Since ologit does not produce an F-statistic I'm not sure you will be able
to calculate its change. However, there are other measures of fit. J.
Scott Long and Jeremy Freese have written up fitstat that maybe of use to
you (type: findit fitstat, to download).
Example:
use "C:\Stata\auto.dta", clear
(1978 Automobile Data)
. quietly ologit rep for mpg [w = length]
. quietly fitstat, saving(mod1)
. quietly ologit rep for mpg price gear_ratio [w = length]
. fitstat, using(mod1)
Measures of Fit for ologit of rep78
Current Saved Difference
Model: ologit ologit
N: 12992 12992 0
Log-Lik Intercept Only: -17515.087 -17515.087 0.000
Log-Lik Full Model: -14431.260 -14678.976 247.716
D: 28862.520(12984) 29357.952(12986) 495.432(2)
LR: 6167.654(4) 5672.221(2) 495.432(2)
Prob > LR: 0.000 0.000 0.000
McFadden's R2: 0.176 0.162 0.014
McFadden's Adj R2: 0.176 0.162 0.014
Maximum Likelihood R2: 0.378 0.354 0.024
Cragg & Uhler's R2: 0.405 0.379 0.026
McKelvey and Zavoina's R2: 0.407 0.375 0.033
Variance of y*: 5.551 5.261 0.290
Variance of error: 3.290 3.290 0.000
Count R2: 0.566 0.528 0.038
Adj Count R2: 0.213 0.145 0.068
AIC: 2.223 2.261 -0.038
AIC*n: 28878.520 29369.952 -491.432
BIC: -94123.084 -93646.596 -476.488
BIC': -6129.765 -5653.277 -476.488
Difference of 476.488 in BIC' provides very strong support for current
model.
Note: p-value for difference in LR is only valid if models are nested.
Scott
----- Original Message -----
From: "Jennifer van Stelle" <[email protected]>
To: <[email protected]>
Sent: Tuesday, July 09, 2002 9:52 PM
Subject: st: How can I compute a change in F-statistic using weighted data
and "ologit"?
> Hi -
>
> I'm a relatively new Stata user, and have never posted to this list
before.
> I've checked the FAQ, however, and don't see an answer to my problem, so I
> thought I'd post it here and hope for a response.
>
> I'm using Stata 7.0 for UNIX, and am doing ordered logit models with
survey
> data, using probability weights and primary sampling unit clustering (data
> is for individuals in 28 different countries). Because I'm doing several
> different models (i.e., Model One with just the control variables, then
> Model Two with the controls plus a subset of independent variables, and
> finally Model Three with the controls and all the independent variables),
> I'd like to be able to tell whether each model is a significant
improvement
> on the previous one. I've been warned about what I should NOT to do (that
> is, don't simply subtract the F-statistic of Model One from Model Two, or
> Model Two from Model Three), but no one has been able to explain to me
what
> I SHOULD do in order to calculate change in F in this particular
situation.
> Can anyone tell me how to do this?
>
> Any help will be gratefully accepted!
>
> Thanks.
>
> - Jen
*
* 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/
