1

My dataset consists of ordinal and nominal variables for independent variables and an ordinal variable for the dependent variable. I have been trying to get odds ratios using the coefficients of independent variables to interpret their impact on the dependent variable. However, there is a strong multicollinearity among ordinal variables. Please find the correlation matrix attached. correlation matrix

I have read two related posts. But, I am not sure how to deal with it. I think I cannot remove all of them except a few. My current option could be using LASSO to penalize the variables based on this answer. Any help is appreciated.

Mehmet Yildirim
  • 113
  • 6
  • In both the posts you link, comments discuss how the right strategy for dealing with multicollinearity (indeed, whether it's a problem at all) depends on what the goals of your analysis are, i.e. whether you are simply trying to predict some outcome or if you want to interpret the coefficients on your regressors in some way, as well as what the regressors in question are (since it might be that several of the highly correlated regressors can be combined to form some sort of 'summary' variable which captures all the information they encode, for instance). Could you please clarify these points? – greggs Jan 07 '21 at 17:54
  • I edited the question. My goal is to interpret odds ratios that are obtained from the coefficients of independent variables. The variables are survey questions and some of them somehow correlated. – Mehmet Yildirim Jan 07 '21 at 18:17
  • Any chance you could clarify the nature of these independent variables? The accepted answer by Stephen Kolassa on the first question you link states that its worth considering why your regressors are correlated. If they ultimately measure the same feature, try to think if there's some way to capture that feature with only one measurement. I think in order to be of more help, we need to know more information about what your independent variables are, and *why* they might be correlated – greggs Jan 08 '21 at 10:43

0 Answers0