The DHS Program User Forum
Discussions regarding The DHS Program data and results
Home » Data » Sampling » Multilevel modelling
Multilevel modelling [message #13121] Sun, 24 September 2017 11:10 Go to next message
denis is currently offline  denis
Messages: 2
Registered: September 2017
Dear DHS team, i am using pooled DHS data from 21 countries. I am doing a multilevel analysis with individuals at the first level and countries at the second level. (Initially, i wanted to do the modelling at 3 levels namely; individual, cluster (PSU) and country level, however, the average number of women per cluster was 19 and the median was 18 which i felt was too small to derive cluster level aggregrates.
My questions are as follow;

(1) Can i do the modelling at 2 levels (individual and countries) since my average cluster size is small or is the cluster size appropriate (average of 19 women per cluster)?

(2) I intend to use some world bank development indicators in my model as country level factors such as GDP per capita, PPP, however, some DHS survey span over 2 years for example Rwanda DHS was done in the year 2014-2015, Congo DR 2013-2014. For these countries which year is appropriate when selecting for instance GDP.

Re: Multilevel modelling [message #13132 is a reply to message #13121] Tue, 26 September 2017 10:49 Go to previous messageGo to next message
Bridgette-DHS is currently offline  Bridgette-DHS
Messages: 2635
Registered: February 2013
Senior Member

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

Yes, you could do a 3-level model. Normally, when using a single survey, we recommend a two-level model with individuals and clusters. There is a limitation, however, that we cannot divide the sampling fraction into separate terms for the selection of clusters and the selection of individuals within clusters.

I recommend that you calculate the mean date of interview using v006 and v007 (or v008). (It would be sufficient to use month and year of interview, but day is available too if you want to include it.) For most external data sets, July 1 of the calendar year is the reference date. I would select the year for which July 1 is closest to the mean date of interview. But if you are using something like a fertility rate for the 3 years before the survey, the midpoint of that reference time interval (which would be 1.5 years before the mean date of interview) would be the appropriate date for matching with other data.

Re: Multilevel modelling [message #13140 is a reply to message #13132] Wed, 27 September 2017 02:29 Go to previous message
denis is currently offline  denis
Messages: 2
Registered: September 2017
Dear DHS team

Thank you very much for your prompt response

Kind regards.
Previous Topic: Stratification in pooled datasets - Tanzania
Next Topic: Pooling male and female files
Goto Forum:

Current Time: Sat Dec 3 20:23:08 Coordinated Universal Time 2022