leader can damage employee's work-related attitudes as much as a disengaged leader who is also abusive.

Limitations

Despite the contributions of this research, some limitations should be acknowledged. First and foremost, our range of leadership styles represented a limited set of popular leadership paradigms. Many other leadership styles have been studied (e.g., Hoch et al., 2018; Wang et al., 2019), and future work is needed to determine if AS co-occurs with alternate configurations of styles. For instance, would followers’ experience of servant, ethical, or authentic leadership change when combined with varying degrees of AS? While this study investigated a limited set of leadership styles (Barling et al., 2011; Dinh et al., 2014), more work that integrates a broader range of leadership styles would help to further develop leadership theory. Our understanding of individual styles and their effects on followers’ outcomes can be informed by knowledge of other styles in the leader's behavioral repertoire. Second, we focused on the employee's perceptions of their leader's behavior and, therefore, lack a relational perspective obtained by collecting data on the predictor or criterion data from both leaders and followers. However, our purpose was to capture followers’ experience of their leaders especially with respect to overt acts of abuse, how AS interacts with other leadership styles, and the implications of these combinations for followers.

Another limitation reflects the challenges inherent to examining conceptually distinct behaviors that are hard to distinguish empirically (e.g., Hoch et al., 2018). Not unlike what has been reported in prior pattern leadership-pattern research (e.g., Arnold et al., 2017; Doucet et al., 2015), we observed high correlations among certain leadership styles, especially TFL and CR. That said, in all four independent samples, our proposed six-factor measurement model was found to be superior to other competing measurement models. In all three studies, we constructed patterns based on a range of positive and negative leadership styles to minimize the overlap (see correlation matrices). Although the positive styles (TFL and CR) moved together within all three patterns, it is noteworthy that we observed much more empirical separation among the negative styles, which was an important focus in this research. Finally, if it makes theoretical sense to predict patterns that differ with respect to these highly correlated styles, researchers might consider a bifactor approach to profile construction (e.g., Morin et al., 2017, Morin & Marsh, 2015).

Another potential limitation is the cross-sectional nature of the data in Studies 1 and 2, which precludes conclusions regarding causality and heightens concerns related to common method bias. We mitigated some of these concerns by introducing a time lag between leadership patterns and follower outcomes (Study 3), using anonymized surveys, ensuring items were worded clearly and concisely using previously validated measures, randomizing question ordering, including attention checks and positively and negatively worded items (Podsakoff et al., 2012). Had common method variance explained our results (whereby all of the leadership-style items loaded on a global latent factor), we likely would have observed patterns that differed in terms of their levels (e.g., all low, all medium, all high) (Morin & Marsh, 2015). Instead, we observed qualitatively distinct, theoretically interpretable patterns that replicated across multiple samples. It is noteworthy that there are some interesting ways that mixture modeling techniques can be used to detect causal relations. For instance, latent transition analysis (LTA) looks at whether membership in latent categories changes over multiple time periods and whether these changes can be associated with the status of auxiliary variables (for an example of LTA in a work context, see Kam et al., 2016). For a more complete discussion of how mixture modeling techniques can be used to test whether membership in a particular latent class moderates known relationships between or among auxiliary variables, we refer readers to McLarnon and O’Neill (2018).

Our reliance on OPD represents another potential limitation. However, we followed best practices (Aguinis et al., 2021) and we note that research comparing different online data-collection platforms has generally found no differences with respect to scale reliability or the ability to reach a diverse participant pool (Porter et al., 2018). Furthermore, the ability to gain access to a sample of participants from a broad range of organizations was of benefit to this research. Sampling broadly across social and organizational context allowed us to capture natural variance among perceptions of the six leadership styles, variance that might otherwise have been attenuated in a single organization where formal controls and cultural norms might have constrained the expression of behavior (especially the negative styles).

Previous research suggests that the presence of certain moderating conditions, such as high job mobility (i.e., you can leave if you desire) and expectations for abusive leadership (i.e., when lived experience confirms expectations) may buffer the emotional impact and resource drain associated with abusive supervisors (Tepper et al., 2017). For instance, employees who have a previous history of aggressive, retaliatory behavior or employees who have worked for abusive leaders in the past might be sensitized to mistreatment and thus cope much better in the presence of an abusive leader (Tepper et al., 2017). These buffering effects may have been present in our sample. Together with the potential for muted effects of AS in field settings, it is possible that future studies may find significant differences in work-related attitudes under passive and passive-abusive leadership. However, other comparative investigations of follower outcomes in response to destructive leadership styles have highlighted the role of attribution processes (e.g., Schyns et al., 2018); therefore, future work is needed to further appreciate the role of cognition in followers’ reactions to negative leader behaviors.

Future Research

Even though leadership has been a thriving area of research for decades, there is still work left to do and much can be gleaned from this data analytic approach and line of inquiry. As such, organizational and management researchers are encouraged to continue using pattern or person-centered approaches to integrate disparate leadership theories and replicate our three-pattern solution on different samples. As alluded to earlier, there is further opportunity to integrate AS with other positive leadership styles, such as ethical or authentic leadership, or destructive styles, such as authoritarian or unethical leadership. Our findings call for further study of how destructive leadership patterns relate to the various facets of psychological well-being. It would also be advantageous to study followers’ perceptions of leader behaviors over a longer timeframe to observe which outcomes are differentially affected as time passes. Recently, researchers have acknowledged that some organizational members hold lay beliefs that enactment of AS is beneficial for the employee and organizational performance (e.g., Watkins et al., 2019). A pattern-oriented analytical approach could be used to probe this issue and further investigate the types of leaders who are most likely to hold this belief or the type of leadership styles that co-occurs with AS most often. The door is also opened to respond to the critiques and calls made by van Knippenberg and Sitkin (2013) to improve construct clarity and theoretical distinctiveness of the dimensions that comprise TFL using new methodologies.

LPA can be applied to identify subpopulations of transformational leaders by investigating the relative strength of idealized influence, inspirational motivation, individualized consideration, and intellectual stimulation. Similarly, well-being outcomes could be examined at the dimension level rather than as a unitary construct as is offered here; for example, physical health dimensions including somatic symptoms, gastrointestinal problems, headaches, sleep disturbances, and respiratory illness (Schat et al., 2005) could be examined separately. Given the importance of leadership for follower well-being, it behooves us to continue this important line of research. Finally, we acknowledge that relations between a leader's behavior and many outcomes are complex and nuanced, as indicated by recent leadership-process studies (e.g., Hackett et al., 2018). Although beyond the scope of the present studies, future research could leverage person-centered analytics to examine how these latent profiles alter or change the strength of known leadership-process models (for a review of how LPA can be used in organizational research, see McLarnon & O’Neill, 2018).