+ Reply to Thread
Results 1 to 3 of 3

Thread: query regarding generalized linear models

  1. #1
    Points: 207, Level: 4
    Level completed: 14%, Points required for next Level: 43

    Posts
    3
    Thanks
    0
    Thanked 0 Times in 0 Posts

    Smile query regarding generalized linear models




    hello all,
    I have a query regarding GZLMs.
    I have a variable A non-normally distributed and I also take the logA, which is normally distributed to use as my dependent variables (one at a time!), and I have e.g. 2 categorical independent variables. I can run two GZLMs: one for A-independents, and one for logA-independents. Since the GZLM does not make assumptions on normality, is there any reason why results would be better for one model over the other, if I specify family(gaussian) and link(identity) in both cases ?

    Thanks in advance

  2. #2
    Omega Contributor
    Points: 38,432, Level: 100
    Level completed: 0%, Points required for next Level: 0
    hlsmith's Avatar
    Location
    Not Ames, IA
    Posts
    7,006
    Thanks
    398
    Thanked 1,186 Times in 1,147 Posts

    Re: query regarding generalized linear models

    The residuals need to be normally distributed not the dependent variable for establishing normality.

    Does the logged model fit any better?
    Stop cowardice, ban guns!

  3. #3
    Points: 207, Level: 4
    Level completed: 14%, Points required for next Level: 43

    Posts
    3
    Thanks
    0
    Thanked 0 Times in 0 Posts

    Smile Re: query regarding generalized linear models


    Quote Originally Posted by hlsmith View Post
    The residuals need to be normally distributed not the dependent variable for establishing normality.

    Does the logged model fit any better?

    hello, thanks for the reply,
    yes I am aware that the normality assumption is with regards to the error terms, but my question is whether using a GZLM with a non-normally distributed variable and again with its normally distributed log would have any reason to give better results in one instance rather than the other. I.e. I'm asking whether there is any theoretical reason to believe that the GZLM will work better or worse with a normally distributed variable. I'm not saying the dependent should be normally distributed, I'm saying IF I happen to have a normally-distributed dependent (logA) will the test perform better or worse than it will if I use the untransformed non-normally distributed values (A). And if yes/no, why?

    thanks again

+ Reply to Thread

           




Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts






Advertise on Talk Stats