BUS603: Managing People

Our dataset had a multilevel structure, given that we asked leaders about their level of mindfulness (i.e., independent variable on Level-2), and we asked at least two followers of those leaders about their leaders' mindfulness in communication and their own satisfaction (i.e., mediator and dependent variable on Level-1). Thus, we first examined the nested structure of our data using R; multilevel. First we examined the variance between the groups of followers reporting to one leader. An ANOVA showed significant differences between the groups of followers. The ICC(1), which is reported in Table 3, indicated that 24–29% of the variance resided between groups. Second, we examined the agreement within groups of followers reporting to one leader. The ICC(2), which is also shown in Table 3, indicated an agreement between 0.48 and 0.54. It is helpful to note that ICC(2) is dependent on the group size. In our study, the average group size was 2.8 and ICC(2) values ranging from 0.48 to 0.54 correspond with estimates about what can be statistically expected.

To test our hypotheses within the multilevel framework, we followed the suggestions for multilevel mediations. Our mediation model reflects a 2–1–1 design with leaders' dispositional mindfulness representing the Level-2-predictor, perceived leaders' mindfulness in communication representing the Level-1 mediator, and followers' satisfaction ratings representing Level-1 outcomes (Figure 1). In 2–1–1 models the within-group effects and between-group effects are confounded – in our case the effect between followers of different leaders on the one hand and within followers of a particular leader on the other hand. To address this problem, Zhang et al. suggested to differentiate the between and the within group effects by inserting the mediator at both levels in the following way: At Level 1, the mediator is centered around the group mean, specifying the within-group effect. At Level 2, the mediator is aggregated for each group using the group mean in order to specify the between-group effect. In line with these recommendations, we included our mediator variable (i.e., perceived leaders' mindfulness in communication) at both levels (i.e., group mean centered at Level 1 and aggregated for the followers of a particular leader at Level 2). Because mindfulness in communication was thought to be a characteristic of each leader, we were especially interested in the effect between followers of different leaders and thus, the effect at Level 2. Consequently, the effects within followers (i.e., Level 1) were treated as control variable "only". Nevertheless, we report the coefficient for both effects (within-groups effects and between-groups effects). Given that we used the aggregated values of mindfulness in communication as our mediator, we calculated the r_wg statisticto assess the appropriateness of aggregating [in addition to relying on the ICC(2) value, which was reported above]. The mean r_wg for perceived mindfulness in communication was 0.74, indicating strong interrater agreement.

The results of the multilevel analysis are reported in Table 4. All calculations were conducted in R using the packages: multilevel, nlme, sjmisc/sjstats, and reghelper.

Table 4. Results of multilevel mediation analyses.