Elevation at work: The effects of leaders’ moral excellence

Abstract Leaders influence followers in many ways; one way is by eliciting positive emotions. In three studies we demonstrate that the nearly unstudied moral emotion of 'elevation' (a reaction to moral excellence) mediates the relations between leaders' and their followers' ethical behavior. Study 1 used scenarios manipulated experimentally; study 2 examined employees' emotional responses to their leaders in a natural work setting; study 3 compared the effects of elevation to those of happiness, serenity, and positive affect. We found that leaders' interpersonal fairness and self-sacrifice are powerful elicitors of elevation, and that this emotion fully mediates leaders' influence on followers' organizational citizenship behavior and affective organizational commitment. In the first study, we also observed a moderation effect of interpersonal fairness on self-sacrifice. Results underline the importance of positive moral emotions in organizations and shed light on the emotional process by which ethical leaders can foster positive organizational outcomes. Keywords: elevationleader self-sacrificeinterpersonal fairnessorganizational citizenship behaviororganizational commitment Notes 1. The performance of the chi-square statistic is affected by several factors, among them sample size, non-normality, and outliers. Applying non-robust estimation methods and test statistics to non-normal data impacts on the estimates, their standard error, and the probability of accepting the model (sometimes increasing and sometimes decreasing it). Yet, though deleting outliers is the best method to deal with them, it is usually looked at with suspicion; so we also estimated a model with outliers in. At this aim, we used the best estimation method for non-normal variables and our sample size (which, according to Bentler and Yuan (Citation1999) is the Asymptotically Distribution Free) and the most robust chi-square test (the Yuan–Bentler asymptotically robust goodness-of-fit test statistic). The full mediation model is still the best ( = 1.7, p = 0.88). 2. The most common method for analyzing data coming from a classic 2 × 2 factorial design with a mediator variable would have been a multivariate analysis of covariance (MANCOVA), with two factors (self-sacrifice and interpersonal fairness), four dependent variables (altruism, courtesy, compliance, and commitment), and a covariate (elevation). There are two important shortcomings deriving by the use of this analysis. (1) MANCOVA would only evaluate moderation effects, whereas our main interest is on the mediation effect of elevation; (2) MANCOVA suffers from stringent limitations and potential ambiguity in interpreting results. One of these limitations is 'homogeneity of regression.' In MANCOVA (and analysis of covariance (ANCOVA) as well), it is assumed that the regression between covariates and dependent variables in one group is the same as the regression in other groups so that using the average regression to adjust for covariates in all groups is reasonable. If heterogeneity of regression is found, and then there is a relation between the independent variables and the covariate, a different adjustment should be made for each group and MANCOVA is inappropriate. In this case, we tested the homogeneity of regression following the procedure described in Tabachnick and Fidell (Citation2001) and found that significant relations between elevation and our independent variables (F (3,101) = 2.80, p = 0.044) would violate MANCOVA's assumptions. Hence, we preferred SEMs and bootstrapped confidence intervals around the indirect effects. 3. Testing the hypothesis that chi square of the full mediation model is different from zero also tests the full mediation model against the partial mediation model, which is a saturated model with all relations hypothesized and estimated, no degrees of freedom and chi square = 0. 4. We chose the version in which the product of the error variances is subtracted, thus: where a is the beta coefficient of the relationship between x and m, b the beta coefficient of the relationship between m and y, SE a the standard error of a, and SE b the standard error of b. 5. As in study 1, the analysis conducted leaving outliers in did not change neither the acceptance of the full mediation model nor the estimates. 6. We also controlled for age, gender, job tenure, and time working with one's supervisor entering them as predictor of IVs and elevation. The only significant relationships we found are between job tenure and fairness (β = −0.18, p < 0.05) and between age and elevation (β = 0.12, p < 0.01). 7. This component might be skipped in retrospective measures, since its relationships with the other components might be negatively affected by different levels of awareness that characterize them (as observed in study 2, physical sensations might be less vivid in respondents' memory).