Estimation with plausible values - PISA data

[Charlotte Rogers]

Hello, I am using PISA data and I want to regress reading test scores on a dummy variable 'year2006' only for the subpopulation of females. For the first plausible value, the command I use is:

svy brr, subpop (female): reg PV1READ year2006

However, PISA provides 5 plausible values and I would like to use them all in my estimation. I have been trying to use pv in Stata and the command:

pv, pv(PV1READ PV2READ PV3READ PV4READ PV5READ) weight(W_FSTUWT) brr rw(W_FSTR*) fays(0.5) subpop(female): reg @pv year2006 [aweight=@w]

The problem is that Stata does not allow me to use the subpop() option. Any idea how can I fix this?

[daniel klein]

pv is user-written (probably from SSC) as you are asked to exaplain. As its documentation shows no subpop() option, it should not be surprising that it does not accept such an option.

Unless I am missing something, a simple if qualifier with the regress command should suffice to restrict the sample to females.

I hope this helps.

Best
Daniel

[Charlotte Rogers]

Thank you Daniel. The "if" qualifier does not work in this case because the randomness in the domain (female) sample size should be reflected in the variance estimate. Therefore, I will obtain incorrect SEs if I perform the regression only for those cases in the population that fulfill female==1 separatedly

[daniel klein]

In that case try combining the pv command with the svy prefix (as also demonstrated in the help file). I see no apparent reason why something along the lines

pv ,pv(pvvarlist) ... : svy ,subpop(female) : reg @pv ...

should not work. But I might be wrong, as I have never used the pv command.

Best
Daniel

[Philip Matthews]

Charlotte, in case you have not found it, there is a package "pisatools" that contains a number of commands for analysing PISA data (use findit pisatools to chase it down). The one named "pisareg" may be of use to you, but I am not sufficently familiar with it to be sure.
Best wishes,
Philip

[wbuchanan]

The five scores are plausible values that you can register as imputed values. Then just use svyset to set the correct weights for your estimations and you should have significantly more flexibility.