ACS Data Users Group

We're trying to better understand estimate reliability/stability for 5-year estimates for sub-county geographies (i.e. places, county subdivisions, and tracts). We noticed that the ACS weighting methodology has minimum population thresholds for the weighting areas. We're curious if this means that estimates for areas with population < 2,500 (i.e. many census tracts, or very small places) will be more modeled because they are smaller than the weighting area? Or, if that's not quite accurate, how might the weighting areas impact estimate reliability? Thanks!

https://www.census.gov/content/dam/Census/library/publications/2010/acs/Chapter_11_RevisedDec2010.pdf

Page 11-16

• "The weighting areas used for the multiyear estimation are generally smaller than those used for the single-year estimation. They are still formed by complete counties or aggregations of counties and they must meet a threshold of 400 unweighted person interviews at the time of their formation. In addition, for the five-year estimation, the weighting area must have a minimum population of 2,500. For the three-year estimation, this generally results in most published counties being defined as their own weighting area as is the case for the one-year estimation. However, since there is no publication threshold for the five-year data product, there will be counties which are not their own weighting area and therefore greater differences between the ACS and PEP estimates of total population may exist. For the formation of the subcounty control areas, the three-year threshold is 8,000 in total population and the five-year threshold is 2,500."

They aren't more modeled, this just means that these small places are combined with their neighbors when getting weighted. I imagine that at such low levels, the only meaningful specific target ACS gets weighted to is the total population projection, so this is basically saying that the counties of size <2,500 will be combined to form the weighting cells for the 5YR estimates.

I'm going from memory on this but I think that the Current Population Survey is used to adjust ACS tables (raking/Iterative proportional fitting ?) totals at the county geography. For very small counties the individual counties are grouped to provide a large enough population. The CPS estimates use births, deaths and population flows to adjust populations from year to year between decennial census years.

The primary data product to which both ACS and CPS calibrate is Population Estimates (https://www.census.gov/programs-surveys/popest.html); that's the one you are thinking of when you talk about births, deaths and population flows. CPS calibrates at the level of their strata (MSAs and the rest of each state beyond MSAs, roughly speaking); ACS can afford to calibrate at lower levels, such as (large enough) counties.

I thought the ACS was weighted and adjusted (controlled) at the county level when it first came out in 2005 but between 2008 and 2009 it was stepped down Places. I never heard or saw a Place level threshold (size) but I do remember Detroit not being happy when their Pop and Race for 2009 was unbelievable. Fortunately with the 2010 around the corner the ACS could correct itself. I looked through some of the documentation but couldn't find a specific Place size other than working like "incorporated places'. www2.census.gov/.../acs_design_methodology_ch11_2022.pdf