@DougSteinley Thanks Doug. @Nate__Haines and @stephenjwild pointed to some resources that illustrated the likelihood function, which really helped. Seeing that the count portion is adjusted for the likelihood of an observation taking on a zero is really u
@Nate__Haines This article linked by @stephenjwild helps too, making the same points. Basically that because it's a mixture, you still need to adjust the count part for the probability of an obs being a zero (e.g. can't treat it as fixed and known). http
@MatthewBJane @KMKing_Psych So I want to amend my answer slightly, because while it's two separate models the Poisson portion is multiplied by (1 - p_i), which I forgot about. The likelihood is in the image. The Poisson portion is also adjusted to account
RT @statwonk: In analytics, if you don’t account for excess or too few zeros your results are suss. https://t.co/0dpT2mK8wq
In analytics, if you don’t account for excess or too few zeros your results are suss. https://t.co/0dpT2mK8wq