The DHS Program User Forum
Discussions regarding The DHS Program data and results
Home » Countries » India » Norms variables [[URGENT]] (district level estimates of varibles related to norms)
Norms variables [[URGENT]] Fri, 24 March 2023 00:35
 niruj Messages: 2Registered: March 2023 Member
Hi, I am trying to get district level estimates/proportions of variables related to norms, domestic violence etc. NFHS-4 and NFHS-5 mentions that these variables are not representative at the district level. Is there any way around this problem ?

---------------------

Subject: Re: Norms variables [[URGENT]] Author: niruj Date: Mon, 27 March 2023 05:22
I tried to get district level proportion for the variable s930 (has a mobile phone that respondent uses). I used the svyset and svy command. Results show the mean by district, but standard errors and confidence interval are missing. Stata produces a note: [Note: Missing standard errors because of stratum with single sampling unit.]

The code I used. (tried both ways of generating stratum.)

* generate stratum
gen stratum = v023

* alternative strata based on region and urban/rural
* egen stratum = group(v024 v025)

gen wt = v005/1000000
svyset v021 [pw=wt], strata(stratum)
svy: mean s930, over(sdistri)

[Updated on: Mon, 27 March 2023 08:36] by Moderator

Report message to a moderator

Re: Norms variables [[URGENT]] [message #26482 is a reply to message #26470] Fri, 24 March 2023 08:38
 Bridgette-DHS Messages: 2809Registered: February 2013 Senior Member

Following is a response from Senior DHS staff member, Tom Pullum:

As stated on page 2 of the final report on the NFHS-5, the full sample was designed to be representative at the district level. This means that for variables included in the full sample, such as number of children, the district-level estimates are unbiases and have standard errors that are below some desirable threshold.

For variables in a subsample, such as the DV variables, the estimates continue to be unbiased (because the subsampling is random) but the standard errors are larger (because of the smaller number of cases). In other words, the caution about "representativeness" relates mainly to uncertainty in the estimate rather than potential bias. There is a possibility that you will incorrectly identify some districts as extreme when it's just that they have few cases.

You can proceed and then calculate (using svyset and svy) the confidence intervals for your district-level estimates. You will see that the confidence intervals are wide, but it's really a judgment call as to whether they are "too" wide. You may be able to group adjacent districts together. Mainly you have to be careful about potentially misdiagnosing specific districts as being very high or very low on your indicators.

Re: Norms variables [[URGENT]] [message #26493 is a reply to message #26482] Sat, 25 March 2023 03:04
 niruj Messages: 2Registered: March 2023 Member