Message-ID: <alpine.DEB.2.02.1209110832290.4898@paninaro.uibk.ac.at>
Date: 2012-09-11T06:37:34Z
From: Achim Zeileis
Subject: Zero inflated Models- pscl package
In-Reply-To: <1347308619164-4642689.post@n4.nabble.com>
On Mon, 10 Sep 2012, Filipa J. wrote:
> Dear R users,
>
> I want to apply zero inflated models with continuous and categorical
> variables and I used pscl package from R and the zeroinf() function. My
> question are the follow:
>
> a) The value of fitted.values is mu or (1-p)*mu? where p is the
> probability of zero came form a zero point mass
See vignette("countreg", package = "pscl"), especially Appendix C.
The default predict(..., type = "response") and fitted(...) both compute
the expected value Equation (8). This corresponds to your (1-p)*mu but
uses a different notation.
> b) If mu is zero, how do i know if it is a zero from the zero point mass
> or from the count process?
You can inspect all component: Your "mu" ("exp(x'b)" in the notation of
the paper) can be computed with predict(..., type = "count") and "p" ("pi"
in the paper) with predict(..., type = "zero").
Best,
Z