Ethical Leadership, Employee Trust, and Organizational Commitment: A Multilevel-Aware Mediation Model and Simulation Study

Abstract

Ethical leadership is consistently associated with favourable employee attitudes, but the mechanisms that link it to employees' attachment to their organisation, and the methodological conditions under which those mechanisms can be estimated credibly, remain insufficiently specified. Integrating social learning and social exchange theory, this study proposes that employees' trust in their leader mediates the relationship between ethical leadership and affective organisational commitment. It also examines how measurement error, team-level clustering and unobserved confounding affect tests of this mediation. A 17-item instrument and a population model with a priori specified effects were defined. Survey data for 360 employees nested in 60 teams were generated by Monte Carlo simulation and analysed with partial least squares structural equation modelling (PLS-SEM), consistent PLS (PLSc), and case-level and team-level (cluster) bootstrapping. Four further experiments, comprising 1,000 to 10,000 replications each, examined parameter recovery across estimators, power as a function of the number of teams, sensitivity to an unobserved confounder, and the Type I error of the direct effect under clustering. The measurement model met conventional criteria (α = .89–.91, composite reliability = .91–.93, AVE = .64–.69, HTMT ≤ .62). Perceptions of ethical leadership showed substantial within-team agreement (ICC(1) = .26, ICC(2) = .68). Ethical leadership was positively related to trust in the leader (β = .561, p < .001), and trust to affective commitment (β = .245, p < .001). The indirect effect (β = .138, cluster-bootstrap 95 % CI [.069, .214]) accounted for 35 % of the total effect, indicating complementary partial mediation. Across replications, PLS-SEM and sum-score regression under-estimated the indirect effect by 22–24 %, whereas PLSc was unbiased. An unobserved confounder of trust and commitment (γ = .30) inflated the indirect effect by 39 % and reduced the direct effect by 37 %. Ignoring clustering raised the false-positive rate for a null direct effect from 5 % to as much as 28 %. Cluster-robust inference largely removed this inflation. The study clarifies the trust-based mechanism linking ethical leadership to commitment and provides evidence-based design guidance for testing it. The guidance covers the choice of estimator, the number of teams, multilevel inference and causal identification. Because the data are simulated, the estimates illustrate the model and the behaviour of the methods. They are not empirical evidence, and the model requires confirmation with field data.