Statalist The Stata Listserver


[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

RE: st: RE: prvalue / dummy variables


From   "Maarten Buis" <M.Buis@fsw.vu.nl>
To   <statalist@hsphsun2.harvard.edu>
Subject   RE: st: RE: prvalue / dummy variables
Date   Tue, 4 Jul 2006 18:26:09 +0200

In order to get probabilities, you will always have to specify values for all variables in your model. This is not some weird thing of -prgen- or Stata, it is a basic characteristic of a multinomial logit model (and other models like logit, probit, poisson, and more). Closest thing to what you want is setting the other variables at their mean, but be prepared for some criticism. I'd guess that Scott Long, Regression models for categorical dependent variables using Stata, Stata Press has a nice discussion of why that is, but I can't check right now since all my books are in boxes (I am moving office sometime soon).

HTH,
Maarten

-----------------------------------------
Maarten L. Buis
Department of Social Research Methodology 
Vrije Universiteit Amsterdam 
Boelelaan 1081 
1081 HV Amsterdam 
The Netherlands

visiting adress:
Buitenveldertselaan 3 (Metropolitan), room Z214 

+31 20 5986715

http://home.fsw.vu.nl/m.buis/
-----------------------------------------

-----Original Message-----
From: owner-statalist@hsphsun2.harvard.edu [mailto:owner-statalist@hsphsun2.harvard.edu]On Behalf Of Jochen Hirschle
Sent: dinsdag 4 juli 2006 17:50
To: statalist@hsphsun2.harvard.edu
Subject: Re: st: RE: prvalue / dummy variables

thank you for your reply Marteen.
Do you (or someone else on this list) know a possibility of how I could predict those probabilities (after multinomial regression) not for individuals - where saying that he / she is 52% female sounds indeed strange - but for two groups which have an identical underlying demographic structure (52% females; 48% males) but completely differ in one other aspect (say the one group containing 100% high educated persons where the other group containing 100% low educated persons)?

Thank you once again.
Jochen

-------- Original-Nachricht --------
Datum: Tue, 4 Jul 2006 11:53:10 +0200
Von: Maarten Buis <M.Buis@fsw.vu.nl>
An: statalist@hsphsun2.harvard.edu
Betreff: st: RE: prvalue / dummy variables

> --- Jochen Hirschle wrote:
> > My question concerns the use of dummy-variables
> > within prvalue. As I only wish to control for
> > certain (dummy)variables (sex, place of birth,
> > etc.) and didn't wanted to produce separate
> > estimations for each value (e.g. sex=0, sex=1),
> > I was using the mean option for the dummy
> > variables just as for metric variables
> > (e.g. age). Does that make sense? How can I
> > interpret the estimated probabilities then? Can
> > I say that these probabilities apply for a group
> > which has an underlying structure concerning sex,
>
> Jochen:
> Say you are calculating the probabilities for
> different values of age but keep the effects of
> other variables at their mean, than you are looking
> at how the probability changes as age changes for
> an otherwise "mean/normal/typical" person. So if
> you keep the variable sex at it's mean you are
> looking at the effect of age for someone who is say
> 52% female. Now, some would argue that someone who
> is 52% female is not very typical, but you can also
> see this as averaging the effect of age over males
> and females. I have no problems with the latter
> interpretation but others do. So it is probably easier
> to use for nominal variables like sex and birthplace
> the mode to choose the "typical" individual, i.e. fix
> continuous variables at their mean and nominal
> variables at their mode.
>
> HTH,
> Maarten
>
> -----------------------------------------
> Maarten L. Buis
> Department of Social Research Methodology
> Vrije Universiteit Amsterdam
> Boelelaan 1081
> 1081 HV Amsterdam
> The Netherlands
>
> visiting adress:
> Buitenveldertselaan 3 (Metropolitan), room Z214
>
> +31 20 5986715
>
> http://home.fsw.vu.nl/m.buis/
> -----------------------------------------
>
>
> *
> *   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/

--
Jochen Hirschle
Sülzburgstr. 40 // 50937 Köln
email j_hirschle@gmx.de
tel 0221.4064541

Echte DSL-Flatrate dauerhaft für 0,- Euro*!
"Feel free" mit GMX DSL! http://www.gmx.net/de/go/dsl
*
*   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/

*
*   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