I was trying to explain ACS to a general population; here's what I wrote:
As you may know, 2000 is scheduled to be the last decennial Census that will gather "long-form" data. The information gathered from the long form of the census covered income/poverty, language, ancestry, education, homeowner/renter status and expenditures for same, plus much more.
Data users were frustrated about having nine-year-old figures in 2009. Thus was born the American Community Survey, which will eventually provide numbers every year starting in 2010.
Because there are different size requirements for geographies, this creates what might be confusion for the user.
2006: 2005 ACS - statistics for entities of 65,000
2007: 2006 ACS - statistics for entities of 65,000
2008: 2007 ACS - statistics for entities of 65,000 PLUS
2005-2007 statistics for entities of 20,000
2009: 2008 ACS - statistics for entities of 65,000 PLUS
2006-2008 statistics for entities of 20,000
2010: 2009 ACS - statistics for entities of 65,000 PLUS
2007-2009 statistics for entities of 20,000
2005-2009 statistics for all entities
Every year after 2010 will have a rolling set three sets of data for the largest geographies. So, in 2010, there will be reported surveys information gathered over 12-, 36- and 60-months periods for Queens County.
But which one should one use?
The multi-year reports are less current. Statistics are gathered over a larger period. However, multi-year reports are more statistically accurate because they include more records, more collected surveys. Particularly for figures with fewer occurrences, e.g., number of grandparents raising their grandchildren, I'd be inclined to opt for the longest period available.
When comparing geographies, obviously one must consider the size of each entity. There are no one-year ACS stats for 24 of New York State's 62 counties because the population of those counties are each below 65,000.
In any case, when doing comparisons, these two rules should be noted:
1. Do not compare one-year stats, three-year stats and five-year stats with each other. Compare one-year with one-year.
2. When comparing multi-year stats, do not compare overlapping years. For instance, do not compare 2005-2007 with 2007-2009 because both have the exact same records from 2007. This is not an issue yet but will be in 2010.
Finally, do all you can to support the 2010 decennial Census, which, like the 2000 survey, is one of the few attempts to get a full count, rather than a sampling. Therefore it becomes the benchmark of other calculations, both for Census products such as the ACS, and for the private firms that estimate demographic and consumer trends, such as Demographics Now and EASI.