Heteroskedasticity, MulticoUmearity, and Common Method Variance

In order to compute regressions, residuals need to be tested for heteroskedasticity (Backhaus, Erichson, Plinke & Weiber, 2008, p. 100). Heteroskedasticity is present if “at each point along the predictor variable, the spread of residuals is different” (Field, 2013, p. 876). In other words, heteroskedasticity accounts for the relationship between residuals and independent variables. The Glejser (1969) test is a predictable measure as it detects multiple presences of heteroskedasticity (Ayoola & Olubusoye, 2012). The Glejser test has been computed for all effects of residuals of predictor variables in the model. The generated plots showed no indication of any pattern which indicates a minimized threat of heteroskedasticity.

To test if predicting variables’ correlation might cause issues for interpretation, tests for multicollinearity are executed. Multicollinearity exists if independent variables correlate in a linear manner with each other. In that case, a redundancy of predictor variables is assumed which would result in unreliableness of the regression calculation (Backhaus et al., 2008, pp. 87-88). Multicollinearity is usually existent to a certain degree without violating assumptions of the model. For this purpose, regression calculations are controlled for the Variance Inflation Factor (VIF). Literature suggests that the VIF should not exceed the value of 10 (Myers, 1990). For the underlying model, the VIF showed values below 2 for few cases, predominantly values were below 1.

Responses for this study were mainly gathered by followers, therefore common method variance could be a potential bias. Harmann’s One-Factor-Test has been described as an examination to detect common method variance in response behavior (Sohnchen, 2009, p. 141). Using exploratory factor analysis, the test measures whether all items load to one factor. Rather than a profound test, the analysis provides an indication for common method bias. The exploratory factor analysis (Promax rotation with Kaiser Normalization) revealed that items for the MLQ 5X short, SLSI, LMX-7, and performance scale load to 16 factors accounting for 70.26% of variance. Results indicate that the data is not subject to common method variance and the risk of single source bias is thus reduced.

summary

This chapter outlined descriptive statistics of respondents. Scale statistics are provided for the MLQ 5X short, SLSI, LMX-7, and performance measure. Ratings of perceptions by followers for both, MLQ 5X short and LMX-7 presented stable Cronbach alpha scores. In the course of the present study, the SLSI has been applied to a larger organizational sample for the first time. The instrument showed high factor loadings and adequate reliability. Particular attention has been paid to the scale structure of the performance measure. Internal consistency turned out to be adequate for the research instrument.