Use or not to use replicate weights, FSS 2001-2020

Hello - I’m looking at cross-sectional trends in food insecurity data. From 2001 to 2009, there are no replicate weights. Then there are 145 replicate weights from 2010-2013, then 160 for 2014 onward.

If I calculate trends asking STATA to use replicate weights when they’re only available for some years, the coefficient of variance balloons for those years without 160 replicate weights (understandably).

Should I:
a) create my own brr replicate weights for 2001 to 2009? What would I do for 2010-2013?
or b) do my analysis without using the replicate weights? If that’s the case, would I just use Taylor linearized variance estimation? Is there something else I should put in the svyset command to control for the survey design? I understood that the replicate weights were a way of accounting for the strata / cluster design since we don’t have those variables

Thank you!

Regarding A, there is no way to create your own replicate weights, because these are created using the full set of sampling design variables (cluster, strata, etc.), which are not included in the public use microdata.

Regarding B, if you are doing the analysis without replicate weights, there is no need to use svyset at all. Just use FSSUPPWT as a sampling weight (pweight in Stata).

My best recommendation is to do the analysis both with and without replicate weights in the 2010-13 and the 2014+ periods. Compare the resulting standard errors. If you see clear patterns, you might consider adjusting your calculated standard errors in the pre-2010 period. Davern et al. (2006, 2007) in the journal Inquiry also lay out a method for improving variance estimation using only variables available in the public use microdata.