Problem #1: Interrelated methodological issues

This model for conceptualizing ineffective leadership collapses antecedents and consequences into a single construct, making it difficult to isolate what leadership actually is. This reflects an ongoing pain point in leadership research, where constructs often conflate leader behaviors and perceptions with their outcomes, thereby undermining construct validity and producing tautological relationships (Antonakis et al., Reference Antonakis, Bendahan, Jacquart and Lalive2010; Fischer et al., Reference Fischer, Dietz and Antonakis2024). When relational states such as trust or psychological safety are embedded in the definition of leadership, empirical relationships become inflated and conceptually circular. Thus, distinguishing forms of leadership through psychological safety and trust may inadvertently reinforce existing measurement problems rather than resolve them.

For instance, this conflation increases the risk of confounding and spurious correlations, as omitted variables, such as actual leader behaviors or contextual factors, may simultaneously influence both perceptions of leadership and relational outcomes (Kotzian, Reference Kotzian2022; Podsakoff et al., Reference Podsakoff, MacKenzie and Podsakoff2016; Wilms et al., Reference Wilms, Mäthner, Winnen and Lanwehr2021). Such conflation undermines construct validity because the same indicators are used to represent both predictors and outcomes, producing tautological relationships that cannot be meaningfully tested (Antonakis et al., Reference Antonakis, Bendahan, Jacquart and Lalive2010; Fischer et al., Reference Fischer, Dietz and Antonakis2024). Empirically, this issue is especially problematic because measures that embed outcomes (e.g., trust, psychological safety) into the definition of leadership will almost inevitably correlate with those same outcomes, inflating effect sizes and creating the illusion of explanatory power and incremental validity. We have seen this same concern arise for other leadership constructs, such as transformational and charismatic leadership; however, we continue to make the same mistake (van Knippenberg & Sitkin, Reference van Knippenberg and Sitkin2013). Ultimately, these issues also severely limit external validity (Messick, Reference Messick1995). If leadership is defined through follower perceptions that are themselves context-dependent, then findings may not generalize beyond the specific context in which they were observed. This aligns with broader critiques that leadership constructs lacking behavioral specificity cannot be reliably transported across settings because they do not capture stable, observable phenomena (Fischer & Sitkin, Reference Fischer and Sitkin2023).

Further, the framework is vulnerable to Goodhart’s law: When relational metrics like psychological safety become targets, leaders may optimize for these indicators rather than engaging in effective leadership behaviors, thereby weakening their validity as measures of leadership itself. Once these outcomes are turned into performance metrics, they can be gamed, as relationships observed in data often break down when directly controlled (Chrystal & Mizen, Reference Chrystal, Mizen and Mizen2003). A clear example comes from education, where equating test scores with learning leads teachers to “teach to the test,” inflating scores while narrowing actual learning. This reflects the broader tendency to encourage “A” while rewarding “B” (Kerr, Reference Kerr1995). Similarly, leaders may engage in impression management or avoid necessary but difficult behaviors (e.g., critical feedback, performance pressure) to maintain perceptions of trust or psychological safety. As a result, these constructs may become decoupled from actual leadership effectiveness, reinforcing the concern that the model classifies leaders based on desirable outcomes rather than the underlying behaviors it aims to capture.

Problem #2: Loss of temporal order and causality

Another issue involves distorted causal effects when considering alternative pathways to these outcomes. Because these variables are mutually reinforcing and typically measured cross-sectionally, from the same source and at the same time, and the framework is highly susceptible to endogeneity, making it unclear whether leadership drives trust and psychological safety or vice versa (Antonakis et al., Reference Antonakis, Bendahan, Jacquart and Lalive2010; Wilms et al., Reference Wilms, Mäthner, Winnen and Lanwehr2021). Indeed, trust, psychological safety, and/or even general liking may influence whether or not one views their leader as competent in the first place. This violates basic assumptions of causal inference and can produce biased and inconsistent estimates, a concern widely noted in leadership research relying on perceptual survey data (Antonakis et al., Reference Antonakis, Bendahan, Jacquart and Lalive2010). This risks circularity, where you end up saying a leader is toxic because trust is low, and trust is low because the leader is toxic. For example, if toxic leadership is defined as that which erodes trust, and trust is then used as evidence of toxicity, the constructs can collapse into one another. The theoretical contribution, then, depends on clearly specifying whether trust and psychological safety are indicators, mechanisms, or defining properties.

Second, failing to account for instrumental leadership behaviors can inflate the apparent effects of other leadership styles, demonstrating how omitted variable bias distorts conclusions about what “works” in leadership (Wilms et al., Reference Wilms, Mäthner, Winnen and Lanwehr2021). Teams may experience high psychological safety due to strong peer relationships, shared norms, or organizational climate, independent of leader behavior. In such cases, attributing psychological safety to leadership risks overstating leader influence and obscuring other causal mechanisms (Antonakis et al., Reference Antonakis, Bendahan, Jacquart and Lalive2010; Meindl et al., Reference Meindl, Ehrlich and Dukerich1985). Although Ducheyne and De Fruyt’s also distinguish incompetent and toxic leadership through competence, benevolence, and integrity, their conceptualization of leadership itself centers on psychological safety and follower trust. According to them, a leader can be incompetent independent of relational perceptions, but we cannot forget that trustworthiness encompasses competence, and competence is a predictor of both trust and psychological safety (Colquitt et al., Reference Colquitt, Scott and LePine2007; Ducheyne & De Fruyt Reference Ducheyne and De Fruyt2026; Mayer et al., Reference Mayer, Davis and Schoorman1995; Zhou & Chen, Reference Zhou and Chen2021), meaning that competence both defines incompetence and predicts relational outcomes, thereby inflating their observed relationships. Additionally, trust itself is a predictor of psychological safety, so if an incompetent leader fails to instill trust, this could likely impact psychological safety as a result (Joo et al., Reference Joo, Yoon and Galbraith2023). More importantly, this assumption can lead to systematic misclassification. If leadership is evaluated primarily through follower perceptions of safety and trust, then any group characterized by strong internal cohesion may produce positive leadership evaluations, regardless of actual leader competence or outcomes.

Problem #3: What is an effective leader?

In defining leadership through psychological safety and trust, Ducheyne and De Fruyt are inherently implying what it means to be an effective leader and then uses deviations from that standard to classify leaders as incompetent or toxic. Consequently, psychological safety and trust primarily function to identify toxic leadership while collapsing effective and incompetent leaders into the same category. Accordingly, this results in the inability to differentiate effective and incompetent leadership. If incompetent leaders can still be trusted and maintain psychological safety, then traditional performance-based evaluations risk systematically conflating incompetence with effectiveness. Conversely, if toxic leaders can sometimes produce short-term performance gains, organizations that rely on outcome metrics alone may fail to detect toxicity. This raises another question: Is the distinction between toxic and incompetent leadership actually a distinction between different forms of ineffectiveness or simply between different deviations from an idealized definition of effective leadership? Are we actually distinguishing between types of leadership or simply between degrees of desirable versus undesirable outcomes? By embedding psychological safety and trust into the definition of effective leadership, incompetence becomes defined as a failure to produce these outcomes, which goes against Ducheyne and De Fruyt’s claims. This highlights a broader issue; psychological safety and trust are not sufficient enough indicators. They are important outcomes but do not span the full domain of leadership effectiveness. Contemporary models emphasize leadership as a dynamic process shaped by interactions among leader, followers, and context, with effectiveness reflected in both task performance and the capacity to respond adaptively to changing environments (Bonini et al., Reference Bonini, Panari, Caricati and Mariani2024; Johns, Reference Johns2024; Oc, Reference Oc2018).

Problem #4: Exclusion of special circumstances

Fourth, it assumes a consistent pairings of poor leadership eroding trust and psychological safety. In reality, those pairings are not stable. Observing a group with high psychological safety does not necessarily imply that the leader is effective. It is possible for groups to report high levels of safety and trust even when leadership is objectively ineffective. Additionally, people may distrust an incompetent leader or trust a toxic but charismatic one. The limitations of this approach become even more apparent when considering extreme cases. For instance, highly cohesive groups, such as cult-like organizations, may report strong internal trust and psychological safety among members. Within such groups, followers may perceive their leader as trustworthy, protective, and even benevolent. Historical examples illustrate this point. Members of groups such as those led by David Koresh or Charles Manson often demonstrated strong loyalty, trust, and perceived psychological safety within the group—even in the presence of harmful or dysfunctional leadership—reflecting the socially constructed and context-dependent nature of trust and group cohesion (e.g., Lalich, Reference Lalich2004; O’Reilly & Chatman, Reference O’Reilly, Chatman, Staw and Cummings1996). If leadership were evaluated solely, or even largely, based on these perceptions, such leaders could be classified as effective or even highly competent. This example is intentionally extreme, but it serves to illustrate a broader conceptual problem: Follower perceptions of safety and trust can and should be decoupled from leadership. The focal article seeks to move beyond perceptions while simultaneously grounding its definitions in them. Although leadership is inherently relational and perceptual, equating leadership with perceptions introduces this risk of misclassification. These examples highlight a broader concern in the leadership literature, where perceptions of safety and trust do not necessarily map onto objective indicators of leadership quality.

Problem #5: Team-level aggregation

The last limitation is that the model aggregates leadership and its effects to the team level, implicitly assuming that leadership operates as a uniform, shared experience. This creates a cross-level inference problem, as it treats relationships observed at the team level as if they reflect individual-level processes (Klein & Kozlowski, Reference Klein, Kozlowski, Klein and Kozlowski2000). However, extensive research demonstrates that leadership is inherently dyadic and varies across followers within the same team; resulting in substantial within-team variability (i.e., leader-member exchange (LMX); Graen & Uhl-Bien, Reference Graen and Uhl-Bien1995). Similarly, empirical work finds that variability in LMX relationships (LMX differentiation) has meaningful consequences for attitudes and performance within teams (Henderson et al., Reference Henderson, Liden, Glibkowski and Chaudhry2009).

Conceptually, this aggregation assumes homogeneity in leader behavior and follower perceptions, overlooking evidence that leaders allocate support, resources, and attention unevenly across team members (Graen & Uhl-Bien, Reference Graen and Uhl-Bien1995). As a result, the model may obscure whether ineffective leadership reflects consistently poor behavior across all followers or highly variable, selectively negative interactions with specific individuals.

Recommendations and future directions

A more coherent approach would be to model an IMOI (i.e., Input → Mediator → Output → Input) approach that distinguishes between leader behaviors and competencies (input), follower perceptions (mediators), and outcomes such as psychological safety and trust (output) that act as a feedback loop (Ilgen et al., Reference Ilgen, Hollenbeck, Johnson and Jundt2005). Such a distinction would allow researchers to examine how leader behaviors influence perceptions, which in turn shape outcomes rather than collapsing these components into a single construct. Maintaining this distinction would clear theoretical boundaries and more accurately classify leadership constructs.

Secondly, large language models (LLMs), natural language processing (NLP), and modern AI systems act as an unutilized tool for improving how leadership is conceptualized and categorized, particularly through data-driven taxonomy construction. Recent advances in LLM-based semantic embeddings provide a scalable method for addressing construct proliferation and conceptual obscurity by placing constructs, scales, and items into a shared semantic space, allowing researchers to quantify conceptual overlap and improve taxonomic organization (Wulff & Mata, Reference Wulff and Mata2025).

Last, agent-based modeling (ABM) acts as a potential tool in which scholars can simulate and examine emergent patterns of specific leadership behaviors before investing the resources toward investigating those ideas. ABM addresses these issues by forcing scholars to formalize constructs as explicit micro-level rules and mechanisms, thereby clarifying conceptual boundaries and reducing ambiguity (Carley, Reference Carley1995; Epstein, Reference Epstein2014). Moreover, because ABMs simulate interactions among heterogeneous agents embedded in dynamic environments, they naturally incorporate context and interdependence, overcoming the tendency of traditional leadership theories to rely on decontextualized, static relationships (Johns, Reference Johns2024; McHugh et al., Reference McHugh, Yammarino, Dionne, Serban, Sayama and Chatterjee2016; Serban et al., Reference Serban, Yammarino, Dionne, Kahai, Hao, McHugh, Sotak, Mushore, Friedrich and Peterson2015). By linking individual-level processes (e.g., perceptions, effort, and interactions) to emergent collective outcomes, ABM enables researchers to test whether purportedly distinct leadership behaviors and constructs produce meaningfully different system-level dynamics.