The DHS Program User Forum
Discussions regarding The DHS Program data and results
Home » Data » Weighting data » Domestic Violence Weighting
Domestic Violence Weighting [message #256] Wed, 03 April 2013 15:21 Go to next message
DHS user is currently offline  DHS user
Messages: 111
Registered: February 2013
Senior Member
We'd like to create an indicator of children ages 0-4 who live in a household where a woman experiences spousal violence. Our method to create this variable is below:
• In the women's (15-49 yr old) file
o Create an indicator of ever experiencing spousal violence for each woman who participated in the domestic violence module.
o Within households, aggregate to the household level and create one indicator that indicates a woman in the household has experienced spousal violence
o Keep one observation per household with the indicator of a woman in the household experienced spousal violence
• In the household member file
o Keep only children ages 0-4
o Merge onto each child observation by household number the household-level spousal violence indicators from the women's file
Now each child aged 0 to 4, should have an indicator of whether a woman in the household has experienced spousal violence. We're then using the household member weight to calculate a prevalence of % of children. However, we're wondering if we should somehow be incorporating the domestic violence weight. Or, would a different method be preferable to obtaining our desired indicator?
Re: Domestic Violence Weighting [message #257 is a reply to message #256] Wed, 03 April 2013 15:25 Go to previous message
Bridgette-DHS is currently offline  Bridgette-DHS
Messages: 3094
Registered: February 2013
Senior Member
Here is a response from one of our DHS experts Kerry MacQuarrie, that should answer your question.

I take it from your summary below that your unit of analysis is individual children age 0-4, and not households, in which case, the appropriate weight would be the individual weight (v005) and not the household weight (hv005). In either case, the domestic violence weight (d005) would not be the appropriate weight to use, as this weight also accounts for the respondent being selected for the domestic violence module from among eligible women in the household, as well as her non-response to the survey. This factor is not an issue in your analysis since your indicator of domestic violence is applied to all children or to the household, and not the just the module's respondents. It is important to note that in some countries, the domestic violence module was administered to a respondent in every sample household, while in other countries, the module was administered only in every second or every third household. However, the v005 weight still applies as these households were not selected in a non-random way that could introduce some bias, but at a systematic interval. The v005 weight is still proportional to the probability of being selected for and completing the interview.

One other note on something which you are likely already aware: The weight variables I mention above are all expressed as an 8 digit number, with 6 implicit decimals. To arrive at the actual weight value, you should divide your chosen weight (i.e. v005) by 1,000,000. E.g., If v005 is 2561000, the actual weight value is 2.561.

Finally, I should mention that a household level indicator of spousal violence calculated from the domestic violence module responses may somewhat underestimate the proportion of households in which spousal violence occurs. That is to say, spousal violence can reasonably be affirmed to occur within the household if the respondent to the domestic violence module reports spousal violence, but spousal violence cannot be affirmed to be absent from households if the respondent to the module reports no spousal violence. This would be the case if there are other eligible women in the household, at least one of whom would have reported spousal violence had she been selected for the module. The use of a conservative measure, though, should reduce the likelihood of a type I error with regards to any statistically significant results you find.

Previous Topic: Questions about weighting when using merged data sets
Next Topic: weighted median
Goto Forum:

Current Time: Sun Jun 23 07:45:58 Coordinated Universal Time 2024