Multilevel modeling with cross-sectional survey data and vce(brr)

[Emily Bonner]

Hello, I am using the Household Pulse Survey data (US Census Bureau) to analyze food insecurity trends in states with grocery taxes during the pandemic. I have created the dataset and weighted it per the US Census Bureau's office of data management direction.
I would like to perform a multi-level analysis of this data to account for household and contextual (state-level) characteristics using the melogit command, on a subpopulation of the dataset that does not include SNAP recipients.

Here is an example of my dataset only including variables of interest:

* Example generated by -dataex-. For more info, type help dataex
clear
input byte cconfINsuf float(reweek rstate) byte rrace2
0 18 0 1
0 18 1 1
1 18 1 3
1 18 1 1
1 18 1 2
1 18 1 1
0 18 1 1
0 18 1 1
0 18 1 1
0 18 1 1
0 18 0 1
0 18 0 1
0 18 0 1
0 18 0 1
0 18 0 1
0 18 0 1
0 18 0 1
1 18 0 3
0 18 0 1
. 18 0 1
end
label values cconfINsuf cconfINsuf
label def cconfINsuf 0 "Enough Food", modify
label def cconfINsuf 1 "Not Always Enough Food", modify
label values rrace2 rrace2
label def rrace2 1 "White", modify
label def rrace2 2 "Black", modify
label def rrace2 3 "Hispanic", modify

CODE for weighting procedure and melogit regression:

svyset [iw=rpweight], brrweight(r2pweight*) fay(0.5) vce(brr) mse


svy brr, subpop(rsnap): melogit cconfINsuf i.rrace2 i.income i.rtenure i.ranywork i.gendA i.reduc i.rms i.rthhld_num i.rthhld_numkid i.reweek##i.r2state || lev2: state

RESULTS:

melogit is not supported by svy with vce(brr); see help svy estimation for a list of Stata estimation commands that are supported by svy
r(322);


The help svy estimation recommended in the error code provides a list of STATA estimation commands that includes melogit, although using melogit with this survey design continues to produce this error.
In STATA, it seems that multilevel modeling cannot accommodate complex survey design using replicate weights.

[Emily Bonner]

For clarification, I would like to perform a multi-level analysis of these data to account for household and contextual (state-level) characteristics using the melogit command on a subpopulation of the dataset that does not include SNAP recipients. The outcome is a binomial random variable.