Page 142 - WHERE WE WORK - Schlegelmilch
P. 142

Ideas in place
 4.3.3 Common method variance
With self-report data, method variance may occur. Therefore, we followed several of Podsakoff’s (2003) procedural remedies to counteract them. We also conducted a Harman single-factor test to determine the extent of method variance in the data. These results indicate that one factor explained 33.53%, thereby being far under the problematic threshold of 50%. This indicated that common method is unlikely to be a serious issue.
4.3.4 Data analysis
We approached the analysis in two steps. First, we used SPSS AMOS 24 to ensure the fit of the two measurement models by conducting two CFA’s, one for each dependent variable. We evaluated each model based on the parameters stated earlier). In each model, we included the respective dependent variable, as well as workplace transparency and workplace flexibility. While idea sharing and workplace flexibility were included as manifest variables (both count variables), idea implementation and workplace transparency were treated as latent variables (for details, see CFA of workplace transparency above). The idea implementation scale consisted of three items, thus it was by definition a just-identified model irrespective of the loading patterns (Malhotra & Sharma, 2008), and fit indices were not available.
Second, we used the statistical analysis software IBM SPSS Statistics 25. For the hypothesized model for the count variable idea sharing, initial analysis indicated that a negative binomial (NB) distribution with a log link was a better fit than the Poisson loglinear because NB does not assume equal mean and variance. We further added an exposure variable (sometimes also called offset; actual weekly work hours), which is common when analyzing count variables to adjust for the possibility of an event occurring. Furthermore, the negative binomial distribution required an estimation of the dispersion, for which we used the conservative estimate of the built-in function. For idea implementation as a dependent variable, we used a linear regression analysis.
140




























































































   140   141   142   143   144