Nonfatal domestic assaults have been undercounted in the NCVS. Of the 4.4 million assault victims identified in the NCVS, only about 355,000 were domestic assault cases. Straus and Gelles (1986) estimated that 2 million people endure severe spousal violence annually (interpreted here as violence with at least minor physical injury), with an additional 2 million subjected to less severe violence. Straus and Gelles' estimates are open to criticism (National Research Council, 1993). A further concern about these estimates is that domestic rape often may be included.

The Straus and Gelles severe violence estimate appears to be better than its critics claim. Under some reasonable assumptions, it proved consistent with NCVS data. Assume the underreporting rate for domestic rape relative to other rape in the NCVS equals the underreporting rate for domestic assault. Under this assumption, physically injured domestic assault victims would number about 1.75 million annually. This estimate should be lower than Straus' and Gelles' estimate; it excludes assaults by unmarried partners, which Straus and Gelles include.

The Straus and Gelles estimates also are consistent with the 1993 Commonwealth Fund survey. In this survey, 3.9 million women reported physical abuse by their partners, with far more reporting verbal or emotional abuse. As a final confirmation of the Straus and Gelles estimates, with the NCVS distribution of medical treatment and victimizations per victim for domestic assault, they imply 1.85 million emergency department visits annually. For women, the total would be 25 percent of injury visits, excluding visits due to motor vehicle crashes--21 percent including these visits (computed from the injury visit count in McCaig, 1994). Four emergency department studies find this percentage ranges from 16 to 30 percent.

Due to their tenuous nature, this study included only the estimates of 2 million spousal victims with severe enough violence to cause injury. This number has been reduced by 355,000 victims of domestic violence estimated in the NCVS data.

Since NCVS undercounts gunshot and knife assaults with physical injury, estimates of gunshot assaults came from nationally representative health care data. The knife assault estimate was computed from the gun assault estimate and the health care data used in the child assault analysis (see Miller and Cohen, 1995).

Drunk driving.Drunk driving is illegal. This study considers it a violent crime when a drunk driver maims or kills innocent victims or damages their property. Some crashes involving a drunk driver, however, would have happened even if the driver had been sober. They might have been due to a mechanical failure, the error of another driver or pedestrian, or an error the drunk driver would have made even when sober. Therefore, simply counting the number of crashes where one driver had a blood alcohol content (BAC) of 0.1 percent or more would overestimate the harm attributable to drunk driving. To estimate attributable crashes, this study relied on the methodology in Levy and Miller (1995). These estimates are somewhat tentative as they were based on a 1962 Michigan study that attempted to differentiate between crashes that were "caused" by drunk driving and those that would have occurred anyway. The U.S. Department of Transportation is currently updating that 1962 study.

Drunk drivers victimize almost 2.9 million innocent people annually. Because they are driving drunk, they kill about 6,150 people, physically injure 500,000, involve another 2.4 million people in sometimes psychologically devastating crashes, and damage 1.15 million vehicles.

Arson.Arson injures relatively few people but can be deadly. The arson counts are from the National Fire Incident Reporting System. They include all arson fires that damaged buildings or vehicles and outdoor fires that resulted in death or physical injury.

Excluded crime.This study focuses on crimes against persons or households. Some incidents--like child neglect--that may not be characterized as crimes in all States and have varying definitions are excluded. Also excluded are crimes against business (theft, fraud, embezzlement, etc.); crimes against the government (regulatory offenses, fraud, tax evasion, etc.); all forms of white collar crime (including fraud); and most "victimless" crimes such as drug offenses, gambling, loan sharking, and prostitution. Here are some of the estimates on the prevalence of these crimes:

Criminal child neglect.This form of child maltreatment is defined by State laws in widely varying ways and is generally a misdemeanor. Several States' neglect statutes, for example, include "environmental neglect," essentially failure to provide shelter. That makes homelessness a crime. Such neglect could be the result of poverty and the inability of the social services system to find the family a home rather than