Replicate Weights Margin of Error


I am curious about the sample error outputs from the microdata replicate weights vs those supplied by the official data tables on I have done a quick tabulation by sex in Philadelphia County. While the estimates are roughly equal to what is showing on, the margins of error are much larger.

I calculated the margin of error with a 90% confidence interval, as suggested by the Census Bureau.

Is this variance between the microdata sample and the official tabulations typical? Are the official margins of error calculated with a different method?

I am not sure exactly which tables you are using from, but would direct you to the most recent ACS data user’s guide for more information on margins of error for tabular data from the Census Bureau. In general, I would expect to see wider confidence intervals using the PUMS microdata than aggregated estimates, though I am surprised at how small the margins of error are in the official estimates you shared. For what it’s worth, I replicated your point estimates and margins of error for a 90% confidence interval using the ACS 2018 5-year data.