Pooling ACS and Census extracts

USA
1

I’m considering pooling some of the ACS and Census samples – in particular, the 2000 ACS with the 2000 5% Census sample, and the 2010 ACS with the 2010 10% Census sample.

Are these samples non-overlapping? If so, I do not think I would want to pool, as some people might be double-counted. If not, I think I would want to adjust the weights in each sample so that the weighted count sums up to the total population count – specifically, I would multiply each weight in each sample by the ratio between the unweighted count in that sample and the unweighted total count across all samples in a given year.

Lastly, are the Census extracts broadly comparable to the ACS in surrounding years? For instance, can I compare the 2009 ACS to the 2010 Census extract?

Are there other issues I’m not considering?

2

Unless you have a distinct need to pool the samples, I don’t recommend doing this because (for most use cases) there isn’t much benefit.

The detailed topical coverage in ACS data is not available in the 2010 decennial (ACS fully replaced long-form decennial by 2010); while both the long-form and ACS were fielded in 2000, ACS was in beta and the sample size is pretty small (both in comparison to the decennial PUMS and subsequent years of ACS). I cannot definitively say that augmenting the number of unweighted cases for short-form questions by including ACS wouldn’t make a difference, but will note that the decennial sample sizes are quite large (1-in-20 for 2000 and 1-in-10 for 2010).

I cannot find any documentation that indicates these samples are non-overlapping (I think they are drawn independently and, therefore, assume that they do overlap given the large sample sizes for decennial PUMS). Here is additional information on the 2000 decennial sample design, the 2010 decennial sample design, and the ACS sample design.

However, if you are interested in pooling ACS and decennial data, there are a few things to keep in mind (hopefully this addresses your question about broad comparability):

  • The reference period varies: point-in-time (decennial) versus rolling data collection (ACS–noting that the public use data do not note the month of data collection)
  • Residence requirements: usual residence (decennial) versus current (ACS)
  • Pay close attention to the universe and question phrasing as they may differ slightly between the ACS and the census (most notably in 2000 when both the long-form decennial and ACS were fielded).

I also recommend reviewing this summary of differences between the ACS and decennial census and this FAQ about comparing ACS data.