Operationalization of Violence
We are concerned with the operationalization of violence in many studies. The public is mainly concerned with the type of violence that rises to the occasion of injury and serious harm. In some studies, such as the National Youth Survey, the use of some scales results in the prevalence of violence being higher than the prevalence of minor crimes such as theft, which indicates that the data are capturing minor acts of violence that may be of little concern to society at large or the criminal justice system. We believe it would be quite difficult to predict minor violence, such as fighting between siblings, and we are not interested in doing so. We recommend that the operationalization of violence take this distinction into consideration and that more researchers collect data that capture the quality of violence.
In offender studies, “violence” is usually operationalized with a dummy code: violent offender vs. nonviolent offender. If researchers wish to compare violent to nonviolent offenders, it is important to make assurances that the “nonviolent” offenders are truly nonviolent people. Just because an offender’s current offense is a nonviolent offense does not mean that individual has no history of violence. In a small number of studies reviewed by us, the authors ensure that the comparison is “pure,” but in many cases, this is not true. This may lead to a type II error, missing an effect that does exist.
Another potential problem with offender studies is that people are not chosen at random from the population of violent and nonviolent offenders for arrest, processing and incarceration. For example, prison samples are likely to have more one-time violent offenders than one-time nonviolent offenders, as severity of the offense and criminal history are weighed in the sentencing process. The ages of violent and nonviolent offenders and their races or other characteristics may differ. If these differences are correlated with potential “causes” of violence, then a simple comparison of violent to nonviolent offenders may be confounded.
A third problem arises when the “cause” of violence being tested is more likely to result in one type of arrest over another. In this book, this arises most obviously in the area of drug use and crime. If drug use itself is a crime, and being arrested for drug possession or trafficking counts as a nonviolent crime, then the data will identify a greater proportion of nonviolent offenders as “drug users” than were caused by drug use to commit nonviolent crime. In this way, the data will make it appear that an association between drug use and nonviolent offending exists, but for some or many, that association will not be in the right causal direction.
If this is the case, then the difference in prevalence of drug users among violent and nonviolent offenders will not be as large as it should be, again leading to a type II error. Further, this problem will arise in different measure across places and samples, due to differences in enforcement.