0
$\begingroup$

Is the following methodology correct:

  1. Fit a multiple linear regression model
  2. Obtain the standardized coefficients
  3. Sum up the absolute value of all standardized coefficients
  4. Divide each individual standardized coefficient estimate by the sum (step 3 above) and multiply by 100 to obtain the % contribution of each predictor variable

If this is incorrect, what is the best way to determine the relative importance of each predictor? In reading other posts, I've found Kruskal's key driver analysis to be one suggestion.....

$\endgroup$
0
$\begingroup$

One method you might consider is dominance analysis or relative weights analysis as an alternative.

The method is implemented for linear regression in many popular packages including R, Stata, and SAS. It will produce a decomposition of the R-square for the model, which can be expressed as a percentage of the R-square explained by each predictor.

In addition, there is a fully online version of relative weights that you could use via Davidson College routed through R.

Give some of these articles a look - they seem to meet the criteria you're looking for in determining importance here.

$\endgroup$

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.