I have a dataset with one dependent (continuous) variable that is not normally distributed and three independent (continuous) variables.I would like to run GLM with one dependent variables against three independent variables. Can I run a GLM without factor (categorical variable) as an independent variable?

  • $\begingroup$ The question is not clear to me. You already mentioned that you've three continuous variable, where does the last factor variable come from? $\endgroup$ – user9292 Jan 6 '17 at 16:02
  • $\begingroup$ Yes, the problem is that I have no factor variable. Therefore, I asked whether I should run a GLM with my current dataset. $\endgroup$ – Bo Sann Jan 6 '17 at 16:08
  • $\begingroup$ OK. What kind of response variable are you dealing with? You already mentioned that it's not normally distributed. $\endgroup$ – user9292 Jan 6 '17 at 16:13
  • $\begingroup$ Response variable is frequency or count variable. So, I choose a log link function as the response variable should follow poisson distribution. Thanks indeed. $\endgroup$ – Bo Sann Jan 6 '17 at 16:16
  • $\begingroup$ f frequency, then you need to use either Poisson or Negative Binomial (check for over dispersion) distribution with a log link function. $\endgroup$ – user9292 Jan 6 '17 at 16:20

Absolutely, you'll want to run a GLM based on a Gaussian distribution for continuous data (For count data you want to use a Poisson distribution). For a Gaussian (normal) distribution it doesn't matters if your continuous variable is not normally distributed. The assumption is the errors from the model are normally distributed. GLM covers regression methods for many types of data: continuous, count, categorical, binomial etc. The important part is that you select the right distribution type for the link function.

See here for more details: https://en.wikipedia.org/wiki/Generalized_linear_model

  • $\begingroup$ Thank you for your answer. My question does not mean to normal or non-normal distribution of the data. It just means to that should I run a GLM without a factor as independent (not meaning dependent) variable. $\endgroup$ – Bo Sann Jan 6 '17 at 16:13
  • $\begingroup$ His response variable is not normally distributed. $\endgroup$ – user9292 Jan 6 '17 at 16:14
  • $\begingroup$ That's not an issue. Yes you don't need categorical variables to run a GLM regression. Just make sure you select the right distribution according to your dependent variable. $\endgroup$ – Morgan Ball Jan 6 '17 at 16:16

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.