Child neglect. As mentioned above, although some preliminary estimates of child neglect were made, they were excluded from total crime estimates due to definitional and consistency concerns. Following the same methodology as for child abuse, the researchers estimated the cost per child neglect case and the aggregate cost of child neglect. The estimated total cost of educational neglect is $3 billion, while the annual cost of other forms of child neglect is $12 billion. These are very tentative estimates.

Special breakdown of cost estimates: domestic crime and crimes against children.

Table 6 provides more details on the incidence and cost of domestic crimes against adults�all of which were also in�cluded in the previous tables. Domestic crime against adults accounts for almost 15 percent of total crime costs�$67 billion per year. Table 7 provides more details about child abuse (ages 0 through 17). Children under age 18 suffer at least 4.9 million personal crimes annually. Losses due to these crimes exceed $164 billion, of which about 40 percent results from domestic crime.

Who pays the crime bill?

As table 8 shows, insurers pay $45 billion annually due to crime. That�s $265 per American adult. Government pays $8 billion annually for restorative and emergency services to victims, plus perhaps one-fourth of the $11 billion in health insurance payments.

Crime victims and their families pay the bill for some crimes, while the public largely pays the bill for others. Tax�payers and insurance purchasers cover almost all the tan�gible victim costs of arson and drunk driving. They cover $9 billion of the $19 billion in tangible nonservice costs of lar�ceny, burglary, and motor vehicle theft. They cover few of the tangible expenses of other crimes.

Victims pay about $44 billion of the $57 billion in tangible nonservice expenses for traditional crimes of violence� murder, rape, robbery, assault, and abuse and neglect. Em�ployers pay almost $5 billion because of these crimes, primarily in their health insurance bills. (This estimate ex�cludes sick leave and disability insurance costs other than workers� compensation.) Government bears the remaining costs through lost tax revenues and Medicare and Medicaid payments.

Uncertainty of the estimates and sensitivity analysis

This section explores the uncertainty in the study's estimates but cannot provide confidence intervals. Since this study relies on a wide range of disparate sources of data, the data are not in a format that would support a systematic study of the confidence interval around the estimates. Although the researchers attempted to be conservative, the estimates have a high degree of uncertainty. Some of the key problems are discussed below.

Number of victimizations

Confidence intervals for NCVS. The NCVS is based on a complex sampling design. Although this research team does not have adequate information to be able to compute standard errors for all its derived estimates, the NCVS series=1 victimization estimates can be examined over the 4-year period (for which this study took the average). These standard errors are relatively small, implying that the NCVS estimates are generally within 5-10 percent of the true population. Although standard errors are low, NCVS estimates are only as good as the survey design and respondents' ability to recall and be truthful. For example, these confidence intervals tell nothing about the true underlying population of rape victims if women are reluctant to volunteer information about rape incidents. Moreover, these confidence intervals apply only to the aggregate victimization count estimates, not for the various breakouts used in this research. For example, although there is a high degree of confidence about the number of robbery victims reported in NCVS, there can be less confidence in the estimated proportion of robbery victims who were hospitalized for broken bones.

Series victimization. Few researchers have used the series victimization counts in the NCVS. This study's team carefully analyzed the raw sample data and checked the series victims for outliers and reasonableness. Although a few of the outliers were obvious miscodings (and thrown out of the sample for purposes of analysis), this study found that these observations generally were quite plausible. Ultimately, when presenting national cost estimates, the researchers decided to truncate the few remaining possible outliers at 10 victimizations in a 6-month period.