I have a dataset containing counts (values from 0 to 30) and different covariates. This is what the distribution of the count data looks like:

enter image description here

After fitting a gamlss boosting model in R (with the function gamboostLSS, zinb distribution) I plotted the probability distribution of one observed subject (out of my dataset) to get a first impression of the model. However, the probability distribution doesn't look at all like the expected distribution (zero-inflated negativ binomial, see plot above).

probability distribution of an observed subject

I plotted several probability distributions for different observed subjects, but they all look like the distribution shown above. After checking the three parameters of the zinb model for the observed subjects, I figured that the overdispersion parameter is quite small, which is the reason for the unexpected probability distribution. What could be the reason the overdispersion parameter is so small / way smaller than expected? Does the plot indicate that my model is bad? To find out the optimal hyperparameters I used a 10-fold crossvalidation. Thank you in advance for your help!



Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.