More Generalized Linear Models (GLM) in R: Poisson, Negative Binomial, and Zero-Inflated Models

แชร์
ฝัง
  • เผยแพร่เมื่อ 10 พ.ย. 2024

ความคิดเห็น • 5

  • @murrayirvine1538
    @murrayirvine1538 6 หลายเดือนก่อน +1

    Why did you not use quasi-poisson to account for the over dispersion rather than a negative binomial model? Or is there not much difference?

    • @WestDesignAnalysis
      @WestDesignAnalysis  6 หลายเดือนก่อน

      Thanks for the question. In this particular case which is atheoretical, I showed poisson, negative binomial, and their zero-inflated counterparts. As you note, negative binomial is typically the route for accounting for overdispersion, but what appears as overdispersion can be due to zero-inflation. And notably, I showed a zero-inflated negative binomial model, which could theoretically account for a case where there is both zero-inflation and overdispersion. Hope that makes sense!

    • @WestDesignAnalysis
      @WestDesignAnalysis  6 หลายเดือนก่อน

      I think I misunderstood. You’re saying use quasi-poisson to address over dispersion? Which type of model are you referring to? Quasi-poisson is often used for under dispersion.

    • @murrayirvine1538
      @murrayirvine1538 6 หลายเดือนก่อน +1

      @@WestDesignAnalysis your first answer clears it up, thanks for the response.
      do you have any videos/articles for obtaining probabilities using GLMs/Mixed Models? Cheers

    • @WestDesignAnalysis
      @WestDesignAnalysis  6 หลายเดือนก่อน

      For generalized linear models you can use the predict function in R. I show that in the “intro to GLM” video. If you’re asking about predicted probabilities in generalized linear mixed effect models, that is a bit more complicated.