josef-pkt
commented
3 years ago more general version with an arbitrary set of truncated points. usage for inflation or deflation models. Parameters of main model can be estimated from the truncated sample, similar to standard hurdle model.
Böhning, Dankmar, and Helen E. Ogden. 2020. “General Flation Models for Count Data.” Metrika, July. https://doi.org/10.1007/s00184-020-00786-y.
score test for overdispersion in zero-truncated model against negative binomial variance/dispersion and mean are not orthogonal as in standard Poisson or GLM-Negbin.
Gurmu, Shiferaw. 1991. “Tests for Detecting Overdispersion in the Positive Poisson Regression Model.” Journal of Business & Economic Statistics 9 (2): 215–22. https://doi.org/10.2307/1391790.
3802 PR for truncated, censored and hurdle models from GSOC 2017, needs rebase, review and finishing up
PR looks pretty good, but will need quite a bit of work, I guess, especially for checking that all main post estimation results work correctly
this issue is to add some notes and references for truncated count models.
One issue in usage is in supporting points that are truncated, e.g. predict for y=0 so we can use it to check zero inflation or deflation. Hurdle model will have full data without truncation.