1
$\begingroup$

I have a non normally distributed variable X (also its residuals are non-normally distributed). I have also two predictor variables, one categorical (two categories) and one discrete. Which regression should I use to test the association between the predictors and X?

I am not a statistics expert, and I am getting lost in a sea of different regression methods.

Thank you!

$\endgroup$
  • 1
    $\begingroup$ I have a feeling that this question is better suited for CrossValidated (the SE site dedicated to statistics). $\endgroup$ – Lord_Farin Jul 1 '13 at 17:23
  • $\begingroup$ (1) Do you expect $X$ to depend linearly on the other variables or not? (2) What do you expect the scatter of the residuals to look like--will it be approximately Normal, or perhaps will it be skewed or expected to exhibit some isolated large variations? Perhaps the amount of scatter will vary with the values of the independent variables? These are basic criteria used to identify appropriate regression procedures. If you let us know where your situation stands with respect to these criteria, you will likely get some focused, relevant answers. $\endgroup$ – whuber Jul 1 '13 at 19:53
1
$\begingroup$

What is X (by the way, the dependent variable is usually called Y). How is it distributed? In what way are residuals from the model non-normal?

If the DV is categorical, you probably want some form of logistic regression.

If it is a count, you might want Poisson, negative binomial or some related model.

If it is continuous, you probably want regular regression, but you might want to transform one or more of your variables. Or, you might need some form of robust regression.

| cite | improve this answer | |
$\endgroup$
  • $\begingroup$ I respond to your questions. The dependent value is slightly skewed towards the left. The scatter respect to the discrete explanatory variable is not necessarily linear, and it does display some isolated large variations. The other explanatory is categorical, with two categories. Thanks Peter, as it is a count, I will look at Poisson or negative binomial. $\endgroup$ – retrot Jul 2 '13 at 8:50

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.