I use an OLS model because the dependent variable is positive and continuous (not binary or
categorical) and because of the complex nature of CSAs. The model was created in Stata 9, and its
tests for multicollinearity show that it is not a problem in the model since (1) the highest correlation
in the correlation matrix for independent variables is 0.41, between “mem_direct_connection” and
“emp_perm_number” while most are much lower and the average correlation (using absolute
values) is 0.16; and (2) the highest variance inflation factor (VIF) is 1.5, while the mean VIF is 1.3.
The bivariate correlations and VIFs are well below the “rules of thumb” that indicate
multicollinearity problems in multiple regressions (Hamilton 2006). Additionally, the included
variable’s coefficients did not change much when individual variables were excluded from the model.