Spss statistics package v 23

Because the JC subscale scores showed a non-ignorable amount of missing data (8.5–9.4%) multiple imputation was applied (Enders, 2010 ). Following recommendations by Graham et al. (2007 (link)), we generated 20 data sets with missing values on JC subscales replaced by the means of a sequential regression method (as implemented in IBM SPSS Statistics package v23, IBM Corp., Armonk, NY, USA). The imputation model included all predictor variables, latent class membership and estimated person parameters in job satisfaction gained from the best class-solution of the mixed IRT model as well as personality traits such as conscientiousness that predicted the missingness (see Part C of the supplementary material for details on the missing analysis). The following analysis was automatically performed on the 20 generated data sets and results were subsequently aggregated. Class membership was predicted in a multinomial logistic regression model. Classification inaccuracy of the rmGPCM was taken into account by using the adjusted three-step method proposed by Vermunt (2010 (link)) that is implemented in Latent GOLD 5.0. For categorical predictors (e.g., job position, organization size) sets of dummy variables were built. To reduce the number of dummy variables, the categories of original predictor variables were regrouped as described above.