children. The new legislation emphasizes desired changes in the behavior of adults, but a very important question is the effects on children. In that regard, one participant spoke of the need for "canaries," that is, outcome measures that can provide early warning signals of problems for children (e.g., children in very poor families in deep poverty), and "crocuses," that is, outcome measures that signal positive trends (e.g., children in families that are no longer very poor).

Workshop participants also noted the need to measure outcomes for both children and adults in families who would have been eligible for benefits under the old program rules but are no longer eligible. Issues of concern include whether previously eligible families will be able to maintain their own households and how they are affected when there is an economic recession. Finally, some participants stressed the importance of improved measures of health and disability status for a range of purposes, such as monitoring the effects of changes to SSI program eligibility, the transformation of public assistance into a work-based program, and, more generally, the aging of the population (see Adler, 1996).

Almost all participants noted the importance of collecting detailed and consistent information on program eligibility criteria and benefit provisions from the states. Given the devolution of responsibility for health and social welfare programs from the federal to state and local governments, program rules with regard to eligibility and benefits will almost certainly differ across states. Program rules may also differ within states, and they may change over time. In the absence of reliable and timely information on these rules, it will not be possible to use national household survey data to accurately simulate program eligibility, calculate participation rates among those who are eligible, or evaluate the accuracy of reporting of participation and benefits. Also, without regularly updated, reliable information about state (and local) program rules, it will not be possible to develop survey questionnaires that ask appropriate questions for respondents in different parts of the country. Thus, a comprehensive national database that provides detailed, consistent, and timely information about how state and county programs are intended to work needs to be developed (see Zedlewski, 1996).

A number of organizations, including the Congressional Research Service, the Urban Institute, and the American Public Welfare Association, are attempting to monitor changes to state welfare program rules; see Table 1. However, these efforts are generally limited in coverage of states or programs (or both). Workshop participants stressed the need for a central source for collection and dissemination of information about program rules.

In addition to discussing changes needed to national household surveys, workshop participants discussed the value of administrative data on current program beneficiaries. Historically, such data have been useful for tracking program participation, simulating the likely effects of program changes that decrease benefits or restrict eligibility, and validating survey-based reports of program participation and benefits. However, administrative data cannot be used to simulate program changes that increase benefits or expand eligibility because they contain no information on nonparticipants. Also, their use for behavioral research is limited because they do not cover people who stop receiving benefits, have limited information for current participants, and, as just noted, have no information for nonparticipants.

With PRWORA now in place, administrative data tracking the AFDC system will be eliminated and replaced with a system that tracks participation in state and local programs funded by the

TANF grant. Thus, time series of administrative data for AFDC will end. Still, administrative data may be useful to evaluate and track caseloads in each state. In fact, PRWORA requires states to file quarterly reports detailing demographic information, employment status, earnings, and forms of assistance provided for a sample of TANF recipients. However, given that current federal regulations provide no matching money for the development of the quarterly reports (the federal government previously shared in the costs of program administration and reporting) and the lack of a quality review process, workshop participants raised questions about how reliable those data will be.

Workshop participants noted that administrative and quality control data on the Food Stamp Program may prove useful for some kinds of analyses of welfare program trends and outcomes. For many food stamp recipients, a time series of administrative data will continue.

The usefulness of administrative data for evaluating health care services, costs, and quality has been affected by changes in the health care industry, particularly the rise in managed care. Traditionally, public health insurance programs have used fee-for-service payment plans that create a paper trail linking payments to services that can be analyzed, and claims data for the Medicaid and Medicare programs have been used to evaluate health care services funded by these programs. However, the growing use of managed care plans, which often charge a flat annual payment for services, means that there are no claims data documenting services and costs for increasing numbers of beneficiaries (see Harvey, 1996).