0
$\begingroup$

I am trying to fit regression model using r for salary on diffrent skills.But model returns regression coefficients as NA for some skills.This is due to high correlation among skills.But I still want to include them in model.Skill score values are between 4 to 8 for all skills. How to solve the problem of NA coefficients in r? Is it appropriate method of analysis?

$\endgroup$
  • $\begingroup$ Welcome to the community. In the above, there's quite a lot that you're not telling us about the problem, including how you determined that multicollinearity was the underlying issue. How can we comment on the appropriateness of the analytic method? Nevertheless, let me ask you a question. I am interested in the model predicting weight from height. I have only three independent variables -- ht1 is height measured in meters, ht2 is height measured in inches and ht3 is height in Angstrom units. Would it make sense for me to require all these height variables in the model? $\endgroup$ – user140401 Jan 4 '17 at 8:12
  • $\begingroup$ NA coefficients might mean that you have more predictor variables than you have cases or that some predictor is a linear combination of the others. Please provide more information about the number of cases, number of predictor variables, whether the skill scores are integer-valued or continuous, and your evidence for multicollinearity. $\endgroup$ – EdM Jan 4 '17 at 15:45
  • $\begingroup$ Thanks. When I searched for problem of NA coefficients it says that it is may be due to multicollinearity. And my data shows high correlation coefficient for some skills. $\endgroup$ – vas17 Jan 5 '17 at 6:09
  • $\begingroup$ For weight and height problem, No need to include all height variables. In my data there are 11 skills.Skill scores are discrete.there are 1000 cases but for some cases some skill scores are not available. $\endgroup$ – vas17 Jan 5 '17 at 6:21
  • $\begingroup$ Skill scores takes few values in rang 4-8 both inclusive and have discrete distribution, $\endgroup$ – vas17 Jan 6 '17 at 9:25
0
$\begingroup$

In very general terms, the easiest ways to get rid of multicollinearity is to try to:

  • Drop one of the highly correlated variables that you think creates the problem
  • You may want to transform the two correlated skills into a ratio and use that instead
  • Collect more data and see if the correlation persists
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.