Multilevel model with region random effects [message #26208] |
Thu, 16 February 2023 09:08 |
sophia614
Messages: 1 Registered: February 2023
|
Member |
|
|
Dear DHS survey specialists,
I am using the women's data in Benin 2018 to run a multilevel model with region random effects in Stata. Most of my variables are at the individual level; however, I do have one measured at the regional level. I tried running the following commands in Stata, but received an error message after the second command (see below).
svyset psu [pw = weight], psu(psu) strata(v023) singleunit(centered)
svyset: xtmixed dep ind $controls || region_recent:
xtmixed is not supported by svy with vce(linearized); see help svy estimation for a list of Stata estimation commands that
are supported by svy
I believe that I need to create level weights, but am not sure how to do so. Do I create them at the region-level (since my random effect is at the region level) or at the psu level?
Any help you can provide is greatly appreciated.
Best wishes,
Sophia
|
|
|
|
Re: Multilevel model with region random effects [message #26532 is a reply to message #26217] |
Wed, 29 March 2023 19:59 |
Aline Semaan
Messages: 3 Registered: March 2023
|
Member |
|
|
Hi Tom,
I am facing a similar situation as Sophia and I found your answer to be quite logical. But I have seen some contradicting information in a previous DHS analytical study (here: https://dhsprogram.com/pubs/pdf/AS26/AS26.pdf) and I would like to get a clarification.
I am conducting an analysis where I plan to study associations between health system capacity (extracted from SPA survey) and length of stay in health facilities after birth (from the DHS). The SPA survey for this country is not a census, hence only a sample domain linkage is possible (DHS Spatial analysis report 10). Thus I plan to link the two databases using the administrative level variable (i.e. Region), and consequently conduct a multi-level modeling approach since the outcome is at the individual level, but some predictors are at the regional level. To quote Wang et. al : "Using the standard statistical approach, questions arise about the appropriate unit of analysis. Analyzing data at the individual level ignores the nesting of people within regions, which results in underestimating the standard errors and increasing the chance of incorrectly rejecting null hypotheses (Raudenbush and Bryk 2002). Alternatively, if the unit of analysis is the region, it will be difficult to include individual-level variables in the analysis. We can address these problems with multilevel modeling and allow for simultaneous investigation of the effects of group-level and individual-level predictors on individual-level outcomes."
In their report, Wang and colleagues do not go into the details of adjusting for sampling weights, and you might be familiar with the fact that running an melogit in Stata fails when the level2 variable is Region because there are no weights for this level.
I failed to find solutions online up to now. Do you have any suggestions on how to conduct this analysis without ignoring the hierarchical structure of the data, and still correctly adjusting for sampling weights? Would including "region" in the multivariable model be sufficient in this case?
Many thanks,
Aline
|
|
|
Re: Multilevel model with region random effects [message #26553 is a reply to message #26532] |
Fri, 31 March 2023 07:57 |
Bridgette-DHS
Messages: 3210 Registered: February 2013
|
Senior Member |
|
|
Following is a response from Senior DHS staff member, Tom Pullum:
I have checked with Senior DHS Researcher, Sara Riese, and we agree that you could use a two-level model with women at level 1 and region at level 2, with 1 as the level 2 weight, because regions were not sampled. If the SPA you are using was not a census, and the facilities were sampled, then there were probably different sampling fractions for different facility types. The facility weights are proportional to the inverse of the sampling fractions for the respective facility types. The client weight accounts for the selection of the facility.
Let us know if you have further questions. For your preliminary analysis, a simplified version of svyset will be ok, but you want to be confident in the final version.
|
|
|