using replicate weights

I am using the 5-years estimates (2010-2014) and I am attempting to make PUMA level estimates for American Indians and Alaska Natives who use Indian Health Services (a combination of the race and health insurance variable), and I am using the replicate weights in STATA. I would like to create estimates of education (5 categories). However, I am receiving a an error message.

Here are the stata commands:

svyset[pweight=perwt], vce(brr) brrweight(repwt1-repwt80) fay(.5)mse

gen ihs_aian=(racamind==2 & hinsihs == 2)

 

set matsize 11000

svy, subpop(ihs_aian): prop education, over(stpuma)

 

Error message:

too many categories

an error occurred when brr executed proportion

r(900);

 

I have not received this message when I use p weights and only with replicate weights. I would appreciate it if you would be able to offer any suggestions.

  • It's a bit strange that you get this message when only using [pw] did not bring the problem up. The categories (cross-tab of PUMA and education) should be the same for each set of replicate weights. (I would suggest just using vce(sdr) sdrweight(repwt1-repwt80), by the way.) If there are about 3,000 PUMAs, then interacting them with 5 categories of education pushes it beyond the 11,000 matrix dimensions that Stata can deal with.

    A simple fix could be to run this in halves, with subpop( if ihs_aian & inrange(statefip,1,30) ) and subpop( if ihs_aian & inrange(statefip,31, 56) ), for instance.

    I wonder though how you are planning to use those results by PUMA. I don't think there will be that many readers of a paper that gives a 3000x5 table of educational attainment :-\.
  • Thank you! It seems to have fixed my problem. I appreciate it! I am creating the estimates to add them to one of my data sets.
  • You may need to drag their standard error around and account for them. If you use the estimates as regressors, the estimates will be subject to the measurement error attenuation bias.