The DHS Program User Forum
Discussions regarding The DHS Program data and results
Home » Countries » Ethiopia » Need help (Multilevel mixed effect logistic regression analysis )
Need help [message #27537] Thu, 31 August 2023 05:31 Go to next message
Jeremy is currently offline  Jeremy
Messages: 9
Registered: February 2022
Dear DHS specialists,
I have a question; I am doing research on Acute Respiratory Infection using the EDHS dataset. I merged the KR file with PR file. For the analysis I use the mixed effect multilevel regression model and do the analysis using R. I adapted the complex sample design provided by your office "DHSdesign<-svydesign(id=dta$v021, strata=dta$v023, weights=dta$wt, data=dta)", as follows:
ARI_TRA1 <- svydesign(id=~V021, weights=~WGT,strata=~V023, nest=TRUE, survey.lonely.psu = "adjust", data=ARI1st)
However, the sum of the estimates in the weighted frequency calculation (9904.89) is not the same as the original sample size (8781). (Descriptive analysis of the alive children aged below 60 months). Why is this difference, please help. Stay Blessed! Stay Safe!
Unweighted frequency
ARI4th %>% freq_table(V024)
Freq. %
0 1650.97 16.67
1 2129.04 21.49
2 2002.36 20.22
3 2168.62 21.89
4 1953.90 19.73
n_total 9904.89 100

weighted frequency
outp1 <- svytable(~Child_age,design=ARI_TRA1)
outp2 <- round(prop.table(svytable(~Child_age,design=ARI_TRA1))*100,d igits=2)
var cat n percent
1 Child_age 0 1443 16.43
2 Child_age 1 1821 20.74
3 Child_age 2 1815 20.67
4 Child_age 3 1923 21.90
5 Child_age 4 1779 20.26
n_total 8781 100.00
Re: Need help [message #27539 is a reply to message #27537] Thu, 31 August 2023 09:15 Go to previous messageGo to next message
Bridgette-DHS is currently offline  Bridgette-DHS
Messages: 3090
Registered: February 2013
Senior Member
Following is a response from Senior DHS staff member, Tom Pullum:

You do not have a problem. First, you seem to expect that the weighted and unweighted numbers of cases should be equal--or equivalently, the mean weight is 1--but that's not necessarily the case. I suggest that you open the KR file and enter these two Stata lines (or their equivalent in R):
summarize v005
summarize v005 if b16>0 & b16<.
You will see that the mean weight is not 1 (or 1000000) for the KR file as a whole or for the subset of children who are in the PR file. That's because v005 is normalized during data processing ONLY to have a mean of 1 (1000000) in the entire IR file. Then the mother's weights are assigned to the children. Women can have any number of children. Subgroups of children, or of women, will therefore have mean weights that differ from 1.

Second, when you estimate a statistical model, at least in Stata, using pweights, Stata will automatically re-normalize the weights so that they have a mean of 1. That's a default that makes sense and that I don't think you could override, even if you wanted to. I expect that R does the same thing.

So, as I said, you do not have a problem.

[Updated on: Thu, 31 August 2023 09:25]

Report message to a moderator

Re: Need help [message #27540 is a reply to message #27539] Thu, 31 August 2023 10:44 Go to previous message
Jeremy is currently offline  Jeremy
Messages: 9
Registered: February 2022
Thank you so much Sir and Dear Madam Bridgette. Have a Great Day. Stay Safe!


Previous Topic: Using separate wealth index
Goto Forum:

Current Time: Thu Jun 20 22:15:46 Coordinated Universal Time 2024