88 Chapter 4 Analytical approach and preliminary analyses Data comprised a multilevel structure (i.e., week-level measurements nested within employees). Therefore, we conducted multilevel analyses with MLwiN in order to test our hypotheses. One series of analyses was conducted with week-level work engagement as the dependent variable, and one series with week-level creativity as dependent variable. Prior to the analyses, we calculated the intra-class correlations for our two dependent variables, which shows the amount of variance attributed to between-level (between persons) variation. This was 56% in week-level work engagement, and 57% in week-level creativity, suggesting that considerable within-level (within persons) variation in the dependent variables remained to be explained by week-level variations in the independent variables. Furthermore, we found that a two-level null model (i.e., a model with the intercept as the only predictor) fit the data better than a one-level null model for both dependent variables, which provides additional justification to the use of multilevel analyses. We conducted multilevel confirmatory factor analyses (MLCFA) in Mplus (Muthén & Muthén, 1998-2011) to test the discriminant validity of proactive vitality management vis-à-vis work engagement. The analyses revealed that a two-factor model, in which all proactive vitality management items loaded on one latent factor, and all work engagement items loaded on a second latent factor, fit substantially better to the data compared to a one-factor model in which all items loaded on the same factor, Δχ2(2) = 654.00, p < .001. This means that the two concepts can clearly be empirically distinguished from each other. Between-level predictors (i.e., goal orientation) were grand-mean centered and withinlevel (week-level) predictors were group-mean centered (Ohly, Sonnentag, Niessen & Zapf, 2010). To control for potential carry-over effects of one’s prior levels of work engagement and creativity, both dependent variables were controlled for their levels of the previous week (i.e., week-level work engagement and week-level creativity were controlled for lagged work engagement and lagged creativity). Following previous practice (Oerlemans & Bakker, 2014), both lagged variables were grand-mean centered. The use of lagged variables rendered the data of one week missing, which resulted in the use of 378 observations for both analyses.
RkJQdWJsaXNoZXIy MTk4NDMw