The DHS Program User Forum
Discussions regarding The DHS Program data and results
Home » Data » Dataset use (other programs) » How to calculate the prevalence of an outcome by SAS for ever married women surveys?
How to calculate the prevalence of an outcome by SAS for ever married women surveys? Fri, 18 December 2020 09:51
 Han Messages: 4Registered: December 2020 Member
I am doing an analysis that looks at smoking prevalence of women from ever married women surveys.

For a "all-women" survey, we can calculate the overall smoking prevalence of women using PROC SURVEYFREQ to in SAS:

weight --> (V005 Women's individual sample weight (6 decimals) / 1,000,000)
cluster --> V021 Primary sampling unit
strata --> V023 Stratification used in sample design
y --> V463a Smokes cigarettes

proc surveyfreq data=a;
strata strata;
cluster cluster;
weight weight;
tables y;
run;

However, if I would like to calculate the overall smoking prevalence of women from an ever married women survey, then what should I do? How should I use the awfactt variable?

Thank you!
Re: How to calculate the prevalence of an outcome by SAS for ever married women surveys? [message #21846 is a reply to message #21781] Tue, 29 December 2020 19:50
 Bridgette-DHS Messages: 2804Registered: February 2013 Senior Member

Following is a response from DHS Research & Data Analysis Director, Tom Pullum:

We do not use SAS and cannot give advice that is specific to SAS. Your larger question is how to calculate the prevalence of an outcome in an EMW survey. What outcome?

Re: How to calculate the prevalence of an outcome by SAS for ever married women surveys? [message #21848 is a reply to message #21846] Tue, 29 December 2020 20:19
 Han Messages: 4Registered: December 2020 Member
Dear Tom Pullum,

The outcome is cigarettes smoking (V463a) prevalence of all women in an EMW survey.
Re: How to calculate the prevalence of an outcome by SAS for ever married women surveys? [message #21852 is a reply to message #21848] Wed, 30 December 2020 19:27
 Bridgette-DHS Messages: 2804Registered: February 2013 Senior Member

Following is a response from DHS Research & Data Analysis Director, Tom Pullum:

For an outcome like smoking, I would just replace v005 with v005*awfactt/100. If you want the breakdown by urban/rural, you use awfactu. By region, awfactr. By education, awfacte. By wealth quintile, awfactw.

This recommendation is based on assuming that, within single years of age, smoking is not related to whether a woman is ever-married or never-married. That may not be a safe assumption. If there IS a relationship, however, you have no way to assess it.

We at DHS much prefer all-women surveys to EMW surveys. The usual justification for the EMW limitation is an assumption that never-married women have had no children, have never used contraception, have no need for contraception, etc. For such outcomes, we follow that assumption (knowing it not completely correct) to expand estimates of fertility, etc., from ever-married women to all women.

An alternative to all-women factors is to merge the women in the IR file with the women in the PR file who are eligible for the individual interview on every criterion OTHER THAN their marital status. For the additional women in the PR file who are brought in with this merge you can impute v201=0 (no children ever born), etc. You can even impute a birth history with no births in it to include them in estimates of fertility rates. But even if you do this, you are limited to covariates that are collected in the household survey.
 Previous Topic: MERGING BR,IR and HR DATA FILES IN R Next Topic: Mixed effect models with complex survey design R
Goto Forum:

Current Time: Fri Jun 2 14:19:32 Coordinated Universal Time 2023