2

I have two models exactly similar, but I’m using a Poisson distribution for one and a Bernoulli distribution for the other. Can I trust the information coming out of loo to help me choose?

The Bernoulli model comes on top.

rbpa <- brm(status ~ ... , family = poisson(), prior = 
            set_prior('normal(0, 4)', class = 'b'),  
            cores = 6, iter = 4000)

rbba <- brm(status ~ ... , family = bernoulli, 
            prior(normal(0, 4), class = b), cores = 6, 
            iter = 4000)

loo(rbpa,rbba)
kjetil b halvorsen
  • 63,378
  • 26
  • 142
  • 467
mountaingoat19
  • 21
  • 1
  • This isn‘t a coding question, but a statistics question. You‘ll probably find answers on Cross Validated. – deschen Nov 30 '21 at 23:57
  • This question was answer here: https://discourse.mc-stan.org/t/using-loo-function-to-help-me-choose-a-distribution/25511 – mountaingoat19 Dec 01 '21 at 16:19
  • 1
    What's your response variable? Bernoulli is for binary data and Poisson is for count data. – Glen Dec 01 '21 at 17:06

1 Answers1

0

Methods for model selection like loo have nothing to do with the problem. Bernoulli distribution is a distribution for the binary variable (0's or 1's) while Poisson distribution can take any non-negative integers like counts. You can't use Bernoulli distribution instead of Poisson or the other way around.

Tim
  • 108,699
  • 20
  • 212
  • 390