I’ve finally found 3 reasons that actually make sense as to why the American Community Survey (ACS) is less reliable than the Census 2000 Long Form. The majority of information below comes from the CTPP Listserve (attributed to Freddie Navarro) and the Census Bureau’s Brian McKenzie’s presentation here.
Why the ACS is Less Reliable Than the Census 2000 Long Form
- When the American Community Survey was in the test period, the mail-back + CATI “cooperation rate” was estimated to be about 78%. This percentage was based on Census 2000 data. But the actual ACS cooperation rate has been closer to 50%.
- The lack of tract-level controls (population totals by age/sex/race) has resulted in an increase of 15-25% in the standard errors.
- The ACS 5 year estimates will include survey responses from 11 million households over 5 years. The Census 2000 Long Form included responses from 18 million households in 1 year. Fewer data points gathered over a longer period of time result in more errors.
In the most current results show that the Coefficient of Variation (CV) for the ACS is 75% higher than the CV from the Census 2000 Long Form. The original estimate was that the ACS CV would be about 33% higher.
Do you know of any other statistical reasons why the ACS data is less reliable than the Census 2000 Long Form data? How will having less reliable demographic data affect you and your projects?