I want to analyze an MSA (from a particular point in time) and I can piece it together from counties but the 3 year 2012 ACS file does not have a COUNTY variable.
The COUNTY variable is constructed by IPUMS-USA using PUMAs. The Census Bureau redefines PUMA boundaries every 10 years, in correspondence with the Decennial Census and the 2012 ACS was the first to use the 2010 Decennial Census Based PUMAs. The 2012 ACS multi-year files, however, maintained the PUMA definitions for the years in which the data were collected (i.e. the 2008 through 2011 data still use the 2000 Decennial Census PUMA definitions). This means that the 2012 multi-year files have two separate PUMA definitions within a single dataset and users should apply PUMA definitions based on the MULTYEAR variable. To define COUNTY in the 2012 multi-year files you can create an extract containing the 2012 and 2011 single-year ACS files and create a crosswalk based on YEAR, STATEFIP, and PUMA relating to COUNTY. In this new extract containing the single-year 2012 and 2011 ACS samples, keep one case for every unique YEAR-STATEFIP-PUMA combination. This will give you a crosswalk between PUMA and COUNTY for both the 2010 and 2000 PUMA boundaries. You can then perform a many to one merge of the new crosswalk onto your muti-year 2012 ACS data file using MULTYEAR-STATEFIP-PUMA (NOTE: you will need the 2008-2011 MULTYEAR value to be 2011 in order to merge with the crosswalk. I recommend creating the variable MERGEYEAR in both your multi-year file and the crosswalk to prevent and confusion).
I hope this helps.