Sticky cost behavior: evidence from small and medium sized enterprises in Turkey

To check the degree of multicollinearity, the variance inflation factor (VIF) for independent variables are calculated. The centered VIF value for the variables is the same since there are only two variables in Model I and it is 1.63. First introduced by Marquardt (1970), VIF value indicates an extreme degree of collinearity when it is higher than 10. Some subsequent studies argue that multicollinearity leads to poorly estimated regression coefficients when VIF values exceed 5 (Paul 2006). While the extreme collinearity is problematic, less than extreme collinearity has less impact on the results (Mason and Perreault 1991). Moreover, even the presence of near extreme multicollinearity (VIFs between 5 and 10) does not violate the least squares estimates’ being the best linear unbiased estimates (BLUE). The presence of multicollinearity inflates the standard errors of the coefficient by the square root of the VIF value estimated for that variable. That is standard errors for the coefficient of both independent variables are 1.28 (square root of 1.63) times as large as they would be if they had been uncorrelated with each other (Allison 1999). Likewise, the unaffected t statistics for the variables would be the reported t-statistics divided by 1.28. Thus, neither the coefficient values, nor their significance values are adversely affected by the presence of collinearity for Model I. Although VIF values and their adjustments are not reported for each estimation result table, the same holds for other models as well.