What We Don't Know About Violence
The impetus for this book was a repeated observation that, in spite of many decades and thousands of studies, we simply do not know as much as we should about the causes of violence. Most lay persons probably assume that criminological science has uncovered the dark secrets of the criminal mastermind. In particular, because violent crime is so egregious, they probably believe that much is known about the causes of violent acts and the brutality of certain, chronically violent individuals. In this book, we will argue that the body of scholarship on violence is woefully incomplete and asystematic and has not clearly identified factors that predict violent as opposed to nonviolent criminality. To be sure, we know many correlates of violence, but we cannot say with any certainty which factors would help us distinguish a violent offender from a nonviolent one.
The problem is not lack of research; there are plenty of studies. The problem stems from a combination of theory and method. It may surprise the lay reader that seminal criminological theories do not usually distinguish between even broad categories of offending like violence and theft. With few exceptions, authors of the classics (social disorganization theory, classical strain theory) attempted no prediction of different crime types, and the more recently published theories (general strain theory, “the general theory” [a.k.a. “self-control theory”], rational choice theory, control theory) also choose not to do so. Even integrated models such as those proposed by Elliott, Huizinga, and Ageton (1985); Thornberry (1996); and Vila (1994) add complexity and nuance not so much by differentiating between crime types but by combining explanatory variables.
Studies that test general theories need only employ measures of criminality or delinquency (at the individual level) or “crime rates” (at the aggregate level). These indices often incorporate a variety of criminal behaviors, which usually include some measure of theft and sometimes include measures of violence. Selfreport studies can include a wide range of delinquent behaviors, such as tobacco smoking and underage drinking, which most people view as minor. Because the number of nonviolent offenses tends to overwhelm the less frequent serious violent acts when summed together, a nominal measure of “criminality” or “delinquency” often measures little more than nonviolent theft and status offenses. Thus, studies that uncover correlations between some construct of interest, such as “self control” and a general measure of delinquency cannot tell us whether individuals who have low self control are more likely to be violent. This lack of specificity impedes our ability to differentiate serious, dangerous offenders from petty delinquents.
One reason for this broad emphasis on total criminality stems from the fact that twentieth-century criminology was characterized by a persistent search for general theories of crime, and the published literature has strongly emphasized testing those theories. General theories are more parsimonious than theories of different crime types, and are, therefore, more desirable by traditional scientific standards. Early criminologists may have been anxious to associate the new field with more highly regarded “hard” sciences, adopting many traditions from the natural sciences, including the emphasis on testing hypotheses based on formal theories. Thus, criminological theory and research have traditionally emphasized a unidimensional “criminality” that makes no distinction between various types of crime or crime severity.
Today, many criminologists accept that there is a unidimensional construct, called variously criminality, deviance, delinquency, conduct disorder, antisociality, and sometimes aggression, which is comprised of criminal activities such as violence but also includes a variety of behaviors including theft, property damage, and drug use. This orientation is not unique to the field of criminology. Even in the field of psychology, there are those who focus on “antisocial” characteristics in their theories and research. The Diagnostic and Statistical Manual of Mental Disorders (DSM-5) (American Psychiatric Association, 2013) lists several diagnoses indicative of general conduct problems such as conduct disorder, oppositional defiant disorder, and antisocial personality disorder.
Bolstering this belief, some prominent and recent empirical reports have concluded that violent offending is simply part of a larger antisocial tendency. Piquero, Farrington, and Blumstein (2007) conclude that “... violent offenders are simply frequent offenders who happen to commit a violent offense during their career” (p. 80). Most studies of serious offenders find that they tend to start offending early, that they commit many types of crimes, and that frequent offending is associated with violent offending. Thus, many criminologists have given up on the idea of distinguishing violent offenders in favor of focusing on chronic or persistent offenders.
There are some exceptions. Felson (2009) has brought his social psychological training to our field and has articulated a “dual conceptualization” of violent crime. He notes that violence is “rule breaking” and should have much in common with other forms of crime, but also "harm-doing," and, as such, requires theories of aggression to explain it. Violent offenders, in his description, are distinct because they want to harm others or, at minimum, are indifferent to doing so. Felson notes that while violent and nonviolent offending are likely to share some common causes (because of the fact that they are both risky and “deviant”), they “should have some different causes as well” (Felson & Osgood, 2008, p. 167).
Of course, many criminologists would point out that there are many researchers interested in violent crime per se, and there are literally thousands of studies related to physically aggressive and violent behavior. Using a measure of violence as the outcome would seem to address our earlier concerns with general measures of criminality. While findings from this type of study are certainly useful, we regret that this approach also poses problems. It is the case that violent offenders tend to commit many types of offenses, and they commit them with regularity. A dependent variable that measures violence, then, is likely to be highly correlated with a measure of general deviant or antisocial behavior. Independent variables in the model that are associated with general antisociality are likely to have a statistically significant association with the violence measure because of the overlap between violence and general deviance. Thus, the predictors of the violent outcome may, in fact, be associated with that general antisocial tendency and not the violent behavior in particular. Imagine that witnessing a parent commit theft causes theft. Imagine that the most prolific thieves also commit violent crime. In such a case, witnessing a parent commit theft would be spuriously correlated with violent crime. We cannot be sure that the correlates of violence discovered in these exercises are predicting violence per se or simply general deviant tendencies.
This sounds like a technical detail, but it is an important one. For example, consider the control for socioeconomic status in studies of single-parent families and delinquency. We employ this control because it could be the case that children from single-parent families may be more likely to commit delinquency because their income is low (not actually because their parent is single) (see Figure 1.1). The reasoning for a control for nonviolent offending is quite similar. In studies of
Figure 1.1 Analogy: An Example of Confounding Factors.
violent behavior and its relation to causative factors such as harsh parenting, we must consider that harsh parenting might be related to general delinquency, and since general delinquency is related to violence, we should be controlling for it in our statistical models. Thus, if we simply look at relationships between risk factors and some violent outcome, we have a model specification problem: a missing variable. Thus, the enormous volume of criminological literature devoted to the causes of violence may or may not teach us something important about distinguishing violent offenders from nonviolent ones.
Figure 1.2 provides a broader visual for the greater problem. Any independent variable of interest, be it harsh parenting, academic achievement, drug abuse, and so on, is expected to have a significant effect on violence. But as the figure shows, it is likely in almost all cases that the variable of interest also has an effect on other forms of criminality, and we know that other forms of criminality are correlated with violence. Studies that look only at general criminality cannot tell us if the direct path to violence is significant; studies only using violence as an outcome cannot tell us if that path would be significant if general criminality were taken into account. Deane, Armstrong, and Felson (2005) capture the crux of the problem in their comment that “The aggregation of criminal behaviors in measuring the dependent variable poses a ... challenge. Any attempt to decompose offending writ large between offense types or blocks of offense types must account for their shared variation, because criminal behaviors are correlated” (p. 956).
One remedy would be to control for general crime rates or a general measure of delinquency in the multivariate model, but we have not seen any studies which
Figure 1.2 Proposed Association Between Various Potential Causes of Violence, General Criminality, and Violent Behavior.
employ this method because few authors are interested in this specific test. We suspect that this approach is why lists of risk factors or correlates of violence read like lists of risk factors for any other type of crime. Prominent reviews of “risk factors for violence” have included factors such as gender, age, ethnicity, school adjustment and academic failure, family background, convictions for nonviolent crime, socioeconomic status, impulsivity, deviant attitudes, family management and discipline, delinquent siblings and peers, and community poverty (Chaiken, Chaiken & Rhodes, 1994; Farrington, Loeber, & Jolliffe, 2008; Hawkins, Herrenkohl, Farrington, Brewer, Catalano, & Harachi, 1998; Lipsey & Derzon, 1998). These would also be on any list of risk factors for delinquency more generally. Literature reviews also include some factors that the trained eye might see as differentially associated with violence, such as violent parents, parent attitudes favorable to violence, prenatal trauma and pregnancy complications, and abuse victimization, but the question remains: how can we know which risk factors are useful in predicting violent behavior? Thus, while this line of research is instructive, and we are certainly grateful for it, we are left with many unanswered questions.