BUS603: Managing People

Validity of measures

Table 3 reports the descriptive statistics and correlations among the study variables.

Prior to testing the hypotheses under investigation, we examined whether the measures we used represented valid tools to assess our target constructs. To this end, we used confirmatory factor analysis (CFA) in AMOS and tested the factorial integrity of our measures. In a first step, we conducted CFA on the item level for each measure separately (i.e., exploitative leadership, abusive supervision, organization-directed destructive leadership) and examined the factor loadings and item reliabilities. While all items of the exploitative leadership measure had excellent psychometric properties, we dropped several items of the other two measures (i.e., abusive supervision, organization-directed destructive leadership) because they did not represent the underlying construct well (i.e., factor loadings were below 0.60 and item reliabilities below 0.40;).

Next, we tested the discriminant validity of our measures. Because of the relatively large number of estimated parameters in the overall model and the small sample size, we created item parcels for all latent leadership constructs. For exploitative leadership, we formed five parcels based on the five dimensions specified by Schmid (i.e., egoism, taking credit, exerting pressure, undermining development, and manipulation). For abusive supervision and organization-directed destructive leadership, we used the factorial algorithm to create parcels. By sequentially including the items with the highest to the lowest factor loadings, while alternating the direction of item selection, three parcels were formed for abusive supervision and two parcels for organization-directed destructive leadership.

On this basis, we tested a series of theoretically viable factor models. Table 4 shows that a three-factor model with the three target constructs as latent variables and parcels as indicators obtained the best model fit and was preferable over alternative solutions. These results provide evidence that our measures captured distinct constructs versus common source effects.

Hypothesis tests concerning followers' emotional reactions

To test our hypotheses, we conducted a series of multiple regression analyses. In addition, given the high correlations among the destructive leadership measures, we followed the procedures suggested by Lorenzo-Seva and applied relative weight analysis. The results of these procedures are depicted in Tables 5, 6.

Abusive supervision was the strongest predictor of overall negative affect (β = 0.50, p < 0.001), followed by exploitative leadership (β = 0.29, p < 0.001), and organization-directed destructive leadership (β = 0.08, ns). Thus, hypothesis 1 was supported. For overall positive affect, abusive supervision (β = −0.27, p < 0.01) and exploitative leadership (β = −0.29, p < 0.05) exerted a similar negative effect, while the effect for organization-directed destructive leadership was not significant (β = 0.09, ns). With regard to the sub-dimensions of negative affect (see Table 7), the following pattern was revealed: the upset dimension was best predicted by abusive supervision (β = 0.47, p < 0.001), followed by exploitative leadership (β = 0.33, p < 0.001). The effect for organization-directed destructive leadership was not significant (β = 0.06, ns). In a similar vein, abusive supervision was the strongest predictor for the afraid dimension (β = 0.46, p < 0.001) followed by exploitative leadership (β = 0.23, p < 0.05). Again, organization-directed destructive leadership had no predictive value here (β = 0.09, ns).

The next set of hypotheses refers to different types of turnover intention. For general turnover intention, the results of regression analysis revealed only exploitative leadership as a significant predictor (β = 0.47, p < 0.001). Relative weight analysis, however, showed that the other two leadership forms also explained variance in general turnover intention (see Table 8); however, exploitative leadership clearly exerted the strongest effect. While these results do not fully confirm hypothesis 2, relative weights analysis does point to an effect in the expected direction.

With regard to calculative turnover, we found a positive effect for exploitative leadership (β = 0.48, p < 0.001), whereas the effect of abusive supervision was negative (β = −0.39, p < 0.001). Given that the two predictor variables were highly correlated (r = 0.75, p < 0.001), while abusive supervision did not correlate with the outcome variable (r = 0.01, ns), this pattern shows the classic signs of a suppression effect. This means that abusive supervision shares no or only little variance directly with the outcome variable but contributes to the regression equation by removing irrelevant variance from the other predictor variables. This is also reflected in the results of relative weight analysis, showing that exploitative leadership explained the major portion of variance in calculative turnover intentions (see Table 6). While hypothesis 2a is again not fully confirmed, taken together, this pattern points to what was predicted.

Interestingly, for immediate turnover, only exploitative leadership was a significant predictor in the regression analysis (β = 0.46, p < 0.001). Again, relative weight analysis revealed that the other two leadership forms also explained variance in immediate turnover intention, yet only to a moderate extent (see Table 6). Thus, hypothesis 2b was not supported.