The DHS Program User Forum
Discussions regarding The DHS Program data and results
Home » Countries » India » Weights for Couple Data NFHS3
Weights for Couple Data NFHS3 Fri, 12 February 2016 11:53
 Isha Messages: 18Registered: February 2016 Member
Hi. I am using the couple recode file from NFHS-3. I had two questions:
1. Should weights be used? Even if they are, will that make the sample representative? Or is it not required to use weights as the sample per se is not supposed to representative of the women and men? Can I do state-wise analysis from the couple data or does it really not say anything per of the state but just about this sample of couples? Is it okay to publish unweighted data mentioning that it doesn't represent the states/country; but then is it advisable to compare states?

2. In case couple weights are to be created, I was able to find only one paper explaining how this can be done from NFHS3 couple data (http://www.ncbi.nlm.nih.gov/pmc/articles/PMC4164230/). If this is the best option, then these probabilities to calculate the weights are available from the authors on request. Or else I could try to replicate it; though I am not a pro in statistical analysis. What do you think is the best thing to do? I am a first year PhD student in Development Sociology and I intend to write some publishable papers with my Advisor using couple data from NFHS-3. I am using Stata.
Re: Weights for Couple Data NFHS3 [message #9173 is a reply to message #9119] Thu, 18 February 2016 08:45
 Bridgette-DHS Messages: 3028Registered: February 2013 Senior Member
Following is a response from Senior DHS Stata Specialist, Tom Pullum:

It is standard practice at DHS to use the husband's weight (mv005) when analyzing the CR files. The husband's weight is preferred rather than the wife's weight because response rates are usually worse for men than for women, so the inclusion of the couple in the sample depends more on getting the husband than on getting the wife.

More complicated calculations of couple weights have been proposed, for example by Professor Stan Becker at the Johns Hopkins School of Public Health. Although theoretically superior, the effect of using these alternatives is small. I recommend just using the husband's weight, as I said.

If you include the national level weights, the national estimates will be unbiased. The state estimates will also be unbiased. The state estimates depend on fewer cases and therefore have larger standard errors and wider confidence intervals, but comparisons across states will be valid.

I do not recommend ever using unweighted estimates, except for data quality checks. Unweighted estimates are biased toward the over-sampled strata and away from the under-sampled strata. Statistical inferences should also include adjustments for clustering and stratification.
Re: Weights for Couple Data NFHS3 [message #9257 is a reply to message #9173] Sun, 28 February 2016 12:15
 Isha Messages: 18Registered: February 2016 Member
Dear Tom,

Thank you for your advice. I will use the men's weight for the couple recode data then. Yes, I saw Stan Becker's work but do not think I will be able to replicate it at this stage.

Thanks once again!
Isha
 Previous Topic: Village Survey: NFHS-2 (1998-99) for India Next Topic: Help with fertility preference variables v627, v628 and v629 and v613
Goto Forum:

Current Time: Sat Apr 13 09:49:18 Coordinated Universal Time 2024