The thing is that the guideline suggests using d005 for weight for domestic violence whereas for all the other variables, I need to use v005.

If that is the case, how do I need to use the weight?

Do I have to use either one of them or create a new weight variable somehow combining those two? (If so, could you let me know how?)

Thank you so much for your support!

]]>

If you are doing a table or regression that includes a DV variable, in any role--either as an outcome or as a predictor, you should use d005. Otherwise, use v005. Are you asking about how to use weights in a logistic (or logit) regression? That depends on what statistical package you are using. Let us know if your question is about weighting in general.

]]>

For descriptive statistics including DV variables, could I also use d005 for weights if DV variables are part of the table?

(In sum, only when variables starting with v are used, I could use v005 for weight, and if DV variables are used along with other variables (starting with v), I need to use d005. Is that right?)

Thank you so much!

]]>

I would describe the underlying principle this way. Think of the women who responded to the DV questions as a subsample. There's some extra nonresponse to these questions--some women who agree to the larger interview will decline to participate in the DV questions, and the refusals vary somewhat from one stratum to another. For that reason, the weights are a little different for the women who do respond to the DV questions than for the full sample of women. So you would use the DV weights for cross tabulations, regressions, correlations, etc., that includes a DV variable. However, if you want to include a descriptive table that gives, for example, the age distribution or education distribution, it would use v005, not d005 for the weights. The distribution of those background variables, on their own, is not affected by nonresponse to the DV module.

]]>

That makes a lot of sense now. Thank you so much!

Just one more follow up question.

My analysis would involve participants with complete data (without missing values) and one of the key variables is domestic violence. If that is the case, could I simply have the participants with the complete data without any missing values in outcome variables or any other covariates and use d005 for the weight?

Thank you so much for your support. It has been really helpful!

]]>

I don't quite understand the question. I would not modify what I said before. For the women who are not asked the DV module, d005 is NA ("." in Stata) and for those who do respond, d005 will be different from v005. ]]>