The DHS Program User Forum
Discussions regarding The DHS Program data and results
Home » Topics » Domestic Violence » calculating prevalence of physical and/or sexual spousal violence
calculating prevalence of physical and/or sexual spousal violence [message #3718] Fri, 30 January 2015 16:53 Go to next message
kdelwiche is currently offline  kdelwiche
Messages: 1
Registered: January 2015
For every DHS country-year that included a domestic violence module, I want to calculate the proportion of women that have experienced any physical and/or sexual violence from their spouse or partner. I am using questions d105a-d105i to construct a binary exposure variable: if the respondent indicated "yes" to one or more of the violent acts, they are "exposed". However, I am unsure as to what to do if a respondent answered some, but not all of the questions about specific acts of violence. I am wondering what the protocol was for calculating this aggregate prevalence metric for the published reports, because the prevalence I'm calculating is typically lower than the physical and/or sexual violence prevalence figures presented in the report tabulations. Are all incomplete responses dropped? Did you impute missing responses? Do you assume missing responses are "No"?
Re: calculating prevalence of physical and/or sexual spousal violence [message #3729 is a reply to message #3718] Mon, 02 February 2015 17:54 Go to previous message
Trevor-DHS is currently offline  Trevor-DHS
Messages: 751
Registered: January 2013
Senior Member
First I suggest checking that you can match the denominator for the tabulation. The domestic violence (DV) questions are not asked for every woman included in the survey. Generally you need to select only the women who were selected for the DV module and who responded. Generally we select women for which v044 == 1. Additionally you may need to select to a subset of these for particular tabulations. For example, it sounds like you would need to select only those women who are married or living in union (v502==1).

Once you can match the denominator, then you will probably find just a few cases where responses were not given to particular questions - code 9 is used for these missing responses. In general we consider these missing responses to be "No" as the number of such cases is few and will have little or no effect on the results and we prefer to be conservative with our estimation of an indicator.
Previous Topic: 2007 Liberia Domestic Violence Module - Response Rates
Next Topic: Logit regressions using two different weights
Goto Forum:

Current Time: Sun Jun 20 18:26:11 Coordinated Universal Time 2021