In IR files, v001 is the cluster number; v002 is the household number (within v001) and v003 is the woman's line number (within v001 and v002). Together, v001 and v002 identify the household; v001 and v002 and v003 identify the individual woman. In IAIR42FL.dta, caseid is a character string that includes v001, v002, and v003. To get a household id, you can remove the last three columns (which give v003) from caseid. That is, use this line in Stata:

gen hhid=substr(caseid,1,12)

In that example, the household has two women, who are on lines 4 and 7. The hhid is a combination of cluster id (hv001 or v001) and household id (hv002 or v002). (The line number is hvidx or v003.)

Most households will have one eligible respondent--if they have any at all. Only a few will have more than one. You will use up all your degrees of freedom if you have fixed effects for household. I would not even try random effects for households. The density of women per household is too low. Fixed or random effects for cluster would be as far down into the data as I would go.

You would definitely use "xi." for fixed effects--no need to construct dummy variables (fortunately!). For random effects, use an "me" model.

However, we really cannot advise on your choice of model, just on data-related issues and to a limited degree on Stata syntax.