PiscesLogoSmallerStill Multicollinearity

Top  Previous  Next

If the environmental variables are highly correlated then results obtained by any multiple regression method can be highly unreliable. ECOM tests for appreciable collinearity between the environmental variables and will give a warning message if appreciable levels of collinearity are present.

 

multicolwarning

 

The results of the multicollinearity test are presented in a tabbed sheet.

 

multicoltest

 

 

A multiple regression is undertaken for each of the environmental variables in turn with all the other environmental variables acting as the independent variables. Values of R-squared close to 1 or Variance inflation factors (VIF) well above 1 are indicative of multicollinearity and you should consider removing one of a group of highly correlated variables from the analysis. High VIF values are highlighted in red. Use stepwise regression and Variable Selection to identify which environmental variables to retain for your analysis

 

If you should obtain very large values for VIF it is likely that some of your variables add up to a constant. This often happens if you are using dummy variables see Linear Combinations of Environmental Variables.