There are some major methodological difficulties in studying this issue. Because violent offenders “... frequently engage in a range of violent and other types of antisocial acts” (Chaiken et al., 1994, p. 218; Moffitt, 2006) it is difficult to disentangle the effects of certain risk factors on violence as compared to nonviolent offending. A report by the Office of Juvenile Justice and Delinquency Prevention states that chronic offenders commit more than half of all serious crimes by juveniles and that “the vast majority” of chronic offenders are serious violent offenders (OJJDP, 1998). However, this leaves at least half of violent crimes to be explained by something other than the causes of chronic offending. Research on child conduct problems often distinguishes between an overtly aggressive and an antisocial but nonviolent “covert” path or group of subjects (e.g., Stanger, Achenbach, & Verhulst, 1997). Stanger et al. (1997) point out that while there are significant correlations between the two sets of behaviors, and together they comprise a higher order externalizing factor, and while some children are categorized as both, many children are categorized as one or the other: overtly aggressive or nonviolently antisocial. Nonetheless, predictors of chronic offending are likely to overlap to a significant extent with predictors of violent offending, and it is a statistical challenge to ensure that the predictors we uncover for violent offending are not simply predictors of frequent delinquency.
Another problem is that there are different kinds of violent behavior. Pulkkinen (1987) discusses offensive versus defensive aggression. Some children bully others, but some use violence to defend themselves and do not attack without reason. There are also distinctions in theory between proactive and reactive aggression, direct and indirect aggression, hostile and instrumental aggression (Farrington, 1991). Although many authors do discuss these distinctions, in practice, violent acts are typically combined without differentiating types of violence. This may result in a loss of power to detect effects. If, for example, a risk factor is associated with offensive bullying, but not with defending oneself, and if there are many instances of defensive aggression in the data set, we might not see a significant association between the risk factor in question and an overall measure of aggression. Of course, we are more concerned with the former than the latter, but our reliance in this volume on published work means that we do not have specific information needed to make such distinctions.
Violent behavior is reportedly intermittent. Most violent people do not commit violent acts all the time, not even regularly-not even every year (Huizinga, Weiher, Espiritu, & Esbensen, 2003; Piquero, 2004). Violent behavior also gets punished severely. A person arrested for a violent crime is more likely than a property offender to be incarcerated, interrupting longitudinal measurement of his “natural” behavior. On the other hand, identification of nonviolent offenders who are truly nonviolent is not perfect, especially since they are often assigned their classification based on their instant offense, rather than their complete criminal record. These two realities increase the potential for error in the measurement of violence and the identification of “violent” and “nonviolent” groups. This has greater implications for the individual studies than for our review overall, but we will return to this issue in the coming chapters.
Relatedly and disturbingly, another challenge for measurement and method is the fact that some violence can be incidental. Miethe and McCorkle (1998) point out that perhaps a third of homicides occur during commission of another felony. They point out that violent crimes are “accidental” in the sense that an otherwise nonviolent burglar may carry a gun and, out of fear, shoot someone in order to get away. Lumping this type of offender with a set of violent offenders and comparing the group to nonviolent offenders may weaken our ability to measure differences between truly violent and nonviolent people. Worse, it is common practice for co-offenders to be charged with a violent offense if they were involved in a crime; all the gang members sitting in a car may be charged with a drive-by shooting when only one of them actually pulled the trigger. Due to this practice, studies using arrest or conviction for violent felonies as indicators of violent behavior may actually categorize some individuals who have not committed a violent act along with those who have. The statistical upshot is a “muddier” group, and smaller differences between a group of supposed violent offenders (which includes some nonviolent individuals) and a group of nonviolent offenders (which likely includes some undetected violent ones). This leads to weak statistical power and the increased likelihood of making a type II error (missing an effect).