Understanding and Preventing Violence: Volume 1 (1993)

Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

EXPANDING THE LIMITS OF UNDERSTANDING AND CONTROL 291 7 Expanding the Limits of Understanding and Control We begin this section by considering the effectiveness of one response to violent crime—incarceration of convicted violent offenders. We then present a framework for broadening perspectives on approaches to prevention. APPROACHES TO CONTROLLING VIOLENCE The Role of Incarceration Incarceration may reduce violent crime through two mechanisms, deterrence and incapacitation. Deterrence operates because incarceration provides a disincentive intended to discourage violent crime.1 In theory, the threat of incarceration deters violent crime by reducing the number of violent offenders or the frequency or seriousness of their violent crimes. In contrast to deterrence, which theoretically alters behavior, incapacitation reduces violent crimes simply by physically isolating persons who engage in violence from the general community. The existence of deterrence and incapacitation effects is not at issue. If incarceration were eliminated entirely, we have no doubt that violent crime would increase (Blumstein et al., 1978). But there is substantial uncertainty about the relevant policy question: What are the marginal effects of changes in either the chance of incarceration per violent crime or the average time served by 

EXPANDING THE LIMITS OF UNDERSTANDING AND CONTROL 292 those incarcerated? These effects cannot be observed directly, because they are measured in terms of potential crimes that did not occur. Rather, their magnitudes must be estimated from data on trends in violent crime and in incarceration levels. Recent Experience The unprecedented increase in prison population since 1975 offers a case study for analyzing the effects of changes in the prison population on levels of violent crime. However, the results remain tentative because the accuracy of the estimates depends on both the accuracy of the data and the validity of assumptions that underlie the analyses. Between 1975 and 1989, the inmate population nearly tripled, while reported annual violent crime levels varied around the level of about 2.9 million. This experience raises an important set of policy issues about how prison population and violent crime levels are related. To examine them the panel commissioned an analysis by Cohen and Canela-Cacho (Volume 4) and also benefited from an analysis by Langan (1991).2 Some pieces of the puzzle can be understood more precisely than others, but one implication of the 15-year experience seems inescapable: if tripling the average length of incarceration per crime had a strong deterrent effect, then violent crime rates should have declined in the absence of other relevant changes. While rates declined during the early 1980s, they generally rose after 1985, suggesting that changes in other factors, including some of those discussed in Part II, may have been causing an increase in potential crimes. Two potential influences on prison population can be discounted. The increases cannot have been due to increases in the probability of arrest per violent crime, because that probability remained essentially constant throughout the period. Similarly, growth in the number of young adults in the aftermath of the baby boom could not by itself have increased prison population (or violent crime) levels by much more than 24 percent, the increase in the U.S. adult population. One relevant factor clearly did change: average prison time served per violent crime approximately tripled between 1975 and 1989, returning to the levels of the 1950s. How did this increase occur? Experience varied somewhat by crime type and state, but the data point to general increases in both the average time served if incarcerated and the chance of imprisonment if arrested. This is consistent with the proliferation of statutes mandating minimum 

EXPANDING THE LIMITS OF UNDERSTANDING AND CONTROL 293 sentences, especially for offenses involving violence or firearms. An increase in average prison time served per crime may prevent crimes either through incapacitation or deterrence. Estimates of incapacitation effects are necessarily imprecise and uncertain because they depend on many specific assumptions. However, under a variety of alternative scenarios (Cohen and Canela-Cacho, Volume 4), the estimated incapacitative effect of tripling the average time served per violent crime was fairly small—preventing on the order of 10 to 15 percent of the crimes that potentially would have been committed otherwise. Two facts seem to explain the limited incapacitative effect of the increase in prison time. First, while the average annual frequency of violent crimes per offender is fairly small (e.g., 5 to 10 robberies per year), a small fraction of offenders commits hundreds of crimes per year. But even before the increase in average prison time per crime, these high-frequency offenders were spending much of their criminal careers in prison—both because their high crime rates presented more opportunities to be arrested and incarcerated and because, under otherwise comparable circumstances, convicted offenders with extensive prior records tend to receive longer prison sentences. Second, incapacitation is subject to diminishing marginal returns, because most criminal careers are fairly short. Successive increases in the per-crime chance of incarceration bring into prison less serious offenders, thus preventing fewer crimes per inmate-year. Successive increases in average time served by those incarcerated allocate larger shares of prison space to offenders who would have stopped committing crimes even if they had been free in the community. The tripling of average prison time served per crime also produces a potential marginal deterrence effect: the additional threat of incarceration presumably deters more people from becoming violent offenders and deters some violent offenders from committing more crimes. Methodological problems prevent sound estimates of deterrence effects from being made (Blumstein et al., 1978); however, the flat trend in violent crimes between 1975 and 1989 in the face of a tripling in the average prison time served per crime is not compatible with any substantial deterrence effect—unless, for reasons unrelated to incarceration, the potential violent crime rate would have substantially increased during the period. Otherwise, the tripling of time served per crime would have been followed by decreases in violent crime levels, which did not 

EXPANDING THE LIMITS OF UNDERSTANDING AND CONTROL 294 occur according to official statistics. (These conclusions are explained in more detail in Cohen and Canela-Cacho, Volume 4.) Alternative Incarceration Policies On the heels of this unprecedented increase in prison population, there is active discussion of pursuing crime control through further increases in the certainty of incarceration following a violent crime, in the term of incarceration, or both (Methvin, 1991). Through incapacitation, increasing the chance of incarceration reduces crime levels more efficiently than does increasing the average time served.3 The different incapacitative effects occur because, when the probability of incarceration following each crime is higher, offenders tend to be incapacitated earlier in their careers. But the longer that time served becomes in relation to career length, the more likely it is that offenders will end their careers while they are still incarcerated and would therefore commit no more crimes in the community even if they were released. For example, projecting the incapacitation effects of 50 percent increases in certainty and in average time served from their base levels for robbery in the late 1970s, Cohen and Canela-Cacho (Volume 4) estimate that a 50 percent increase in the chance of incarceration prevents twice as much crime as a 50 percent increase in the average time served by those incarcerated. Deterrence theory also suggests that increasing certainty reduces crime more efficiently than increasing severity, but for different reasons. Theoretically, the disincentive effect of increasing the probability of incarceration begins at the time the offender enters prison. The disincentive effect of adding additional expected time served begins later, after the preexisting average time served is reached, and is therefore theoretically reduced because of “time discounting"—events farther in the future receive less weight in the offender's decision making. One might expect that weakening restrictions on the use of improperly obtained evidence would increase the certainty of incarceration for each crime. The effect on the probability of arrest per violent crime is unclear, since it is not known how frequently such arrests are not made solely because of officers' concerns over dismissal because evidence was obtained improperly. But because problems of due process virtually never arise as a reason for declinations or dismissals in violent offense cases in large urban prosecutors' offices (Boland et al., 1992), loosening restrictions on evidence should not be expected to perceptibly increase the certainty of incarceration given arrest for a violent crime. 

EXPANDING THE LIMITS OF UNDERSTANDING AND CONTROL 295 A Broader Perspective We have noted an important finding: recent trends in crime and criminal justice policy responses have resulted in the near-tripling of prison populations between 1975 and 1989—with no apparent decrease in levels of violent crime. This strongly suggests that other factors that tend to increase violent crime levels —some of which can be changed by governments or individuals—are at work. Taking a broader perspective shifts attention in several directions. One is to understand and ultimately modify the psychosocial and biomedical processes through which individuals develop potentials for violent behavior. Others are to consider the settings—communities, markets, gangs, families, places, and encounters—in which violent events occur, looking for ways to intervene. In this report we have looked at violent events from all these perspectives. Too often, theories or interventions grounded in different perspectives are treated as if they were mutually exclusive. This is unfortunate, because each one contributes significant facets to the understanding and control of violence. In part this is because the concept of violence lumps together many different types of behavior and a great diversity of events. As this volume demonstrates, for every type of violent event, there are many risk factors; consequently, there are opportunities for intervention at many different points. Situations differ in terms of triggering elements for violent events; settings differ in terms of the situations that occur in them and the violence potentials of the people who congregate in them. Exploring all these relationships expands the range of promising but untested preventive opportunities—particularly because of the diversity of violent behaviors and events that has been documented in previous chapters. Except perhaps when self-preservation requires it, no situation—no dispute, no opportunity for gain or power, no degree of anger or sexual arousal—provokes violent behavior in a majority of people. Some predisposing individual characteristic must be present in most events. Nevertheless, some situations and encounters provoke violent behavior from more people than do others, and identifying such situational hazards for violence is a first step in eliminating them. Similarly, very few predisposing individual characteristics provoke violent outbursts regardless of the situation. Nevertheless, identifying and modifying predisposing individual characteristics for some violent behavior are a step toward preventing some violent events. 

EXPANDING THE LIMITS OF UNDERSTANDING AND CONTROL 296 Predisposing factors for violence vary in terms of their proximity to the violent event. The development of some relevant individual characteristics, such as a temperamental disposition that favors high thresholds for anxiety or fear, begins before birth or in early childhood. But some individuals' potentials for violence are raised by experiences in adulthood, such as the accumulation of frustration or anger. Similarly, some societal conditions, such as widespread acceptance of violence as an appropriate dispute resolution technique, may take centuries to develop; these are sometimes dismissed as immutable root causes. Other societal risk factors, such as the concentration of poverty in a particular geographic area, may develop over only a few years. A MATRIX FOR UNDERSTANDING AND CONTROL To give ourselves a system for thinking about potential interventions to control violence, we found it useful to classify risk factors for violent events using the matrix in Table 7-1. The columns in the matrix classify risk factors by their proximity to a violent act: from left to right, as predisposing factors and processes, situational elements. or triggering events. These categories span four levels of social and individual description that are commonly used in studies of violence. Studies of social factors are concerned with macrosocial institutions, such as societies and communities, as well as with microsocial encounters among pairs or small groups either as individuals or as members of gangs or families. At the individual level, studies are concerned with both psychosocial and biological components of behavior. Of course, the factors in each cell of Table 7-1 are only illustrative examples. Far more are discussed in Chapter 3. We do not assume that any single level is more fundamental than the others in explaining a particular type of violence. Rather, as Chapter 3 explains, violent events and community violence levels arise out of interactions across the levels, and these interactive processes differ from one type of violence to another. Because these interactions are so poorly understood, modifying risk factors and observing responses is the most promising method of simultaneously accumulating understanding while building a capacity for violence prevention. Research into causes of violence and evaluations of preventive interventions are usually compartmentalized. Academic disciplines concerned with violent behavior tend to focus on one level of description at a time. Public agencies tend to modify only one 

EXPANDING THE LIMITS OF UNDERSTANDING AND CONTROL 297 

EXPANDING THE LIMITS OF UNDERSTANDING AND CONTROL 298 cluster of risk factors at a time—police agencies rarely offer professional psychological counseling; juvenile courts rarely offer preventive interventions in early childhood. Although some specialization is essential, it creates a fundamental difficulty for understanding and controlling violence. We believe that interactions and feedback loops across the cells of the matrix have important effects on violence levels, but with existing research we cannot be sure. Do drug markets attract people with high potentials for violence, do characteristics of drug markets raise the violence potentials of buyers and sellers, or both? Did a theoretically promising cognitive-behavioral preventive intervention fail to change some children's behavior because of neurological deficits that kept them from comprehending the intervention, or because the proffered rewards had no status in the local neighborhood culture? Because of specialization, when effects such as these are recognized at all, they usually show up in research reports as "untested rival hypotheses" and in evaluations as "unexpected consequences." Better understanding requires more research that follows up such hypotheses even when they cross academic disciplinary boundaries, and unexpected consequences even when they lead beyond an agency's traditional array of interventions. Better understanding and better violence control both require more evaluations that compare different agencies' intervention techniques, using outcome measures at multiple levels of description. We take up this matter later in the chapter. The following pages illustrate the processes and risk factors arrayed in the matrix. A violent event requires the conjunction of a person with some (high or low) predisposing potential for violent behavior, a situation with elements that create some risk of violent events, and usually a triggering event. Development of an individual's potential for violence may have begun before birth: perhaps with conception involving an alcoholic father, or through abnormal prenatal neural development. It may have begun during early childhood in a violent household, or through school failure, or through frequent exposure to violence in the neighborhood or from the media. A hazardous situation for violence could involve a dispute, perhaps aggravated by a miscommunication in a bar because of loud background noise, which was misinterpreted as an insult because of intoxication and escalated because participants were afraid of losing face in bystanders' eyes. The surrounding community could be gang turf, the site of illegal drug or gun markets, or a neighborhood 

EXPANDING THE LIMITS OF UNDERSTANDING AND CONTROL 299 where large numbers of unsupervised teenagers reside. It may be the scene of recent aggravating events such as police brutality, or of frequent brawls between members of different ethnic groups. The neighborhood may be experiencing social disruption as stable families move to the suburbs, as businesses close, and as public services decline. Risk factors such as these make differential contributions to different categories of violent events and suggest different interventions for control. Obviously, not all of them contribute to any single event. A major problem in understanding violence is to describe the probability distributions of predisposing factors, situational elements, and triggering events at the biological, psychosocial, microsocial, and macrosocial levels. The problem in controlling violence is to choose among possible interventions. This choice, indeed the decision whether to intervene at all in a triggering event, a hazardous situation, or a predisposing process, properly depends on both understanding and values. The knowledge requirements include predictive accuracy in assessing the risk of violence, evidence that the intervention prevents violence, and evidence about side effects. One barrier to achieving predictive accuracy is that many risk factors have been identified in relation to aggressive behavior, not violent behavior, and the relationships between the two are not well understood. Another barrier is the aforementioned feedback loops and interactions, which are poorly understood and may cross the boundaries of the cells in the matrix—risk factors such as family size and maternal drug abuse during pregnancy may be stronger predictors in poor neighborhoods than in wealthy neighborhoods, for example. Consequently, the panel found scientific bases for very few unambiguous policy recommendations. However, we found a number of promising candidate strategies. The ones that we believe would also satisfy public criteria of fairness are recommended in Chapter 8 for development, evaluation, and refinement in violence problem-solving initiatives. Because investigations of violent events commonly discover triggering events first, then contributing situational elements, then (if at all) predisposing processes, we discuss them below in that order. Triggering Events Describing an event that triggers violence does not provide a basis for inferring motivation for the violent behavior. But in the 

EXPANDING THE LIMITS OF UNDERSTANDING AND CONTROL 300 American legal system, elements of the triggering event determine which authorities, if any, are empowered to intervene in the event or punish those involved. Analyses of events may also suggest useful clues to how high-risk events might be prevented in the future. Triggering events associated with violence between spouses and between competing crack dealers are very different, and many interventions intended to prevent one type of violence have little relevance for the other. As shown in Table 7-1, events that trigger violent behavior can be observed at all four levels of description. At the biological level, for example, an error in processing social signals can trigger a violent reaction that would not have occurred had the signal been correctly processed. Brain activity during the aura that precedes a seizure or during the interictal period between seizures can trigger violent behavior. At the psychosocial level, a violent act may arise out of heightened impulsivity or, in the case of premeditation, from the recognition of an opportunity. Most events that trigger violence, however, reflect none of those, but rather a sequence of information processing operations, described by Dodge (1986) as encoding, interpretation, response search, response decision, and enactment. At the microsocial level, triggering events occur in the dynamic communications with others that generate a continuing flow of information to be processed. At the macrosocial level, political assassinations, killings by police officers, and controversial jury verdicts are examples of catalytic social events that have triggered major episodes of violence. Awareness of all four levels of triggering events suggests a range of preventive strategies. Examples include pharmacological interventions, teaching children to process information in ways that lead them to choose nonviolent responses, and preventive responses immediately following catalytic social events—the difficulty lies in developing tactics and demonstrating their effectiveness. Situational Elements In most violent events, contributing situational elements are most visible in the microsocial encounter that precedes the event. These elements include the dynamics of communications among participants, such as disputes, threats and counterthreats, exchanges of insults, robbery and resistance, and the urgings of bystanders. Both the nature and the interpretation of these exchanges may be conditioned by preexisting social relationships among participants: an intimate relationship, a power or status hierarchy (e.g., guard/inmate, 

EXPANDING THE LIMITS OF UNDERSTANDING AND CONTROL 301 weaker/stronger, armed/unarmed, shielded/vulnerable), or a culturally defined relationship (e.g., membership in rival gangs, or in different ethnic groups, or in a "hate group" and the category against which it is prejudiced). The risk of violence in an exchange may be increased by communication impairments that are due, for example, to language barriers, to culturally defined connotations of words, to different definitions of insults, or to the influence of alcohol or other drugs. But the outcome of the microsocial encounter is not independent of participants' psychological and biological state at the time. Accumulated or "flash" emotions—anger, frustration, stress, fear, for example—all contribute. Premeditation, sexual arousal, or urges to "prove one's manhood" sometimes play roles. Though these psychological states must usually be inferred by observing external behavior, each of them has an associated neurological profile, which could in principle be measured in terms of electrical, hormonal, and neurochemical activity in different regions in the brain. If these processes are altered by consumption of alcohol, other drugs, toxic substances, or injury, the resulting behavioral changes may alter the risk of violence. At the macrosocial level, interactions between characteristics of places and people create situations, elements of which influence the probability and consequences of violence. The routine activities of the people involved—whether drug dealers and customers, or cab drivers and convenience store clerks working alone at night—may create occasions for violence. But whether violence actually occurs depends also on physical characteristics of the place (e.g., lighting, alarms, shields), and on the proximity of responsible monitors (e.g., police, respected "old heads" in the neighborhood, passersby, guard dogs). If violence does occur, its consequences will depend on the nature of accessible weapons. The prospects for limiting damage often depend on the promptness and quality of emergency medical services that respond. Because situational elements from all levels contribute to the outcome, the possibility exists that even without full causal understanding, altering one link in a chain of events might have prevented a violent event or prevented an assault from becoming a homicide. Predisposing Factors Beyond the immediate situation, factors and processes at all four levels create predisposing conditions for violence. We discuss 

EXPANDING THE LIMITS OF UNDERSTANDING AND CONTROL 302 first the psychological and biological processes that interact to determine an individual's potentials for different violent behaviors, and second the social processes that make some environments more likely than others to become locations of violence. Biological and Psychosocial Processes Modern perspectives see some violent behavior in part as the result of complex, long-term developmental processes that shape the personality and character of the violent actor. Basic descriptions of these processes often emphasize childhood learning of individual behavior: children are assumed to learn "scripts" of what events are about to occur, how one should react, and what the outcome will be. These scripts guide behavior and are more likely to elicit violent behavior if they include violent acts that are rewarded rather than punished. More recent refinements have expanded on this basic description in at least three ways. First, not every violent act can be traced back to scripts; certain situations, which are beyond most people's range of experience, would provoke violent behavior in nearly everyone. Second, while the existence of scripts is nearly universal, their content will depend on chance events during childhood and adolescence. The family, peer groups, and the macrosocial environment all influence an individual's range of experience. Examples of potentially relevant developmental events include the amount of violence seen in the home, through the news and entertainment media, or in the neighborhood; the economic and social rewards to those who commit violence, those who seek to prevent it, and those who become its victims; and cultural practices that tend to instill or disable empathy for victims. Third, the explanation is becoming enriched through attention to biological influences. A person's nervous system is constantly shaping and being shaped by its environs. Each individual's nervous system responds to a given event in a slightly different way. Information processing in the nervous system is based on excitatory and inhibitory neural signals in sensory and motor pathways that interact with arousing, affective, and memory processes that involve chemical neurotransmitters and hormones. For example, alcohol's action on GABAA or serotonin receptors may differ in violence-prone individuals due to genetic or maturational factors or due to past exposure to environmental or pharmacological stressors. In short, the processes that develop an individual's potentials for different violent behaviors are not understandable exclusively 

EXPANDING THE LIMITS OF UNDERSTANDING AND CONTROL 303 in any of the following terms: learning, the macrosocial environment, or a biologically determined temperament. Yet each is relevant. Of the many people with the capacity for violent behavior, the few who display it are more likely to have grown up in families and neighborhoods in which it was easy to learn that violence was not negatively sanctioned (or was rewarded), or in which opportunities for violent expressions of frustration occurred frequently. Of the many individuals who live in such high-risk environments, those with certain temperaments will be more likely to act on impulses toward violent behavior, and especially to develop patterns of chronic violent behavior. Regardless of the roles played by learning, the environment, or temperament, when such people commit violent acts, they enter a category that is sometimes assigned such labels as dangerous offenders. Macrosocial and Microsocial Processes There is tremendous variation in violence rates over short and long time periods and across macrosocial units—nations, cities, communities within cities, and addresses within communities. Variations over time are too large and transient to be explicable simply by genetic processes over time. Variations across areas are too large to be attributable solely to the congregation of individuals with certain temperaments in certain places. Logically, predisposing factors at the social level must be important determinants of violence levels. The difficulty is in determining what the relevant factors are—a search that is made more difficult by the fact that so many candidates tend to occur together in communities that are plagued by high violence rates. In communities, one such candidate is the structure of nonviolent channels for acquiring money and power, for dealing with anger, and for achieving sexual gratification. A general lack of legitimate economic opportunity; a widespread, persistent feeling of powerlessness; the decline of social capital that would otherwise pass nonviolent norms from one generation to the next; the lack of constructive recreational activities; a high prevalence of disrupted intimate relationships—all these are symptoms of a macrosocial structure that fails to offer legitimate opportunities to achieve the purposes for which violence is used. Cultural and subcultural norms, such as acceptance of behavior degrading to women or alienation among members of some ethnic categories against social control agencies managed by members of others, may reduce inhibitions against violence. Cultural practices 

EXPANDING THE LIMITS OF UNDERSTANDING AND CONTROL 304 within a subculture (e.g., trading insults, termed “jonin'" or "playing the dozens" by black teen-age males) sometimes create encounters that culminate in violence. Spike Lee's movie Do the Right Thing provides a fictional but plausible account of how conflicts between ethnic subcultures may create a backdrop against which a small altercation triggers a violent confrontation. Recent research discussed in Chapter 3 focuses attention on what has been called the concentration of poverty—and its consequences such as single-parent households, unsupervised unemployed teenagers, and crowded housing—and its relationship to violence. Enhanced exposure to potentially addictive and expensive drugs without a means of earning the money to buy them may be a potential macrosocial risk factor. And, as discussed in Chapter 6, widespread firearm availability may not affect violence rates but clearly increases the lethal effects of violent events. Interactions and Interventions As Chapter 3 illustrates, the difficulty—for both understanding and control —is that predisposing processes interact across all four levels to determine violence levels. A family's place in the economic opportunity structure influences both the risk that a child's neural system will be damaged through premature birth or head injury and the family's access to medical care that might fix the damage. A young adult male's position in the economic structure influences his chance of being chosen as a marriage partner—a fact that influences the risk of violent behavior by both the young adult male and the children he fathers. Neurological problems, by limiting a child's cognitive or communications capacity, make early school failure more likely—perhaps setting the stage for frustration, exclusion from upwardly mobile peer groups, and a low position in the adult opportunity structure. How frequently individuals engage in acts of violence depends not only on their underlying potentials but also on how frequently they encounter occasions for violence, on the social and legal rewards and punishments for violent behavior, and on interactions between occasions for violence and individual potentials. Whether temperament that predisposes a person to violent behavior is actualized at a particular point in time depends not only on physiology but also on recent exposure to stress, on the chronicity and intensity of recent intoxication, and on other conditions that are influenced by the surrounding environment. This raises the possibility that so-called dangerous offenders 

EXPANDING THE LIMITS OF UNDERSTANDING AND CONTROL 305 emerge from social structures that generate frequent occasions for violence as well as from high individual potentials toward violent behavior. Moreover, it is likely that their experiences—through the media, within the family, in the community—with violence and its rewards and punishments will be incorporated in their development and will influence their future likelihoods of violence. Effects on the Environment Even the preceding discussion oversimplifies by implicitly assuming that the environment is fixed as individuals move through life. In reality, of course, the number and concentration of people pursuing different developmental paths—and the violence level itself—have profound effects on the environment. Unpunished violence may increase the probability of future episodes not only by providing examples that get absorbed into children's development, but also by weakening neighborhood and community institutions or by motivating neighborhood residents to arm themselves. Chapter 4 recounts an example of how crime and violence may have changed the social structure in some communities in ways that contributed to increases in homicide rates that accompanied the crack epidemics in many cities but did not generally decline as the crack epidemics subsided. In brief, the story runs as follows: the ease of turning powdered cocaine into crack allowed small- scale amateurs to decentralize and expand drug markets that had long been monopolized by organized crime. The market expansion, turf battles, and evolution of the "rules of the game" sparked increases in violence. The resulting fear provided a predisposition, and the crack profits a means, to purchase firearms. In the aggregate, the escalation of violence choked out legitimate economic opportunities and encouraged a mass exodus to the suburbs of families with attachments to the legitimate economy and with the means to relocate. In the eyes of teenagers in the families who stayed behind, the social status of the few remaining "old heads" fell beneath that of successful teenage drug entrepreneurs —leaving open-air drug markets, boarded-up vacant houses, and unemployed youth armed with semiautomatic weapons in communities where homes, thriving businesses, orderly schools, and churches once stood. The evidence that supports this explanation is fragmentary and inconclusive. Nevertheless, as a theory of violence levels, population 

EXPANDING THE LIMITS OF UNDERSTANDING AND CONTROL 306 movement, and economic decline, it illustrates the point that social-level processes are more than the sum of individuals' actions. In the aggregate, the distributions of individuals' potentials for violence and of occasions for violence damage communities' stocks of social capital—the families, markets, churches, social networks, and other organizations and social institutions that individuals build, maintain, or destroy. IMPROVING UNDERSTANDING AND CONTROL THROUGH EVALUATION Table 7-1 presents a scheme for classifying risk factors related to violent behavior. What is required to use it to reduce violence are better data and research on how individuals and communities respond to interventions that modify the risk factors. Because of poorly understood processes that cut across cells of the matrix, neither theoretical nor empirical research confined to a single cell or descriptive level is fully adequate for policy prescriptions. Progress in learning what works depends on sound evaluations of specific policy innovations. If an evaluation can answer a simple question—What effect did this particular innovation have on violence in this time and place?—then, if utilized, that answer becomes a building block for both a more generic understanding and an improved capacity to control violence. An analogy may be useful. In the pharmaceutical industry, the development of new therapies proceeds by trial and error within general guidelines offered by basic science and experience. Many promising drugs are tested in this process, but only a few of them are ever approved for general therapeutic use. The Food and Drug Administration requires that any new drug be subjected to extensive testing and proven to be both effective and safe. The selection principle there gives systematic evaluation a primary role that is currently lacking in social program innovations. In our view there should be no greater presumption in social programs than in pharmaceuticals that a new intervention is either safe or effective. Yet, as with pharmaceutical development, the absence of that presumption should be a stimulus for problem solving—iterations of diagnosis, intervention design, evaluation, and intervention refinement in light of evaluation results—not an excuse for inaction. 

EXPANDING THE LIMITS OF UNDERSTANDING AND CONTROL 307 Learning from Interventions Three categories of interventions can yield basic lessons concerning the causes and control of violence. The first category includes all those interventions for which violence is the central concern—examples could include legal restrictions on carrying guns in public, school curricula intended to teach conflict resolution skills, incarceration, shelters for battered women, and restrictions on the sale of violent pornography. The case examples of evaluations presented later in this chapter are drawn from this category. Policy interventions in this category may reduce or mitigate violence through mechanisms such as the following: deterrence through the threat of punishment; incapacitation by isolating people who have demonstrated patterns of violent behavior; opportunity reduction whereby potential targets for robbery, rape, and other violent crimes are rendered less vulnerable to victimization; restriction of violence-related commodities such as guns, some psychoactive drugs, and perhaps sadistic pornographic materials; mitigation of the consequences of violent attack through improved emergency medical response, counseling for rape victims, and so forth; and education in techniques for avoiding violent confrontations. The second category includes programs and policies that have some effect on risk factors for violent behavior—examples are prenatal care, early childhood education, drug treatment, welfare programs, abortion funding, the minimum legal purchase age for alcohol, and the design and control of public housing. Although these are sometimes advocated as part of a general effort to make streets and homes safer, their other effects are more direct and generally come first in any evaluation of these programs. The third category includes interventions that have pervasive influence on the quality of life—examples are macroeconomic stabilization policy, zoning rules, and minimum wage legislation. While there is an occasional congressional hearing on how the condition of the economy influences the homicide rate, or how international conflict affects domestic violence, those possible connections have negligible influence on policy making in those areas.4 Methods What methods are available to evaluators of specific interventions? Suppose that a state legislature forms a study commission 

EXPANDING THE LIMITS OF UNDERSTANDING AND CONTROL 308 to consider the possibility of amending the criminal sentencing code to institute enhanced sentencing provisions for gun use in violent crime. One important consideration is whether such enhanced sentences would reduce the murder rate in robberies and other felonies. To form a judgment on the matter, there are various possibilities: • informed speculation, • extrapolation from interview data, • study of natural variation, • evaluation of changes in law or policy, and • randomized controlled field experiment. While each of these approaches has been widely used in other contexts, and each may yield persuasive results under some circumstances, they are not equal: arguably the five research designs form a hierarchy of increasing validity. The randomized experiment generally provides a more valid basis for reaching conclusions concerning effects of the intervention. The use of randomized field experiments in evaluating crime control and public health measures has been endorsed by various groups of experts (Farrington et al., 1986; Reiss and Boruch, 1991; Coyle et al., 1989), and despite the practical difficulties there have been a number of such experiments successfully completed. The methodological superiority of these studies has tended to enhance their influence with researchers and policy makers. One notable example, the Minneapolis domestic violence experiment, is discussed later in this chapter. The methodological strengths of randomized controlled field experiments may be unattainable in practice. Ethical, constitutional, or legal restrictions properly prevent some experiments from being undertaken. When experiments are undertaken, difficulties are likely to arise in preserving and monitoring the intended treatments for the experimental and control groups, especially since these treatments typically require that the implementing agency relinquish part of its ordinary discretion and accede to the artificial dictates of the experimental design. Furthermore, collecting complete and reliable data on outcomes may prove difficult due to nonrandom attrition of sample members, and the possibility that the measurement process will be influenced by the experimental interventions. These problems are of course also present in nonexperimental evaluations (Berk, 1989). A more serious problem with field experiments may arise in extrapolating from the experimental setting to full-scale implementation 

EXPANDING THE LIMITS OF UNDERSTANDING AND CONTROL 309 (Manski, 1990). An experiment is conducted under somewhat artificial conditions and usually on a limited scale. In going from experimental trial to policy, the intervention may change in important ways that influence its effects: administrative and funding arrangements change, the scale of the program increases, the interaction between the program and the sociolegal environment may differ in practice from what it was in the experiment. These and other changes associated with operationalizing experiments often change their outcomes. These problems tend to argue in favor of evaluating "real" changes, using quasi-experimental methods (Cook and Campbell, 1979), paying close attention to understanding the process by which the treatment was "assigned" (Heckman and Hotz, 1989). In many cases the option of conducting a true experiment is not available, and even if it is possible to conduct some sort of experiment, it is unlikely to preempt other approaches. Evaluation research using rigorous quasi-experimental designs is an important tool for assessing the effects of violence control interventions. Organizational Contexts Two communities of actors and interests are involved in evaluation research. The community of policy makers and program managers includes legislators, police chiefs, school officials, store owners, television producers, and others responsible for formulating and implementing policies that may influence violence. The community of evaluation researchers involves both the discipline and the experts who do this work and promulgate its norms. Productive evaluation research is of course fostered by a cooperative relationship between these two communities—but that is sometimes difficult to achieve. Evaluation researchers tend to share certain beliefs about what constitutes persuasive evidence, beliefs that are sometimes at odds with those of the policy- making community. In the public forum in which suggestions for innovations are exchanged, an assertion of efficacy is usually evaluated by whether those making the assertion have suitable credentials, whether the basic idea makes sense, and whether the suggested innovation is responsive to pressing needs of the relevant agency. The research community is also influenced by these considerations but is inclined to withhold judgment until efficacy has been demonstrated through systematic evaluation. Thus the norms of researched evaluation require a wait- and-see attitude that often conflicts with the needs of policy 

EXPANDING THE LIMITS OF UNDERSTANDING AND CONTROL 310 makers to commit to some action quickly in response to public concern and to protect past actions from criticism. In the course of innovation, implementers are usually more exposed than evaluators to external pressures to conform to ideological agendas and to minimize exposure to risks of public failure. A productive relationship between the two communities requires that these differences in timing and vulnerabilities be acknowledged.5 Agencies have much to gain from successful evaluations, regardless of the findings. The evaluation is a source of information that is of potential value to the agency concerning the efficacy of a new policy to which there is no long-standing organizational commitment. In some cases the evaluation may be part of a political stratagem for reform, or simply a means for garnering public attention for a new program. If funding for the evaluation comes from outside the agency, it may increase the agency's budget and help support activities that are of value to the agency. The chief executive may receive recognition outside the agency for his or her participation. Cooperation is fostered by a relationship of mutual trust between the agency and the evaluator, by keeping the financial and organizational burden on the agency small, and by providing the relevant actors in the agency something of value in return for their cooperation. Both agency and evaluator must deal with organizational constraints. The burdens and rewards of conducting the experiment may be unequally distributed among the different layers of the agency organization; in such circumstances, agreement in principle between the agency executive and the evaluators may be insufficient to ensure proper implementation. It will be necessary to secure the compliance of those who are immediately responsible for implementation—the cop on the beat, the classroom teacher, the assistant district attorney. Another important institutional constraint on evaluations is the availability and quality of data from official records. The opportunity to evaluate an intervention is limited by the quality of data available on outcome measures. While in some cases the evaluator can gather some data with interviews and other procedures, it is usually true that data from official records play an important part in the evaluation of violence countermeasures. Evaluation Case Studies The general observations in the preceding discussion are illustrated here with four cases, each of which involves an evaluation 

EXPANDING THE LIMITS OF UNDERSTANDING AND CONTROL 311 of an intervention intended to reduce violence. Table 7-2 summarizes key aspects of these cases. They differ with respect to the mechanism by which the intervention is intended to reduce violence, by the evaluation strategy, and by the nature of the interaction between evaluator and policy makers. In each case the interventions have been considered for adoption in a number of jurisdictions. These four cases were not selected on the basis of either methodological excellence or findings of success—they illustrate wide variation in these terms. Rather, this group was assembled to illustrate the importance of sound evaluation methodology, the need for long-term commitments, the likelihood of mixed results, the need for replication, the need for cross-disciplinary collaboration, and the mutually reinforcing roles of basic science and evaluation research. These issues are discussed at the conclusion of this chapter. Controlling Domestic Violence In the early 1980s Sherman and Berk (1984) designed and implemented an experiment concerning the consequences of each of the three common police actions in domestic assault cases: arrest, mediation, and ordering the offender off the premises.6 The experiment was conducted in Minneapolis, using police officers who were assigned to target areas in the city and who volunteered to participate. A participating officer investigating a domestic violence call first determined whether it fit the definition of experimental cases (i.e., misdemeanor assault in which the victim did not want to seek an arrest warrant). If so, he followed one of the three responses designated in advance, depending on the outcome of a random drawing. The researchers followed up each of these experimental cases for six months with biweekly interviews of the victim (whenever possible) and by monitoring police records to determine if the suspect had been rearrested for a subsequent offense. As it turned out, the rearrest rate was about 50 percent lower for the arrest group than for the other two groups combined (Sherman and Berk, 1984). The policy importance of this result persuaded the National Institute of Justice to fund several replications in other cities. The results of these follow-up experiments in Omaha, Charlotte, and Milwaukee have been substantially different from those of the Minneapolis experiment. In particular, the replications offered no evidence that in these three cities an on-scene arrest reduces the average recidivism rate of suspects relative to other 

EXPANDING THE LIMITS OF UNDERSTANDING AND CONTROL 312 TABLE 7-2 Summary of Four Case Studies Sanctions for Mandatory Violence Two-Clerk Domestic Sentences for Prevention Requirement Violence Gun Crimes Curriculum Intervention Alternative Law 10-hour Law police mandating curriculum to mandating responses to longer teach youths two clerks in misdemeanor sentences if about violence stores at domestic felony night assaults involving a gun Violence Specific General Enhance skills Reduce reduction deterrence deterrence in avoiding criminal mechanism and violent opportunity incapacitation encounters Implementing Police Courts Schools Convenience agency stores Evaluation Randomized Quasi Controlled Quasi method controlled experiment field experiment field (multiple experiment experiment sites) (multiple sites) Data Interviews Vital statistics Questionnaires Police records Police Police and records court records 

EXPANDING THE LIMITS OF UNDERSTANDING AND CONTROL 313 police actions (Dunford et al., 1990; Hirschel et al., 1990; Sherman and Smith, 1991). Analysis of the Milwaukee data suggests that the consequences of arresting a domestic violence suspect may differ depending on his socioeconomic status, with positive effects for some groups canceling negative effects for others (Sherman and Smith, 1991). The results in Colorado Springs and Dade County, Florida, showed a deterrent effect of arrest in victim interview measures but not in official data (Sherman, 1992). The Minneapolis experiment was originally justified in terms of its potential contribution to basic deterrence research, rather than as an evaluation of specific policy alternatives for domestic violence (Sherman, 1980). The National Institute of Justice funded it as part of its Crime Control Theory Program. As a random assignment experiment in the use of arrest, it is an important methodological innovation in the study of specific deterrence. The results of the Minneapolis experiment and its replications are widely cited by criminologists seeking general principles concerning the effects of sanctions on the subsequent behavior of criminals. As it turned out, however, the Minneapolis results also figured importantly in the policy arena, being frequently cited as justification for the police to adopt a more punitive approach in domestic violence cases (Lempert, 1989). Minneapolis was chosen as the site of the original experiment because the new police chief there, Anthony Bouza, believed in the importance of field research and was a friend and mentor of the principal investigator, Lawrence Sherman. The Minneapolis City Council members welcomed the project as a federally financed mechanism for training police to be more sensitive to domestic violence victims. And although the midlevel management in the police force actively opposed the experiment, a number of patrol officers were recruited to participate after an intensive "marketing" effort by the evaluator. Maintaining their cooperation proved to be one of the most difficult challenges facing the evaluators. Similar problems were encountered in the replication cities. Despite these problems of implementation, the experiments were able to preserve the experimental manipulation and generate an adequate number of cases for statistical purposes. From the perspective of science, then, the Minneapolis experiment and its successors have been valuable. They have demonstrated that within the range of sanctions that were studied, the special deterrence mechanism of arrest is only conditionally effective in reducing the average recidivism rate of domestic violence suspects, and the necessary conditions are not yet known. 

EXPANDING THE LIMITS OF UNDERSTANDING AND CONTROL 314 The strength of the experimental design was in the internal validity of the results—yet there are important questions that could not be addressed within this approach. Police departments that increase the use of arrest in domestic violence cases may produce a general deterrent effect, whether or not there is any effect on recidivism rates. Measuring this general effect requires a different approach than measuring recidivism. The focus shifts from the level of the individual to the level of the jurisdiction, and a true experiment would require implementation of new police procedures in samples of randomly selected jurisdictions rather than cases—an approach that would raise truly formidable implementation problems in the absence of an agency authorized to impose such procedures on jurisdictions. Fortunately, data are available on jurisdiction-level victimization rates for domestic violence, so it is possible to employ quasi-experimental methods to evaluate the general deterrence effect of changes in police practice and related interventions. This type of evaluation has not been done in the case of domestic violence, but has been widely employed elsewhere. The basic approach is illustrated by the next case. Mandatory Sentences for Gun Crimes In 1975 Florida implemented a highly publicized law that required three- year mandatory sentences for carrying a firearm while committing a felony. A number of other states followed suit in subsequent years. These laws, and the means by which they were implemented, were sufficiently alike in Florida, Michigan, and Pennsylvania that they can be viewed as near-identical “treatments" in a natural experiment. Colin Loftin and his associates (Loftin and McDowall, 1981; Loftin et al., 1991) evaluated the impact of these laws in each of six cities, using a quasi-experimental statistical method. They found clear evidence that implementation of these mandatory sentencing provisions tended to reduce the gun homicide rate while having little or no effect on the nongun homicide rate. This result is important evidence on the general deterrent effect of sentence severity.7 This result is particularly interesting given the near-consensus by criminologists that sentence severity does not matter much in deterring violent crime (Cook, 1980). In their evaluation of the mandatory sentencing law, Loftin et al. analyzed monthly data on homicides for several years before and after the law's implementation in each of six cities. Their 

EXPANDING THE LIMITS OF UNDERSTANDING AND CONTROL 315 evidence for its effectiveness is that the monthly gun homicide counts tend to be lower following the intervention than would be expected based on an extrapolation (using the ARIMA technique) of the monthly series before the intervention. While it is possible that the change in gun homicide rates was the result of some other change in policy or environment that happened to coincide with the implementation of the gun law, that seems unlikely.8 First, while such a coincidence could happen once, the results are based on six cities in three different states. Second, nongun homicide rates did not shift (either up or down) at the time of the interventions, as they most likely would have if there had been some other change in the factors influencing criminal violence. The domestic violence experiments required a close working relationship between the evaluator and the agency, with patrol officers agreeing to relinquish their discretion to the dictates of the experiment. In contrast, the quasi- experimental evaluation of mandatory sentencing laws was carried out almost entirely at arm's length. The evaluators made no effort to manipulate the behavior of the front-line agencies. The evaluation was conducted on the basis of data generated by the routine reporting systems of police departments and the medical examiner's office (for homicide data). Of course, the quality of these data limits the potential of the quasi- experimental approach. The monthly homicide counts were reliable. But in assessing the effect of mandatory sentencing on other types of violent crime, the evaluators faced serious data limitations. First, the Uniform Crime Reports did not distinguish between gun robbery and other armed robbery prior to 1975. Second, it appeared to the investigators that the time series data for both assault and robbery were seriously flawed by variations in reporting and recordkeeping practices. They conclude that their only trustworthy results are for homicide. Even with that limitation, this quasi-experimental evaluation offers credible evidence concerning some of the effects of this type of intervention, and hence about the basic mechanism of general deterrence. The Two-Clerk Ordinance In April 1987, the city of Gainesville, Florida, began requiring convenience stores to employ two clerks during nighttime hours of operation. This ordinance was adopted in response to a detailed analysis by the Gainesville Police Department of convenience store robberies. The department's analysis, which was 

EXPANDING THE LIMITS OF UNDERSTANDING AND CONTROL 316 based on interviews with robbers and other types of information, concluded that having two clerks on the premises was a substantial deterrent to robbers.9 In the six months following implementation of the rule, the number of convenience store robberies was less than half the number in the corresponding period of the previous year ((Wilson, 1990). The apparent success of the two-clerk rule in Gainesville was touted by the police department and later encouraged the state legislature in Florida to consider making that city's requirement statewide (Kilborn, 1991). The Gainesville intervention is of interest both as an antidote to convenience store robbery and more generally as an example of the opportunity reduction mechanism for curtailing violent crime. A convenience store with two clerks may be a less attractive robbery target than one with a single clerk because of the increased likelihood of victim resistance and the increased chance that the victims will be effective witnesses in the subsequent police investigation. Other tactics for reducing convenience store robbery through opportunity reduction have also been tried. Indeed, five months before Gainesville implemented the two-clerk requirement, an ordinance took effect there that required convenience stores to install security cameras, to train night employees in robbery prevention, to limit cash on hand, and to maintain visibility between the interior of the store and the parking lot (Wilson, 1990)—requirements that previously had been adopted nationwide by the Southland Corporation for its 7-Eleven stores. Opportunity reduction tactics have also been instituted in public transportation and other commercial venues to reduce robbery victimization rates (Cook, 1986). Evidence on the effectiveness of such tactics has practical application and also provides important data for scholars concerned with criminal motivation and cognition. Evaluating the consequences of the two-clerk rule has proven difficult, for reasons that illustrate the importance of good evaluation practice. The Gainesville Police Department concluded that the rule caused a dramatic reduction in convenience store robbery, basing its conclusion on the fact that the convenience store robbery rate was substantially lower in the months following the two-clerk rule than it had been during the preceding year. Such a simple before-and-after approach to measuring an effect is unreliable and, in the Gainesville case, yielded an erroneous conclusion. A practitioner, retired police chief Jerry Wilson (1990) provided a more complete evaluation funded by the National Association 

EXPANDING THE LIMITS OF UNDERSTANDING AND CONTROL 317 of Convenience Stores. First, he demonstrated that there had been a sharp spurt in the convenience store robbery rate during fall 1986; this spurt ended in December with the arrest of three men suspected of being responsible for a great many robberies. The two-clerk rule went into effect the following April and had no discernible effect on monthly rates around that time. (There were two robberies each month from January through August). Thus the fact that there were many fewer convenience store robberies in 1987 than 1986 can be more credibly explained by the arrests in December than by the implementation of the two-clerk rule. This conclusion gains support from the fact that the convenience store robbery rate in the surrounding county followed the same temporal pattern as in Gainesville, even though the stores in the county were not subject to the two-clerk requirement. Thus the conclusion that the two-clerk rule was effective fails when rival hypotheses for the observed pattern are considered.10 The lesson is that, as in any other enterprise, evaluation can be done well or poorly, and the results may be sensitive to the quality of the work. But while the two-clerk rule did not have the large impact that the Gainesville Police Department claimed for it, there remains a possibility that it had a small deterrent effect. Reliably detecting a small effect (say, 5 or 10%) is very difficult in a city as small as Gainesville. Temporal patterns are difficult to ascertain simply because the numbers are so small. So the ultimate conclusion from the Gainesville story is not that there was no effect, but rather that there was no miracle. A careful evaluation does not rule out the existence of a small effect or the possibility that the rule is worth promulgating. The setting and circumstances of this intervention simply do not allow for a more definite conclusion. Violence Prevention Curriculum for Adolescents The development of the Violence Prevention Curriculum for Adolescents (Prothrow-Stith, 1987) was motivated by the high rate of intentional violence among youthful acquaintances and a belief that youths could be trained to defuse potentially violent confrontations. Originally developed for use in tenth-grade health classes in Boston high schools, in a series of 10 sessions, it informs participants about their risks for interpersonal violence, teaches them strategies for reducing that risk, and uses videotape both to analyze other violent encounters and to role-play the risk reduction strategies. The developers of the curriculum later adapted it for 

EXPANDING THE LIMITS OF UNDERSTANDING AND CONTROL 318 use in community settings outside the school and launched a comprehensive Violence Prevention Project that involved community-based organizations and mass media campaigns (Prothrow-Stith, 1987). The aim of the program is to change both attitudes and behavior. In 1987 the Education Development Center, Inc. (EDC) conducted a field- test evaluation of its violence prevention curriculum, with funding from the National Institute of Justice.11 The results of this evaluation offer only weak support for the claim that this curriculum is effective in reducing violent behavior or victimization. Nevertheless, by the end of 1990, more than 4,000 copies of the curriculum had been sold. While the dissemination of the evaluation results apparently helped stimulate sales and adoptions, the widespread interest in this curriculum is not due to systematic evaluation results so much as other factors: the concern by school administrators that something be done about adolescent violence, the lack of an alternative intervention clearly demonstrated to be effective, and the impressive credentials and visibility of the author, Deborah Prothrow-Stith, a public health professional. The evaluation of the original curriculum was conducted in six schools across the United States. These schools were recruited primarily from among those that had contacted the EDC concerning the curriculum. In each school one teacher of tenth-grade classes was selected for training, and his or her classes were divided between treatment and control conditions. Participating teachers received one day of training at EDC in teaching the curriculum. The evaluation was primarily based on responses to student questionnaires about attitudes and behaviors, administered before the onset of the program and two months after its completion. A longer-term follow-up was ruled out by the unwillingness of teachers to devote time to administering another questionnaire. Because no posttests were administered in two of the schools, pre-post comparisons were possible for only four schools. Of the seven questions relating to behavior, for only one did the curriculum appear to make a difference: in all four schools, students who received the curriculum reported having fewer fights during the past week than did those in the comparison classes. It is not clear whether this difference in self-reported behavior was due to real changes in behavior or to changes in the students' attitudes, which caused them to be less willing to admit being in a fight. The evaluators did not make use of official school or police records to validate self- report data.12 

EXPANDING THE LIMITS OF UNDERSTANDING AND CONTROL 319 The results of this evaluation are not persuasive that this approach is helpful in reducing aggressive behavior by high school students. As discussed in Chapter 3, accumulated evidence on the stability of aggressive behavior after middle childhood leaves open the possibility that with suitable changes, a program that applied these principles in the early grades might have shown more success. A more informative evaluation would require more resources. In the absence of clear evidence of effectiveness, it is notable that the curriculum has enjoyed such widespread success and has engendered the more comprehensive community- based program. In any event, the ongoing evaluation of that new, expanded version of the intervention will be of great interest. Lessons from the Case Studies These case studies illustrate that successful interventions, sound evaluations, and basic science are mutually supportive but difficult to coordinate and carry out. There is little disagreement with the axiom, "The best way to find out if X affects Y is to manipulate X and measure the change in Y." Unless that is done effectively, neither researchers nor policy makers are entitled to make many definitive statements about what works and what does not. Yet more is required when Y is violence in homes and communities, and X is a law, police procedure, school curriculum, community-based activity, traditional mode of discipline, or therapeutic intervention. The risk factors discussed in Part II and arrayed in Table 7-1 suggest promising Xs at different levels of description and at different stages in causal chains. But turning those suggestions into workable and effective solutions to violence requires a problem-solving approach that includes designing publicly acceptable interventions, evaluating them, using the results to refine the intervention, and replicating the evaluation. That process can contribute to improved violence control capability while it contributes to scientific understanding of violence. But it takes commitment by policy makers and the research community alike to the principles outlined below. The Importance of Sound Methodology Good intentions and a plausible story are not enough to ensure the success of an intervention. The only way to determine if something works is to try it, in a way that lends itself to reliable 

EXPANDING THE LIMITS OF UNDERSTANDING AND CONTROL 320 evaluation. Systematic testing and evaluation are essential to progress in reducing violent behavior. Evaluation research often depends on existing data systems, and it is limited by the quality of such systems, the breadth of the data collected, and the compatibility with other systems. A particular problem is difficulty of linking data on intentional injuries and other consequences to data on the violent events that cause them. Randomized controlled field experiments usually have important advantages as an evaluation strategy, yet in some circumstances it is possible to learn a great deal from well-designed "uncontrolled" interventions. Regardless of the basic approach, quality control is as important in evaluation as in other tasks. Seemingly minor methodological flaws in evaluations have occasionally led to advocacy of unnecessary or overly costly interventions and to embarrassment of the advocates; more frequently, they have created uncertainties that led to a misleading conclusion that "nothing works" or that some promising intervention strategy should be abandoned rather than revised and reevaluated. Long-Term Commitments Because the predisposing processes relevant to violence work over periods of years, some evaluations will require long-term commitments from implementing agencies, evaluators, and sponsors. Sponsors' commitments of program and project funding must therefore be long term to encourage the best researchers to invest their careers in them. Researchers who join such efforts must make long-term career commitments and maintain them despite publication pressures and the emergence of alternative short-term opportunities—an undertaking that young researchers may find easier as civil servants than in academic settings. And agency administrators who collaborate in problem-solving initiatives must limit their commitments to those they can maintain through budget cuts, changes in political priorities, and other pressures. Anticipating Mixed Results Evaluation results are rarely clear-cut. More commonly, an intervention produces the anticipated effect on some outcome measures but not others, or it affects some subjects but not others, or the results are partially obscured by some measurement problem, by some breakdown in implementing the intervention or evaluation, or by some unexpected intervening event. 

EXPANDING THE LIMITS OF UNDERSTANDING AND CONTROL 321 Such findings should suggest further analyses of the available data, changes in the intervention, or both. The results of reanalyses, while not as conclusive as are the basic results of a randomized experiment, may still provide valuable suggestions for revising the intervention or its implementation, for applying it to a different target population, or for adapting it to some specific subcultural or community context. These suggestions can and should be tested by revising and replicating the evaluation under the different conditions. Replication Promising interventions are rarely subjected to more than a few replications before adoption, yet replication is even more critical in evaluation than in the laboratory sciences. Because the effects of most interventions depend on local context, several replications will usually be needed even to distinguish "usually successful" interventions from "usually unsuccessful" ones, a point noted by Reiss and Boruch (1991). Once replications have established that some intervention works under at least some conditions, more broad-based comparative replications can help to determine what works best in these conditions. Often a comparison of interventions by different professions or control agencies—arrest versus alcohol abuse treatment, for example—is needed. These require more interagency collaboration than is traditional. Because violent events are so diverse, no single institution can expect to prevent more than a small fraction of them. This should be no cause for despair, but rather a call for a broadly diversified strategy of violence control analogous to diversified strategies as pursued in disease control. Communicating Across Disciplines Reflecting on its own experience, the panel believes that the difficulty of communication across disciplines and violence control agencies is a major barrier to developing effective interventions. Interdisciplinary communication requires each researcher to invest substantial time and effort in learning one another's vocabularies, in learning how phenomena at different levels of description are measured and classified, and in learning about the fundamental problems and current priorities in other violence-related disciplines. It also requires forums for interdisciplinary exchanges of research findings—something that does not naturally occur in most professional meetings. And it would be facilitated 

EXPANDING THE LIMITS OF UNDERSTANDING AND CONTROL 322 by a clearinghouse in which information could be easily accessed and shared about the current state of the art, disputes, and gaps in knowledge in the many fields concerned with a particular problem in violence research. Building Science Through Evaluation Evaluation research has two overlapping but distinct purposes. The first is reactive: to evaluate innovations that have become important because they are receiving consideration for widespread adoption. The second is proactive: to gain knowledge about the effectiveness of various mechanisms for violence control. The reactive agenda is important because of its immediate relevance to ongoing policy decisions. The proactive agenda is important if criminological theory is to be grounded in all available evidence—specifically evidence about relevant causal relationships that cross traditional disciplinary boundaries, particularly between biological and social levels of description. Evaluation evidence that an intervention grounded in some widely accepted theory failed may be the first sign that some such cross-cutting relationship exists. Evaluation research can be a rich source of evidence to guide theory development, and theory should provide some guidance to policy choices. Innovations in agriculture, medicine, and industry are routinely subjected to testing—sometimes on thousands of subjects over a decade or more—before adoption and dissemination. This practice is much weaker in social interventions, and the results are haphazard policies and a dearth of systematic knowledge. Basic science, policy development, and evaluation research should not be isolated activities. Rather, they comprise the fundamental triad of a problem- solving approach to violence. NOTES 1 Deterrence theories differ with respect to their explanations of how people perceive and weigh different sanctions, how alternatives to crime affect choices, the crime types that are most responsive to incentives, and other factors that affect how sanctions influence crimes. Reviews of this literature can be found in Blumstein et al. (1978), Cook (1977), Geerken and Gove (1975), Gibbs (1975), Klepper and Nagin (1989), Paternoster (1987), Tittle (1980), and Zimring and Hawkins (1971). 2The analysis by Cohen and Canela-Cacho is limited to the post-1975 period because pre-1975 data on individual frequencies of violent offending are not available. 

EXPANDING THE LIMITS OF UNDERSTANDING AND CONTROL 323 3 Analyzing the criminal careers of a sample of Denver convicts, Petersilia and Greenwood (1978) reported that, if all members of the sample had received one-year terms (i.e., a policy of relatively high certainty of a short sentence), their crimes would have been reduced by 15 percent through incapacitation at a cost of a 50 percent increase in prison population. Mandatory minimum four-year terms only for those with prior convictions (i.e., lower certainty of a long sentence) would have achieved the same crime reduction, but with a 150 percent population increase. 4 For example, Cook and Zarkin (1985) used data on the last nine business cycles to demonstrate that robbery is somewhat countercyclical while criminal homicide is not. 5 Paul Offner describes a life cycle theory of innovations in the welfare reform area. When a reform is being launched, the parties involved are caught up in the excitement and resist research considerations. Administrators are wary of the mysteries of evaluation methods and in any case feel sure of what works and what doesn't. Advocates are by definition convinced that they know what needs to be done; moreover, they are likely to object to experimental designs that require different treatments for similar recipients. Later, when the results almost invariably fall short of heady expectations, research questions are asked concerning costs and benefits (Manski, 1990). 6 This account is informed by a narrative prepared at the panel's request by Lawrence Sherman. 7 An increase in sentence severity may also reduce crime rates through the mechanism of incapacitation. If that were the only effective mechanism, then the observed reduction in crime would occur gradually. For example, if a mandatory sentencing law increased the minimum sentence from one year to three, then there would be no effect on crime during the first year following its implementation. But the evidence suggests an immediate effect, which suggests that the deterrence mechanism was important. 8 It has been pointed out that a shift in public attitudes may have preceded enactment of the new laws, and that the analysis by Loftin et al. could not eliminate the possibility that such a shift, rather than the law, caused the decrease in gun homicides. While that is technically possible, we consider it unlikely that the attitudes of community residents who were contemplating gun homicides coincide with those of the residents whose efforts brought about the new law. 9A detailed account of this rather unique effort on the part of the police department is given in Goldstein (1990). 10 Wilson (1990) suggests two other rival hypotheses. First is the implementation in November 1986 of the Gainesville ordinance requiring that convenience stores institute various robbery- prevention measures. Second is the possibility that the data on convenience store robberies are affected by the two-clerk rule; some convenience store robbery reports are false, intended to cover up a theft by an employee, and such false reports would be less likely if two clerks were on the premises. 

EXPANDING THE LIMITS OF UNDERSTANDING AND CONTROL 324 11 This account is based to some extent on a narrative provided at the panel's request by the Education Development Center. 12 School and police records provide incomplete counts of fights, because many fights occur off school grounds and most fights do not lead to arrest. However, these records might have been useful in validating self-reports of some events. References Berk, Richard A. 1989 What Your Mother Never Told You about Randomized Field Experiments. UCLA Program in Social Statistics Working Paper. Blumstein, Alfred, Jacqueline Cohen, and Daniel Nagin, eds. 1978 Deterrence and Incapacitation: Estimating the Effects of Criminal Sanctions on Crime Rates. Washington, D.C.: National Academy of Sciences. Boland, Barbara, Paul Mahanna, and Ronald Sones 1992 The Prosecution of Felony Arrests, 1988. NCJ-130914. Washington, D.C.: U.S. Government Printing Office. Cook, Philip J. 1977 Punishment and crime: A critique of current findings concerning the preventive effects of punishment. Law and Contemporary Problems 41:164-204. 1980 Research in criminal deterrence: Laying the groundwork for the second decade. In N. Morris and M. Tonry, eds., Crime and Justice: An Annual Review of Research. Vol. 2. Chicago: University of Chicago Press. 1986 The demand and supply of criminal opportunities. In M. Tonry and N. Morris, eds., Crime and Justice: An Annual Review of Research. Vol. 7. Chicago: University of Chicago Press. Cook, Philip J., and Gary A. Zarkin 1985 Crime and the business cycle. Journal of Legal Studies XIV:115-128. Cook, Thomas D., and Donald T. Campbell 1979 Quasi-Experimentation: Design and Analysis Issues for Field Settings. Boston: Houghton Mifflin. Coyle, Susan L., Robert F. Boruch, and Charles F. Turner, eds. 1989 Evaluating AIDS Prevention Programs. Washington, D.C.: National Academy Press. Dodge, K.A. 1986 A social information processing model of social competence in children. Pp. 77-125 in M. Perlmutter, ed., Minnesota Symposium in Child Psychology. Vol. 18. Hillsdale, N.J.: Erlbaum. Dunford, Franklyn A., David Huizinga, and Delbert Elliott 1990 The role of arrest in domestic assault: The Omaha police experiment. Criminology 28:183-206. 

EXPANDING THE LIMITS OF UNDERSTANDING AND CONTROL 325 Farrington, David P., Lloyd E. Ohlin, and James Q. Wilson 1986 Understanding and Controlling Crime: Toward a New Research Strategy. New York: Springer Verlag. Geerken, Michael R., and Walter R. Gove 1975 Deterrence: Some theoretical considerations. Law and Society Review 9:497-513. Gibbs, Jack P. 1975 Crime, Punishment, and Deterrence. New York: Elsevier. Goldstein, Herman 1990 Problem-Oriented Policing. New York: McGraw-Hill. Heckman, James J., and V. Joseph Hotz 1989 Choosing among alternative nonexperimental methods for estimating the impact of social programs: The case of manpower training. Journal of the American Statistical Association 84(408):862-880. Hirschel, J. David, Ira W. Hutchison III, Charles W. Dean, Joseph J. Kelly, and Carolyn E. Pesackis 1990 Charlotte Spouse Assault Replication Project: Final Report. Washington, D.C.: National Institute of Justice. Kilborn, Peter T. 1991 Rethinking safety in workplace that lures crime. New York Times , April 7, pp. 1, 27. Klepper, Steven, and Daniel Nagin 1989 The criminal deterrence literature: Implications for research on taxpayer compliance. Pp. 126-155 in J.A. Roth and J.T. Scholz, eds., Taxpayer Compliance. Volume 2: Social Science Perspectives. Washington, D.C.: National Academy Press. Langan, Patrick A. 1991 America's soaring prison population. Science 251:1568-1573. Lempert, Richard 1989 Humility is a virtue: On the publication of policy-relevant research. Law & Society Review 23(1):145-161. Loftin, Colin, and David McDowall 1981 “One with a gun gets you two": Mandatory sentencing and firearms violence in Detroit. Annals of the American Academy of Political and Social Science 150:455. Loftin, Colin, David McDowall, and Brian Wiersema 1991 A Comparative Study of the Preventive Effects of Mandatory Sentencing Laws for Gun Crimes. Unpublished manuscript. College Park, Md.: University of Maryland Institute of Criminal Justice and Criminology. Manski, Charles F. 1990 Where we are in the evaluation of federal social welfare programs. Focus 12(4) Methvin, Eugene H. 1991 An anti-crime solution: Lock up more criminals. Washington Post , October 27, p. C1, C4. 

EXPANDING THE LIMITS OF UNDERSTANDING AND CONTROL 326 Paternoster, R. 1987 The deterrent effect of the perceived certainty and severity of punishment: A review of the evidence and issues. Justice Quarterly 4(2):101-146. Petersilia, J., and P.W. Greenwood 1978 Mandatory prison sentences: Their projected effects on crime and prison population. Journal of Criminal Law and Criminology 69(4):604-615. Prothrow-Stith, D. 1987 Violence Prevention Curriculum for Adolescents. Newton, Mass.: Education Development Center. Reiss, A.J., Jr., and R. Boruch 1991 The program review team approach and multisite experiments: The spouse assault replication program. Pp. 33-44 in R.S. Turpin and J.M. Sinscare, eds., Multisite Evaluations. San Francisco: Jossey Bass. Sherman, Lawrence 1980 The Specific Deterrent Effects of Arrest for Domestic Violence: A Randomized Field Experiment. Proposal submitted to the National Institute of Justice. Washington, D.C.: Police Foundation. 1992 Policing Domestic Violence: Experiments and Dilemmas. New York: Free Press. Sherman, Lawrence, and Richard A. Berk 1984 The specific deterrent effects of arrest for domestic assault. American Sociological Review 49:261-271. Sherman, Lawrence, and Douglas A. Smith 1991 Interaction Effects of Formal and Informal Social Control for Domestic Violence. Unpublished manuscript. College Park, Md.: University of Maryland. Tittle, C.R. 1980 Sanctions and Social Deviance: The Question of Deterrence. New York: Praeger. Wilson, Jerry V. 1990 Convenience Store Robberies: The Gainesville, Florida 2-Clerk Law. Washington, D.C.: Crime Control Research Corp. Zimring, F., and Hawkins, G. 1971 The legal threat as an instrument of social change. Journal of Social Issues 27:33.