Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: st: zoib
Maarten Buis <firstname.lastname@example.org>
Re: st: zoib
Mon, 31 Oct 2011 16:48:31 +0100
On Mon, Oct 31, 2011 at 3:08 PM, Jet wrote:
> Hello, Maarten and other experts in zoib regression for proportion outcomes:
Remember, that you _must_ say where you got a user written program
from. This is not some silly rule, it is in your own interest, as it
makes your question answerable, and I guess you want to ask an
answerable question. I am assuming you are referring to -zoib-
available from SSC.
> For the 0<DV<1 model, do we interpret the exp(B) as changes in odds
> ratio (in having value of 1) or changes in proportion?
Neither, it is similar but not the same as an odds ratio. I called
that statistic relative proportion ratios, a description can be found
in -help betafit-, which is also available from SSC.
> For the
> estimated margins in the post-estimation function [predict(prcond)],
> do we interpret the ouptut as the predicted proportion or probability?
prcond, refers to the proportion conditional on being more than 0 and
less than 1. If you want the probability of getting a 0 or 1 you
choose pr0 or pr1 respectively.
However, in both cases you are interpreting your results conditional
on not being a 0 or 1, and I would expect that that in most cases
makes no sense whatsoever. I would (almost) always use
-predict(proportion)- instead, that would tell you the effect on the
proportion regardless whether it is a 0, 1, or something in between.
If you are really only interested in the proportions between 0 and 1,
than you can just ignore all observations with 0s or 1s and analyse
the model with -betafit-, it will give you exactly the same results.
-zoib- only makes sense when you do not want to ignore the 0s and 1s.
Hope this helps,
Maarten L. Buis
Institut fuer Soziologie
* For searches and help try: