Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: st: Brant's Test for Parallel Lines
From
Richard Williams <[email protected]>
To
[email protected], <[email protected]>
Subject
Re: st: Brant's Test for Parallel Lines
Date
Fri, 14 Oct 2011 17:03:42 -0400
At 04:36 PM 10/14/2011, Stephen Clark wrote:
Hello,
I have an ordered logit model that uses a large amount of data, 2.5 million
cases. Brant's test for parallel lines is significant (p>chi2 = 0.00) which
suggest that the model I am using is inappropriate. However, I have found
various references that suggest the test is pessimistic with very large
samples - highlighting small differences that are not actually critical. I
cannot get access to Brant original paper so I would welcome some advice as
to whether this is likely to be the case here. Is there a better test
available for large samples?
Probably the same can be said of most statistical tests -- if you
have 2.5 million cases, even substantively trivial deviations from
the null will be statistically significant.
You might consider using a BIC test instead, which is possible using
the -gologit2- program available from SSC. Do something like
use "http://www.stata-press.com/data/r12/nhanes2f";, clear
gologit2 health female black rural, pl sto(ologit)
gologit2 health female black rural, npl sto(gologit)
lrtest ologit gologit, stats
The last command yields
Likelihood-ratio test LR chi2(9) = 56.11
(Assumption: ologit nested in gologit) Prob > chi2 = 0.0000
-----------------------------------------------------------------------------
Model | Obs ll(null) ll(model) df AIC BIC
-------------+---------------------------------------------------------------
ologit | 10335 -15764.4 -15614.23 7 31242.46 31293.16
gologit | 10335 -15764.4 -15586.17 16 31204.35 31320.24
-----------------------------------------------------------------------------
Note: N=Obs used in calculating BIC; see [R] BIC note
Based on a brant test (or the very similar chi-square test reported
above) you would conclude the assumptions are violated. But based on
the BIC test, you would stick with the ordered logit model.
The -gamma- option on gologit2 also lets you see how big the
deviations from proportionality are, which you might look at as a way
of assessing how substantively important they are.
-------------------------------------------
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/help.cgi?search
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/