I have used GAMLSS to fit a zero-inflated model. However, when I then turn around and FIT that model, it produces absolutely no zeroes at all (even though the data used to fit the model is about 40 % zeroes).
What's going on here?
Asked
Active
Viewed 49 times
0
-
Not sure why you were downvoted, it's a perfectly okay question, but this has already been answered in detail on site. Let me see if I can find it. – Glen_b May 15 '17 at 02:55
-
See [Can a Tweedie glm predict exact zeros?](https://stats.stackexchange.com/questions/174121/can-a-tweedie-glm-predict-exact-zeros) -- as pointed out in the answer there, this question could arise in any model where your model includes some proportion of exact 0's -- for some reason it's a not-so-uncommon confusion across a variety of such models. Note that *fitted values* are estimated *means* and the fitted mean will generally not be zero unless all the observations are 0 -- for nonnegative data even a single non-zero value will yield a mean that's larger than 0. – Glen_b May 15 '17 at 03:02
-
Perhaps you're expecting your model to predict the conditional *mode* or a conditional *median*. It would be a reasonable thing to want but that's not what you get when you ask for the fitted value. – Glen_b May 15 '17 at 03:16