From the VRE documentation: "The variance replicate estimate tables are intended for advanced users who are adding ACS data within a table or between geographies"
I was wondering if this also holds true between tables. My guess is no, but if the sample drawn for each replicate is always the same regardless of table, then this would be true.
I tried testing it empirically but got peculiar results. I compared tables 19001 (income) and 25075 (housing value), which I know to be highly correlated, and ranked the 80 replications for various rows and combinations of rows. If replicate #27 had more wealthy people than any other replicate, it would also have greater numbers of high-end homes, on average. I found the correlations here to be around 0.2, which is much higher than chance but much lower than I'd expect if my original proposition was true.
- Frank Boscoe
The replicate weighting factors are applied to cases (persons and households), so I'd be seriously surprised if these did not work out between tables. (I mean, if I were doing them, it would have been SO MUCH EASIER to just do this uniformly between tables. But I am not the Census employee... just your friendly survey statistician.)
I'm curious whether there was ever an official answer found for this question. I'm wondering something similar. I'd like to calculate the MOE for vacant - seasonal as a percent of total housing units which requires using tables B25002 and B25004. Would I just use the variance replicate estimate tables for both to calculate the var_rep to use in the formula?