pdf - Nyenrode Business Universiteit
pdf - Nyenrode Business Universiteit
pdf - Nyenrode Business Universiteit
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
4.4. ANALYSES 107<br />
The model does show significant relations for the antecedents for the room for managerial<br />
opportunism. The room for managerial opportunism is influenced by information asymmetry<br />
(positively), goal ambiguity (positively), measurability of outputs (negatively), and<br />
decentralization of decision rights (positively). The directions of these relations are consistent<br />
with my expectations. The size and industry control variables do not have significant<br />
effects on the room for managerial opportunism.<br />
To evaluate the fit of the model, I calculated various fit indexes. Assessing the fit of a SEM<br />
is not as straightforward as, for example, an OLS regression analysis. The evaluation of<br />
the statistical properties of various fit indexes is still an active research topic, and no single<br />
measure provides a definite answer with regard to the fit between a model and the data, as a<br />
wide variety of goodness-of-fit statistics are available for SEM (Kline 2005). Kline suggests<br />
that multiple measures must be assessed to evaluate the goodness-of-fit. As the indicators<br />
of model fit, I use the chi 2 , the goodness-of-fit index (GFI), the adjusted goodness-of-fit<br />
index (AGFI), the comparative fit index (CFI), the root mean square residual (RMR), and<br />
the root mean square error of approximation (RMSEA). Kline (2005) suggests that the<br />
following factors indicate a good fit: an insignificant chi 2 ; a GFI, AGFI, and CFI greater<br />
than 0.90; a RMR close to 0; and an RMSEA less than 0.045. As presented in table 4.12<br />
- panel B, the indicators show good model fit.<br />
Despite the good fit, the model shows potential non-normality issues. The multivariate<br />
kurtosis measure lies above the threshold value of ten suggested by Kline (2005). Kline<br />
(2005) argues that kurtosis greater than ten may suggest a problem with the data. The<br />
kurtosis in the current model is 13.524. The high kurtosis is due to the inclusion of control<br />
variables for business unit and firm size 12 . Because multivariate normality is a strict<br />
assumption of maximum likelihood estimation (MLE) in SEM models, I use bootstrapping<br />
with 500 samples, which does not assume multivariate normality, to analyse whether the<br />
kurtosis affects the MLE estimates (Kline 2005, Bryne 2001). I find that the results are<br />
qualitatively similar, and that normality issues did not drive the results. Additionally,<br />
the Bollen-Stine bootstrap p does not reject the model. The Bollen-Stine bootstrap is a<br />
bootstrap modification of model chi 2 that is used to test model fit by, for example, adjusting<br />
for the lack of multivariate normality. Similar to the chi 2 , the Bollen-Stine bootstrap is<br />
insignificant at p = 0.701, which indicates a fit between the data and the model. The<br />
bootstrap statistics are presented in table 4.12 - panel C.<br />
4.4.2.2 Analyses with RPE-based-Targets Measure<br />
To buttress the findings of the base model analyses, I run a robustness check. Here I follow<br />
the strategy of chapters 2 and 3 by rerunning the analysis with a different operationalization<br />
of RPE use: RPE-based-Targets.<br />
12 Excluding the size controls from the model reduces the kurtosis to 3.928, but does not yield qualita-<br />
tively different conclusions.