Yahoo Answers is shutting down on May 4th, 2021 (Eastern Time) and beginning April 20th, 2021 (Eastern Time) the Yahoo Answers website will be in read-only mode. There will be no changes to other Yahoo properties or services, or your Yahoo account. You can find more information about the Yahoo Answers shutdown and how to download your data on this help page.
I was doing a logistic regression but the model was not performing well. I think I might have to check data..?
My question is.....Is it necessary to transform variables into Log/ Sqr / Sqr root/ etc.... this type of transformation we generally do in regression analysis. So will it work same as in Logistic Regression?
- OPMLv 77 years agoFavorite Answer
You are asking the wrong question entirely. You should never take the log, square root, etc simply to make the regression work. It may well be true that there is no relationship between the data.
Making a transform can have unintended consequences. Likewise, it is also true that transformations can do very positive things.
Ask yourself a very different question. Ask, from a theoretical perspective, whether y=mx+b makes more sense than y=m*SQRT(x)+b. (linear regression example)
Best fit models are often over-fit models. It is common the best fit model is also a wrong model. Work out what makes sense. If you use a transformation then it is important to mentally work through the mathematical implication of the model. Fit is a poor criterion. If it were a Bayesian model I would say to use Bayes factors as that would tend to resolve the model that is best fit to nature, but it is unlikely you are using a Bayesian method based on the question.
Outside Bayesian methods there really are no good model selection methods EXCEPT rationality. Be certain your model makes sense, then do transforms. Some models make no sense without transformation.
- John MLv 77 years ago
yes, you need to use the natural log function to convert your independent variables into continuous data. To convert back, use the exponential function.