Re: st: Svy: ordinal regression assumptions

 From Richard Williams <[email protected]> To [email protected], [email protected] Subject Re: st: Svy: ordinal regression assumptions Date Thu, 29 May 2008 15:36:18 -0500

```At 12:34 PM 5/29/2008, Simon, Alan (CDC/CCHIS/NCHS) wrote:
```
```Hi all,

I have a complex sample design survey (svy), and would like to use an
ordinal regression.  Does anyone know how to test the assumption of
proportional odds that is made with ordinal regression if I also have a
complex sample design.  Without complex sample design, it seems easy
enough to use the brant test, but this doesn't seem to work after
ordinal regression in svy.  Any suggestions?
```
There are various ways to do this with gologit2, which can be downloaded from SSC. gologit2 also lets you estimate partial proportional odds models, which can produce more parsimonious models. Example:

. webuse nhanes2f

. gologit2 health female black, svy

Generalized Ordered Logit Estimates

Number of strata = 31 Number of obs = 10335
Number of PSUs = 62 Population size = 116997257
Design df = 31
F( 8, 24) = 27.06
Prob > F = 0.0000

------------------------------------------------------------------------------
| Linearized
health | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
poor |
female | .1451402 .098885 1.47 0.152 -.056537 .3468174
black | -.9556602 .1435276 -6.66 0.000 -1.248387 -.6629337
_cons | 3.06297 .114295 26.80 0.000 2.829864 3.296076
-------------+----------------------------------------------------------------
fair |
female | -.1731967 .0663503 -2.61 0.014 -.3085191 -.0378743
black | -.8136862 .0911644 -8.93 0.000 -.9996173 -.6277551
_cons | 1.773817 .0700574 25.32 0.000 1.630934 1.9167
-------------+----------------------------------------------------------------
average |
female | -.1726385 .0576992 -2.99 0.005 -.2903167 -.0549602
black | -.8834612 .0709884 -12.45 0.000 -1.028243 -.7386795
_cons | .3752842 .0436688 8.59 0.000 .2862212 .4643472
-------------+----------------------------------------------------------------
good |
female | -.2333665 .0614472 -3.80 0.001 -.3586889 -.108044
black | -.7486971 .1310825 -5.71 0.000 -1.016042 -.4813526
_cons | -.7903677 .042234 -18.71 0.000 -.8765045 -.704231
------------------------------------------------------------------------------

. test [#1 = #2 = #3 =#4]

( 1) [poor]female - [fair]female = 0
( 2) [poor]black - [fair]black = 0
( 3) [poor]female - [average]female = 0
( 4) [poor]black - [average]black = 0
( 5) [poor]female - [good]female = 0
( 6) [poor]black - [good]black = 0

F( 6, 26) = 6.65
Prob > F = 0.0002

. test [#1 = #2 = #3 =#4]:female

( 1) [poor]female - [fair]female = 0
( 2) [poor]female - [average]female = 0
( 3) [poor]female - [good]female = 0

F( 3, 29) = 7.19
Prob > F = 0.0009

. test [#1 = #2 = #3 =#4]:black

( 1) [poor]black - [fair]black = 0
( 2) [poor]black - [average]black = 0
( 3) [poor]black - [good]black = 0

F( 3, 29) = 0.89
Prob > F = 0.4579

The proportional odds assumption is violated in this case (but only for the female variable, not black).

For more, see the gologit2 support page at

http://www.nd.edu/~rwilliam/gologit2/index.html

-------------------------------------------
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/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/