The DHS Program User Forum: Weighting data » Unable to recreate numbers from DHS Final Survey

Home » Data » Weighting data » Unable to recreate numbers from DHS Final Survey

Show: Today's Messages :: Show Polls :: Message Navigator

Unable to recreate numbers from DHS Final Survey [message #5598]

Sun, 14 June 2015 11:38

duke2015
Messages: 27
Registered: March 2015
Location: United States

Member

Hi there,

I am interested in recreating the table on disposal of child stools from Uganda 2011 DHS final report. The variable is v465 from the children's recode and is asked about the youngest child in the household. First, I used this code for the survey adjustments:

egen strata=group(v024 v025)
gen sampwt=v005/1000000
svyset v021 [pweight=sampwt], strata(strata) singleunit(centered)

When I run svy: tab v465, my proportion estimates are a bit different from what is stated in the DHS final report. Do you have any explanation for this discrepancy?

Many thanks!

Report message to a moderator

Re: Unable to recreate numbers from DHS Final Survey [message #5623 is a reply to message #5598]

Wed, 17 June 2015 14:15

Bridgette-DHS
Messages: 3214
Registered: February 2013

Senior Member

Following is a response from Senior DHS Stata Specialist, Tom Pullum:

I think you are referring to table 10.11 on page 138 of the report on the Uganda 2011 survey. It's interesting that the label for v465 in the KR file is "disposal of youngest child's stools when not using toilet", but the question was not just asked about the youngest child! You have to specify that hidx is 1. If you enter "tab v465 if hidx==1 [iweight=v005/1000000]" you will get the distribution you are looking for. Note also that I am not using svy. For a table, the cluster and stratum adjustments have no effect, and iweight will work fine.

Report message to a moderator

Re: Unable to recreate numbers from DHS Final Survey [message #6695 is a reply to message #5623]

Mon, 29 June 2015 10:32

duke2015
Messages: 27
Registered: March 2015
Location: United States

Member

Thank you!

If I am looking to run a regression predicting child HAZ from this variable, will this coding in Stata 13 work for the svy adjustment?:

egen strata=group(v024 v025)
gen sampwt=v005/1000000

svyset v021 [pweight=sampwt], strata(strata) singleunit(centered)

Thanks in advance!

Report message to a moderator

Re: Unable to recreate numbers from DHS Final Survey [message #6734 is a reply to message #6695]

Mon, 06 July 2015 08:55

Bridgette-DHS
Messages: 3214
Registered: February 2013

Senior Member

Following is a response from Senior DHS Stata Specialist, Tom Pullum:

You did not say what survey you are using. There are a few surveys for which v024 x v025 will not correctly give the strata. In general, however, this will be correct.

It is not necessary to divide v005 by 1000000. Stata always normalizes pweights to have a mean of 1. The following would be equivalent:

egen strata=group(v024 v025)

svyset v021 [pweight=v005], strata(strata) singleunit(centered)

Report message to a moderator

Re: Unable to recreate numbers from DHS Final Survey [message #8482 is a reply to message #6695]

Mon, 02 November 2015 15:12

duke2015
Messages: 27
Registered: March 2015
Location: United States

Member

I am using the Uganda 2011 DHS survey. Please confirm if the egen strata=group (v024 v025) coding in Stata 13 will give the correct strata.

Thank you!!

Report message to a moderator

Re: Unable to recreate numbers from DHS Final Survey [message #8490 is a reply to message #8482]

Tue, 03 November 2015 12:36

luke
Messages: 4
Registered: November 2015
Location: Baltimore MD

Member

Duke2015 - I posted about this same issue yesterday and the information may help you. I don't think that using residence (urban/rural - v025) is appropriate for stratification for the 2011 Uganda DHS, as you suggest. This survey drew a random sample of clusters from the 2010 Uganda National Health Survey, which was stratified according to region only.

For the time being, I am using region (v024) as the strata for Uganda 2011. I'm still not sure this is the correct approach. Is a random sample of clusters of a stratified sample also stratified in the same way as the original sample?

Here is my post from yesterday:

http://userforum.dhsprogram.com/index.php?t=tree&th=4622 &goto=8481&S=39796439d8b4cca149d1e3c9b2778a7f#msg_84 81

Report message to a moderator

Re: Unable to recreate numbers from DHS Final Survey [message #8491 is a reply to message #8490]

Tue, 03 November 2015 13:03

duke2015
Messages: 27
Registered: March 2015
Location: United States

Member

Thanks, Luke.

I really don't know the right approach. So you are using region as the strata and not using rural/urban at all? Can you show me your coding for the svyset command?

Thanks!

Report message to a moderator

Re: Unable to recreate numbers from DHS Final Survey [message #8492 is a reply to message #8491]

Tue, 03 November 2015 13:19

luke
Messages: 4
Registered: November 2015
Location: Baltimore MD

Member

Sure. This works for most countries I am using then you will see a way to handle the exception for Uganda:

gen psu = hv021
gen strata = hv022
replace strata = hv024 if (hv000=="UG6" ) // Uganda 6 has no strata. Sampled clusters from UNHS 2010 clusters, which were stratified according to region.
gen sampwt = hv005/1000000
svyset psu [pw = sampwt], strata(strata)

BTW I incorrectly wrote UNHS as Uganda National Health Survey in my last message. The H is for Household.

Report message to a moderator

Previous Topic:	Setting svyset for KDHS 2008
Next Topic:	how the sample increase when we use weights

Goto Forum:

-=] Back to Top [=-

[ Syndicate this forum (XML) ] [

] [

]

Current Time: Sun Dec 22 07:54:40 Coordinated Universal Time 2024