Although the SSDI is not the “gold standard,” it nevertheless offers certain information not provided by NDI (such as Social Security numbers for name matches), permitting a check on assumptions as to which combination of identifying characteristics results in an acceptable match.
The Social Security number was found to be a necessary, but not sufficient, requirement for a match. The addition of last name with minimal additional identifiers (as seen in group 192–197) was found to be inadequate. This group, however, was rare (5 instances out of 3,505 candidate matches). Possibly when the last name is present, other identifiers are present also.
The addition of last name, together with some other unique information (first name, day of birth, and month of birth), or Social Security number with day of birth and month of birth was, however, likely to provide a good match.
We found errors on 6% of the Social Security numbers evaluated. They included a transposition error and identification of a spouse, a parent, or somebody quite different. Our cohort was born before Social Security was instituted, some (in particular, women) may not have worked outside the home and so did not obtain personal Social Security numbers, and others may have worked in occupations that were not included in Social Security at that time. Some may be entitled to Medicare through a spouse and so present the spouse's number as their own. Such conditions create problems in identifying people by Social Security number alone. With certain exceptions (state employees in certain states, railroad retirees), Social Security numbers are now required for all and are issued to newborns. Accurate matching for those born more recently should therefore be easier.
Additional issues may create problems for matching. We found that NDI, SSDI, and EPESE dates of birth did not always agree. It is unclear whether this reflects inaccuracy by EPESE interviewers, by persons completing the death certificate, or informant error. In none of these cases were source documents (birth certificates) or derived sources (driver's license, passport) consistently checked. If the SSDI day of birth derives from Social Security information, it may be the more accurate source, but it is unclear whether SSDI uses this source. So, interviewer error, subject error, and entry error may all be implicated, not just for dates of birth, but for the Social Security number also.
Nevertheless, parsimonious information (Social Security number plus additional personal information) could adequately distinguish survivors from decedents. In epidemiologic studies, the NDI is commonly used to identify survival status, date of death, and death certificate-determined cause of death, all matters of importance when planning health services for an aging society. Although matching is often mentioned, the criteria and their level of accuracy are rarely indicated. It is important that such matching be accurate, and it is helpful if the criteria used are readily available and easy to apply. We offer an approach that meets these requirements.