The DHS Program User Forum      
Discussions regarding The DHS Program data and results
Home » Data » Weighting data » Weighting data for descriptive analysis of background characteristics
Weighting data for descriptive analysis of background characteristics [message #11939] Thu, 09 March 2017 15:47 Go to next message
chichi is currently offline  chichi
Messages: 9
Registered: March 2017
Member
Hello, I am starting to work with DHS data and stata for the first time. I often read that DHS data must be weighted for every analyses. I am using the Namibian DHS 2013. I appended the men's file to the women's file and merged then this combined women's/men's file with the HIV file. At the beginning of my analysis I want to make some descriptive analyses. Now my question is, whether I have to weight the data for descriptive analyses of respondents background characteristics?
Re: Weighting data for descriptive analysis of background characteristics [message #11941 is a reply to message #11939] Fri, 10 March 2017 10:05 Go to previous messageGo to next message
Bridgette-DHS is currently offline  Bridgette-DHS
Messages: 1073
Registered: February 2013
Senior Member
Following is a response from Senior DHS Stata Specialist, Tom Pullum:

If you have correctly merged the files, you will only be working with cases for which hiv03 (the test result) is not coded ".", and for those cases the correct weight to use is hiv05.

When you are checking recodes, checking data quality, doing some initial exploration of the data, you can omit the weights. Otherwise, we recommend that you use weights (in this instance, hiv05) for descriptive purposes, as well as for statistical models, because then your results will be representative of the household population.

Re: Weighting data for descriptive analysis of background characteristics [message #11982 is a reply to message #11941] Wed, 15 March 2017 12:50 Go to previous messageGo to next message
chichi is currently offline  chichi
Messages: 9
Registered: March 2017
Member
Thank you very much for your response!
Re: Weighting data for descriptive analysis of background characteristics [message #12509 is a reply to message #11939] Wed, 31 May 2017 12:13 Go to previous messageGo to next message
atikaukh is currently offline  atikaukh
Messages: 2
Registered: January 2017
Location: Jakarta
Member

I want to analyze about determinants for unmet need spacing in Indonesia using Indonesia DHS 2012. If I use binary logistic regression, should I use weighted data? Is there any difference in interpretation the result if I do use unweighted data?

Please give me some advices , which one is better: use a weighted data, or unweighted data. thank you :)
Re: Weighting data for descriptive analysis of background characteristics [message #12510 is a reply to message #12509] Wed, 31 May 2017 12:25 Go to previous messageGo to next message
Bridgette-DHS is currently offline  Bridgette-DHS
Messages: 1073
Registered: February 2013
Senior Member
Following is a response from Senior DHS Stata Specialist, Tom Pullum:

Please look at earlier postings on this subject. You should always use the weights. In your case, use "svyset", including "[pweight=v005]", and then "svy: logit...." If you do not include the weights, then your estimates will be biased toward the strata that were over-sampled and away from the strata that were under-sampled.

Re: Weighting data for descriptive analysis of background characteristics [message #12512 is a reply to message #12510] Wed, 31 May 2017 13:26 Go to previous message
atikaukh is currently offline  atikaukh
Messages: 2
Registered: January 2017
Location: Jakarta
Member

thank you for the explanation:)

but unfortunately now I am using SPSS. I have a problem with their output when I run my binary logistic regression with weighted data and then the output still unweighted data.


I am running a logistic regression in SPSS with a sample that uses survey weights. The sample size is 1000 and the weights are along the lines of .86 or 1.23 depending on the case. I am using the weights option in the Data command to weight my sample before I run my logistic regression but then I get the unweighted frequencies in the Categorical Variables Coding part of the output.

At the beginning my output gives me a Case Processing Summary, which says Unweighted cases with a superscript "a" which then says "If weight is in effect see classification table for the total number of cases." My classification tables all are properly saying 1000 cases which is right. I have 7 predictor variables (all categorical) and one outcome variable (dichotomous). My output seems to make sense, but I am troubled that the frequencies are all the unweighted frequencies. Are my results still legitimate? I am afraid that I may not understand the weighting quite properly.

If I use stata, is that problem can be solved?
Previous Topic: Weights for analysis of HIV+/HIV- urban and rural populations
Next Topic: Weighting the husband/partner
Goto Forum:
  


Current Time: Fri Jul 28 15:01:19 Eastern Daylight Time 2017