concerns about participating. The impact of such waivers on the response rates and the resultant bias introduced by refusals also should be taken into consideration and evaluated during the research and testing phase of the survey.

The presence in a regulatory environment of statistical databases linked to earnings and other administrative files is precarious. It is not clear whether individuals in the sample would have legal protections from regulatory uses of their survey data when they are matched to the longitudinal earnings file. But they deserve such protection.

There are several accepted models of matching that remove from the regulatory agency direct access to the information that would allow actions to be taken against individuals. SSA is urged to review the procedures used in the Health and Retirement Survey sponsored by the National Institute on Aging and other relevant surveys to protect the respondents' confidentiality and yet permit linkage with other data files. For example, the matching could be performed by the contractor, and a file could be returned to SSA stripped of all identifiers. The contractor also could update the match over the years. Other arrangements to guarantee that matched files cannot be abused deserve attention.

RECOMMENDATION 4-8. The committee recommends that the Social Security Administration enhance its safeguards for matched data according to accepted practices by employing procedures used in recent federal surveys, and that it take into consideration the effect of such procedures on response rates.