Why are some MSA identifications missing in 2012 ACS data for Virginia (e.g. Charlottesville)?

I am interested in examining observations that (with high probability) are in the Charlottesville MSA. This MSA contains one PUMA (51090) in its entirety-- but all observations coded into this PUMA take on the value “Not in identifiable area” for the MSA variable, met2013. Similarly for the two PUMAs (51089 and 51105) that overlap, but are not exclusively contained within, the Charlottesville MSA.

Why can some observations be identified, but not these? Is there a work-around?

Charlottesville, VA is an example of how the new MET2013 protocol for identifying metropolitan areas differs from the protocol used in METAREA. Under the METAREA protocol, Charlottesville would have been considered an identifiable MSA because, as you pointed out, PUMA 51090 lies entirely within a single metro area, and thus has 0% commission (no persons residing outside a metro area are identified as living in it). However, MET2013 takes both commission and omission (where a MET2013 code is not assigned to some residents of the corresponding MSA). So, while you can confidently state that every person living in PUMA 51090 is also living in the Charlottesville MSA, the sizable portion of the Charlottesville MSA that lies in other PUMAs (which themselves include areas outside of the MSA) means that a frequency or analysis of MSA would not accurately represent the Charlottesville population.

If you are willing to accept the omission error, you can use the Crosswalk between 2013 MSAs and 2010 PUMAs spreadsheet available at the bottom of the MET2013 description page to identify PUMAs with 100% of their population within a single MSA.
I hope this helps.