A simulated dataset with a strong zero-inflation component, designed to
illustrate where zeroinflTweedieGLM() fits clearly better than the base
tweedieGLM(). The count component depends only on x1; the
zero-inflation component is driven solely by x2 (which is independent of
x1). Generated from a compound Poisson-Gamma (Tweedie, p = 1.5) with
structural zeros, then ceiling()-ed to non-negative integers.
A data frame with 400 rows and 3 columns:
- y
Non-negative integer count response.
- x1
Continuous predictor for the count component.
- x2
Continuous predictor for the zero-inflation component
(independent of x1).
Source
Simulated; see data-raw/DATASET.R in the package source for the
generating script.
Details
True data-generating model:
Count component: log(mu) = 1.8 + 0.9 * x1, with phi = 2.0
and Tweedie power p = 1.5.
Zero-inflation component: logit(pi) = 0.5 - 2.5 * x2.
Approximately 61% of observed responses are zero.