Nonresponse in Employee Attitude Surveys: A Group-Level Analysis

We conducted hierarchical MGCFA for heterogeneous groups to evaluate the psychometric equivalence of the measure across companies, allowing for different items in the different companies (Hattrup et al. 2007). In a first step, each item was allowed to load on its respective factor in each group. In a second step, we constrained the loadings of the shared items to equality across companies. The overall model across all four samples showed adequate fit with the data (CFI = 0.912, IFI = 0.912, RMSEA = 0.036), and the restriction on item loadings did not result in a significant decrease in model fit (ΔCFI = 0.003) in accordance with Cheung and Rensvold (2002). Thus, results of the MGCFA analyses support the overall equivalence of scales across companies.

We chose to compute heterogeneity indices as indicators for group composition instead of group mean levels, given the potential theoretical importance of group heterogeneity. We also, however, repeated the analysis with the group mean descriptors (mean levels for age and tenure and proportions for gender and employment status) instead of the heterogeneity indexes. That is, we conducted a two-step hierarchical linear regression analysis for sample 4, using aggregate job satisfaction, group function and the group means as independent variables. In a first step we entered the means for gender, age, tenure, employment status, and group size. Again, all variables were centered at their grand means. Group size and tenure were found to be significantly related to group response rates. Entering aggregate job satisfaction in a second step resulted in a significant increase in R ² (ΔR ² = 0.02, ΔF = 4.41, p = 0.04). Once aggregate job satisfaction was entered in the regression, however, the only significant covariate was group size.