Does character matter in Ukraine? Leader character and employee outcomes in a non-WEIRD context

Introduction

Value-based approaches to leadership have reached a critical juncture. A first set of critics claim that established constructs, such as authentic leadership, draw on classical and existentialist philosophy without engaging deeply with its substance (Alvesson and Einola, 2019). A second criticism holds that authentic leadership studies, and studies of similar positive constructs, conflate what leaders do with how followers evaluate them (Banks et al., 2023; Fischer and Sitkin, 2023). That conflation produces causal illusions in the resulting evidence (Fischer et al., 2024). To address these critiques, Lux and Lowe (2024) proposed reframing authentic leadership through a signaling theory lens (Connelly et al., 2011; Spence, 1978). That move redefines authenticity as the concordance between a leader's values and the signals they emit, regardless of whether those values are virtuous or vicious.

The Leader Character Framework (LCF) offers a complement to concordant signaling. Grounded in Aristotelian virtue ethics, the LCF conceptualizes character as a volitional habit of being that informs leaders' decisions and actions. Character is not a collection of leader attributes, such as traits, internalized values, or behavioral styles. This habit of being reflects an interconnected system of character elements clustering into character dimensions (Crossan et al., 2017). However, every character dimension, Judgment included, can operate as a vice when it is imbalanced or not supported by the others (Crossan et al., 2024). Figure 1 illustrates the 62 elements and 11 dimensions of the LCF.

The LCF avoids these critiques for different reasons. The framework operates following the virtue-ethics tradition rather than invoking it, meeting the first critique. Critics also warn that leadership studies routinely pair positively valenced constructs with favorable outcomes, making the positive association close to guaranteed (Alvesson, 2020; Alvesson and Einola, 2019). The LCF does not treat more of a dimension as better than less because the golden mean doctrine makes excess and deficiency equivalent failures (vices). The LCF also breaks the valence-based conflation. The standard for virtuousness does not depend on followers' approval or their attributions because virtuous character is defined independently of its outcomes (van Knippenberg and Sitkin, 2013). Virtuousness is defined by its moral goodness and pursued as an end in itself, not for the returns it brings (Bright et al., 2006). This independence is what makes character the kind of quality a signal can carry. A signal carries information about a quality only if that quality is defined apart from the receiver's response (Connelly et al., 2011). Whereas, Lux and Lowe (2024) specify the mechanism through which authenticity is signaled, the LCF specifies the virtuous content that concordant signaling may carry.

Yet the empirical evidence linking the LCF to employee outcomes has lagged its conceptualization. Existing studies typically relate character to one outcome at a time (e.g., Seijts et al., 2024). To our knowledge, only Monzani et al. (2021) have modeled leader character against multiple employee outcomes simultaneously. They mapped leader character and four employee outcomes into a single network joined by a small number of bridging ties. They also measured leader character at the signal's source (leaders). A signaling account is a claim about reception, so it cannot be tested from the sender's vantage alone.

Reception is where our design departs from that work. We measured character at the receiving end, treating reception as part of the account. How followers decode a leader's conduct is culturally embedded, and signals that convey one quality to one group may convey something else to another (Lux and Lowe, 2024). Most leadership evidence accumulates in Western, Educated, Industrialized, Rich, and Democratic (WEIRD) societies (Henrich et al., 2010). Whether leader character associations with employee outcomes hold elsewhere remains an open research question.

We conducted our study in Ukraine in 2021. Ukraine diverges from the WEIRD profile on economic conditions and institutional stability rather than on education or industrialization. Tang and Crossan (2026) argue that leader character matters most when a context weakly enforces its espoused norms. At the time of our study, Ukraine ranked 122nd of 180 countries on the Corruption Perceptions Index (Transparency International, 2022). Ukrainian organizations also faced the compounding pressures of the COVID-19 pandemic and disruption following the 2014 conflict (Seijts et al., 2025a,b).

The present study extends Monzani et al. (2021) in a non-WEIRD setting, with three design changes. First, we shift from character signal emission to its reception, replacing leader self-ratings with follower ratings. Second, we drop resilience, retain three of the four earlier outcomes, and add three more. Third, we specify every model at the construct and dimension levels; Monzani et al. (2021) modeled the dimension level only. With the receiving side of leader character now in view, we ask:

RQ: Is perceived leader character associated with employee outcomes? If so, are the associations direct and independent, does one outcome connect leader character to the rest, or is there some intermediate organization?

We answer our research question in two steps. The first step maps the associations that remain when the full set of variables is modeled jointly. The second step compares competing accounts of how leader character perceptions relate to the six outcomes. Signaling theory gives that comparison a directional expectation. Character signals are sender-referential, so an evaluative attitude toward the leader should be the proximal outcome. Signaling theory constrains the entry point, not the ordering among the remaining outcomes. That limit is why we proceed abductively and frame the study as exploratory.

Our study makes three contributions. The first is theoretical and spans both sides of the signaling exchange. On the sending side, we reposition leader character as an antecedent of authentic signaling, supplying the normative content and the cost structure that concordance-based accounts leave unspecified (Lux and Lowe, 2024). On the receiving side, we extend the logic of the Causal Attitude Network (CAN) model beyond work attitudes to psychological states, mindsets, and behavioral intentions (Carter et al., 2020; Dalege et al., 2016). The dominant approach treats these outcomes as independent endpoints of leader influence. We argue instead that they form a structured cluster whose organization shapes where leader character's associations appear and how far they extend. Together, we specify whether character relates to employee outcomes, where those associations connect, and why.

Our second contribution is methodological. We offer a disciplined analytic strategy for choosing among rival orderings when theory permits more than one. Leadership research is fragmented, with many overlapping frameworks competing to explain the same outcomes (Leroy et al., 2025; Meuser et al., 2016). Several accounts can motivate a plausible ordering among a set of employee outcomes, and standard structural equation practice lets each one appear defensible. This invites confirmation bias because the order a researcher tests is the order that survives.

Pairing psychometric network analysis with competing structural models constrains this freedom. The network first estimates the conditional dependency structure without imposing direction, narrowing the orderings worth testing. Competing structural models then adjudicate among theoretically grounded specifications against that scaffold rather than theory alone. Because both are estimated on the same sample, the pairing disciplines model choice rather than validating it. The strategy suits any system of constructs.

Our third contribution is contextual. We extend the LCF's evidence base beyond the WEIRD samples in which it was developed. The LCF holds that character is universal yet reflexive to context (Crossan et al., 2024). Ukraine in 2021 departs from the conditions in which the framework was built. A signaling account sharpens what that setting tests. Signals are informative only when sender and receiver share a common reading heuristic. The universality claim is a claim about such a shared code, and it has been asserted more often than examined. A setting that lacks the code would show dimensions cohering differently than WEIRD samples and character entering the outcome cluster elsewhere. The recovered network and its outcome associations resemble patterns reported in prior samples. We did not formally test that resemblance, so we read it as suggestive rather than as evidence of invariance.

Theoretical background

The LCF defines character as

“an interconnected suite of embodied and virtuous habits. These habits are virtuous insofar as they collectively influence judgment and human flourishing.” (Crossan et al., 2024, p. 653)

Three properties ground the framework theoretically. First, character is habitual in the Aristotelian sense, meaning that it is built and strengthened through repeated exercise rather than fixed at birth. The more consistently these habits are practiced, the more integrated and reliably expressed they become. This malleability is what makes character developable in a way that personality traits are not (Crossan et al., 2021b).

A second property deepens the contrast between leader character and personality traits. Personality traits describe relatively stable tendencies that are automatically activated in response to stimuli. Character dimensions are instead volitionally engaged and evaluatively directional. They are consciously chosen and effortfully enacted, and they describe what a person strives to be when pursuing virtuous conduct (Crossan et al., 2023, 2024). The dimensions are not value-neutral but directionally good in their typical expression, although they can become vices when enacted in excess or deficiency. This evaluative directionality, together with the framework's grounding in practical wisdom (Schwartz and Sharpe, 2006), situates the LCF conceptually closer to the virtue ethics tradition than to personality psychology (McCrae and Costa, 1997, 2003).

Third, in the LCF, character is structured as a system rather than a collection of orthogonal traits. The 11 dimensions do not function independently; their capacity to sustain leader effectiveness depends on how they are enacted in concert with one another (Monzani et al., 2021). A strong Judgment reflects the exercise of phronesis, the human capacity to discern what is called for in a particular situation and to deploy the appropriate dimensions, in the appropriate degree (Schwartz and Sharpe, 2006). This coordinative role leads us to expect Judgment to occupy a central position within the character network. Judgment showed the highest betweenness centrality in three prior samples (Crossan et al., 2017; Monzani et al., 2021). We revisit that expectation in our exploratory network analysis.

Together, these three properties distinguish leader character from personality traits. Empirical evidence supports the distinction between character and personality in its relation to other constructs. For example, in two field studies of law enforcement personnel, leader character predicted motivation to lead and active problem-solving. Those associations held after controlling for the Big Five traits and psychological capital. Leader character was also indirectly related to other coping responses through a motivation to lead (Seijts et al., 2024).

Leader character as a costly signal

Lux and Lowe's (2024) editorial proposes redefining authentic leadership as the concordant, value-based signaling of the four dimensions that Walumbwa et al. (2008) established, namely, self-awareness, an internalized moral perspective, balanced processing, and relational transparency. Their definition deliberately focuses on signal fidelity rather than positive moral content or the expression of a “true self,” an intentional neutrality that answers critics who identify an arbitrary positive bias in the construct (Alvesson and Einola, 2019; Einola and Alvesson, 2021). That neutrality, however, leaves open the question of what content merits signaling in the first place.

What Lux and Lowe's (2024) redefinition gains in precision regarding the signaling mechanism, it cedes in normative content. Concordance is normatively silent. A leader can signal a self-serving identity as faithfully as a virtuous one, and the mechanism cannot tell the difference. The LCF supplies the missing standard. Its 11 dimensions specify which habits evidence excellence when leading, and its golden-mean grounding sets the criterion virtue as the mean between excess and deficiency (Crossan et al., 2024). Moreover, these dimensions are universal at the abstract virtue level, even though their situated enactment varies across cultures and historical periods. Courage names a virtue recognizable across contexts, but what courageous conduct looks like depends on the specific situation. Phronesis, or practical wisdom in the Aristotelian sense, is what maps the abstract virtue onto the concrete case.

This virtue mapping is what separates the LCF from internalization-based accounts. In Walumbwa et al. (2008) definition and Lux and Lowe's (2024) redefinition, moral content becomes authoritative once it is internalized. Internalization fixes the values a leader signals; it asks neither whether those values are worthy of endorsement nor what enacting them well requires in the situation at hand. The LCF asks both, the first through its virtue criterion and the second through phronesis.

A recent theoretical account reinforces this distinction. Leaders with strong character exercise sound judgment and engage in bright-side leadership. Having a weak or imbalanced character invites excess even when the leader acts concordantly with internalized values (Tang and Crossan, 2026). The LCF specifies which habits merit being called character (only those that are virtuous), treats their enactment as phronetically calibrated rather than fixed, and provides a failure vocabulary in the form of excess, deficiency, and judgment failure. Courage can fall into recklessness or cowardice; humility, into obsequiousness or arrogance. The LCF thus supplies normative content on principled rather than ideological grounds.

Tang and Crossan (2026) further hold that context integrity, meaning a context's effectiveness in enforcing its espoused norms, moderates how strongly character predicts leader behavior. Where enforcement is weak, norm-following is less rewarded, so behavior tracks the leader's own habits more closely. Context integrity is thus a property of the signaling environment; it sets the conditions under which norm-consistent conduct pays. We extend their claim from leader behavior to employee outcomes and from the strategic leaders they theorize to the direct supervisors our respondents rated.

Authentic signaling does not by itself guarantee that followers will endorse the signal. Concordance between a leader's inner values and outward conduct is unobservable. Followers must instead infer it, much as they attribute charisma to those leaders they perceive as embodying valued qualities (Platow et al., 2006). Any such inference needs an evidentiary base in conduct the follower can observe. Signaling theory calls that base signal reliability, also termed credibility, the combination of a signaler's honesty and the signal's correspondence to the underlying quality (Connelly et al., 2011). Credibility holds only where signal costs are structured so that dishonest signaling does not pay. A signal is credible when emitting it is cheap for senders who hold the underlying quality and expensive for senders who do not.

That cost asymmetry is what makes the signal worth attending to. Character, understood as habituated conduct, meets this condition. Virtuous habits are inexpensive to sustain for leaders who have cultivated them and prohibitively costly to simulate for leaders who have not. Simulation demands consistency across situations, and over time, the property that separates a credible signal from an isolated display. Character, therefore, stands as an antecedent of authentic signaling. Habituated conduct leaves little discrepancy to manage and supplies the observable evidence on which an attribution of authenticity rests; endorsement follows from that attribution. This antecedent placement also answers the valence-based conflation critique because it separates the leader's habitual conduct, the signaling of that conduct, and the attribution that followers form about their leader's intentions (Fischer and Sitkin, 2023).

The costly signal logic underlying charismatic signaling extends to character (Akstinaite et al., 2024). The cost, however, falls elsewhere. Charismatic signaling is costly at the moment of production because it demands creativity, symbolic communication, and expertise. Character signaling is costly upstream. Virtuous habits are cultivated over time and erode without deliberate practice, so the cost recurs rather than being paid once (Crossan et al., 2023). Those habits are also exercised by choice rather than triggered by the situation, which keeps the conduct controllable in the sense that signaling theory requires. Both costs are differential because both are lower for senders who already hold the quality. Together, control and cost separate character conduct from a mere cue, information that leaks from a sender without being sent (Akstinaite et al., 2024).

If character operates as a costly signal, signaling theory also indicates where its associations with employee outcomes should first appear. Character signals are sender-referential because their content is the leader's own habits rather than a vision, a strategy, or the organization. Followers who decode such a signal are, in the same act, forming either a positive or negative evaluation of the person who sent it (i.e., an attitude). A decoded character signal should therefore appear first in the follower's evaluation of the supervisor. The next section develops the outcome cluster within which this ordering can be examined.

From independent employee outcomes to an interconnected cluster

We extend the logic of the CAN model beyond work attitudes (Carter et al., 2020; Dalege et al., 2016). Following this logic, we argue that classical employee outcomes in the organizational psychology literature are structured as a network of four interrelated construct classes. These constructs are work attitudes, psychological states, mindsets, and behavioral intentions, each with a distinct theoretical identity.

Job attitudes are psychological tendencies expressed by evaluating a particular entity with some degree of favor or disfavor, integrating cognitive and affective components that are practically inseparable in operation (Eagly and Chaiken, 1993; Judge and Kammeyer-Mueller, 2012). Within the work attitudes class, job satisfaction is the evaluative response to the job and its facets (Spector, 1997), whereas interpersonal workplace norms capture evaluative judgments about the behavioral standards governing respectful conduct in the relational work environment (Derayeh, 2008; Hershcovis, 2011; Martin and Cullen, 2006).

A second class comprises psychological states. Whereas, attitudes evaluate an object, states describe the motivational and emotional condition the person is in (Judge and Kammeyer-Mueller, 2012). Work engagement is the positive and fulfilling state of vigor, dedication, and absorption in one's work (Schaufeli and Bakker, 2004; Mazzetti et al., 2023). Subjective wellbeing is emotional, psychological, and social flourishing (Keyes, 2002). Both are psychological states in this sense.

A third class collects mindsets. Mindsets are binding forces that commit employees to courses of action relevant to a target (Meyer and Herscovitch, 2001). They differ from attitudes, which evaluate an object rather than direct conduct toward it. They also differ from psychological states, which describe a person's condition, whereas mindsets bind conduct to a chosen target. Recent reviews sometimes group mindsets, such as commitment, among job attitudes (Woznyj et al., 2022). We retain the mindset formulation because what this construct contributes to our cluster is binding rather than evaluation. Organizational commitment instantiates the class through three components (Meyer et al., 2002). Affective commitment rests on desire and is associated with autonomous motivation and positive affect. Normative commitment rests on felt obligation and is accompanied by pride or guilt. Continuance commitment rests on perceived cost of leaving and is associated with security anxiety (Meyer, 2014; Meyer et al., 2006). Profiles dominated by affective and normative commitment are associated with the highest-quality outcomes, a combination termed value-based commitment. Continuance-dominant profiles mainly predict compliance (Meyer, 2014).

Finally, a fourth class comprises behavioral intentions. Distinct from attitudes (which evaluate), states (which describe a condition), and mindsets (which bind action to a target), intentions capture a person's readiness to act (Ajzen, 1991). An intention reflects how much effort a person is willing to invest in a behavior, integrating evaluative and normative inputs. Intentions are motivational constructs rather than forecasts of what a person will do. Turnover intentions instantiate this class within our cluster. They are where an unfavorable evaluative attitude toward the job consolidates into readiness to leave. Tett and Meyer (1993) formalize that endpoint as the downstream consolidation of quit thoughts, job search activity, and comparisons of alternatives.

Table 1 summarizes the theoretical mechanisms and meta-analytic evidence for the pairwise relations among these classes. Under this perspective, it follows that work attitudes, psychological states, mindsets, and behavioral intentions form an empirically interconnected cluster rather than a set of independent endpoints. This cluster is conceptually distinct from the leader character dimensions that may relate to it. Within the cluster, some associations are likely to be more proximal and others more downstream. In this way, the table delineates a theoretically patterned space of pathways through which leader character may connect to employee functioning. The question for the present study is where leader character's associations enter this outcome cluster and how far they extend. The four pathways developed in the next section provide the theoretical rationale for alternative entry points and downstream patterns within it. These accounts inform the competing structural specifications estimated later.

Table 1. Meta-analytic benchmarks and network partial correlations for pairwise associations among focal outcomes.

A, Attitude; PS, psychological state; M, mindset; JS, job satisfaction; WEG, work engagement; TI, turnover intentions; WPN, interpersonal workplace norms; OC, organizational commitment; SWB, subjective wellbeing; JFA, job facets aggregation model (Spector, 1985); TAM, tripartite attitude model (Rosenberg and Hovland, 1960); TPM, Turnover Process model (Tett and Meyer, 1993); SIT, social identity theory as integrated with workplace commitment (Meyer et al., 2006); JD-R, job demands–resources model; COR, Conservation of Resources theory.

^†Life satisfaction proxy; Mazzetti et al. (2023) did not use Keyes's Mental Health Continuum directly.

^‡Martin and Cullen (2006) explicitly identify turnover intention as a research gap in the ethical work climate literature; no meta-analytic estimate is available.

^§Not all meta-analyses reported the mean true score correlation (ρ), so the metric column also includes the sample-size weighted mean observed correlation (r_w) when ρ is not reported.

^aCredibility interval; the confidence interval for this estimate is narrower than display precision.

The psychological mechanisms connecting leader character and employee outcomes

Leader character is relevant to all four construct types in the employee outcome cluster. Yet the mechanisms differ by construct type and the dependencies among them constrain how character's associations are organized across the cluster. The following four theoretical pathways develop the attitudinal, motivational, volitional, and decisional connections in turn.

The attitudinal pathway locates leader character's most immediate associations in employees' evaluative appraisals of their work context, beginning with the supervisor. Character's enactment supplies the observable conduct from which that appraisal forms. This is the entry point the signaling account predicts. Leadership behaviors that demonstrate consideration for employees are among the most powerful situational antecedents of job satisfaction (Judge and Kammeyer-Mueller, 2012). The dimensions that constitute leader character are one source of those behaviors. Each dimension gives employees something specific to decode. They infer competence and results orientation from perceptions of judgment and drive. They infer fair and consistent treatment from perceptions of justice and accountability. They infer genuine concern for their wellbeing from perceptions of humanity and humility. These inferences aggregate into job satisfaction (Spector, 1997).

Leader character is also associated with employees' evaluative judgments about the behavioral standards governing respectful conduct in their work unit. A leader who habitually enacts humility, justice, and humanity signals that respectful and collegial conduct is expected and enforced. That signal gives employees a behavioral reference point. Against it, each employee assesses what is and is not acceptable in the workplace (Hershcovis, 2011; Martin and Cullen, 2006). These individual assessments constitute interpersonal workplace norms as Derayeh (2008) operationalized them. They are shaped not through any single interaction but through the accumulated behavioral template that character-consistent leadership provides over time. Interpersonal workplace norms are therefore attitudinal in the same sense job satisfaction is, differing in target rather than in kind. Both reflect employees' direct evaluative response to the leader's habitual conduct.

The motivational pathway addresses character's relevance to psychological states. Through exemplary role modeling, leaders with strong character establish themselves as identification targets who embody values and behavioral standards that followers can internalize as self-determined guides for their own conduct (Monzani and Van Dick, 2020). Transcendence carries a purposive, future-oriented, and inspired orientation; drive carries passion and vigor. Followers who internalize these dimensions activate a corresponding motivational state in themselves. Transcendence's inspired element aligns with the motivational reading of leadership in the job demands–resources tradition, in which leadership fosters engagement by enhancing job resources (Mazzetti et al., 2023). Drive's vigor element energizes the vigor, dedication, and absorption that constitute the state (Schaufeli and Bakker, 2004). At the same time, the sense of meaning, purpose, and social contribution generated by this internalization process supports the emotional, psychological, and social flourishing that constitutes subjective wellbeing (Keyes, 2002; Ryan and Deci, 2000; Seijts et al., 2022). Both psychological states relate to character through a motivational route rather than as direct evaluative responses to it.

The volitional pathway addresses how leader character relates to employees' mindsets. The identification process associated with character does not stop at investment in the work role. It also accompanies deeper cognitive and emotional alignment with the organization as a collective. Leaders represent the organization in their followers' eyes. The values embodied in a leader's habitual conduct become the proximal signal through which employees calibrate their attachment to the organization itself. That calibration runs through the leader, regardless of whether the organization's formal values are equally virtuous. Meyer et al. (2006) distinguish identities grounded in shared values from identities grounded in mere membership. They propose that the first kind activates the desire (affective) and obligation (normative) mindsets behind value-based commitment. A leader whose habitual conduct embodies virtuous values gives employees such an identity to hold. Value-based commitment therefore captures the identity-based bonds associated with character-based leadership, rather than an evaluative or motivational response.

The decisional pathway concerns how leader character may relate to behavioral intentions, and what the turnover literature reports about their antecedents. The turnover process model argues that unfavorable job evaluations initiate a sequential withdrawal process in which dissatisfied employees generate exit cognitions, engage in job search activity, and eventually consolidate those cognitions into turnover intentions (Tett and Meyer, 1993). Job satisfaction is an attitudinal antecedent of this process. Satisfaction and organizational commitment each contribute independently to turnover intentions rather than one working through the other. Those intentions in turn mediate nearly all the attitudinal association with turnover itself.

The four pathways map onto four competing specifications. The attitudinal pathway puts an evaluative response closest to leader character in a parallel and a serial arrangement. The motivational and volitional pathways put psychological states and mindsets in that position. The decisional pathway nominates the endpoint rather than an entry because the turnover process model defines intention as the consolidation of prior evaluations. A fourth specification reverses that nomination and treats turnover intention as the proximal correlate, so the nomination is tested rather than assumed. We weigh all four against the conditional structure the network recovers.

Method

Sample

In all, 120 Ukrainian employees were invited to participate in an online survey (July–December 2021). Of these, 108 completed the survey in full (10% attrition), of which 14 were excluded before the analysis because of careless responding or failing attention checks. Therefore, the final analytic sample was 94 (61.7% female; modal age bracket 35–44 years, 39.4%; 66.7% private sector). Approximately 70.2% of respondents were in the Kyiv region and 27.7% in the Lviv region (Internet Protocol geolocation). Missing data on focal variables were minimal (0%−1.1%), with only sector affiliation showing any missing values. We used full-information maximum likelihood as the primary approach (Enders and Bandalos, 2001; Yuan and Bentler, 2000). Results were essentially unchanged under listwise deletion, so conclusions do not depend on the missing-data method (see Supplementary material).

Procedure

We administered the survey in Ukrainian through Qualtrics in 2021, recruiting currently employed adults aged 18 or older through professional networks in the Kyiv and Lviv regions. This convenience approach is efficient but not a probability sample, so we treat the findings as exploratory and limit generalization to similar employed populations. Each participant rated their direct supervisor's leader character using the Leader Character Insight Assessment (LCIA) and reported on their work context, job-related attitudes, and subjective wellbeing. An attention-check item embedded in the organizational commitment scale (“Please choose ‘Somewhat agree' for this item”) flagged careless responding.

Data collection during the COVID-19 pandemic relied on a single self-reported online source at one time point. This cross-sectional, single-source design cannot rule out endogeneity concerns such as omitted variables, simultaneity, and common-method variance (Antonakis et al., 2010). We therefore frame all associations as correlational and refrain from causal claims.

Study procedures followed established ethical principles for research with human participants (American Psychological Association, 2017), including voluntary participation, anonymity, and the right to decline responding to any question. All participants provided their informed consent by clicking on an item before the survey started. Similarly, all scales included a “decline to answer” option, which was treated as missing. An anonymized data set, analysis code, and Supplemental materials are available on the Open Science Framework at https://osf.io/2nwh5/.

Measures

All our measures were translated into Ukrainian using Brislin's (1980) approach. In short, first, the study content was translated into Ukrainian by a professional English translator. Then, the Ukrainian version was back-translated into English by a second independent translator. Finally, both translations were checked for inconsistencies.

Leader character (other-reported)

Participants were asked about the likelihood that their leaders would demonstrate 62 specific behaviors that evidenced character in a leadership position employing the LCIA (Crossan et al., 2014). Each behavioral indicator is classified into one of 11 dimensions shown in Figure 1, namely, Transcendence, Drive, Collaboration, Humanity, Humility, Integrity, Temperance, Justice, Accountability, Courage, and Judgment.

Sample items are “Holds and pursues high standards of performance” (Strives for Excellence–Drive) and “Let his or her accomplishments speak for themselves” (Modesty–Humility). The Likert-scale scores ranged from 1 (extremely unlikely) to 5 (extremely likely). Given the sample size, the 62 items were aggregated into 11 dimension-level parcels (Little et al., 2002), one per character dimension, which served as the indicators in all subsequent analyses. The scale also offered “don't know” and “decline to answer” options, which were treated as missing values.

Job satisfaction

A 35-item scale assessed nine facets of job satisfaction (Spector, 1985). The dimensions were satisfaction with pay, promotion, supervisor, benefits, contingent rewards, coworkers, nature of work, communications, and operating conditions, except nature of work, which consisted of three items, one of which was reverse-scored. The fourth nature of work item was inadvertently omitted when the online survey was set up in Qualtrics.

Items were rated on a 5-point Likert scale from 1 (Strongly disagree) to 5 (Strongly agree). Some example items are “I feel I am being paid a fair amount for the work I do” (pay), “Those who do well on the job stand a fair chance of being promoted” (promotion), “My supervisor is quite competent in doing his/her job” (supervision), “The benefits we receive are as good as most other organizations offer” (benefits), “When I do a good job, I receive the recognition for it that I should receive” (rewards), “I like the people I work with” (coworkers), “I like doing the things I do at work” (nature of work), “Communications seem good within this organization” (communications), and “I have too much paperwork” (operating conditions).

Interpersonal workplace norms

The Interpersonal Workplace Norms Scale is a nine-item unidimensional scale, which was employed to measure how employees perceived their workplace norms (Derayeh, 2008). The scale uses a 7-point Likert scale from 1 (strongly disagree) to 7 (strongly agree). Four items were reverse-coded. An example item is “There are generally no serious consequences for crude behavior in my department or business unit” (reverse-coded).

Organizational commitment

We measured organizational commitment with the 18-item revised Three-Component Model scale (Meyer et al., 1993). Affective, normative, and continuance commitment were each assessed using six items. Responses were rated on a 7-point Likert scale from 1 (strongly disagree) to 7 (strongly agree). An example item is “I feel a strong sense of ‘belonging' to this organization” (affective commitment).

Work engagement

The nine-item Utrecht Work Engagement Scale (Schaufeli et al., 2006) was used to measure three facets of work engagement, namely, vigor, dedication, and absorption. Responses are rated on a 7-point Likert scale, ranging from 1 (never) to 7 (always). An example item is “I am immersed in my work” (Absorption).

Subjective wellbeing

We used the 14-item short-form of the Mental Health Continuum (Keyes et al., 2008). The scale scores ranged from 1 (not true at all) to 5 (frequently, if not always). Sample items are “How frequently do you feel that... your life has a sense of direction or meaning to it” (Psychological Wellbeing), “you had something important to contribute to society” (Social Wellbeing), and “happy” (Emotional Wellbeing).

Turnover intentions

A four-item scale was used to measure turnover intentions (Chalykoff and Kochan, 1989). The scale used a 7-point Likert anchor ranging from 1 (Strongly Disagree) to 7 (Strongly Agree). An example item is “Before long, I will be leaving this organization.”

Analytical strategy

All analyses were conducted in R Version 4.4.3 (R Core Team, 2024), using the packages lavaan (Rosseel, 2012), qgraph (Epskamp et al., 2012), and BayesFactor (Morey and Rouder, 2024). The analytic strategy proceeded in five stages: (1) construct validity via confirmatory factor analysis and dimension-level network analysis; (2) bivariate associations between leader character and each employee outcome using frequentist and Bayesian methods; (3) psychometric network analysis at the dimension and construct levels to estimate the conditional dependency structure linking character to the outcome cluster (Costantini et al., 2015; Epskamp and Fried, 2018; Lauritzen, 1996); (4) theory-anchored, network-informed structural equation models comparing alternative directional specifications of that structure (Kline, 2023; Satorra and Bentler, 2001); and (5) robustness checks addressing common method variance (Podsakoff et al., 2003; Richardson et al., 2009), an alternative two-indicator commitment specification, demographic covariates (Becker, 2005; Bernerth and Aguinis, 2016), reverse-specified models, and a full-sample analysis. Full estimation details, including network regularization and stability diagnostics, structural model specification, and the robustness procedures, are reported in Supplementary material.

These five stages form one abductive sequence rather than five separate analyses. The Stage 3 network does the central work. It narrows the orderings worth testing by showing which conditional links are present and which are not. Psychometric networks have precedent in this literature. Carter et al. (2020) modeled job satisfaction at the facet level, and network models predicted turnover better than latent-variable models. We nonetheless retain a reflective model for character. Character is rated conduct rather than an attitude, so the network critique of latent attitude models does not transfer. Stage 4 then translates that structure into competing directional accounts. Theory supplies each account's entry point, and the network supplies the downstream arrangement. This logic is neither purely confirmatory nor purely data-driven. The network constrains theory, and theory interprets the network, while the competing models adjudicate accounts that theory alone would leave undecided.

Results

Table 2 reports means, standard deviations, and Pearson correlations in the lower diagonal. EBICglasso partial correlations (Chen and Chen, 2008; Foygel and Drton, 2010; Friedman et al., 2008) are shown in the upper diagonal. Cronbach's alphas are shown on the main diagonal.

Table 2. Descriptive statistics, reliabilities, zero-order correlations, and EBICglasso partial correlations with bootstrap 95% confidence intervals.

N = 94. Cronbach's alpha on the diagonal in parentheses. Lower triangle: zero-order Pearson correlations. Upper triangle: edge weights (w), the partial correlations from the construct-level EBICglasso network at γ = 0.50 (non-thresholded; for comparison with the lower triangle), with 95% bootstrap confidence intervals from 10,000 non-parametric resamples (Epskamp et al., 2018). Partial correlations whose confidence interval excludes zero are shown in bold; those whose confidence intervals includes zero are shown in italics. Em-dashes in the upper triangle indicate edges shrunk to zero in the regularized network. Gender and Sector were not included in the network.

^*p < 0.05; ^**p < 0.01; ^***p < 0.001.

Stage 1. Is there a coherent signal to decode?

The shared confirmatory factor analysis showed adequate fit to proceed with the focal measurement model, and all four latent factors met conventional convergent and discriminant validity standards (Anderson and Gerbing, 1988; Fornell and Larcker, 1981; Henseler et al., 2015). The higher-order character factor was strongly saturated and clearly separable from the three employee-state factors of work engagement, value-based commitment, and subjective wellbeing, which supports treating character as a composite in the analyses that follow. Full standardized loadings, composite reliability, average variance extracted, the inter-factor correlation matrix, and the small-sample interpretation of the root mean square error of approximation (RMSEA) are reported in Supplementary material (Tables A1, A2).

Character-only network analysis

The character-only network (11 nodes, EBICglasso, γ = 0.50, thresholded; see Figure 2) was fully connected after regularization, with every dimension retaining at least one edge. The topology was identical across γ ϵ {0.25, 0.50, 0.75}, indicating that the connectivity pattern is robust to the choice of sparsity tuning parameter. Descriptively, Judgment had the highest point estimate for node strength among the 11 dimensions, although this observation cannot be treated as a reliable ranking (see Figure 3). Bootstrapped difference tests showed that no centrality advantage was statistically reliable, with all 95% confidence intervals including zero [range (−1.06, 0.35)], and the correlation stability (CS) coefficient for strength was 0.13, below the 0.25 minimum interpretability threshold (Epskamp et al., 2018). We report the ordering descriptively and draw no inferences from centrality rankings in this sample.

In a network this dense, large stable differences in node centrality are not expected. The primary finding is full connectivity rather than the dominance of any single dimension. That absence is itself theoretically informative. Full connectivity is what the framework's systemic claim predicts, and that claim does not require any dimension to dominate. This pattern corroborates the higher-order structure identified in the confirmatory factor analysis (CFA; composite reliability [CR] = 0.98, average variance extracted [AVE] = 0.82) and aligns with the dense interconnectivity reported by Monzani et al. (2021).

Community detection on the character-only network (walktrap, Louvain, and fast-greedy algorithms) yielded three sub-clusters, with substantial agreement on three robust pairings (Justice–Integrity, Humanity–Humility, Drive–Temperance) but unstable boundaries given the limited power of community detection at N = 94. Substantively, all 11 dimensions remained part of a single overall network (see Supplementary material). These results justify using a leader character composite in Stages 2 and 3 and as a latent factor in Stage 4. A coherent, densely interconnected character signal exists on the sender side. Whether the employee outcome cluster registers it, and where, is the next question.

Stage 2. How far do character's associations reach?

We regressed each of the 24 outcome scores, comprising the six construct composites and their 18 subscales, on overall leader character. Leader character (overall) was significantly related to all outcomes, with the strongest effects on job satisfaction, interpersonal workplace norms, and the commitment and work engagement facets, and the weakest on continuance commitment and subjective wellbeing facets (see Figure 4).

Bayes factors for the dimension-level correlations (Ly et al., 2016) corroborated this pattern, with decisive evidence for links between all 11 character dimensions and job satisfaction and substantial to decisive evidence for work engagement and organizational commitment (see Figure 5; Supplementary material).

This breadth is informative about magnitude and reach, and uninformative about structure. A positively framed leadership construct relates positively to desirable outcomes and negatively to undesirable ones almost by design (Alvesson, 2020). A pattern in which every pathway looks supported cannot say which pathway carries the association. The outcomes are also related to one another (Table 2), so any single association may be borrowed from a neighboring construct. The network analysis addresses this next.

Stage 3. Which associations survive joint modeling?

Relative to the broad pairwise associations summarized in Table 2, Stage 3 yielded a narrower conditional structure once the full outcome cluster was modeled jointly. Community detection on the non-thresholded 31-node extended network converged on three communities across all three algorithms, with the 11 character dimensions clustering together and separating from most of the outcome variables (Figure 6). This grouping supports the Stage 1 result.

The network findings admit a loose and a strict reading. Read loosely, the non-thresholded panel kept weak direct edges from character to several outcomes, including value-based commitment and work engagement. Turnover intention was the only outcome not connected to character in either panel. Read strictly, regularized thresholding removed all but one. The thresholded construct-level network retained 10 edges, with topology stable across all γ values. After regularization, job satisfaction was the only construct retaining a direct edge with leader character; every other direct edge shrank to zero (see Figure 7).

Beyond that single link, the network separated into a psychological state–mindset region and an attitude–behavioral intention region, consistent with the findings shown in Table 1. At the dimension level, the same concentration appeared at a single facet. In the 31-node network, the satisfaction with the supervision subscale clustered with the 11 character dimensions and bridged them to the remaining outcomes (see Supplementary material). The network fixes where character associations enter the cluster. It cannot fix how the remaining construct types are ordered downstream of that entry because its edges are undirected. The structural models take up that question.

Stage 4. How is the cluster ordered?

The structural models address the ordering question the network left open. Four models were estimated and compared. Because the data are cross-sectional, we interpret the indirect effects as a decomposition of observed associations under an assumed ordering rather than as estimated causal mediation (Antonakis et al., 2010).

Each model was anchored in the entry points developed in the Theoretical Background section. Model A (Attitudinal–Parallel) specified job satisfaction and interpersonal workplace norms as parallel proximal mediators of leader character. Model B (Attitudinal–Volitional–Motivational) specified job satisfaction as the sole proximal correlate, with both commitment forms downstream of satisfaction, work engagement downstream of commitment, and subjective wellbeing downstream of engagement. Interpersonal workplace norms and turnover intention followed from satisfaction, with turnover intention also fed by value-based commitment. This ordering follows Meyer et al. (2006), who hold that commitment develops from identification and carries, at least in part, its effects on motivation. The reverse ordering, which places engagement ahead of commitment, remains defensible on motivational grounds and is retained in Supplementary material as Model B1 (Attitudinal–Motivational–Volitional).

Model C (Motivational–Volitional–Parallel) specified work engagement and value-based commitment as parallel proximal mediators, corresponding to the motivational and volitional entry-point accounts. Model D (Intention–Entry) reverses the decisional pathway's nomination and treats turnover intention as the proximal correlate, with satisfaction and the commitment-first chain downstream. Theory disfavors this arrangement; estimating it lets the data speak to that claim rather than assume it. Two serial orderings of the motivational family, Models C2 and C3, are sensitivity specifications in Supplementary material, where neither improves on the attitudinal-entry models. Table 3 reports the χ² test, the χ²/df ratio, and derived goodness-of-fit indices (Marsh et al., 2004).

Table 3. Structural equation model fit indices across specifications.

N = 94. Focal models: A = job satisfaction and workplace norms as parallel mediators; B = job satisfaction as the sole proximal mediator; C = work engagement and value-based commitment as parallel mediators. The Commitment measurement structure block compares how the three commitment components are modeled, whether as a single factor, three correlated factors, or a value-based factor (affective and normative) with continuance as a separate construct. ULC, unmeasured latent common method factor. χ² statistics are Satorra-Bentler scaled (Satorra and Bentler, 2001) under MLR estimation. χ²/df ratios below 3 indicate acceptable fit and below 2 indicate good fit. Δχ² difference tests are Satorra-Ben ed. AIC and BIC in the Discriminant validity and Commitment measurement structure blocks reflect CFA-only specifications and are not directly comparable to those in the Focal and Robustness blocks. Within the Commitment measurement structure block the three rows share identical observed variables, so their AIC and BIC are directly comparable, and the value-based specification attains the lowest values. Information criteria are comparable only among models that share the same observed variables, so the rows with demographic covariates should be read within their own grouping rather than against the focal models because the covariate models add gender and sector to the modeled variables. Em-dashes, not applicable; CFA, confirmatory factor analysis; CFI, comparative fit index; TLI, Tucker-Lewis index; RMSEA, root mean square error of approximation; SRMR, standardized root mean square residual; AIC, Akaike information criterion; BIC, Bayesian information criterion; MLR, robust maximum likelihood estimation.

^***p < 0.001.

Model A's parallel claim failed on its second branch. The job satisfaction pathway was supported, but neither the path from leader character to workplace norms nor the paths from workplace norms to subjective wellbeing and turnover intention reached significance. Model A's direct effects on the distal outcomes were likewise non-significant (Table 4). All supported character-outcome associations operated through job satisfaction.

Table 4. Structural path estimates across Models A, B, C, and D.

N = 94. Stage 1 = character-to-mediator paths (a); Stage 2 = mediator-to-outcome paths (b); Stage 3 = direct character paths (c′). Model B specifies no direct character paths; its serial mediator paths appear in Stage 2. Model D's estimates are reported for completeness; its global misfit precludes interpretation. Model B1 is reported in Supplementary material; covariate sensitivity in Table 6, Panel B. b = unstandardized estimate; β = STDYX standardized estimate; confidence intervals are robust under MLR. Dashes = path not specified. CHAR = leader character; JS, job satisfaction; VBC, value-based commitment; CC, continuance commitment; WEG, work engagement; SWB = subjective wellbeing; TI, turnover intention; WPN, interpersonal workplace norms.

^*p < 0.05; ^***p < 0.001. ns, not significant (p>0.05).

Model B fit better than Model A on five additional degrees of freedom, and every one of its nine structural paths was significant (Table 4). Job satisfaction related positively to value-based commitment and negatively to continuance commitment. The commitment forms related to engagement, positively from the value-based and negatively from the continuance form. Engagement related to subjective wellbeing, and satisfaction carried the associations with workplace norms and turnover intention, the latter with a second route through value-based commitment. All bootstrapped indirect effects excluded zero (Table 5). Explained variance ranged from 0.14 for continuance commitment to 0.60 for work engagement.

Table 5. Standardized indirect and total effects of leader character on outcomes across Models A, B, and C with bootstrap 95% confidence intervals.

N = 94. Cells show standardized indirect and total effects with bias-corrected bootstrap 95% confidence intervals from 10,000 resamples (MacKinnon et al., 2004). Model B retains no direct character paths, so each of its total effects equals its total indirect effect. Model D is excluded; its global misfit precludes interpreting indirect effects. Model B1 is reported in Supplementary material. Asterisks = bootstrap CI excludes zero. Dashes = path not specified. CHAR = leader character; JS, job satisfaction; VBC, value-based commitment; CC, continuance commitment; WEG, work engagement; SWB, subjective wellbeing; TI, turnover intention; WPN, interpersonal workplace norms.

^***p < 0.001.

The motivational entry fared worst among the entry accounts. Model C fit worse than Models A and B on both information criteria and on the comparative fit index (CFI), the Tucker-Lewis index (TLI), and RMSEA, although its standardized root mean square residual (SRMR) was marginally lower. Both a-paths from leader character to the proximal mediators were significant, but five of eight mediator-to-outcome paths were not. Among these was a negative coefficient from value-based commitment to subjective wellbeing. How commitment was specified did not account for this weaker performance. The outcome corroborates the network's elimination of a motivational entry point rather than merely restating it.

Model D failed as theory predicted, and in an instructive way. Every one of its eight structural paths was significant, including the entry path from character to intention (β = −0.37, p < 0.001). The arrangement still could not reproduce the cluster's covariances, χ²(224) = 421.0, CFI = 0.912, RMSEA = 0.096, SRMR = 0.17, trailing Model B by 49 Akaike information criterion (AIC) and 42 Bayesian information criterion (BIC) points. Character explained 14% of the variance in intention as an entry, against 48% for satisfaction in Model B. The misfit concentrated exactly where the network located the doorway. The largest residual correlations sat between the character dimensions and job satisfaction (0.36–0.46), the association Model D forces through the intention bottleneck. Significant paths do not certify an arrangement. The decisional pathway describes where the cluster ends, not where character enters it.

Taken together, the four-model comparison identified Model B as the most parsimonious specification and the one most consistent with the thresholded network result that job satisfaction retained the only direct edge with leader character (see Figure 8). Two cautions attach to that consistency. Model B's structure aligns one-to-one with the network estimated on this same sample, so part of its fit advantage reflects that dependence. Its path from continuance commitment to engagement rests on an edge that is absent in the full sample. The alternative ordering, Model B1, fit worse on every index; the comparison is reported in Supplementary material. Whether this structure survives variation in method, measurement, covariates, causal direction, and sample composition is the remaining question.

Stage 5. Does the structure hold?

We conducted five robustness checks, with full results in Supplementary material. The continuance-excluded specification and the demographic-covariate models appear alongside the focal models in Table 3. Table 6 summarizes the common method variance, sample sensitivity, and reverse-causality analyses. Across all five robustness checks, the structure recovered in Stage 4 held.

Table 6. Robustness summary: common method variance, sample sensitivity, and reverse-causality fit.

N = 94. CMV, common method variance. ULC, unmeasured latent common method factor. Panel A: focal vs. ULC specification; Δ = ULC – focal. Panel B: primary sample (N = 94) vs. full sample (N = 108); Δ = full – primary. Panel C: reverse specifications, estimated on the earlier three-indicator commitment block (Supplementary material). ✓ = direction and significance verdict (α = 0.05) retained under the alternative; No = verdict changed. Attenuation alone does not flag a path; the ULC is conservative and absorbs shared variance. In Model C the engagement-to-wellbeing coefficient exceeds 1 under the ULC, a suppression artifact consistent with that model's weaker fit. CHAR, leader character; WEG, work engagement; JS, job satisfaction; VBC, value-based commitment; CC, continuance commitment; SWB, subjective wellbeing; TI, turnover intention; WPN, workplace norms.

Model B carried its advantage through the method-factor and covariate checks, with two qualifications reported in Supplementary material. Its commitment-to-turnover indirect effects lost significance once demographic covariates entered, and the continuance-to-engagement path is the sample-sensitive one noted earlier. The reverse specifications fit worse throughout. The entry point proved more stable than any ordering. In every specification we estimated, job satisfaction remained the proximal link between leader character and the rest of the cluster. This pattern corroborates the attitudinal entry point without demonstrating it.

Discussion

The present study abductively explored how leader character's associations with six employee outcomes are structured using a Ukrainian sample shaped by the compounding pressures of the COVID-19 pandemic and ongoing geopolitical disruption. Combining psychometric network analysis with competing structural equation models, we recovered the conditional dependency structure of the focal constructs. We then adjudicated among four theoretically grounded specifications of how character's associations are organized within that structure.

Leader character was positively associated with job satisfaction, interpersonal workplace norms, work engagement, subjective wellbeing, and value-based commitment, and negatively associated with continuance commitment and turnover intention. A serial specification was preferred, with job satisfaction as the proximal correlate through which leader character's associations extended to the broader outcome cluster. Of its two orderings, Model B placed both commitment forms between satisfaction and work engagement. Leader character was related to turnover intention indirectly rather than directly, with its negative association running through job satisfaction once the other outcomes were modeled jointly. By contrast, interpersonal workplace norms did not retain a direct link to leader character once job satisfaction was accounted for. These results contribute to the leadership literature in three ways.

The first contribution is theoretical, and it spans both sides of the signaling exchange. On the sending side, we repositioned leader character as an antecedent of authentic signaling, supplying the normative content and the cost structure that concordance-based accounts leave unspecified (Lux and Lowe, 2024). On the receiving side, we modeled employee outcomes as a target-organized cluster rather than as independent endpoints, extending the CAN logic beyond work attitudes (Carter et al., 2020; Dalege et al., 2016). The two halves met in the data. When the six outcomes were modeled jointly, leader character's associations concentrated in satisfaction with the supervisor, the facet whose target matches the sender. The broad associations reported in one-at-a-time designs may largely reflect one target-matched association carried across an unmodeled cluster.

The second contribution is methodological. Network structure can discipline the choice among outcome orderings that theory leaves open, a common situation in a fragmented field with many overlapping frameworks (Meuser et al., 2016). Pairing psychometric network analysis with competing structural models narrows the specifications worth testing. Theory then decides among the few that remain. The strategy is portable to other construct systems for which theory does not fix the ordering.

The third contribution is contextual. We extend the LCF's evidence base beyond its original context. The dimensional network and its associations with the employee outcome cluster in Ukraine in 2021 resemble patterns previously reported in WEIRD samples. We did not test this resemblance formally, so we read it as suggestive rather than as evidence of invariance. This extension is deliberately bounded. It does not establish that high-demand settings amplify character's associations because that claim requires comparison against a low-demand benchmark. Theory gave us grounds to expect amplification. Tang and Crossan (2026) predict, for strategic leaders' behavior, that character matters more where norm enforcement is weak. Signaling theory converges on the same expectation from a different direction. Signal cost is what separates honest from dishonest emission (Connelly et al., 2011). A context that penalizes virtuous conduct raises that cost. Character is also theorized to supply the agency needed to act against such pressure (Crossan et al., 2023). We apply these claims to supervisor-level perceptions without testing them directly.

Relation to prior findings

Monzani et al. (2021) modeled the 11 dimensions of character with subjective wellbeing, resilience, organizational commitment, and work engagement. Mindset-specific ties (drive–affective commitment; judgment–continuance commitment) made organizational commitment the node connecting leaders' character ratings to the other outcomes. In our study, the bridging role then moves from a mindset (commitment) to a work attitude (job satisfaction), with value-based commitment, work engagement, and subjective wellbeing now following from job satisfaction. The two findings are complementary, as the mindset bridge Monzani et al. (2021) recovered is itself nested within a broader attitudinal structure that becomes visible only when work attitudes are entered into the extended construct network.

The weaker performance of the motivational–volitional specification (Model C) aligns with independent evidence that character's motivational correlates are unstable. Seijts et al. (2024) found that motivation to lead accounted for the association between character and some coping responses. The direct association with active problem-solving persisted even after controlling for the Big Five and psychological capital. In Model C, leader character was related to motivational mediators, but their paths to other outcomes were largely non-significant.

Both serial specifications order leader character upstream of job satisfaction within the turnover process. Tett and Meyer (1993) and Hur and Abner (2024) identify job satisfaction as an attitudinal antecedent of withdrawal. When the other outcomes are modeled jointly, our model represents leader character's negative association with turnover intention as indirect, through job satisfaction. To our knowledge, no prior LCF study has examined turnover intention as an outcome. Across these three comparisons, job satisfaction emerges as the stable attitudinal correlate, while mindset and motivational correlates play a less consistent role.

Theoretical implications

Leader character is theorized as foundational to the broader family of positive leadership styles, which also includes authentic, transformational, and servant leadership (Monzani and Van Dick, 2020; Sturm et al., 2017). Our findings bear on that position through a connected sequence. Character's enactment functions as a concordant signal in Lux and Lowe's (2024) sense. We extend their account by adding a cost condition they do not specify. A signal that is expensive to produce without the underlying quality is an honest indicator of a characteristic of the signaler (Akstinaite et al., 2024). Our results show where that signal registers in followers' attitudes, and that the pattern appears outside the settings in which the framework was built. We describe the sequence in the order our models recovered, not because our design identifies it. Cross-sectional and single-source data cannot establish that ordering (Antonakis et al., 2010).

Where the character signal registers

Our signaling premise carries a directional assumption about where character's associations should appear. Character signals are sender-referential because their content is the leader's own habits. Followers who decode such a signal are also evaluating the person who sent it. Such evaluation should be where the signal first registers. When the outcome cluster was modeled jointly, character's associations concentrated at the facet whose target matches the leader, satisfaction with the supervisor. Leader character's remaining direct associations with the distal outcomes were weak or non-significant once job satisfaction entered the model.

The finding is also more informative than a broad association between leader character and job satisfaction would have been. Perceived character and satisfaction with the supervisor are strongly related, yet they remain conceptually distinct. Character is the broader construct, a multidimensional perception of the leader's virtues, whereas satisfaction is a single evaluative summary of a role (e.g., supervisor) within a specific context (e.g., organization). The two are separable in content, and the signal still lands on the narrower one (Supplementary material). The content that the character signal carries is virtue, and virtue is valued across cultures and historical epochs (Dahlsgaard et al., 2005). The implication is that perceived virtue pays off relationally, by being associated with other positive work attitudes, mindsets, psychological states, and behavioral intentions. Because character is rated by followers, that payoff attaches to being perceived as virtuous, not to virtue that goes unobserved.

The attitude literature predicts the same concentration. Attitudes relate most strongly to outcomes that match them in target (Fishbein and Ajzen, 2011). Work attitudes are themselves organized by target. Employees hold distinguishable attitudes toward the organization, the supervisor, and coworkers (Lavelle et al., 2009; Meyer et al., 2006). Carter et al. (2020) further distinguish symbolic from instrumental facets of job satisfaction. Symbolic facets, including satisfaction with the supervisor, evaluate a person rather than a resource. Signaling theory and attitude theory converge here from different premises. We note in the Limitations section that this principle also supplies an alternative to our reading.

The rest of the cluster sits downstream of that evaluation. In Model B, job satisfaction related positively to value-based commitment and negatively to continuance commitment. Both commitment forms, in turn, related to work engagement, positively from the value-based form and negatively from the continuance form. Engagement carried the association onward to subjective wellbeing. Job satisfaction also carried the negative association with turnover intention, with a second route running through value-based commitment. Interpersonal workplace norms likewise followed from satisfaction rather than from character directly. The ordering within this chain is less settled than the chain itself, and the alternative ordering (Model B1) cannot be excluded by a cross-sectional design. The cluster behaves as a chain rather than a fan, with one target-matched attitude doing the connecting work.

Three lines of evidence converge on this account. The network, which imposes no direction, left character one doorway into the cluster, at job satisfaction. The model comparison, which imposes direction, preferred the specification that walks through that doorway and orders commitment before engagement, the direction Meyer et al. (2006) propose. The two arrangements that contradict the account failed in the predicted ways. The motivational entry lost its downstream paths. The intention entry could not reproduce the cluster, and its largest residuals fell on the very character-satisfaction association the network had isolated. Each method is fallible in a different way, so their agreement carries more weight than any method alone. The convergence corroborates the account without demonstrating it.

Evidence from a non-WEIRD context

Our signaling account rests on a premise about decoding. A signal works only if receivers share a rubric for reading it. We argued that the framework's dimensions are universal at the level of the abstract virtue. The framework locates the adaptation to local conditions in Judgment (Crossan et al., 2017). That allows structure to hold where enactment differs. In 2021, Ukraine let us put weight on the premise. A coherent attitudinal bridge emerged, together with character-dimension interconnectivity consistent with Monzani et al. (2021) findings. Both indicate that the structure holds in a markedly different environment. We read this as consistency with the decoding premise, not as a demonstration of it. One national sample cannot establish generality. Whether high-demand environments amplify character's associations is a separate question. Answering it requires benchmarks from lower-demand environments and falls outside the scope of the present data.

Practical implications

The recovered ordering has practical implications for how organizations act on leader character. We frame our findings within the three human resource (HR) practice areas that Seijts et al. (2017) identified, namely, selection, performance management, and talent development. A signaling account maps onto these three areas. Selection screens the signal, performance management reads whether and how followers decode it, and development invests in its production cost. We frame these as implications of our associational results rather than causal claims.

Embed character in manager selection alongside competence

Because leader character came first in the chain our results recovered, selection and promotion are the earliest point at which an organization can act on it. Competence-based systems built around behavioral event interviewing, assessment centers, and structured competency models are well established. Still, they assess what a candidate can do rather than who the candidate is. Character-based selection, as piloted by the Canada Revenue Agency (Crossan et al., 2021a), shows that character can be embedded into job profiles, screening criteria, structured interviews, and reference checks alongside competency assessment rather than replacing it.

Use satisfaction with the supervisor to calibrate character-based feedback

Seijts et al. (2017) identify performance management as a second practice in which to embed character, particularly through behaviorally anchored feedback on character-relevant conduct. Satisfaction with the supervisor co-clustered with managers' character and was the attitude most closely tied to the other outcomes. As a result, standard engagement surveys already collect the relevant information in their supervisor-satisfaction items. When a full assessment of managers' character is not feasible, HR can read employees' satisfaction with supervision as an indicator of how direct reports experience their managers' character. Where LCIA-based assessment exists, these items can complement it in appraisal conversations.

Invest in developing managers' character to support employee outcomes

The dominant HR paradigm treats work engagement, organizational commitment, and retention as direct intervention targets, funding engagement programs, retention bonuses, and culture campaigns aimed squarely at them. The recovered ordering suggests this sequence is inverted. Those outcomes tracked job satisfaction, and specifically satisfaction with the supervisor. Programs aimed at lifting engagement and commitment without also addressing satisfaction with the manager and the job are therefore likely to be less effective. In practical terms, an engagement program competes with the employee's daily experience of the manager. Gains are unlikely to last while that experience stays the same. The most direct investment is the manager through character-focused development and supervisory coaching (Seijts et al., 2017).

Limitations

First, the data set is single-source and cross-sectional. Pandemic constraints precluded multisource or experimental designs. Leader character was rated by one follower per supervisor. The construct thus reflects followers' perceptions of supervisor character rather than a multisource aggregate. The single-source design also raises common-method and reverse-causation concerns. The reverse-causation models fit worse than Model B, yet cross-sectional fit comparisons cannot fully separate near-equivalent orderings. Experimental or quasi-experimental designs could establish causal relations (Antonakis et al., 2010). Woznyj et al. (2022) call for the causal attitude network model to be applied across time. Therefore, future research should conduct a longitudinal replication, adding a marker variable and an a priori measured cause variable (Williams and McGonagle, 2016).

Second, the focal finding has an alternative reading that our design cannot exclude. Leader character and satisfaction with the supervisor share their source and target, both reported by the same employee about the same supervisor. Target matching alone predicts that character's association concentrates at the supervisor-directed facet. Our signaling account predicts the same concentration for a different reason. The two readings are observationally equivalent for this edge; we do not claim to have separated them. An extended measurement model quantifies this entanglement (Supplementary material). Character and overall job satisfaction remained separable at the construct level. Character and satisfaction with the supervisor did not separate. All structural models operate at the construct level; the entanglement bears only on the descriptive network. Target matching is also silent on the serial structure that follows, in which satisfaction with the supervisor connects to engagement, commitment, wellbeing, and turnover intention. Separating the two accounts would require an outcome sharing the leader's target but not the rater's or a design that manipulates the signal.

Third, the sample came from a single national context during acute disruption and from organizations recruited through professional networks (Seijts et al., 2025a,b). Even so, generalization to other contexts, sectors, and stable conditions will require replication. Employees who agree to rate a supervisor likely hold a favorable view of that supervisor. Recruitment may thus have compressed variance at the low end of the character distribution. It may also underrepresent employees in small or informal firms.

Fourth, the LCIA operationalizes the virtuous expression of each dimension, not its excess or deficient forms. Our models recover the degree of virtuous enactment that followers report, not whether that degree was right for the situation. Non-monotonicity remains a property we claim for the framework rather than one these data test. Future research will need instruments that directly score excess and deficiency.

Finally, N = 94 limits the network and structural analyses. The strength-centrality CS coefficient (0.13) fell below the 0.25 interpretability threshold, and the betweenness coefficient was 0.00. This leaves the framework's prediction that Judgment coordinates the other dimensions (Crossan et al., 2017, 2024) untested here rather than unsupported. Judgment's first-place ordering on strength was invariant to the regularization setting and including the excluded cases. Invariance to a tuning parameter is not stability under resampling, and only the latter licenses a ranking claim. A larger sample is required to adjudicate either prediction.

Conclusion

In a Ukrainian sample surveyed in 2021, far from the WEIRD settings where the framework was built, job satisfaction retained the only direct link with perceived leader character. The network narrowed the orderings, and theory chose among those that remained. That pattern speaks to the debate this article opened with. Critics argue that positive leadership research conflates what leaders do with how followers evaluate them. Our findings sit close to that charge because the association we recover is itself an evaluation of the leader.

Signaling theory keeps the terms apart. It separates the habit a leader has cultivated, the conduct through which that habit becomes observable, and the attribution a follower forms. Character supplies the content of the signal; authenticity is the attribution that follows. Measuring the received signal and the receivers' responses to it does not dissolve the conflation problem, but it names which part of it we observed.

One limitation marks the next step. A single cross-sectional sample of 94 employees cannot establish the order in which these evaluations form; doing so would require repeated measurement and a rater separate from the outcome. The receiving side of leader character has been assumed more often than examined. Our Ukrainian data suggest that it is worth examining.

Generative AI statement

The author(s) declared that Generative AI was used in the creation of this manuscript. Generative AI assisted in writing and refining the R code used for the data analysis and in editing and drafting the manuscript text. The author(s) designed the study, made all analytic decisions, verified all results, and take full responsibility for the content.

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.

Supplementary material

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/forgp.2026.1900531/full#supplementary-material

Supplementary material Supplementary material at the source

Footnotes

Edited by: Rashimah Rajah, University of Twente, Netherlands

Reviewed by: Zu Chenxi, Third Affiliated Hospital of Sun Yat-sen University, China

Fidan Alhas, Independent Researcher, Kars, Türkiye