Closed sfaridimehr closed 5 years ago
The call predict(..., type = "zero_part")
will return you predictions for the log odds (if type_pred = "link"
) or the probability (if type_pred = "response"
) for a zero count coming from the degenerate distribution at zero (i.e., not from the Poisson distribution). Note, that these log odds or probabilities are based only on the fixed effects of the zero_part.
All other option for the type
argument in predict()
for zero-inflated or hurdle models give (1 - pi) * exp_count
, where pi
is the probability of coming from the degenerate part (i.e., the same as above), and exp_coint
is the expected counts from the Poisson part.
The hurdle model from pscl package gives this option to predict response, count, probability and zero (which is the ratio that is multiplied by count and gives response). Does "zero_part" from GLMMadaptive package give probability from Poisson or probability from binary logistic regression?