Stata The Stata listserver
[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

st: RE: Cutpoints and Data


From   "Nick Cox" <n.j.cox@durham.ac.uk>
To   <statalist@hsphsun2.harvard.edu>
Subject   st: RE: Cutpoints and Data
Date   Mon, 1 Dec 2003 11:13:22 -0000

Tania Polcyn

> After many unfruitful hours perusing STATA manuals and the 
> Archives, I thought 
> it might be best to query the knowledgeable group on the listserve.
> 
> One of my independent variables for my panel dataset ranges 
> from a scale of 1 
> to 7.  My professor therefore suggested that I investigate 
> ologit and oprobit 
> models for my regressions.  Unfortunately, my dependent 
> variables are 
> uncooperative in this respect because they all have a range 
> of more than 50 
> categories.
> 
> I have looked at a kdensity of my dependent variables and 
> used the tabulate 
> command for assistance in determining where to impose cut 
> points.  I am aware 
> of "hordered" which performs a Hausman-type specification 
> test whether 
> categories in a ologit or oprobit model can be joined or 
> dropped. To this end 
> it estimates another ologit/oprobit model with a collaped response.
> 
> My question though is how do I go about grouping my data? 
> How do I divide my 
> data into for example three ordered groups? I think I am 
> missing some basic STATA manipulation....

I don't follow the logic here. The choice of -ologit- 
or -oprobit- as model should be based partly on 
what is appropriate scientifically and partly on the nature 
of the dependent variable (and partly on other matters). 
The nature of the independent variables is secondary. 

Given something which has >50 more categories, 
a first question here is whether that variable 
is _ordered_, not clear from that you say. 
Also, if it is categorical, I can't see that
-kdensity- is appropriate as a way of looking 
at the distribution. This takes your underlying 
numbers very literally and smooths something 
that quite possibly should not be smoothed. 

The second question is -- irrespective of how 
you do it, on which others have advised -- how 
much information will this throw away when 
you degrade >50 categories to 3? 

Nick 
n.j.cox@durham.ac.uk 
*
*   For searches and help try:
*   http://www.stata.com/support/faqs/res/findit.html
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/



© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index