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

st: Multicollinearity tests for multinomial logit models

From   "Anderson, William D" <>
To   <>
Subject   st: Multicollinearity tests for multinomial logit models
Date   Fri, 3 Aug 2007 11:17:11 -0500


I am running a multinomial logit model and am concerned about
multicollinearity among two of my covariates. While there's limited
information out there about how to handle this problem, I've attempted
to use the Hendrickx -perturb- ado to test for multicollinearity.
Unfortunately, I've run into an error. I'm wondering if the user group
can give me some guidance about potential fixes or other approaches.

My command line is:

perturb: mlogit dependent daytermpct2 preswlra year1 prespop presstaff
fornvote ideoldiff divgov timsempct2 gdpchan2 unempch,
cluster(prescluster) baseout(1), if year<1997,
poptions(pvars(daytermpct2 timsempct2) prange(0.68 0.825))

When I run that command line, Stata generates ML estimates and begins
what appears to be the iteration process to estimate the perturbation
effects. I believe this because the following output (following the MNL
output) is generated:

Perturb variables:
daytermpct2                       normal(0,.68)
timsempct2                        normal(0,.825)

There is a significant pause and then the following appears:

variable daytermpct2 not found
variable timsempct2 not found

This persists even after removing the -cluster- and other conditional
commands (if year<1997, for example).

Does anyone have have suggestions about why this might be the case and
how I might use -perturb- in light of this problem?

Or, even better, does anyone have suggestions about how to test for
multicollinearity in MNL models that might be more appropriate than the
Hendrickx -perturb- ado?

Regards, Bill Anderson

William D. Anderson, Ph.D.
Director, USD Government Research Bureau
Assistant Professor, Department of Political Science
414 East Clark Street
Vermillion, South Dakota  57069
(605) 677-5708

*   For searches and help try:

© Copyright 1996–2017 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index