Yi-fu <[email protected]> made the following query about adjacent logits:
>I want to run a four-category multinomial logistic model and I find the
>comparison will be more interesting for me if I use adjacent logits.
>Does anyone know how to do multinomial logistic regression with adjacent
>logits?
Constantine <[email protected]> replied with the following:
> In an older version of Stata I did this through -mlogit- and using
> -constraint- for the constraints b_j = (J-j)*b. I think this is still what
> you need to do.
>
> Example: Y is the outcome (let's say it takes on values 0, 1, and 2) and X
> and Z are predictors.
>
> . constraint define 1 [2]x =2*[1]x
>
> . constraint define 2 [2]z =2*[1]z
>
> . mlogit y x z, basecategory(0) constraint(1 2)
In Stata 9 we now have -slogit- for fitting just such a model. For example
. webuse sysdsn3
(Health insurance data)
. slogit insure age male nonwhite site2 site3, dim(1) base(1) nolog
Stereotype logistic regression Number of obs = 615
Wald chi2(5) = 28.20
Log likelihood = -539.75205 Prob > chi2 = 0.0000
( 1) [phi1_2]_cons = 1
------------------------------------------------------------------------------
insure | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
age | .0108366 .0061918 1.75 0.080 -.0012992 .0229723
male | -.5032537 .2078171 -2.42 0.015 -.9105678 -.0959396
nonwhite | -.9480351 .2340604 -4.05 0.000 -1.406785 -.489285
site2 | -.2444316 .2246366 -1.09 0.277 -.6847113 .1958481
site3 | .556665 .2243799 2.48 0.013 .1168886 .9964415
-------------+----------------------------------------------------------------
/phi1_1 | 0 (base outcome)
/phi1_2 | 1 . . . . .
/phi1_3 | .0383539 .4079705 0.09 0.925 -.7612535 .8379613
-------------+----------------------------------------------------------------
/theta1 | 0 (base outcome)
/theta2 | .187542 .3303847 0.57 0.570 -.4600001 .835084
/theta3 | -1.860134 .2158898 -8.62 0.000 -2.28327 -1.436997
------------------------------------------------------------------------------
(insure=Indemnity is the base outcome)
The phi's are the constant multipliers for the regression coefficients and
the theta's are the constants. Here instead of fixing the mulitiplier to
2 we add it to the model as a parameter to be estimated.
I believe this is what Yu-fu is looking for.
-Rich
[email protected]
*
* 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/
