The DHS Program User Forum
Discussions regarding The DHS Program data and results
Home » Data » Weighting data » Weighting DRC Domestive Violence Data in Multilevel Analysis
Weighting DRC Domestive Violence Data in Multilevel Analysis [message #4151] Wed, 08 April 2015 23:27 Go to next message
carlson_monica is currently offline  carlson_monica
Messages: 9
Registered: November 2014
Location: United States
Member
Hello,
I am about to run my analysis using the DRC 2013 data -- looking specifically at drivers of GBV and using survey data of women selected for the domestic violence module. A key part of my multilevel analysis is looking at variables at the individual data as well as variables I have aggregated to the cluster and province levels. My understanding is that I should use the DV weight, but I have also come across some published papers using DHS domestic violence data that did not weight the data. I'm using SPSS v22, from what I have read, the GENLINMIXED command does not recognize the "WEIGHT BY" syntax. At this point, I am considering my options as 1) running the multilevel model without the D005 weight applied or 2) switching to STATA which allows for weighting in multilevel analysis. Unless, one of your SPSS experts is aware of how to add the DV weight to the GLMM procedure in SPSS. My questions therefore are:

1) If I am just looking at significance of predictors in my model and not producing population estimates, do I absolutely have to use DV weights in my multilevel analysis? This seems to be a grey area in the literature I've read on using sample weights in multilevel analyses.

2) If you highly recommend using the weights, are you aware of how to do this in SPSS GENLINMIXED? Or do you recommend I switch to analysis in STATA? Or do you recommend I run a single-level analysis in SPSS complex samples procedure to account for the hierarchical sampling design of the DHS?

Thank you so much in advance for any assistance on this.
Re: Weighting DRC Domestive Violence Data in Multilevel Analysis [message #4192 is a reply to message #4151] Tue, 14 April 2015 13:09 Go to previous messageGo to next message
Sarah B is currently offline  Sarah B
Messages: 23
Registered: June 2013
Member
Hi Monica,

I agree that the use of weights in multilevel models is a grey area in the literature -- I've discussed this with one of the experts in this field, Sophia Rabe-Hesketh, and she agrees there's not one clear way to go.

That said, weights may be more important when using DV weights rather than regular (v005) weights because of the differential selection probabilities and non-response with regard to the DV module.

I'm afraid I'm not familiar with multilevel modeling in SPSS. In general, I'd say the importance of using a multilevel vs. single level depends on your research question. Assuming the multilevel nature is important for your research Q, switching to Stata may be the best way to go, especially if you have access to Stata v 13 or 14 which allow pweights with the mixed/melogit/meprobit/etc. commands.
Re: Weighting DRC Domestive Violence Data in Multilevel Analysis [message #14819 is a reply to message #4192] Mon, 07 May 2018 11:13 Go to previous message
Hassen
Messages: 121
Registered: April 2018
Location: Ethiopia,Africa
Senior Member
Dear all,Interesting issue!!
Thank you !!


Hassen Ali(Chief Public Health Professional Specialist)
Previous Topic: AIC and BIC model fit tests on weighted DHS data
Next Topic: Country-level weights for MLM
Goto Forum:
  


Current Time: Thu Mar 28 17:31:05 Coordinated Universal Time 2024