Why good teams struggle under narcissistic CEOs?

Top management teams (TMTs) are central to strategic decision-making, innovation and firm performance, and Upper Echelons Theory has long held that executives' psychological traits – especially the CEO's – shape how those decisions get made (Abatecola and Cristofaro, 2020; Hambrick, 2007; Hambrick and Mason, 1984). Among these traits, Chief Executive Officer (CEO) narcissism is one of the most consequential and most contested (Giardino et al., 2025): associated with bolder strategic moves and heightened firm visibility on one hand, and with relational damage and degraded collaboration on the other (Chatterjee and Pollock, 2017; Cragun et al., 2020; Kavusan et al., 2025; Tuggle et al., 2024).

Recent research has advanced rapidly but along parallel tracks. Structural composition work shows how narcissistic CEOs reshape who sits on the executive team and board (Junge et al., 2025; Kruse et al., 2025); firm-level outcome work documents downstream damage to credit risk, internal control quality and reporting integrity (Almaleki et al., 2021; Chen et al., 2026; Salehi et al., 2021a, b); and subordinate-level work shows how narcissistic leadership erodes middle managers' identification and engagement (Li et al., 2025). Each stream documents what is altered at one organizational level, but leaves unspecified the intra-team relational and cognitive process that connects an individual-level executive trait to the collective TMT outcomes documented downstream. These streams share a common site – the CEO–TMT interface (Georgakakis et al., 2022; Simsek et al., 2018) – yet examine its inputs (who is appointed) and outputs (firm-level consequences) rather than the process within it. SM1 organizes this 2024–2026 body of work by level of analysis and locates the present study within it, in what Georgakakis et al. (2022) term the social-interactionist specification of the interface – foregrounding relational processes between the CEO and other executives rather than team composition. We therefore integrate Upper Echelons Theory (UET) (Hambrick, 2007) with group process theory (Lawrence, 1997) and information processing theory (Galbraith, 1973) to ask: How does CEO narcissism shape conflict dynamics within top management teams, and under what conditions does conflict influence team effectiveness?

We test a moderated-mediation model in which team conflict mediates the relationship between CEO narcissism and team effectiveness, and hypothesize that four team-level conditions moderate that mediation across two stages. At the first stage – the formation of conflict in response to narcissistic leadership – team trust and decision commitment are hypothesized to buffer. At the second stage – the conversion of emergent conflict into effectiveness loss – cooperative conflict management and agreement-seeking behavior are hypothesized to buffer against this conversion. The four moderators are differentiated along two dimensions: temporal focus (reactive vs proactive) and object (regulating the conflict process vs converging on the decision outcome). This scheme makes explicit the distinction between cooperative conflict management and agreement-seeking behavior, which is often conflated in prior work. The model is tested on survey data from 115 TMT members in mainland Chinese firms, with CEO narcissism assessed through other-ratings rather than archival or self-report measures. The mediation model and the second-stage moderation hypotheses are supported; the first-stage interactions are statistically significant but opposite in sign to the buffering we predicted – an asymmetry we examine in detail.

The contribution is threefold. First, we look inside the executive team to identify team conflict as a proximal mechanism through which part of the CEO narcissism's association with team effectiveness is transmitted, going beyond prior work that linked narcissism to conflict and to firm performance only as separate associations (Chatterjee and Pollock, 2017; Cristofaro et al., 2025). Second, we test the two-stage conditional process architecture as an integrated system – estimating indices of moderated mediation rather than relying solely on piecewise moderation tests – distinguishing conditions that shape the formation of conflict from those that shape its conversion into effectiveness loss, and explicitly differentiating cooperative conflict management from agreement-seeking behavior. Third, our results reveal that the two stages are asymmetric: downstream integrative mechanisms buffer conflict's negative association with effectiveness, whereas the upstream relational conditions long assumed protective – team trust and decision commitment – are associated with lower conflict overall but with a stronger, not weaker, narcissism–conflict link. This sharpens the contingency view of narcissistic leadership and cautions against treating relational strength as automatic protection. What is new is the joint two-stage moderation specification, the use of other-rated CEO narcissism at the TMT level and the focus on intra-team relational and cognitive processes in a mainland Chinese context.

This study draws on UET (Hambrick, 2007; Hambrick and Mason, 1984), in which executives' characteristics and cognitive frames partly shape organizational outcomes; narcissism stands out for its capacity to distort information processing, escalate conflict and undermine collaboration within TMTs (Chatterjee and Hambrick, 2007; Resick et al., 2009). Recent reviews identify relational and team-process mechanisms as the priority frontier (Abatecola and Cristofaro, 2020; Effah et al., 2025; Honsová et al., 2025), yet how narcissistic tendencies translate into team-level processes remains underspecified. We therefore integrate two complementary perspectives with UET: group process theory (Lawrence, 1997), which illuminates conflict dynamics within TMTs and information processing theory (Galbraith, 1973), which explains how trust and decision commitment shape the exchange, interpretation and enactment of information in complex decisions (Olson et al., 2007a; Parayitam and Dooley, 2009).

Three recent developments sharpen these lenses. Within UET, the interface reframes strategic leadership as interactions between the CEO and the TMT rather than a property of either alone (Georgakakis et al., 2022; Simsek et al., 2018), which behavioral-strategy work connects to the microfoundations of executive decision-making (Cristofaro et al., 2024). Within group process theory, evidence on group-level narcissism shows narcissistic actors heighten competitive perceptions and dysfunctional dynamics (Du et al., 2024; Park et al., 2025). Within information processing theory, trust and decision commitment are increasingly understood as interpretive conditions governing how executives encode and act on the same leader behavior, not merely as outcomes (Olson et al., 2007a). Together, the three lenses specify how CEO narcissism enters the team (group process), how it is filtered (information processing) and why its collective consequences follow (upper echelons) – the intra-team process left open by the streams summarized in SM1.

We define the constructs as follows. CEO narcissism reflects egotism, self-admiration and a strong desire for attention and influence; narcissistic CEOs overestimate their abilities, emphasize visibility and pursue bolder strategic actions, particularly under expanded discretion (Kavusan et al., 2025; Resick et al., 2009; Tuggle et al., 2024). Team conflict captures cognitive (task-related differences), affective (interpersonal tension) and process (disagreements over roles and execution) dimensions (Amason, 1996; Jehn, 1994; Parayitam and Dooley, 2009); because cognitive conflict's benefits typically erode when affective or process conflict accompanies it (Olson et al., 2007b), we adopt an integrative conceptualization in which the three dimensions interact and can escalate. Team effectiveness refers to TMT members' collective engagement and coordination toward organizational goals (Van der Vegt et al., 2000). Team trust is members' willingness to be vulnerable based on perceived competence, integrity, and benevolence (McAllister, 1995). Decision commitment reflects the extent to which members support and implement collective decisions after evaluating alternatives (Wooldridge and Floyd, 1990). Cooperative conflict management treats conflict as a shared problem and seeks integrative solutions; it becomes operative when conflict emerges (Tjosvold, 1998). Agreement-seeking behavior describes the TMT's inclination to reach consensus by incorporating diverse inputs before decisions are finalized; it operates around decision closure (Knight et al., 1999). While both share an integrative orientation, they differ in temporal focus (reactive vs proactive), object (the conflict process vs the decision outcome) and mechanism (mitigating affective fallout vs channeling cognitive content toward consensus).

CEO personality shapes how information is conveyed to, interpreted by and acted upon by TMT members (Hambrick and Mason, 1984). Although certain narcissistic traits – confidence, grandiosity, a desire for admiration – may foster bold leadership, narcissism also entails a lack of empathy, resistance to criticism and exploitative tendencies that undermine collective functioning (Sedikides et al., 2004). While some suggest a curvilinear relationship with performance (Uppal, 2020), a substantial body of research documents predominantly negative effects on team effectiveness, firm performance and counterproductive behavior (Amernic and Craig, 2010; Braun et al., 2018; Schmid et al., 2021), reinforced by recent firm-level evidence on weaker internal controls, distorted financial-statement comparability, elevated risk-taking and heightened credit risk (Almaleki et al., 2021; Chen et al., 2026; Salehi et al., 2021a, b). Notably, Chen et al. (2026) show that the credit-risk effect is amplified when the TMT shares working experience and the CEO has a broad network – indicating that team-level conditions shape the consequences of narcissistic leadership – and Li et al. (2025) show that narcissism erodes middle managers' identification and divergent strategic engagement. Even where narcissistic CEOs appear decisive (Engelen et al., 2016), the cumulative evidence converges on negative effects on collective functioning:

How TMT members respond to a CEO's directives depends on interpersonal style, and narcissism fosters a negative climate, especially among high-status members less inclined to defer (Chatterjee and Pollock, 2017). Three dynamics link it to elevated conflict. First, narcissistic leaders display inflated self-views and biased information processing (Campbell and Foster, 2007; Gardner and Pierce, 2011) and experience frustration over unmet admiration needs, which they displace onto others (Mormile et al., 2023). Second, their preference for dominance manifests in bold, high-risk, rapid decisions with limited openness to alternatives (She et al., 2020), amplified under high discretion (Kavusan et al., 2025); such unilateral choices provoke cognitive conflict (Nemeth, 1986) that spills into affective and process conflict as members perceive exclusion or unfairness. Third, narcissistic CEOs reshape the TMT to satisfy psychological needs – favoring shorter role tenures and higher firm-tenure separation, configurations tied to weaker behavioral integration (Junge et al., 2025; Kruse et al., 2025) – and even appointing loyal allies (Chatterjee and Pollock, 2017; Zhu and Chen, 2015) displaces rather than eliminates tension, encouraging coalition formation and information withholding (Carnevale et al., 2018a, b; Park et al., 2025). Collectively:

Although cognitive conflict can improve decision quality, the three dimensions frequently overlap in practice, making it hard to disentangle productive disagreement from accompanying emotional and relational tension (Olson et al., 2007b; Parayitam and Dooley, 2009). This is magnified under narcissistic leadership: bold, high-risk decisions raise strategic uncertainty and emotional stakes (Kavusan et al., 2025), and when decisions are perceived as unilateral, task disagreements escalate into affective or process conflict, eroding cognitive conflict's benefits – consistent with evidence that task conflict and effectiveness follow an inverted-U (Luo and Lu, 2020; Parayitam and Dooley, 2011). Because effective strategic decision-making requires coordinated, emotionally balanced information exchange (Cooke et al., 2015), unresolved or escalating conflict impairs effectiveness:

Hypotheses 1–3 jointly imply that CEO narcissism's influence on team effectiveness is transmitted through the relational and cognitive dynamics it activates. Sedikides and Campbell's (2017) Energy Clash Model frames narcissistic leaders as disruptive forces that destabilize relational equilibria, triggering conflict and subsequent attempts to restore balance. In the context of narcissistic leadership, conflict's dysfunctional manifestations dominate – particularly when narcissistic CEOs interact with non-narcissistic members whose values and tolerance for unilateral decision-making differ. Because collaboration, open discussion and commitment are essential for effectiveness, conflict is a key pathway through which CEO narcissism undermines collective functioning:

The mediated effect is unlikely to operate identically across TMTs. From an information-processing perspective (Galbraith, 1973), trust functions as an interpretive filter on CEO behavior: under high trust, assertive or ego-centric behaviors may be construed as confidence or vision; under low trust, the same behaviors are more likely to be read as manipulative or self-serving, increasing conflict. Narcissistic leaders are frequently perceived as lacking integrity, a core component of trustworthiness (Olson et al., 2007b; Savchuk, 2019); low trust heightens sensitivity to relational slights, whereas high trust inclines members to grant the benefit of the doubt and process disagreements constructively (Lewis, 2004). Accordingly:

Where trust governs how CEO behavior is interpreted, decision commitment governs how the TMT relates to the decisions that emerge (Amason, 1996; Zhang and Kwan, 2019). Under high commitment, members align behind decisions influenced by a narcissistic CEO and emphasize shared responsibility, limiting escalation into overt conflict. Under low commitment, behaviors such as dominating discussions or overruling others are more likely to be read as illegitimate; limited psychological ownership inclines members to challenge outcomes, dissent or build coalitions (Ghazinejad et al., 2018):

Whereas trust and commitment shape whether narcissism produces conflict, two mechanisms shape whether emergent conflict damages effectiveness. The first, cooperative conflict management, is a process-oriented approach activated when disagreement surfaces. Conflict management theory distinguishes between competitive approaches that heighten tension and cooperative approaches that engage divergent views, seek integrative solutions and prioritize shared objectives (Tjosvold, 1998; Tjosvold et al., 2006), thereby preserving cognitive diversity while preventing relational breakdown (Lovelace et al., 2001; Pappenhausen and Parayitam, 2015). Under narcissistic leadership, a cooperative orientation reframes disagreement as joint problem-solving so that conflict can enhance rather than erode functioning; in its absence, conflict escalates into antagonism or disengagement:

Although both second-stage moderators share an integrative orientation, they differ in three respects (Knight et al., 1999; Parayitam et al., 2012): in temporal focus (cooperative conflict management is reactive, activated when conflict surfaces; agreement-seeking is proactive, operating around decision closure), in object (processing disagreements vs converging on shared decisions) and in mechanism (mitigating affective fallout vs channeling cognitive content toward consensus). A team can manage conflict cooperatively without converging, or pursue consensus without processing the affective dimension, leaving residual tension. Under narcissistic leadership, agreement-seeking behavior tempers defensiveness and structures deliberation toward closure, preventing escalation into factionalism or paralysis; in its absence, conflict generates fragmentation:

The conceptual model is presented in Figure 1. By integrating Upper Echelons Theory with group process and information processing theories, it offers a multi-level explanation of how CEO narcissism links to team processes and TMT effectiveness.

The hypothesized relationships are theoretically bounded by cultural context, especially power distance and collectivism (Hofstede, 2001; House et al., 2004). In high-power-distance settings, assertive senior-leader behavior is often normalized as expected authority rather than problematic dominance (Farh and Cheng, 2000; Tsui et al., 2007); in collectivist settings, harmony norms, face-preservation (mianzi) and relational obligation (guanxi) can suppress overt disagreement even where tension exists (Cai et al., 2012; Tjosvold et al., 2006). If such norms suppress reported conflict, the estimates here would be conservative relative to low-power-distance or individualist settings – a proposition for future comparative work, not a claim tested here. Mainland China combines high power distance, collectivism and a paternalistic-leadership tradition (Farh and Cheng, 2000; Hofstede, 2001); although Confucian norms traditionally discouraged overt narcissistic display, recent evidence suggests narcissism has grown among Chinese executives and that hierarchical norms may facilitate rather than constrain its exercise once it emerges (Cai et al., 2012; Song et al., 2025; Xu et al., 2025). The relationships should thus be read as bounded within a high-power-distance, collectivist Chinese context, with the moderators themselves likely culturally inflected (Wang, 2021).

We collected data via a structured survey of TMT members in mainland Chinese firms using a non-probability convenience sample. Of 162 responses received, 41 incomplete surveys and 6 with missing data on CEO narcissism were excluded, yielding 115 useable responses – a completion rate of 71.0% among returned surveys, acceptable given the elite nature of TMT research. Because this figure is computed on returned surveys rather than initial invitations, we report it as a completion rate rather than a response rate. The choice of mainland China was theoretically motivated (Section 3.9).

The CEO was not the focal respondent; we examined senior executives' perceptions of the CEO's narcissistic tendencies and TMT functioning. Although multi-respondent designs are preferable for collective constructs, prior TMT research has relied on knowledgeable single informants when full-team access is difficult (Miller et al., 1988; Janssen et al., 1999; Olson et al., 2007a), and we treat informant qualification as a design feature, not a sampling convenience (Kumar et al., 1993). The constructs are informed by perceptual assessments from experienced senior executives, not aggregated team-level measures: by design, no within-team aggregation is performed, and no aggregation statistics (rwg, ICC) are computed, as each observation is a qualified informant's assessment of the CEO and the focal decision (Section 7.3).

To reduce socially desirable responding, participation was voluntary and confidential, and responses were used for academic purposes only; identifying information collected for recruitment was accessible only to the data-collection team, and neither organizations nor CEOs saw individual responses (Latkin et al., 2017). Survey sections were staggered, and construct blocks were separated to reduce respondents' tendency to infer relationships among variables.

Data collection involved qualifying stages. Respondents confirmed direct participation in a strategic decision-making process with the CEO. They recalled a specific decision – defined, following Mintzberg et al. (1976), as non-routine, consequential and resource-intensive – most concerned with restructuring, M&A, new product development and resource reallocation. Following Parayitam and Dooley (2009, 2011), only decisions involving at least four participating executives, including the CEO, were considered; the mean was seven (SD = 4.6). The survey was translated into Mandarin using translation–back–translation, verified by three experts and administered by a Mandarin-fluent Chinese co-author. Collection occurred between mid-May and mid-August 2022; the relationships examined are not tied to transient market conditions, though replication with more recent data is desirable (Section 7.3).

Respondent and firm characteristics were recorded during recruitment and screening. The sample was drawn from firms in Beijing, Shanghai and Nanjing. The average respondent was 39 years old; 62.6% were male and 37.4% female. Firms were 9% state-owned, 73% private and 18% publicly listed; by industry, 25% real estate, 36% manufacturing, 28% services and 11% other; by size, 12% large (>1,000 employees), 67% medium (300–1,000) and 21% small (<300).

CEO narcissism and team conflict were measured on five-point Likert scales; team trust, decision commitment, agreement-seeking behavior, cooperative conflict management and team effectiveness on seven-point scales. Minor wording adjustments aligned items with the strategic decision-making context while preserving conceptual meaning (Honsová et al., 2025). Full items, loadings, reliability, average variance extracted (AVE) and validity diagnostics are reported in the Supplementary Materials (SM2–SM5).

CEO narcissism used 12 adjective-based items adapted from Resick et al. (2009) and used in executive research – e.g. “conceited and egotistical,” “likes to be the center of attention.” Although such measures do not capture clinical narcissism, they are widely used to assess subclinical tendencies observable in managerial behavior, appropriate when clinical assessment is infeasible; all 12 retained (α = 0.94). Team conflict used ten items adapted from Jehn (1994) to capture cognitive, affective and process dimensions (Parayitam and Dooley, 2009); all were retained (α = 0.88). Team trust used 11 items from McAllister (1995), all retained (α = 0.84). Decision commitment was measured using seven items adapted from Wooldridge and Floyd (1990), one (D3) was excluded due to a low loading, leaving six (α = 0.89). Agreement-seeking behavior was measured using the six-item Knight et al. (1999) scale; two items were excluded, leaving four (α = 0.75). Cooperative conflict management used five items adapted from Tjosvold et al. (2006), all retained (α = 0.82). Team effectiveness used five items from Van der Vegt et al. (2000), all retained (α = 0.81). In total, three indicators (D3 and two agreement-seeking items) were excluded for weak loadings, all before composite scores were calculated and before any hypothesis test; the excluded items are listed verbatim with the retention criterion in SM2.

Our strategy followed a two-step approach (Anderson and Gerbing, 1988). First, the measurement model reliability, convergent validity and discriminant validity – was validated in SmartPLS. Second, composite scores were extracted and entered into Hayes's (2018) PROCESS macro to test mediation and moderation through bootstrapped conditional process analyses. No latent structural model with seven simultaneous constructs was estimated; each PROCESS model estimates at most four regression coefficients on composite scores.

Following Hair et al. (2011), Cronbach's alpha values above 0.70 indicate acceptable internal consistency; across constructs, alphas ranged from 0.75 to 0.94. Convergent validity was assessed through standardized loadings and AVE: all loadings were significant (p < 0.001), and most exceeded 0.70, and all constructs were modeled reflectively. The AVE exceeded 0.50 for all constructs except team conflict (AVE = 0.49). Although marginally below the threshold, the conflict construct was retained because reliability remained high (α = 0.88), loadings were acceptable and discriminant diagnostics supported adequacy; prior work notes that AVE slightly below 0.50 may be acceptable when other indicators are satisfactory (Fornell and Larcker, 1981). We return to this in Section 7.3. Discriminant validity was examined in two ways. First, following Fornell and Larcker (1981), AVEs exceeded the squared inter-construct correlations in all cases: the largest correlation was between decision commitment and team effectiveness (r = 0.49; squared = 0.24), below the AVE of each (0.65 and 0.57); for CEO narcissism and team conflict, the squared correlation (0.12) was well below both AVEs (0.63 and 0.49). Second, all heterotrait–monotrait (HTMT) ratios were below the conservative 0.85 threshold, the highest at 0.79 (Henseler et al., 2015). Full measurement results, loadings, AVEs and the complete HTMT matrix are provided in SM2 and SM5.

We assessed multicollinearity using outer and inner variance inflation factors (VIF). Following Hair et al. (2011), VIF values below 5 indicate that multicollinearity is unlikely to distort estimates. As reported in SM3 and SM4, all inner VIF values remained well below thresholds; three narcissism indicators marginally exceeded 5 at the item level, reflecting expected semantic redundancy within a 12-adjective trait scale.

Because the study relies on single-informant perceptual data, common method bias (CMB) is a concern, which is addressed both procedurally and statistically. Procedurally, respondents were assured of confidentiality, and survey sections were staggered, with sensitive items separated (Podsakoff et al., 2003). Statistically, five checks were run. Harman's single-factor test returned less than 25% of variance on one factor (indicative but not sufficient alone; Howard et al., 2024); Kock's (2015) full-collinearity construct-level VIFs were all below 3.3; and the hypothesized seven-factor model fit substantially better than a one-factor model. More diagnostically, common method variance generally attenuates rather than inflates interactions (Siemsen et al., 2010), so the four significant interaction terms – two opposite in sign to our predictions (Section 5.4) – and the near-zero correlation between agreement-seeking behavior and team conflict (r = −0.03, n.s.; Table 1) are difficult to reconcile with a pervasive uniform method factor. Finally, no unrelated marker variable was included; as a conservative sensitivity check in the spirit of Lindell and Whitney (2001), we partialled the smallest observed correlation (|r| = 0.03) from all zero-order correlations, which changed them by at most 0.03 and left every significant correlation significant (SM8). We treat this as a sensitivity benchmark, not a definitive marker test. These diagnostics suggest CMB is not severe, but perceptual inflation among relational constructs cannot be ruled out, given the cross-sectional, single-informant design (Podsakoff et al., 2024), so findings warrant appropriate caution.

Means, standard deviations, and zero-order correlations computed from the final composites are reported in Table 1. The largest correlation was between decision commitment and team effectiveness (r = 0.49). Correlations of 0.8 or higher may signal problematic multicollinearity (Kennedy, 2008; Tsui et al., 1995); none exceeded this level, and the inner VIFs were below 5 (SM4).

To test H1–H4, we used PROCESS v4.2 (Model 4); the moderation hypotheses were tested with Model 1, estimated separately for each interaction. All analyses were run on composite scores in their original metric (predictors were not mean-centered), with 20,000 bootstrap resamples and 95% bias-corrected confidence intervals throughout. Mediation results appear in Table 2 and moderation results in Table 3; full conditional-effect and Johnson–Neyman output appears in SM6 (Spiller et al., 2013).

Step 1 of Table 2 shows the total effect of CEO narcissism on team effectiveness was negative and significant (b = −0.206, SE = 0.061, t = −3.35, p = 0.001; 95% CI [−0.327, −0.084]), explaining 9.0% of the variance (R² = 0.090; F(1, 113) = 11.22, p = 0.001), a small effect (f² = 0.10; following Cohen (1991), f² of 0.02–0.15 is small, 0.15–0.35 medium, above 0.35 large). H1 is supported. Step 2 shows CEO narcissism on team conflict was positive and significant (b = 0.262, SE = 0.066, t = 3.97, p < 0.001; 95% CI [0.132, 0.393]; R² = 0.122; F (1, 113) = 15.77, p < 0.001; f² = 0.14). H2 is supported. Step 3 shows team conflict on team effectiveness was negative and significant (b = −0.214, SE = 0.085, t = −2.51, p = 0.014; R² = 0.138; F (2, 112) = 9.01, p < 0.001; f² = 0.16). H3 is supported. For H4, the total effect (−0.206) decomposed into a direct effect (−0.149, p = 0.021) and an indirect effect through team conflict (−0.056 = 0.262 × −0.214); the bootstrap estimate (Boot SE = 0.023) yielded a 95% CI of [−0.105, −0.015], excluding zero. H4 is supported. The mediation is partial and modest: the indirect effect accounts for approximately 27% of the total effect, and the direct effect remains significant, so team conflict is best described as a proximal mechanism that transmits part of the association, not its sole conduit.

As shown in Table 3, the CEO narcissism × team trust interaction on team conflict was significant (b = 0.146, SE = 0.073, t = 2.02, p = 0.046; 95% CI [0.003, 0.290]; R² = 0.233; f² = 0.30; ΔR² = 0.028, p < 0.05). However, the coefficient is positive whereas H2a predicted a negative (buffering) interaction: because the focal narcissism–conflict slope is positive, a positive product term means the slope strengthens as trust rises. Simple slopes confirm this – the conditional effect was 0.085 (n.s.) at −1 SD, 0.215 (p = 0.001) at the mean, and 0.345 (p < 0.001) at +1 SD; the Johnson–Neyman region of significance lies above a trust value of 4.60, covering 69.6% of the sample (SM6). H2a is not supported.

Figure 2 illustrates the pattern: high-trust teams report lower conflict at every level of CEO narcissism. Still, the narcissism–conflict line is nearly flat under low trust and steepest under high trust – so trust is associated with less conflict overall yet a stronger narcissism–conflict association.

The CEO narcissism × decision commitment interaction was likewise significant and positive (b = 0.163, SE = 0.080, t = 2.05, p = 0.042; 95% CI [0.006, 0.321]; R² = 0.253; f² = 0.34; ΔR² = 0.029, p < 0.05). The conditional effect was 0.049 (n.s.) at −1 SD, 0.161 (p = 0.020) at the mean, and 0.273 (p = 0.004) at +1 SD; the Johnson–Neyman region begins at a commitment value of 5.11, covering 67.0% of the sample (SM6). H2b is not supported.

Figure 3 shows the analogous pattern: high-commitment teams report lower conflict at every level of narcissism, but their narcissism–conflict slope is steeper, so the gap narrows as narcissism increases.

Turning to the second stage, the team conflict × cooperative conflict management interaction on team effectiveness was significant (b = 0.118, SE = 0.048, t = 2.48, p = 0.015; 95% CI [0.024, 0.213]; R² = 0.195; f² = 0.24; ΔR² = 0.044, p < 0.05). Because the focal conflict–effectiveness slope is negative, the positive product term indicates buffering, as H3a predicts. The conditional effect was −0.493 (p < 0.001) at −1 SD, −0.330 (p < 0.001) at the mean, and −0.168 (n.s., p = 0.074) at +1 SD; the Johnson–Neyman boundary lies at 5.04, with the negative effect significant for the 87.0% of the sample below it (SM6). H3a is supported.

Figure 4 shows the buffering pattern: effectiveness declines steeply with conflict when cooperative conflict management is low and is nearly flat when it is high.

Finally, the team conflict × agreement-seeking behavior interaction was significant (b = 0.193, SE = 0.090, t = 2.14, p = 0.034; 95% CI [0.015, 0.371]; R² = 0.236; f² = 0.31; ΔR² = 0.032, p < 0.05), again indicating buffering. The conditional effect was −0.400 (p < 0.001) at −1 SD, −0.244 (p = 0.002) at the mean, and −0.088 (n.s.) at +1 SD; the Johnson–Neyman boundary lies at 5.18, with the negative effect significant for the 71.3% of the sample below it (SM6). H3b is supported. Among the supported second-stage moderators, agreement-seeking behavior shows the largest interaction coefficient and, as Section 5.5 shows, the largest index of moderated mediation.

Figure 5 shows that effectiveness declines steeply with conflict when agreement-seeking behavior is low and is nearly flat when it is high. In sum, the mediation model (H1–H4) and both second-stage buffering hypotheses (H3a, H3b) are supported, whereas both first-stage interactions (H2a, H2b) are significant but in the opposite direction of prediction and therefore not supported. Section 6 interprets this asymmetry.

The hypotheses specify a two-stage architecture, but moderated mediation was not formally hypothesized. To evaluate the architecture as an integrated system, we estimated post hoc conditional process models – PROCESS Model 7 for each first-stage moderator, Model 14 for each second-stage moderator – reporting the index of moderated mediation and conditional indirect effects (20,000 resamples; full results in SM7). The indices mirror the moderation results. For team trust, the index was −0.031 (95% CI [−0.081, −0.001]), indicating a significantly stronger negative indirect association at higher trust. For decision commitment, the index was −0.035 (95% CI [−0.088, 0.007]); because the interval includes zero, formal moderated mediation was not obtained, though the conditional indirect effect was significant at high commitment. At the second stage, both indices were positive and significant, indicating the indirect effect weakens as the moderator rises: 0.028 (95% CI [0.007, 0.059]) for cooperative conflict management and 0.057 (95% CI [0.011, 0.125]) for agreement-seeking behavior, with the conditional indirect effect non-significant at high levels of either. The integrated models thus corroborate the asymmetry: transmission of CEO narcissism into lower effectiveness through conflict is buffered downstream and, if anything, amplified upstream.

Given the sample of 115, we conducted two power diagnostics (SM9). A sensitivity analysis for the regression models (α = 0.05, power = 0.80, one tested predictor) indicates a minimum detectable effect of approximately f² = 0.07 (Faul et al., 2009); every focal model exceeds this (f² between 0.10 and 0.34), although interaction increments smaller than the threshold could go undetected. A Monte Carlo simulation for the indirect effect (Schoemann et al., 2017), using the observed path coefficients and standard errors (a = 0.262, SE = 0.066; b = −0.214, SE = 0.085), yields an estimated power of approximately 0.69. Power for the mediation test is therefore modest, and the indirect effect should be read with corresponding caution.

Grounded in Upper Echelons, group process and information processing theories, our findings support the mediation model and the second-stage moderation hypotheses, while the first-stage interactions were significant but in the opposite direction of prediction. The intra-team correlates of CEO narcissism are thus systematic, conditional and asymmetric across the two stages.

The direct and mediated effects (H1–H4) indicate that CEO narcissism's negative association with effectiveness is partially transmitted through team conflict, with the indirect path accounting for approximately 27% of the total effect. This is consistent with evidence that narcissistic leaders provoke tension through dominance, defensiveness and limited empathy (Carnevale et al., 2018a, b; Mormile et al., 2023; She et al., 2020), and adds a process-level account to the firm-level damage documented in recent work on credit risk, internal controls and reporting quality (Almaleki et al., 2021; Chen et al., 2026; Salehi et al., 2021a, b). Where prior research shows what narcissistic CEOs are associated with at the firm level and who is affected across the hierarchy (Junge et al., 2025; Kruse et al., 2025; Li et al., 2025), our mediation finding suggests one path along which such damage may propagate inside the team. All these associations are perception-level and cross-sectional, reported by single senior informants (Section 7.3).

The moderation findings qualify this account. Downstream, the buffering we predicted materialized: cooperative conflict management and agreement-seeking behavior attenuate the negative conflict–effectiveness association, which becomes non-significant at high levels of either, and both second-stage indices of moderated mediation are significant. Upstream, the prediction failed informatively: team trust and decision commitment are associated with lower conflict at every level of CEO narcissism (both correlate negatively with conflict; Table 1), yet the narcissism–conflict slope strengthens rather than weakens as they rise, and that slope is significant only in the upper region of each moderator.

We offer a post hoc interpretation of the first-stage reversal, to be treated as a hypothesis for future research rather than a finding: trust and commitment may govern the expression of disagreement, not only its occurrence. In higher-trust teams, executives may feel safe enough to voice opposition openly under narcissistic leadership – converting narcissistic behavior into overtly reported conflict – whereas in lower-trust teams, disagreement may be suppressed even as latent tension persists (cf. Carnevale et al., 2018b). Decision commitment may operate similarly by raising the stakes: when executives are strongly invested in an agreed course, narcissistic attempts to redirect it may provoke explicit resistance rather than quiet accommodation. This reading fits the harmony and face-preservation norms of the study's context (Section 3.9), under which reported conflict plausibly tracks expressed rather than latent disagreement; designs separating the two are needed to test it. Trust's level effect and its slope effect thus simply differ – trust is not “harmful,” but neither is it the upstream shield the buffering hypothesis assumed.

Several alternative explanations merit consideration. A dissatisfied-informant account – a disaffected executive rating the CEO as narcissistic, the team as conflictual, and effectiveness as low – predicts uniformly negative associations and, at most, inflated main effects. It does not predict the pattern observed here: the near-zero agreement-seeking–conflict correlation (r = −0.03, n.s.), a theoretically structured set of four interactions, or first-stage interactions running against our own predictions. Method or affect variance likewise cannot manufacture interactions and generally attenuates them (Siemsen et al., 2010). We cannot rule out informant-level bias with these data – only multi-informant or temporally separated designs can – but the data resist the simplest versions of that account. Two boundaries deserve emphasis: the findings sit within a high-power-distance, collectivist, Confucian-influenced context in which face-preservation may lead to underreporting of latent conflict, making the observed narcissism–conflict association conservative relative to other settings (Cai et al., 2012; Wang, 2021); and although narcissistic CEOs pursue bolder strategic actions under favorable conditions (Kavusan et al., 2025), our data do not show that boldness yields productive intra-team dynamics, so claims about a strategic “upside” warrant caution when team functioning is the outcome.

In sum, this study contributes a perception-based, process-level account to the literature on CEO narcissism. Intra-team conflict appears to be a proximal mechanism linking an individual-level executive trait to collective TMT outcomes as perceived by senior informants, and the conditions that shape this mechanism differ by stage: downstream integrative mechanisms contain it, while upstream relational conditions coincide with lower conflict levels but a stronger association between narcissism and conflict. Rather than treating narcissism as inherently paradoxical, the findings advance a contingency-based view in which where the team intervenes matters as much as whether it does.

Recent research on CEO narcissism has advanced along composition, firm-level and subordinate-level lines (SM1), each documenting what narcissistic leadership alters at a particular level while leaving the connecting intra-team process underspecified. Addressing that gap yields three contributions: each moves from a theoretical premise to a finding to what the finding does to the theory, and closes with a corollary for practice.

First, we identify team conflict as a proximal mechanism connecting CEO narcissism to TMT effectiveness. Prior research has separately linked narcissism to conflict and to firm outcomes (Chatterjee and Pollock, 2017) but has not modeled conflict as the bridge between CEO traits and team outcomes. Our mediation finding (H4) shows the negative association is partially transmitted through team conflict, with a significant direct component remaining, responding to calls within Upper Echelons Theory to specify the team processes that connect executive personality to collective outcomes (Hambrick, 2007; Cristofaro et al., 2024) and suggesting one path through which firm-level damage may emerge.

Second, we extend strategic leadership theory by theorizing two channels through which narcissism generates conflict: interpersonal friction and procedural exclusion. Prior accounts emphasize relational damage – dominance, defensiveness, lack of empathy (Carnevale et al., 2018a, b; Mormile et al., 2023); our H2 finding is consistent with both channels, and recent strategy research documents the procedural one – narcissistic CEOs' preference for bold, unilateral moves (Kavusan et al., 2025; Tuggle et al., 2024) can generate conflict by cutting off participative deliberation, independently of interpersonal volatility. Because our design measures conflict as an integrated construct, it cannot empirically separate the channels; we develop the distinction theoretically because it directs attention to different team-level responses and reframes narcissistic leadership as a decision-making-style, not only a trait.

Third, we move boundary-condition research beyond single-buffer treatments (Ong et al., 2016) by specifying two stages at which team conditions operate – and showing the stages are not symmetric. Downstream, the integrative mechanisms behaved as contingency theory expects; upstream, the relational conditions long assumed to be protective did not buffer, being associated with lower overall conflict but a stronger narcissism–conflict association, with formal moderated mediation through decision commitment not observed. This stage-and-type distinction makes the contingency view more precise and less comfortable: the conditions that contain narcissism's correlates are not interchangeable, the stage of intervention matters; and upstream relational strength is no guarantee against narcissism-related conflict. Our post hoc voice-based interpretation (Section 6) offers a concrete, testable proposition for the first stage.

Read together, the three contributions trace a single loop from theory through evidence to practice. The mediation pathway (upper echelons) links an executive trait to a collective outcome through conflict (indirect effect −0.056), implying boards should treat CEO personality as a team-process risk, not only a firm-level one. The first-stage conditions (information processing) act as interpretive filters – lowering overall conflict yet coinciding with a stronger narcissism–conflict association – implying that managers should read visible disagreement in high-trust teams as information, not failure. The second-stage conditions (group process) determine whether emergent conflict erodes effectiveness, suggesting that investing in integrative, consensus-oriented routines is the most defensible lever once a narcissistic CEO is in place. Table 4 renders these corollaries as role-differentiated actions (Section 7.2). Together they shift CEO-narcissism research from whether narcissism is harmful to how its correlates spread inside the team, which conditions contain them and which, contrary to expectation, do not.

Two sentences of scope first. The guidance below distinguishes findings-based levers – the moderating mechanisms our estimates directly address – from literature-informed extrapolations, such as governance and screening actions, which draw on the cited assessment and governance literature rather than on our data; our study did not examine boards, CEO selection or intervention effects. With that distinction in place, the findings carry implications for boards, CEOs and TMT members under three headings.

Executive selection and screening. Boards may treat CEO personality as a strategic risk factor, not merely a style preference: narcissism's correlates are not confined to firm-level outcomes such as credit risk (Chen et al., 2026), financial-statement comparability (Almaleki et al., 2021), internal controls (Salehi et al., 2021a) or risk-taking (Salehi et al., 2021b), but extend to intra-team dynamics before those consequences materialize. Pre-appointment screening with validated instruments – HEXACO Honesty-Humility (Ashton and Lee, 2009) and behavior-anchored narcissism measures (Honsová et al., 2025; Resick et al., 2009) – may help identify candidates posing elevated TMT-process risk; this draws on the cited assessment literature, not our estimates. Where a sitting CEO already scores high, attention shifts to mitigation through the team-level conditions below.

Building team-level conditions. Our estimates speak most directly to the two downstream mechanisms: cooperative conflict management and agreement-seeking behavior buffered the negative conflict–effectiveness association, which became non-significant at high levels of either, with agreement-seeking behavior showing the larger interaction coefficient and index of moderated mediation. Investment in consensus-oriented decision processes and integrative conflict-management capability, therefore, appears the most defensible lever for teams led by narcissistic CEOs. Trust and decision commitment warrant a more nuanced message: both are associated with lower overall conflict and remain worth cultivating, but our results caution against relying on them to neutralize narcissism-related conflict, since that association was stronger, not weaker, at higher trust and commitment. One practical reading is diagnostic – in high-trust, high-commitment teams, narcissism-related friction may surface visibly, which boards and TMTs can treat as information rather than as evidence that trust has failed. None of these conditions is self-sustaining under narcissistic leadership: boards can protect TMT-member voice through governance structures, CEOs can signal openness to dissent and pre-commit to participative processes, and members can adopt cooperative-conflict norms and “disagree-then-commit” practices.

Practical action matrix. Table 4 summarizes these recommendations as a decision aid: each row corresponds to a moderating mechanism, with columns for the diagnostic signal that it is weak, the governance action boards may consider, the leadership action the CEO can take, and the team-level practice members can adopt. The table note separates cells linked to our estimates from those that extend beyond them.

A note of caution closes this section. Although our findings show that the conflict–effectiveness association can be buffered by downstream mechanisms, they do not show that narcissism's correlates can be eliminated – and, as the first-stage results show, relational strength alone is no guarantee. Boards should not treat the moderating conditions as a substitute for careful CEO selection, and TMT members should not assume these conditions emerge spontaneously. The buffering documented here is partial, conditional and effortful – cultivated rather than given.

We begin with the design constraints, because they license the claims. This study supports perception-level, cross-sectional associations among constructs reported by single senior informants in one national context. It does not support causal claims, team-level aggregation or generalization beyond the sampled setting. Findings should be read as associations among informed perceptions – the construed reality on which upper echelons theory holds that executives act (Hambrick and Mason, 1984) – not as team-level facts.

Within that boundary, several limitations follow. Although the diagnostics in Section 5.2 indicate CMB is unlikely to substantially distort the findings (Kock, 2015; Podsakoff et al., 2003, 2024), perceptual inflation among relational constructs cannot be ruled out; other-ratings reduce social desirability bias on the focal predictor but not on the team-process measures. The convenience sample of 115 limits power – minimum detectable effect near f² = 0.07 and Monte Carlo power for the indirect effect near 0.69 (Section 5.6; SM9) – so small effects may have gone undetected and the mediation estimate carries a wide interval. The team-conflict measure sits marginally below the AVE threshold (0.49); its retention rests on reliability, loadings and discriminant diagnostics (Section 5.1), and future work should refine the scale. Because the archived analytic file retained the substantive responses but not the recruitment-stage demographics, control-variable models (firm size, ownership, industry, age, gender) could not be estimated here; robustness to such controls is an open question for replication. Multi-respondent designs enabling within-team agreement (rwg, ICC) and aggregation are the most direct extension, and longitudinal or multi-wave designs would help disentangle causal direction.

A second cluster concerns cultural contingency. The hypotheses were developed and tested within mainland Chinese TMTs, characterized by high power distance, collectivism and Confucian norms that historically discourage overt narcissistic display while sustaining strong hierarchical authority (Cai et al., 2012; Farh and Cheng, 2000; Hofstede, 2001). If face-preservation and harmony norms suppress the open expression of disagreement, observed conflict would underrepresent latent conflict, making the positive narcissism–conflict association conservative; high power distance may attenuate the observable consequences of narcissistic dominance (Tsui et al., 2007; Wang, 2021), so the relationship may be stronger in low-power-distance settings; and the moderators are themselves culturally inflected (Knight et al., 1999; Tjosvold et al., 2006). Cross-cultural replication – in low-power-distance/individualist settings, in other high-power-distance/collectivist settings and in cross-national MNE samples (Lee et al., 2025) – is a priority next step. Four further extensions appear promising: examining how perceived CEO narcissism shapes attitudes and behaviors among middle managers and front-line employees (Li et al., 2025); integrating emotion-related processes such as TMT emotional intelligence and exhaustion into the conflict mechanism; studying decision-making under crisis or strategic uncertainty as a boundary condition (Mormile et al., 2023); and pairing survey measures with observational or archival indicators of dissent to test whether trust and commitment license the voicing of disagreement.

CEO narcissism is one of the most consequential – and most contested – traits in contemporary executive leadership. This study clarifies how its correlates unfold within the top management team and identifies the conditions under which they are contained. The moderated-mediation model shows narcissism's negative association with team effectiveness is partial and indirect, transmitted in part through team conflict and that transmission is not uniform. Downstream, teams that manage conflict cooperatively and actively seek agreement show no significant conflict–effectiveness penalty at high levels of those mechanisms. Upstream, the protection expected from trust and decision commitment did not materialize: both accompany lower conflict overall, yet the narcissism–conflict association strengthens as they rise – a caution against equating relational strength with immunity. These conditions do not arise spontaneously under narcissistic leadership; they require deliberate cultivation by boards, CEOs and members alike. The broader implication stands, sharpened: the strategic costs of narcissistic leadership are not fixed features of the trait but properties of the team that surrounds it – and knowing which properties help, at which stage, is what this study contributes.

The supplementary material for this article can be found online.