interpretation and production of multi-year ACS

I am not quite sure where to post this, but since this is a relatively active group, I just hope I will get an answer here a bit quicker :).

My questions are:

1. What is the interpretation of a multi-year population? If we are talking about ACS 2010-2014, what is the population that it is really representing? The underlying finite population of people who live in the U.S. keeps changing: new people are born or migrate in; other people die or migrate out; people move within the country, or change their GQ status. So the finite population that ACS 2010-2014 has in mind must be some rather hypothetical population somewhere in the middle of that time range. In particular, the weights must sum up to the number that is resembling the 2012 population count more than the 2010 or the 2014 population count.

2. Is there any technical documentation on how the multiyear weights are produced? For a 5-year data product, you probably need to divide some intermediate weights by 5 at some stage of the process, or do something more elaborate if you changed your stratification schemes / probabilities of selection.

  • The multi-year estimates are generally the average of data collected throughout the time period. This document may be helpful:
  • Perhaps of most use in answering your question (though the answer may not be particularly satisfying) is this excerpt:
    "In general, ACS estimates are period estimates that describe the average characteristics of population and housing over a period of data collection. The 2011-2013 ACS 3-year estimates are averages over the period from January 1, 2011 to December 31, 2013, and the 2009-2013 ACS 5-year estimates from January 1, 2009 through December 31, 2013, respectively. Multiyear estimates cannot be used to say what is going on in any particular year in the period, only what the average value is over the full period. "
  • Thanks, Beth. This is probably the more recent one: They discuss the weighting but not as much the interpretation of the target population (which is conceptually tricky). Roberts and Binder (2009) talked about these difficulties, see