Help: "maxvar too small"

[Abhishek Patil]

I am trying to run a regression that accounts for sample selection bias with the use Heckman, twostep. I want to add a series of dummy variables (i.HHID) to the model to account for the households (household fixed-effects). The problem I am facing is that I have 3,495 unique households. I am using Stata/IC 16.1 for Mac (Intel 64-bit), which has a maxvar limit of 2048 which is not "settable in this version of Stata". Is there any way I can work around this problem. I have tried to use the "absorb" function in my regression command.

I am running the following:

   heckman ln_WKHOURLY potential_experience potential_experience2 i.EDUC7 i.ED2 i.ED3 i.RO3 i.Primary_Activity_cat i.HHID, select(potential_experience potential_experience2 i.EDUC7 i.ED2 i.ED3 i.RO3 i.Primary_Activity_cat i.HHID NPERSONS NCHILD NELDER) twostep

The message returned by Stata: "maxvar too small"

[FernandoRios]

You will need an alternative methodology for that.
Do you really need hhid as an explanatory variable? Couldn't you cluster by (hhid) instead?
Alternatively, is there enough intra hhid variation to estimate the model (how many households have more than 1 working member)
Perhaps you may need to readdress if the method you are using, is the most appropriate for your research question.
Fernando

[li qizheng]

have you solved this problem~