Using replicate weights in simulations

1

Is it possible to use replicate weights in simulation-based modeling?

The idea would be:

(1) Use the replicate weights to estimate total counts and SE for the population being simulated, per normal intended use.

(2) Draw a random value from a normal distribution using the estimate and SE values from (1).

(3) Find the replicate weight whose total is closest to the value drawn in (2).

(4) Use that particular replicate as the weight for that simulation iteration.

The intent is to propagage measurement error while still preserving information from the replicates.

2

From the basic steps you’ve described here, I think this should work just fine. In general, I would not expect the results using replicate weights to differ too much from results using the standard person or household level weights. Depending on the specifics of your calculations, however, the replicate weights could generate larger standard errors and slightly influence interpretation.