survey will be to avoid double counting with the provider survey, to count volunteer time, and to track compensation for victimization (which affects who pays rather than the total cost) separately from resource costs.

Most wage losses due to crime result from the occasional injury that causes permanent disability. The NCVS questions the same respondent in six time periods. Since permanent disability is rare, the NCVS might collect these rare but important costs for crimes described in prior surveys. NCVS estimates also probably would improve if respondents reported medical and mental health care costs, rather than just medical costs (which some respondents may interpret as including psychiatric care).

This study dealt only peripherally with costs to third-party victims (e.g., fear of crime among the general population or crime victims' families, post-traumatic stress disorders of emergency service personnel). These costs merit focused study. This study also barely touched on crime-induced suicide, crime-induced criminality, or AIDS transmission through rape. These events could potentially be large contributors to victimization costs. Conversely, explicit suicide costs easily could double count since both pain and suffering and mental health treatment already are incorporated in this study's framework.

Lost quality of life is the largest cost component. It also is the least easily measured. Research should continue on measurement methods. A National Highway Traffic Safety Administration project (MacKenzie et al., 1994) has produced new physician estimates of impairment following injury that will improve the willingness-to-pay approach. A regression analysis of published values of statistical life estimates could better separate out modeling effects, potentially yielding more credible fatal risk reduction values.

In a rapidly growing number of States, cause-coded hospital discharge data represent a rich source for data on the most serious injuries resulting from crime. In the States analyzed for this study, the health care system appeared to capture data on many more victimizations than the NCVS. These data lack details about crime context, but that information might be added by linking them with police crime reports. A demonstration is needed to assess the value of linked data and the difficulty/cost of linkage.

Applying the cost estimates. Criminologists and public policy researchers are beginning to use crime cost estimates in many new ways to help shed light on the desirability of various policy options. A next step would be to compute cost estimates for defined populations--the elderly, youthful offenders, women, low-income people, minorities, and police-reported victimizations. Although these breakouts might prove useful, care must be taken in deriving them from existing data sets. For example, although it might be reasonable to estimate the percentage of victims who are Hispanic and their age distribution, at some point the sample cells become too low to make meaningful estimates. It might not be reasonable to estimate the number of Hispanic females from the ages of 18 to 24 who suffer from broken bones due to physical nondomestic assaults, for example.

Costs per police-reported crime are higher than costs of the average crime. These costs can be derived with this study's costing system and data. For analyses of police resource allocation, they would be more appropriate than this study's cost estimates. Conversely, the costs here are most appropriate to use in assessing the wisdom of early offender release and diversion programs.

The recent health care reform debate emphasized proven interventions, ones that had benefits exceeding their costs. Data on proven violence prevention approaches were notably lacking in the debate. These costs provide a basis for such analyses.