I am investigating NPSV over the last 12 months in NFHS-4 using variables d124 and d117a. However, given that these questions were only asked to people who replied "yes" to questions d125 and d115y, the universes for these variables is much smaller.

Has anyone run analyses with these variables (d124 and d117a), and if so did you choose to code the missing observations not in the universe as "no" responses (and is that a valid method of statistical analysis), or exclude them from your analysis entirely? If they are excluded from the analysis, then there are only about 800 observations (for d124) and 6000 observations (for d117a), which is much smaller than the 79,000 women who were included in the domestic violence module.

I would appreciate insight into what other people have done to study these variables! ]]>

To get the correct denominator you have to go back to the beginning of the filter or skip. It is not necessarily the full sample--usually this is a module for a subsample. Then you can construct a new variable that has, say, zero rather than NA for the cases in the correct denominator who were not asked the questions that lead to d124 or d117a. It may help to look at the questionnaire to figure out the sequence.

]]>

In this context, the denominator is the population that is asked the question. You could construct a composite variable that included all women who were asked d125. The following would include all the possible combinations of d124 and d125. For the new variable, codes 0 and 6 are the same as for d125. d124 is only applicable if d125=1. The three possible responses to d124 are given codes 1, 2, 3 in the composite variable.

* d124 is only asked if d125 is yes (d125=1)

* Construct a composite variable, call it d124_d125

gen d124_d125=.

replace d125_d124=0 if d125==0

replace d125_d125=1 if d125==1 & d124==0

replace d125_d125=2 if d125==1 & d124==1

replace d125_d125=3 if d125==1 & d124==6

replace d125_d124=6 if d125==6

]]>

And how would you apply the DV weights to this to make the variables usable in a regression of d124 as the Y variable?]]>

The regression would include [pweight=d005]. You can get specifics with "help weight".

]]>

I have used both of the following STATA codes, and am not sure which is correct. Also, how would I take into account the strata in the code for the following regressions?

reg d124 hv205 [pw=d005], cluster(idhspsu)

reg d124 hv205 [pw=d005], r

]]>

You have to specify svyset and then put "svy:" in front of estimation commands. Many examples have been posted on the forum. This should work:

svyset IDHSPSU [pw=d005], strata(IDHSSTRATA) singleunit(centered)

svy: reg d124 hv205

]]>