Re: st: comparing coefficients

[Richard Williams <richardwilliams.ndu@gmail.com>]

At 07:07 AM 1/9/2013, David Hoaglin wrote:
> John,
>
> What's wrong with including both genders in a single model and
> expanding the list of predictors to include gender and the interaction
> of gender with each of the current predictors?

Actually, that has some problems of its own. Suppose you ran

use "http://www.indiana.edu/~jslsoc/stata/spex_data/ordwarm2.dta";, clear
reg warm i.yr89 i.white c.age c.ed if male==0
reg warm i.yr89 i.white c.age c.ed if male==1
reg warm i.male##(i.yr89 i.white c.age c.ed)

It isn't hard to see that the last model with interactions gives you 
the exact same results as the 2 groups run separately. Things are 
just parameterized differently.

But if instead you run

ologit warm i.yr89 i.white c.age c.ed if male==0, nolog
ologit warm i.yr89 i.white c.age c.ed if male==1, nolog
ologit warm i.male##(i.yr89 i.white c.age c.ed), nolog

You see the final model does not give exactly the same results as the 
first two models.

Why not? In the final ologit model, you are allowing all the 
variables to have different effects in each group. BUT, you are 
constraining the cut points to be the same for each group, whereas 
when you run the model separately for the two groups the cutpoints 
can differ. That may not be a bad assumption to make, but nonetheless 
it is an assumption.

Probably more important problems are outlined at 
http://www3.nd.edu/~rwilliam/xsoc73994/L31.pdf. The key issue is 
that, unless the two groups have the same residual variability, tests 
of equality across groups are distorted.

The more I work with ordinal models, the more I think you should do 
everything you can to get continuous dependent variables. There are a 
lot of things you can do in OLS regression that don't generalize as 
you would expect to logit and ologit models.

One other point: as far as I know this command from the original post 
is not legitimate, because there is no store option in ologit:

ologit y1 educationmedium educationhigh age income class household 
country if gender=0, or store(step1a)


> David Hoaglin
>
> On Wed, Jan 9, 2013 at 4:30 AM, John Stymans <johnstymans@hotmail.be> wrote:
> > Dear Statalist Users,
> > I have a problem when comparing coefficients across groups.
> > I am using Stata 12.
> > I first run an ordered logit model for two groups (males and females).
> >
> > ologit y1 i.education age income class household country if 
> gender=0, or store(step1a)
> > ologit y1 i.education age income class household country if 
> gender=1, or store(step1b)
> > goal is to compare across gender the effect of education which is 
> a categorical variable with 3 values, hence the use of i..
> >
> > when I use suest:
> >
> > suest step1a step1b
> >
> > test [step1a _y1]i.education=[step1b_y1]i.education
> >
> > the command does not work. I am however able to split education 
> in three values an add two of them to the regression (omitting educationlow);
> >
> > ologit y1 educationmedium educationhigh age income class 
> household country if gender=0, or store(step1a)
> > ologit y1 educationmedium educationhigh age income class 
> household country if gender=1, or store(step1b)
> > and than run suest:
> >
> > suest step1a step1b
> >
> > test [step1a _y1]loweducation=[step1b_y1]loweducation
> >
> > Yet I am unable to compare the difference in significance of 
> education across gender in general. Does anyone know how to solve this issue?
> >
> > Many thanks in advance and best regards,
> >
> > John
> *
> *   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/

-------------------------------------------
Richard Williams, Notre Dame Dept of Sociology
OFFICE: (574)631-6668, (574)631-6463
HOME:   (574)289-5227
EMAIL:  Richard.A.Williams.5@ND.Edu
WWW:    http://www.nd.edu/~rwilliam

*
*   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/