American Community Survey Data Users Group
ACS Data Products & Resources
ACS Public Use Microdata Samples (PUMS)
ACS Summary Files
Application Programming Interface (API)
ACS Data Issues
ACS Data for Small Geographic Areas & Population Subgroups
Mapping ACS Data/GIS Applications
Measuring Trends Over Time with ACS Data
Measuring Trends Over Time…
ACS vs CPS at the state l…
This group requires membership for participation - click to join
Dec 11, 2013 2:34 PM
ACS vs CPS at the state level
Which is more 'accurate' at the state level? Single year ACS or CPS? In particular, I'm looking at trends in the number of people in poverty over time. Some times the two different trends seem to go in opposite directions. My concern is espectially for 2012 where the ACS shows poverty not growing, but the CPS shows a pretty sharp increase. Maybe the general trends are similar, but year to year they seem different.
I have two sources
1. Table 21. Number of Poor and Poverty Rate, by State
Historical poverty tables. These tables are based on the CPS
2. ACS, S1701, Poverty Status in the Past 12 Months, American Community Survey, various years.
Not sure how well this table will stay formatted, but I've attached a spreadsheet too.
below ACS CPS
2012 3,025,016 3,328,036
2011 3,027,342 3,085,000
2010 2,821,470 3,062,000
2009 2,691,757 3,018,000
2008 2,581,491 2,734,000
2007 2,570,014 2,757,000
2006 2,662,199 2,668,000
2005 2,565,836 2,760,000
Dec 12, 2013 11:00 AM
My understanding is that the ACS is much better/more accurate for state-level estimates purely because the sample size is so much larger than the CPS. To take one example, Alabama, the CPS estimate is a 16.2% poverty rate in 2012, and a 1.9% standard error (the 90% margin of error, which is what the ACS reports rather than standard error, would be +/- 3.1%). The ACS for the same year, by contrast, has a higher point estimate but a much smaller margin of error -- 19.0% and +/-0.4%, respectively. In fact, the CPS estimate of Alabama's 2012 poverty rate, due to the large margin of error, isn't significantly different than the ACS estimate, because the upper value of the CPS confidence interval is 19.33%. So I would defer to the ACS estimate for state-level analysis to ensure greater confidence in the point estimate, which is what generally gets the focus.