1
$\begingroup$

From what I've studied so far, GLM's are to be used when the error term of a response variable is not assumed to be normally distributed. However, I also read that sometimes a transformation of a response variable is used (after which regular linear regression is applied) to "normalize" it? If so, then what is the difference between those two modelling approaches?

$\endgroup$
4
  • 3
    $\begingroup$ See stats.stackexchange.com/questions/67626/understanding-of-glm $\endgroup$
    – Nick Cox
    Nov 14, 2018 at 18:41
  • $\begingroup$ If GLM means Generalized Linear Model, then linear regression is a spacial case of GLM. It seems you exclude the linear regression from GLM. $\endgroup$
    – user158565
    Nov 15, 2018 at 2:31
  • $\begingroup$ GLMs are designed to capture nonlinear relationships . Apparently, the assumption cited by you in the initial part of your question invokes the idea of non-linear/random effects. The transformation followed by regression has a different perspective and thus, the two modelling approaches differ from each other substantively. $\endgroup$
    – user10619
    Nov 15, 2018 at 9:54
  • $\begingroup$ However, I also read that sometimes a transformation of a response variable is used (after which regular linear regression is applied) to "normalize" it? What is the source for it.This part seems to create confusion to the main theme that is invoked by GLM. Regression is not an integral part of GlM. $\endgroup$
    – user10619
    Nov 15, 2018 at 11:41

0