I have a dependent variable which I have to predict based on several independent variables. These independent variables are several types like one is numeric, one is categorical other is in terms of percentages another one in the units of dollars ,etc. Also,the dependent variable may be categorical , percentage, numeric or any other type. In all these situations is it correct apply any standardization method on all independent variables and then predictive model?


1 Answer 1


Ignoring the categorical variable for a moment: standardizing variables is only necessary when you have interaction terms where not doing so introduces multicollinearity.

Choosing to standardize only changes the interpretation of your coefficients. Gelman has a nice post on whether to standardize or not here: http://andrewgelman.com/2010/04/12/a_question_abou_9/

Categorical variables logically can't be standardized but you can choose different coding schemes. Again your choice of coding scheme (dummy, effect, orthogonal) won't change the results but will change the ease with which you can test hypotheses and interpret your results. Here's a long worked example of regression using all three schemes: http://faculty.cas.usf.edu/mbrannick/regression/anova1.html


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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