The DHS Program User Forum
Discussions regarding The DHS Program data and results
Home » Data » Weighting data » How to use weights when using different samples of women
How to use weights when using different samples of women [message #23373] Mon, 30 August 2021 16:14 Go to next message
Jennie is currently offline  Jennie
Messages: 11
Registered: March 2021
Member
Dear all,

I use the DHS datasets for the Uganda survey year 2006 (dataset UGIR52DT) and 2011 (dataset UGIR61DT). I use a fixed effect model, and the treatment varies at the district level. My main outcome variables are women's outcomes such as employment, household decision-making, and domestic violence.

I have tried to find an answer here about how to use weights (pweight=v005). However, my concern is that I am using different samples:

- The variable employment includes all women; I think I should use (pweight=v005).
- The variable household decision-making only includes married women. So, in this case, I am not sure if I could use v005?
- Domestic violence includes a subgroup of women and women that are ever married. Do you know how I should use weights in this case?

Moreover, since my treatment varies at the district level, are there any problems with using weights? My outcome variables are at the individual level.

Kind regards,
Jennie
Re: How to use weights when using different samples of women [message #23378 is a reply to message #23373] Tue, 31 August 2021 08:20 Go to previous messageGo to next message
Shireen-DHS is currently offline  Shireen-DHS
Messages: 91
Registered: August 2020
Location: USA
Senior Member
Hello Jennie,

Since you have variables related to domestic violence in your analysis, you must use the domestic violence weight for all the analysis (d005). This is because only a specific sample of women are selected for the domestic violence questions and the weight takes care of this selection. I don't believe this is a problem with using this weight if you have district level variables.

You also must account for the survey sample design by also accounting for the primary sampling unit (v001) and the strata (usually v023 but for these specific surveys it's the grouping of v024 and v025). In Stata you do this using svy commands and svyset. In SPSS you would need to use a complex survey package that you would need to purchase. If you do not, then your sampling errors will be underestimated. Please refer to the complex survey design section in our Guide to DHS Statistics: https://www.dhsprogram.com/Data/Guide-to-DHS-Statistics/inde x.htm#t=Analyzing_DHS_Data.htm&rhsearch=svyset&rhhlt erm=svyset&rhsyns=%20

Best,
Shireen Assaf
The DHS Program

Re: How to use weights when using different samples of women [message #23386 is a reply to message #23378] Thu, 02 September 2021 04:37 Go to previous messageGo to next message
Jennie is currently offline  Jennie
Messages: 11
Registered: March 2021
Member
Dear Shireen,

Thank you very much for your reply and the information. Maybe this is already clear, but I have to make sure that I don't misunderstand this. Do you mean that I should use the domestic violence weight for all variables, even variables such as women's employment? For example:

reg employment X [pweight=d005]
reg household_decision_making X [pweight=d005]
reg domestic_violence X [pweight=d005]

or using separate weight for each

reg employment X [pweight=v005]
reg household_decision_making X [pweight=?]
reg domestic_violence X [pweight=d005]

In the latter case. If that is the correct way, what should I use as a weight for household decision-making, that only includes married women?

When I use svyset, I need to specify the weight. So then, I should use [pweight=d005], right?

Thanks for your help!

Kind regards,
Jennie
Re: How to use weights when using different samples of women [message #23388 is a reply to message #23386] Thu, 02 September 2021 08:53 Go to previous messageGo to next message
Shireen-DHS is currently offline  Shireen-DHS
Messages: 91
Registered: August 2020
Location: USA
Senior Member
Hello Jennie,

The use of the domestic violence weight is meant to adjust your estimates for the women in the domestic violence sample. I am unsure what your analysis involves but if you have a domestic violence variable in your analysis (regressions, crosstabulations, etc) then you need to use d005 and also you should divide this by 1 million , i.e.
gen dwt=d005/1000000

If you are just tabulating employment alone, are you trying describe this variable among all women or only for women in the domestic violence sample? If it's all women then you should use v005 if it's for women in the domestic violence sample then use d005. For household decision making you would use v005 if you want to describe it for all women eligible for these questions, the construction of the variable already takes care of the fact that it's among ever-married women, the denominator would differ from all women sample (i.e. you should have missing values for single women in the sample).

For me it sounds like if you are including these variables in your analysis, you should restrict your analysis to ever-married women so you have the same denominator and use the domestic violence weight also in the svyset. Again you need to divide any weight you use by 1 million.

If you are analyzing each of these separately, then you would use v005 for employment and household decision making and d005 for domestic violence. You will have three separate denominators in this case.

Hope this helps.

Best,
Shireen
Re: How to use weights when using different samples of women [message #23391 is a reply to message #23388] Fri, 03 September 2021 16:15 Go to previous message
Jennie is currently offline  Jennie
Messages: 11
Registered: March 2021
Member
Dear Shireen,

This information was helpful. Thank you!

Best,
Jennie
Previous Topic: Issues with weighting NEW Ethiopian DHS-2019
Next Topic: Weighting appended samples using different countries' KR data files.
Goto Forum:
  


Current Time: Mon Sep 27 03:13:06 Coordinated Universal Time 2021