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: Count data with known upper limit
From
Austin Nichols <[email protected]>
To
"[email protected]" <[email protected]>
Subject
Re: st: Count data with known upper limit
Date
Thu, 19 Sep 2013 11:53:29 -0400
Sébastien Fosse <[email protected]> :
Simply model expected y=proportion passing, ranging from 0 to 1
instead of 0 to 5, as invlogit(Xb), using -glm- with a logit link and
appropriately robust SE. See also
http://www.stata.com/statalist/archive/2008-12/msg01005.html
You could also set up a system of correlated binary regressions, if
you are worried about aggregating across distinct pass/fail criteria,
or explore IRT models. See also -help sem-
On Thu, Sep 19, 2013 at 11:27 AM, Sébastien Fosse
<[email protected]> wrote:
> Dear Statalisters,
> I am trying to analyze count data where the outcome is the sum of 5 dichotomous variables -each of these 5 variables is associated with a specific “Fail/Pass” criterion. Given this upper limit (of 5), such data don´t seem to be “very Poisson-like”.
> Alternatively, I have also thought of an ordered logit model with 5 ordered categories (that is, one category for each possible count). However, I am not sure that those categories would have much sense given the sum structure of those outcome data (that is, a given sum outcome may result from various combinations). To possibly solve this issue, I have heard that the homogeneity of those 5 constructed categories may be checked but I am unsure about the way to do such a check.
> Any suggestion or references to select the appropriate model (and Stata commands) for those count data with a known upper limit would be very helpful.
> Many thanks.
> Sébastien
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/faqs/resources/statalist-faq/
* http://www.ats.ucla.edu/stat/stata/