Census to Eliminate ACS 3-year Estimates

The Census Bureau has announced their intention to eliminate the American Community Survey 3-year estimates products starting with the upcoming 2012-2014 estimates series. This is an important data produce that well impact small to medium sized communities nationwide.

If you aren’t familiar with the ACS product stream, annual period estimates are reported for 1-year, 3-year, and 5-year periods based on the population size of governmental and administrative geographies. This is because the ACS sample simply isn’t large enough to provide 1-year data for all areas. Areas of 65,000 or more can receive estimates based on one year of data collection. Areas between 20,000 and 64,999 receive estimates based on three years of data collection and all areas regardless of size receive estimates based on the full five year data collection period. The elimination of the 3-year estimates will impact a governmental and administrative units with greater than 20,000 population.

Here are a number of important issues that arise as a result of this decision and reasons why the Census Bureau should reconsider this cut of data products.

  • Nationwide, 39 million people or 12.2 percent of the nation’s population live in counties between 20,000 and 64,999 and 55 million people or 17.5 percent live in medium sized cities, villages and Census Designated Places. In seven states, over 30% of the population live in counties in the 20,000-64,999 range. In Vermont, 72 percent of the state’s population lives in counties that would be affected. All of these areas would lose this valuable data source.
  • The 3-year estimates are better than both the 1-year and 5-year data for time series comparisons. By the Census Bureau’s own recommendations, the use of non-overlapping periods for time series comparisons means that the 5-year estimates must span a 10 year period to allow for proper comparisons. The 3-year estimates allow for more frequent analysis and avoid much of the irregularity of the 1-year estimates. For example, the 3-year estimates allow for comparison of data from 2005-2007 to 2008-2010 and 2011-2013 (pre-recession, recession, and post-recession). Elimination of the 3-year estimates means such analysis has to rely on periods 2005-2009 and 2010-2014, none of which are useful for analyzing the impacts of the recent recession. This really provides the best example of the negative impact of reliance on the 5-year estimates. The 3-year estimates allow for analysis of the social and economic climate before and after the recession while the 5-year data would totally mask such sudden periods of economic distress. Analysis of any future recessionary periods would be greatly hampered by reliance on the 5-year estimates.
  • The 3-year estimates provide needed stability in the estimates over the 1-year data while still reflecting a reasonable time period for analysis. A simple measure like the poverty rate for the City of Albany, New York (population 97,000) shows considerable year-to-year variation that is not statistically significant and can be very misleading to data users. The 3-year estimates show a much more stable level of poverty and avoid the mistake of placing too great an emphasis on the reported data without verifying the margins of error, which unfortunately, is a problem for too many users.
  • Data suppression as a result of filtering is also an issue. Characteristic detail can be lost when suppression occurs and the 3-year data products have the advantage of larger sample sizes. Thus even the 1-year estimates, available for areas with higher population thresholds, can yield less data than the 3-year estimates. Again, resulting in a net loss of characteristic detail that the 3-year estimates provide.
  • Changes in questionnaire content, definitions and wording can negatively impact comparisons over time. Reliance on the 5-year estimates means that any changes in content require a prolonged period of data collection before being able to use comparable data items.
  • Just when data users have become comfortable with the ACS product stream and have learned how to use, and not use, the data, the Census Bureau is eliminating one of the most useful products. Part of the Census Bureau’s stated rationale for this cut is that the 3-year estimates were never intended to be a permanent part of the product stream. If that is the case, that position was never stated to the data user community. The 3-year estimates have become an important analytical tool for communities in that 20,000 to 65,000 range and the 5-year estimates simply don’t provide the needed portrait the ACS has provided for the past seven years.

These issues all reflect the importance of the 3-year estimates and their use in population analysis. The 3-year estimates are important for medium sized areas of 20,000 to 65,000 but they are also important for improved analysis in areas of over 65,000 where the 1-year data exhibits considerable variability and inconsistent trends.

You can comment on this proposed cut by directing correspondence to Mr. John Thompson, Director of the Census Bureau, Ms. Katherine Wallman, Chief Statistician at the Office of Management and Budget, and your local representatives.

This entry was posted in ACS Data Releases, Census Data. Bookmark the permalink.

Leave a Reply

Your email address will not be published. Required fields are marked *