Statalist


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

RE: st: Question about multinomial logistic regression and random effects with multiply imputed data


From   "Barksdale, Crystal" <cbarksda@jhsph.edu>
To   <statalist@hsphsun2.harvard.edu>
Subject   RE: st: Question about multinomial logistic regression and random effects with multiply imputed data
Date   Mon, 4 Aug 2008 11:29:36 -0400

Dear Maarten,

Thank you very much for your suggestions and assistance!  This has truly been helpful.  I have, hopefully, one last question pertaining to this matter.  When I run the commands as you provided, and then display the estimates, they appear ordered.  Is there a way to display the estimates so that they correspond to the way the variables are entered into the model?  Also, I am guessing that your example could also be used/modified to use with gllamm, as that will allow me to the random effects modeling? 

Thanks again for your invaluable assistance!
Crystal 

Date: Thu, 31 Jul 2008 16:33:55 +0100 (BST)
From: Maarten buis <maartenbuis@yahoo.co.uk>
Subject: RE: st: Question about multinomial logistic regression and random effects with multiply imputed data
 
- --- "Barksdale, Crystal" <cbarksda@jhsph.edu> wrote:
> We are considering running the analyses on each multiply imputed
> dataset (separately), and then trying to combine the results
manually;
> however, we are having some difficulty figuring out how to store the
> parameter estimates and standard errors for each variable and it's
> corresponding category. For example, if I run a multinomial logistic
> regression with an outcome variable with 3 categories, I will get
> three separate estimates of the predictor. It is not immediately 
> clear how to save these three separate estimates and standard errors,
> to use ultimately in combining the estimates across the multiply
> imputed datasets. 
 
In the example below I use the rules from
http://www.stat.psu.edu/~jls/mifaq.html (the rules for computing the
degrees of freedom differs from the one used in mim, it is up to you to
do some reading and find out which one you like best)
 
*--------------------- begin example ---------------------------
sysuse auto, clear
recode rep78 1/2=3
ice rep78 foreign mpg gear_ratio, clear m(5)
 
mlogit rep78 foreign mpg gear_ratio if _mj == 1
matrix b = e(b)'
matrix v = e(V)
matrix v = vecdiag(v)'
forvalues i = 2/5 {
	mlogit rep78 foreign mpg gear_ratio if _mj == `i'
	matrix b = b, e(b)'
	matrix vmat = e(V)
	matrix v = v, vecdiag(vmat)'
}
drop _all
svmat b
svmat v
egen qbar = rowmean(b*)
egen ubar = rowmean(v*)
egen b = rowsd(b*)
replace b = b^2
gen t = ubar + (1 + 1/5)*b
gen se = sqrt(t)
gen df = (5-1)*(1 + 5*ubar/(5+1)*b)^2
*------------------------- end example ------------------------
(For more on how to use examples I sent to the Statalist, see
http://home.fsw.vu.nl/m.buis/stata/exampleFAQ.html )
 
Hope this helps,
Maarten
 
- -----------------------------------------
Maarten L. Buis
Department of Social Research Methodology
Vrije Universiteit Amsterdam
Boelelaan 1081
1081 HV Amsterdam
The Netherlands
 
visiting address:
Buitenveldertselaan 3 (Metropolitan), room Z434
 
+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 Maarten buis
Sent: Tuesday, July 29, 2008 6:04 PM
To: statalist@hsphsun2.harvard.edu
Subject: Re: st: Question about multinomial logistic regression and random effects with multiply imputed data

--- Crystal Barksdale <cbarksda@jhsph.edu> wrote:
> I am wondering how (and if) I can do a multinomial random effects
> logistic regression with multiply imputed data. 

Multinomial random effects logistic regression is discussed in (Haan &
Uhlendorf 2006). A nice point of Multiple Imputation is that it is not
method-specific, all it requires is that the sampling distribution of
the parameters is (approximately) Gaussian / normal. As long as this is
(approximately) true, you can use Multiple Imputation for your
multinomial random effects logistic regression. 

There are ofcourse other assumptions like the missing data needs to be
MAR, and if you don't know what that means you'll have to start reading
before you touch -ice- or -mim-. A good starting point is (Allison
2001).

-- Maarten

Allison, P. (2001) Missing Data, Thousand Oaks: Sage.

Haan, P. and Uhlendorf, A. (2006) Estimation of multinomial logit
models with unobserved heterogeneity using maximum simulated
likelihood. The Stata Journal, 6(2): 229--245.

http://www.stata-journal.com/article.html?article=st0104

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

visiting address:
Buitenveldertselaan 3 (Metropolitan), room Z434

+31 20 5986715

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


      __________________________________________________________
Not happy with your email address?.
Get the one you really want - millions of new email addresses available now at Yahoo! http://uk.docs.yahoo.com/ymail/new.html
*
*   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