art photo



Can one interpolate between the 2000 and 5-year ACS (2005-2009) to get characteristics of census tracts? In other words, can the 5-year data be thought of as a snapshot for 2010?


The 5-year data represent years 2005-2009, so 2010 is not even the endpoint, 2009 is. Regardless, these are not supposed to be seen as representing the end point; they are to be treated as a period estimate for the period represented – 1 yr, 3yr or 5yr. From the ACS website:

"American Community Survey 1-, 3-, and 5-year estimates are period estimates, which means they represent the characteristics of the population and housing over a specific data collection period."

If the geography of an area changes over the period, e.g., Phoenix absorbs a suburb, the 5-year data for Phoenix will be based on the terminal year geography. Likewise, that is my understanding of the Ns as well although you are not really supposed to use the Ns as counts.

So, if one wanted to create an interpolated estimate she could create data for 2000 – 2007 with something like this:

2001 = [(1 x 2000) + (0 x ACS2005-2009)]/2
2003 = [(.8 x 2000) + (.2 x ACS2005-2009)]/2
2005 = [(.2 x 2000) + (.8 x ACS2005-2009)]/2
2007 = [(0 x 2000) + (1 x ACS2005-2009)]/2

Obviously, my weights are arbitrary.

The ACS data are not quite as large as planned and they had particularly low mail response rates in some areas. The end results is that some of the census tract data are quite dicey – e.g., large margins of error.

The ACS Compass products have a nice appendix that describes how to combine units of geography and calculate new margins of error. Here is a link to the appendix:

Appendix to ACS Compass Products

The relevant page is 14, although I would recommend glancing at the entire 27 page appendix. It is written for non-statisticians, but not the average public.

Related Question Groups: