+ Reply to Thread
Results 1 to 4 of 4

Thread: Poisson data pairwise comparisons valid?

  1. #1
    Points: 3,789, Level: 38
    Level completed: 93%, Points required for next Level: 11

    Posts
    27
    Thanks
    3
    Thanked 1 Time in 1 Post

    Poisson data pairwise comparisons valid?




    Hi everyone,

    I'm a bit confused if what I am doing is valid. I thought I would get your two cents worth.

    I have done a poisson linear regression using a GLM, comparing three categories.
    I did this because of an excess of zero values.

    My data fit the model (omnibus test p<0.001). And I am finding a significant effct of groups.

    I asked for pairwise comparisons and means. My question is, are these pairwise comparisons valid to use even though my data does not fit a normal distribution? Is the computer program compensating for the fact that I am asking for the means while using a poisson regression? What about reporting the means?

    Thanks a lot for any help.

    lken

  2. #2
    Devorador de queso
    Points: 95,819, Level: 100
    Level completed: 0%, Points required for next Level: 0
    Awards:
    Posting AwardCommunity AwardDiscussion EnderFrequent Poster
    Dason's Avatar
    Location
    Tampa, FL
    Posts
    12,935
    Thanks
    307
    Thanked 2,629 Times in 2,245 Posts

    Re: Poisson data pairwise comparisons valid?

    Well we really don't know what program you're using.

    Or what type of data you have. You said you used a poisson GLM because of an excess of zero values... but that's not a good reason to use a poisson.

    Can you just describe what you did in a little more detail?
    I don't have emotions and sometimes that makes me very sad.

  3. #3
    Points: 3,789, Level: 38
    Level completed: 93%, Points required for next Level: 11

    Posts
    27
    Thanks
    3
    Thanked 1 Time in 1 Post

    Re: Poisson data pairwise comparisons valid?

    Hi Dason,

    Thanks for taking the time to go through this with me.

    I have data on the amount of rot in trees in three different climatic areas (6-9 sites each, about 4-6 trees a site). I want to figure out if the different climatic areas have different amounts of rot in the trees, while accounting for factors like the diameter of the tree and how many years the tree has been dead for. I am using SPSS.

    I first tried a univariate analysis with the cm of rot as the outcome, climatic zone as the predictor, and diameter and years dead as covariates.
    However, my data did not pass Levene's test for equality of error variances (p=0.004). I beleive this is because many trees had no rot in them at all, and my data were zero inflated.

    I then used a GLM with a poisson distribution. I was following instructions from the UCLA website. However, my Chi-squared statistic was about 3, indicating my model was overdispersed. I then used a negative binomial distribution, and my Chi-squared is now 1.3. I don't know how to further check if my model fits other than looking at the omnibus test (p<0.001)...

    I asked for pairwise comparisons between climatic zones. Would it be valid to graph the means with the standard errors, explain that diameter and years since death were held at constants, and report p values?

    Perhaps I should go to a non-parametric model? But then I don't know how to control for my covariates.

  4. #4
    Points: 3,789, Level: 38
    Level completed: 93%, Points required for next Level: 11

    Posts
    27
    Thanks
    3
    Thanked 1 Time in 1 Post

    Re: Poisson data pairwise comparisons valid?


    Sorry, my Pearson Chi-Square is 1.067, so I think I can say my data fit. If that is the case, can I then graph the means for my three climatic zones?

+ 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