Statalist


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

st: RE: Treatment of outcome categories in Ologit


From   "Lee, Albert" <Albert.Lee@hud.gov>
To   "'statalist@hsphsun2.harvard.edu'" <statalist@hsphsun2.harvard.edu>
Subject   st: RE: Treatment of outcome categories in Ologit
Date   Mon, 28 Sep 2009 17:29:43 -0400

You may only want to do an ologit if the numeric value represent ordered categories, and the distance between these categories does not matter.  For example, 3.6 is preferred to 3.5; and 4.2 is preferred to 3.6.  However, the degree of preference in the latter case is not six time higher than in the former case.  If orders matter, and not the degree from one category to another, then you can group your score into categories, i.e.,

egen ord_cat_score=group(score)

The ord_cat_score will recode your scores in order into 1, 2, 3, etc... regardless of cutoff.  You can run an ologit on the ord_cat_score.

I suggest you look into Long and Freese, who has a nice treatment on ologit.  http://www.stata.com/bookstore/regmodcdvs.html 

If the order and the difference between the score matter, then you may want to consider a fractional logit.  To do that you may want to rescale your scores from 0 to 1 by dividing your scores by 50, i.e., 0.07=3.5/50.  Fractional logit is a better option than OLS for the reasons you mentioned below.  This link provides some details: http://www.stata.com/statalist/archive/2005-10/msg00883.html

I hope it helps.

Albert.

-----Original Message-----
From: owner-statalist@hsphsun2.harvard.edu [mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of Rijo John
Sent: Monday, September 28, 2009 4:56 PM
To: stata
Subject: st: Treatment of outcome categories in Ologit

Hi statalist,

I have a dependent variable (a score) which is ordinal in character.
However, it has around 30 or so categories the value of which are
continuous ranging anywhere from 0 to 50. There could be values like
3.5, 3.6, 4.2 etc. My questions are:

1) Is Ologit the right estimation method for this type of decision
variables? ( I should say that I am not particularly interested in
analyzing each cutpoints. I am only interested in how the changes in
explanatory variables affect the over all score. I dont expect OLS to
return good results because the dep var is both bounded and ordinal in
character.
2) Given that ologit is the right method, how does stata treat the
values in my score variable while estimating? Does it round off the
scores into whole numbers before estimating or does it treat them as
they are. If it rounds off obviously 3.5 score and 3.6 score will be
treated same. Stata description on it tells me "The actual values
taken on by the dependent variable are irrelevant, except that larger
values are assumed to correspond to "higher" outcomes." My estimation
gives me as many coefficients for cut-points as there are categories,
which makes me assume that it must not be rounding those numbers. Can
someone clarify?

Thanks.
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   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