The 1875 thing seems... unlikely. While it is possible that the SSA decided to create an epoch of 1875, I don't believe COBOL itself has such an epoch.<p>Moreover, the SSA's actual Numerical Identification (NUMIDENT) files, the master records for applications/deaths/claims, appears to stored dates in text as either YYYYMMDD or MMDDYYYY strings according to their archived versions.<p><a href="https://www.openicpsr.org/openicpsr/project/207202/" rel="nofollow">https://www.openicpsr.org/openicpsr/project/207202/</a><p>I think it's more likely that there are missing death records for some entries (which isn't really a problem unless there are active claims), data entry error, someone didn't get the most recent NUMIDENT iteration (the records are essentially append-only) or did something silly like tried to treat social security numbers as unique ids.