The DHS Program User Forum
Discussions regarding The DHS Program data and results
Home » Topics » Unmet Need » Unmet need groups
Unmet need groups [message #1355] Mon, 17 February 2014 09:18 Go to next message
smgwu is currently offline  smgwu
Messages: 4
Registered: October 2013
Location: Washington, D.C.
I would like some advise on this:
If I were to run regression analyses for unmet need, would it be best to do logistic regression, making the dependent variable binary by excluding all women who are not at risk for pregnancy (i.e. unmarried or not living with a partner, not sexually active, infecund, etc) or to run multivariate logistic regression with the dependent variable having three outcomes - unmet need, met need and no met need (including the women who are not at risk for pregnancy or who do not have unmet need)?

Re: Unmet need groups [message #1361 is a reply to message #1355] Mon, 17 February 2014 18:58 Go to previous message
Messages: 292
Registered: March 2013
Senior Member

It might be helpful if you could give some more detail - exactly what are you trying to understand using your regression analysis?

I would say that the main issue when deciding about whether to run your regression on the subsample of women who "need" v. all women is who your population of interest is. If you are just interested in whether or not married and child-bearing-age women have access/met-need, then use only them in the analysis. If your question is more like "what is the fraction of all women who have unmet need", then include everyone. But in general, based on this really short description, it sounds like you are only interested in the "at risk for unmet need" group, which would point towards using the subsample. On the other hand - unmarried women may very well be at risk, so...

One other thing to think about. The two regressions could yield the same results if you specified them so that all covariates had a different coefficient for at-risk/no-risk women respectively. By including the not-at-risk women and not allowing for different coefficients, you are essentially including a constraint that those women have the same determinants of unmet need as at-risk women. Without more information, I have no idea if that makes sense, but it is one way to think about the difference.

Happy to follow-up - unmet need isn't my main area of interest, but econometric specification is something I work a lot on.
Previous Topic: Help with Unique Identifier Kenya 2010 SPA
Next Topic: Data not matching
Goto Forum:

Current Time: Wed Jul 24 18:44:14 Coordinated Universal Time 2024