Bookmark and Share

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: comparing probit coefficients across two groups


From   Ebru Ozturk <[email protected]>
To   <[email protected]>
Subject   RE: st: comparing probit coefficients across two groups
Date   Mon, 7 Jan 2013 12:33:07 +0200

But you said: "after controlling for hetero, you can test to see whether interaction terms, e.g. Gender*whatever, are significant"

So oglm produces interaction terms. How do we interpret it?

----------------------------------------
> From: [email protected]
> Date: Sun, 6 Jan 2013 17:32:15 -0500
> Subject: Re: st: comparing probit coefficients across two groups
> To: [email protected]
>
> Again I don't quite understand the interest in marginal effects for
> interaction terms. For whatever reason, Stata does not provide them.
> The Norton article cited earlier makes the case that you actually
> ought to compute a marginal effect for every case, since sometimes the
> effect will be positive, sometimes negative, sometimes significant,
> sometimes not significant. You might be able to figure out how to
> adapt the FAQ, but even if you could you would only be getting a
> marginal effect at the mean. Being stuck with Stata 10 further
> handicaps you because you can't use the margins command.
>
> The request for marginal effects of interaction terms has come up
> several times, but I have never heard a good explanation of what is
> wrong with the way margins does it, e.g. If the model includes A, B,
> and A*B, you get marginal effects for A and B but not A*B. Basically
> the interaction term is taken into account when computing the marginal
> effects for A and B. This makes sense to me since the interaction term
> cannot vary independently of the variables used to compute it.
>
> But if I am missing something I would like to hear what it is.
>
> Sent from my iPad
>
> On Jan 6, 2013, at 4:19 PM, Ebru Ozturk <[email protected]> wrote:
>
> > Thank you very much for this extra explanation. I have one more question, it might be a wrong question though. When we compare coefficients across groups we don't produce marginal effects for interaction terms or for main effects. In non-linear models, the issue of interaction terms is obvious. So while comparing coefficients by heterogeneous choice models how do we deal with this problem?
> >
> > ----------------------------------------
> >> From: [email protected]
> >> Date: Sun, 6 Jan 2013 14:01:45 -0500
> >> Subject: Re: st: comparing probit coefficients across two groups
> >> To: [email protected]
> >>
> >> In all modesty I agree with Bill. ;-) I think my slides are pretty
> >> clear on how to proceed, but if not you can read the articles the
> >> slides draw from. But just to be explicit,
> >>
> >> * after controlling for hetero, you can test to see whether
> >> interaction terms, e.g. Gender*whatever, are significant
> >>
> >> * you can try to make the hetero problem go away by modifying the
> >> model, e.g. In my example I showed how adding articles^2 made a hetero
> >> equation unnecessary
> >>
> >> * you can instead take Long's approach and just compare predicted
> >> probabilities for the two groups,
> >>
> >> * and, although not in my notes, Maarten's suggestion of looking at
> >> odds ratios has been suggested by others.
> >>
> >> So in short, a lot of information is already out there, and there is
> >> code that shows how to do things in Stata (some of it is in my slides
> >> and other code can be found in the articles that are mentioned). I
> >> don't think there is much to be gained by retyping things that have
> >> already been written up so I think the thing to do now is just look
> >> over the materials already suggested more carefully.
> >>
> >> Sent from my iPad
> >>
> >> On Jan 6, 2013, at 12:22 PM, William Buchanan
> >> <[email protected]> wrote:
> >>
> >>> Hi Ebru,
> >>>
> >>> If you looked at the slide shows that Richard directed you towards it seems pretty clear and gives you working syntax that you could likely generalize to your specific research/variables.
> >>>
> >>> <from below>
> >>>>> http://www3.nd.edu/~rwilliam/xsoc73994/L31.pdf
> >>>>> http://www3.nd.edu/~rwilliam/xsoc73994/L31H.pdf
> >>>
> >>>
> >>> HTH,
> >>> Billy
> >>>
> >>>
> >>> On Jan 6, 2013, at 9:08 AM, Ebru Ozturk wrote:
> >>>
> >>>> The thing is I still do not get how you compare coefficients across groups (i.e. female versus male). The papers generally explain how to compare across models not groups. For instance, how do we interpret the outcome of oglm estimation with heterogeneous choices? How do we compare male and female? Do we need to run two different models one with female and the other with male? Then compare the coefficients?
> >>>>
> >>>> oglm warm yr89 male white age ed prst, het(yr89 male)
> >>>> oglm warm yr89 female white age ed prst, het(yr89 female)
> >>>>
> >>>> ----------------------------------------
> >>>>> From: [email protected]
> >>>>> Date: Fri, 4 Jan 2013 16:22:46 -0500
> >>>>> Subject: Re: st: comparing probit coefficients across two groups
> >>>>> To: [email protected]
> >>>>>
> >>>>> Well, yes. That is why I called it "Using heterogeneous choice models
> >>>>> to compare logit and probit coefficients across groups." ;-) It builds
> >>>>> on Allison's example and gives various others. But see my handout for
> >>>>> other takes on the problems and possible solutions.
> >>>>>
> >>>>> Sent from my iPad
> >>>>>
> >>>>> On Jan 4, 2013, at 4:13 PM, Ebru Ozturk <[email protected]> wrote:
> >>>>>
> >>>>>> Allison's (1999) paper can answer the question of e.g. is the effect of education on income greater for men than it is for women?. But when I read the paper titled "Using Heterogeneous Choice Models To Compare Logit and Probit Coefficients Across Groups" (Williams, 2009) do you think it also focuses on comparison of coefficients across groups?
> >>>>>>
> >>>>>> ----------------------------------------
> >>>>>>> Date: Fri, 4 Jan 2013 15:24:03 -0500
> >>>>>>> To: [email protected]; [email protected]
> >>>>>>> From: [email protected]
> >>>>>>> Subject: Re: st: comparing probit coefficients across two groups
> >>>>>>>
> >>>>>>> At 01:25 PM 1/4/2013, Ebru Ozturk wrote:
> >>>>>>>
> >>>>>>>> Dear All,
> >>>>>>>>
> >>>>>>>> I run two models below with Probit estimation on Stata 10. First
> >>>>>>>> model covers firms that collaborate and the second model covers
> >>>>>>>> firms that do not collaborate. I want to compare the coefficients'
> >>>>>>>> of "s_breadth" variable. I cannot understand quite that whether
> >>>>>>>> -oglm- command applies to this type of problem. Can anyone make it clear?
> >>>>>>>>
> >>>>>>>>
> >>>>>>>> ///// example //////
> >>>>>>>>
> >>>>>>>> probit radical_d businessgrp logemp continuous_rd process_inn
> >>>>>>>> product_inn total_innv ind1 s_breadth if collab_developd==1
> >>>>>>>> estimates store collab1
> >>>>>>>> probit radical_d businessgrp logemp continuous_rd process_inn
> >>>>>>>> product_inn total_innv ind1 s_breadth if collab_developd==0
> >>>>>>>> estimates store collab0
> >>>>>>>> suest collab1 collab0
> >>>>>>>> test [collab1]s_breadth =
> >>>>>>>> [collab0]s_breadth
> >>>>>>>
> >>>>>>> First off, given dichotomous DV and probit link, you might as well
> >>>>>>> use -hetprob- as opposed to -oglm-.
> >>>>>>>
> >>>>>>> Second there are various issues involved in comparing coefficients
> >>>>>>> across groups. Various solutions have been proposed. Personally I
> >>>>>>> think the discussions of the problems may be stronger than the
> >>>>>>> discussions of the solutions. For a summary, see
> >>>>>>>
> >>>>>>> http://www3.nd.edu/~rwilliam/xsoc73994/L31.pdf
> >>>>>>>
> >>>>>>> http://www3.nd.edu/~rwilliam/xsoc73994/L31H.pdf
> >>>>>>>
> >>>>>>> But, reading the articles listed at the end (starting with Allison's
> >>>>>>> 1999 paper) will yield a clearer picture.
> >>>>>>>
> >>>>>>>
> >>>>>>> -------------------------------------------
> >>>>>>> Richard Williams, Notre Dame Dept of Sociology
> >>>>>>> OFFICE: (574)631-6668, (574)631-6463
> >>>>>>> HOME: (574)289-5227
> >>>>>>> EMAIL: [email protected]
> >>>>>>> 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/
> >>>>>> *
> >>>>>> * 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/
> >>>>>
> >>>>> *
> >>>>> * 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/
> >>>> *
> >>>> * 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/
> >>>
> >>>
> >>> *
> >>> * 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/
> >>
> >> *
> >> * 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/
> > *
> > * 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/
>
> *
> * 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/ 		 	   		  
*
*   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/


© Copyright 1996–2018 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   Site index