A certain amount of difference in population estimates are understandable across aggregated sets. After all, the data captured are for different year intervals. What about large swings? For example, weighted (perwt) estimates for ACS IPUMS CPS in PUMA = 11300 (Grays Harbor & Mason Counties, Washington State) for 1-year (2012) is 132,246; for 3-year (2010-2012) is 29,883; and for 5-year(2008-2012) is 23,404. Any tips on this puzzle?
I believe the answer to your question can be found in the “Note Regarding Multi-Year Samples” on the PUMA variable description page.
I hope this helps.