Help with Likelihood Ratio Test

[Shakib Ishfaq]

I keep encountering an error, due to the number of observations differing. I have ran the two regressions on the restricted and unrestricted model. My Professor in his words has said, "Change the sample so that you are running them on the same sample. If you test the restriction on the estimate from the unrestricted sample, it will do it automatically." My issue is, how do I do this? I have seen tutorials online and other examples, they follow the method I have been doing. So where am I going wrong?

I have uploaded my question sheet also, with the two different models.

Please any help is deeply appreciated. Coursework for Econometrics.pdf

[Bruce Weaver]

Hello Shakib. I have not looked at your PDF, but perhaps the following example will help. It sounds like you are having the same problem that arises for the second version of my reduced model.

clear
webuse lbw
tab smoke

* Generate some missing data for variable smoke so that the full
* model does not use all cases in the file.

replace smoke = . if inlist(_n,25,50,75,100,125,150,175)
tab smoke

logit low age lwt i.race smoke ptl ht ui // The full model
generate byte fullsample = e(sample) // Flag cases used for full model
estimates store full

logit low age lwt i.race if fullsample // Reduced model using full model sample
estimates store reduced1

logit low age lwt i.race // Reduced model using all available cases
estimates store reduced2

lrtest full reduced1, stats
lrtest full reduced2, stats

Here is the output from the two lrtest commands:

. lrtest full reduced1, stats

Likelihood-ratio test                                 LR chi2(4)  =     21.79
(Assumption: reduced1 nested in full)                 Prob > chi2 =    0.0002

Akaike's information criterion and Bayesian information criterion

-----------------------------------------------------------------------------
       Model |        Obs  ll(null)  ll(model)      df         AIC        BIC
-------------+---------------------------------------------------------------
    reduced1 |        182 -113.1361  -107.3981       5    224.7962   240.8162
        full |        182 -113.1361  -96.50243       9    211.0049   239.8409
-----------------------------------------------------------------------------
               Note: N=Obs used in calculating BIC; see [R] BIC note.

. lrtest full reduced2, stats
observations differ: 182 vs. 189
r(498);

[Shakib Ishfaq]

Thanks for the reply Bruce, I will try something similar to this and see my results.

I have a question however. My models include 1 period lags and differences, could this be the reason?

[Shakib Ishfaq]

Now I am getting the following:

logit rl lr1 drl r lr0 dr0

outcome does not vary; remember:
0 = negative outcome,
all other nonmissing values = positive outcome
r(2000);

end of do-file

r(2000);

[Carlo Lazzaro]

Shakib:
if you do not have any variation in your dependent variable, -logit- gasps.