The DHS Program User Forum
Discussions regarding The DHS Program data and results
Home » Countries » Ethiopia » Weights not allowed for Model analysis??
Weights not allowed for Model analysis?? [message #14755] Thu, 03 May 2018 15:28 Go to next message
sko16 is currently offline  sko16
Messages: 21
Registered: June 2016
Location: Addis Ababa, Ethiopia
Dear DHS Experts,
I am using the Individual (women) 2016, Ethiopian DHS data set. I want to look Frequency of ANC (Antenatal Care) Services by re-categorizing into:
Inadequate visits (< 4 visits) = 0
Adequate visits ( 4 and more visits) =1
against women sociodemographic characteristics such as with Age, Educational status, working status ... etc applying Logistic regression using STATA version 15. But the output gave me error message as 'weights not allowed'. For instance, I used the following command to weight the data: xtmelogit FANC MABC [iweight = wgt]; here, FANC is the dependent variable that stands for Frequency of ANC (Antenatal Care) Service and MABC stands for Mothers age in categoery.
My question is, is it possible to analysis models like Logistic regression by weighting the DHS data or not? If yes, would you give me a direction to walk, please?

Re: Weights not allowed for Model analysis?? [message #14770 is a reply to message #14755] Fri, 04 May 2018 10:52 Go to previous message
Bridgette-DHS is currently offline  Bridgette-DHS
Messages: 2253
Registered: February 2013
Senior Member

Following is a response from Senior DHS Stata Specialist, Tom Pullum:

When you get that message, it means that the current version of the estimation command does not allow weights. You then have only two options. One is to proceed without weights, in which case your estimates will be biased toward the over-sampled strata in the survey. The other option is to use a less advanced model. For example, try using just logit, or melogit, with svy. Logit will certainly work and I think melogit will work. I would take the second option, because I think the bias from ignoring the weights is more serious than the fine tuning of the standard errors, which is the main role of the more advanced models. But not everyone would agree. You have to decide--but it's one or the other.

In the history of statistical software, many procedures that now include weights and the other svy adjustments first appeared in versions that did not initially include them. Who knows, maybe Stata 16 will allow weights for this command.
Previous Topic: Result Difference
Next Topic: Calculating Timing of First PNC check up from EDHS 2011
Goto Forum:

Current Time: Wed Apr 21 12:42:13 Coordinated Universal Time 2021