Frank, Elaine, Pete and others all gave spot-on assessments.  Thank you!

I’m a frequent ACS PUMS user -- so I’ll contribute what I’ve noticed as I’ve compared ACS PUMS 1-Year datasets vs ACS PUMS 3-Year.  The microdata at record-level should be almost identical. You should see the same numbers of records, and also almost identical data taken from respondent questionnaires.  One exception would be: where the coding of responses has changed (for example, the coding of “YearBuilt” changed after 2005). 

Beyond that – another exception – Bureau-assigned data elements could change.  Most notably, you can detect slight adjustments to the weights applied to individual household and person records. This happens for the reason described by Pete – the updating of annual population totals.

Whether you’re working with PUMS or summary data, the Bureau’s tweaking of annual population totals should be fairly small… for now… 

The really big series break will be in 2011 – when the decennial Census reveals just how far off those annual estimates have been, which should cascade over into a re-benchmarking of ACS 1-Year, 3-Year, and 5-Year tables  published in Fall 2011.

Consider: The Census Bureau has estimated Minnesota's 2007 population at 5,182,000.  Meanwhile, the Minnesota State Demographer and Metropolitan Council also publish annual estimates (our official numbers for State government purposes), putting Minnesota's 2007 population at 5,263,000, or 1.6% higher.  This is not a small discrepancy – and I strongly suspect that the Bureau’s efforts to make everything fit leads to seriously unreliable housing occupancy (vacancy) rates in a lot of counties and cities.

Can I ask: Do FSCPE analysts in other states see similar discrepancies between the Bureau’s annual estimates and “alternative” (State Demographer) annual estimates or annual projections??  Very curious to know.

-- Todd Graham

   Metropolitan Council Research

   651/602-1322