LRTEST poisson vs negative binomial loop

[jose antonio]

Hello everyone, im testing over dispersion in a count variable. I'm using Hilbe's book "Negative binomial regression" for the analysis. He used the formula LR = −2(LP − LNB), but calculate it by hand and then used chiprob(1, result)/2. All my independent variables are presence or ausence of drug consumption, that means i have to adjust models separately.

local sus binoh bintab binmar bincoc binpas
foreach var of local sus {
poisson num_psex `var' sexo i.or_sexual i.tramo_etario edad_inicio nse
estimates store m1
nbreg num_psex `var' sexo i.or_sexual i.tramo_etario edad_inicio nse
estimates store m2
lrtest m1 m2
}

test involves different estimators: nbreg vs. poisson

this is what i got. And it make sense because im not using the same function. I need to store the log likelihood of my models and then use the formula, but i dont know how to make it efficient.

Please, i need some ideas. Thank you very much.

[jose antonio]

LP=Log likelihood poisson; LNB=Log likelihood negative binomial

[Andrew Musau]

You do not need to do the test by hand, just use the -force- option to override the error.

lrtest m1 m2, force

[jose antonio]

Thank you Andrew, that worked for me. I improved my code and this is what i got.

local sus binoh bintab binmar bincoc binpas
foreach var of local sus {
poisson num_psex `var' sexo i.or_sexual i.tramo_etario edad_inicio nse
scalar m1 = e(11)
nbreg num_psex `var' sexo i.or_sexual i.tramo_etario edad_inicio nse
scalar m2 = e(11)
lrtest m1 m2, force
}

11 invalid name

I got this error. it seems that i cant store e(11) in a loop, im missing something important.

[Andrew Musau]

You are now mixing the manual approach and my suggestion. m1 and m2 refer to models in the lrtest syntax, not scalars.

local sus binoh bintab binmar bincoc binpas
foreach var of local sus{
    poisson num_psex `var' sexo i.or_sexual i.tramo_etario edad_inicio nse
    estimates store m1
    nbreg num_psex `var' sexo i.or_sexual i.tramo_etario edad_inicio nse
    estimates store m2
   lrtest m1 m2, force
}

[jose antonio]

Thank you Andrew, you were right, i was mixing de approaches. Now it worked perfectly.