And the IRS has three times confused me with my father, who shares the same name (and, yes, we long ago shared the same address). He is not a US citizen, however. My theory is that, confronted with two similar records, one with a SSN and one without, they assume the one without is just missing it, and merge them.

To the OP - this is a very difficult problem, IMHO. So you need to think about what you can do if you have a low confidence in your match - is the information still useful?

There may be also be many times when you have two datasets that appear to match, but in reality refer to very different beings. I guess it all depends on your data.

There are various tools to canonicalize an address, but I don't know of any free ones that aren't for personal use only. See http://www.cedar.buffalo.edu/adserv.html