OUR APPROACH, GOALS, AND PURPOSE
In order to disentangle the causes of violence from those of an omnibus criminal trait, we will explore the literature on violence in a systematic way. In the following chapters we will first provide background, discussing the development of the violent person (Chapter 2), and motivation and context (Chapter 3) to lay out our reasoning for choosing the “good prospects” for predicting violent, over nonviolent, offending.
In subsequent chapters, we will then focus on a series of potential causes of violence. Pilot work for this project indicated that several factors would be “good prospects” for distinguishing between violent and nonviolent offending. This pilot work included a systematic examination of reviews of the literature, a sample of studies in criminology journals, as well as our reading of the literature during the course of our careers. The same “suspects” keep appearing, sometimes within the same studies. The categories are: intelligence, education factors, parental attachment, parental warmth and rejection, abuse and trauma, poverty, community and culture, and substance use. This list does not contradict findings from important longitudinal studies. Farrington (1991) attempted to distinguish violent adult offenders from frequent offenders. In his analysis, factors such as poverty, parenting, abuse, education and intelligence emerged. In Farrington’s (1997) study of the biosocial bases of violence, he finds that intelligence, socioeconomic status, child-rearing, parental violence and school performance are among the main variables related to violence. Most of the factors we have selected appear in Hawkins’ list of risk factors for youth violence (see also Lipsey and Derzon, 1998). They have also appeared in our own analyses of National Youth Survey data, and occasionally other data sets. We had to limit ourselves to a manageable number of risk factors for this book, so we did not cover other good prospects which include: myriad biological factors, interparent conflict, and some psychiatric illnesses.
In each chapter, we will assess evidence related to whether or not the factor under study can help us make our distinction. Each chapter opens with a review of the literature, and is followed by the systematic analysis of findings. In some chapters, we will include substantial review of studies on physical aggression and violence alone, because they are informative, and analogous nonviolent outcomes are not available (for example, in studies of attachment and aggression it is rare that nonviolent outcomes are available for children). For some topics, too, reviews on associations with violence are scant (e.g., parental warmth and rejection), so we first provide evidence of any association with physical aggression and violence, before narrowing our focus to the differential etiology of violence. In all chapters, our analysis will rely most heavily on the close investigation of certain studies that allow us to compare statistical relationships between risk factors and violent and nonviolent outcomes.
The “null hypothesis,” then, is that the coefficients for associations between any risk factor and violent behavior will be the same as the ones for nonviolent crime in direction and magnitude. In studies of offenders, the null hypothesis will be that violent offenders will not have significantly different scores on measures of risk. In studies where a control for nonviolent offending is applied, the null hypothesis will be that the effect of risk on violence will not be statistically significant. Any evidence that allows us to reject these null hypotheses would suggest that general theories are inadequate. In order for us to conclude that a factor remains a “good prospect” for further evaluation as a likely differential predictor of violence, however, we require a pattern of findings that makes it visually evident that this is true.
When possible we will compare significance levels and effect sizes. Each chapter will vary based on the methodologies used in the studies under scrutiny. In some cases, we may use common data sets and display our own comparisons. Rather than provide a formal meta-analysis for each topic, which would be beyond the scope of this volume, we employ vote count analysis followed by a deeper evaluation of the subject matter. In each chapter we provide a sense of the proportion of coefficients reported by authors which are consistent with the differential etiology of violence hypothesis, and, because some studies report a great deal more coefficients than others, and may have an inordinate effect on those effect “counts,” we also provide study-level counts. In those, we will refer to a “preponderance of comparisons,” which means that more than half of the coefficients are in the predicted direction and statistically significant. This is a very simple way to convey to the reader the state of the literature, but we do acknowledge that our choice of cut-point is somewhat arbitrary and conservative.