I am doing multivariate analysis using logistic regression to see the relationship between one categorical outcome variable and a group of continuous and categorical explanatory variables. I did preliminary explanatory analysis using chi-square for the categorical covariates and t-tests and Mann-Whitney tests for the continuous variables based on the type of the distribution.

When I did the univariate analysis using binary logistic regression for the same variables, the results are different for the skewed data (previously analysed by Mann-Whitney) and the same for the normal data (previously analysed by t-test).

Should I stick to the logistic regression for the univariate analysis, or should I do either transformation or categorization for the skewed data before launching the multivariate analysis?

  • 3
    $\begingroup$ Why are you doing the univariate analysis? Is it for a table? Curiosity? Hypothesis test? Or what? $\endgroup$ – Peter Flom Mar 17 '14 at 10:16
  • 3
    $\begingroup$ To add to Peter's questions, univariate analysis can cause an amazing amount of damage when done before multivariable analysis, because there is a temptation to use the uivariate results in guiding model building. $\endgroup$ – Frank Harrell Mar 17 '14 at 11:59
  • $\begingroup$ I am doing the univariate analysis to decide which variables be included in the model using the p value of 0.25 as a cut off point $\endgroup$ – shebl Mar 18 '14 at 11:32
  • $\begingroup$ When not used to decide on inclusion/exclusion of variables, it can be useful in understanding results. $\endgroup$ – kjetil b halvorsen Oct 9 '17 at 2:08

Your Answer

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

Browse other questions tagged or ask your own question.