Violent Repeat Victimization: Prospects and Challenges for Research and Practice

John Laub: Good morning. I'd like to welcome you to today's installment of Research for the Real World, NIJ's translational criminology seminar series. My name is John Laub, and I'm the Direct of the National Institute of Justice. I just want to thank you all for being here today.

As you know, today's presentation is entitled “Violent Repeat Victimization: Prospect and Challenges for Research and Practice.” We'll feature Dr. Janet Lauritsen from the University of Missouri at St. Louis. In her presentation, Dr. Lauritsen will summarize existing research on repeat violent victimization, examine the potential impact that reducing repeat victimization might have on violent crime rates, and discuss factors that could be used to predict whether revictimization is likely to occur.

As a criminologist who studied repeat offending for a long time, I've always been struck by the fact that the same kinds of intellectual questions we pose about offending we rarely pose about victims of crime. It may well be due to the general neglect of victims in our field, but I do think this issue of repeat chronic victimization is one that deserves attention.

In keeping with the spirit of translational criminology, Dr. Lauritsen will also suggest how we could use this research on repeat victimization to inform and develop effective policy and practice.

It is now my pleasure to introduce Dr. Janet Lauritsen. Dr. Lauritsen's a professor of criminology and criminal justice at the University of Missouri St. Louis. Her research focuses on the causes and consequences of victimization, the social and historical context of crime and victimization, and quantitative research methodologies. She is currently a visiting fellow at the Bureau of Justice Statistics, one of our NIJ sister agencies here in the Office of Justice Programs, where she's working on data from the National Crime Victimization Survey to measure patterns and trends in repeat victimization. They have a recent report that just came out on this in April 2012, looking at measuring repeat victimization in the NCVS regarding how to count the series victimizations as their called.

Dr. Lauritsen is a member of the Committee on Crime Law and Justice for the Division of the Behavioral and Social Sciences in education at the National Academy of Sciences. She also serves on the editorial boards of several journals, including Criminology, the Journal of Quantitative Criminology, Criminology and Public Policy, and the American Journal of Sociology.

Dr. Lauritsen earned her Ph.D. degree in sociology from the University of Illinois, Urbana.

I've known Janet for a very long time, and she's been a dear friend and colleague. I was trying to recall the first time I met Janet. The easy answer is 1989, when she was hired as a postdoc on a project that Rob Sampson and I were working on related to victimization and offending overlap.

However, I have this very nagging recollection that I actually met Janet, this very young student at the University of Illinois, in 1983 when she was working at the Survey Research Center when I was visiting my friend Mike Offerson, who was a professor or assistant professor at the University of Illinois. I seem to recall that Janet was either working on a project with Mike and Marcus Felson at the time or not. But it may have been just the fleeting image. So I've known Janet for a long time, and much to my surprise, discovered that she is the person I have co-authored the most with, second only to Rob Sampson.

I guess when I think about Janet, one of the things that comes to mind is that Janet is a true scientist. One of the things that we think about when we think about science, I think, is everyone knows good scientists have the ability to pose good research questions that could be reasonably answered with available data. Analytic skill is something that a true scientist has.

I think there are two other traits that don't get enough attention in my view as to traits of a good scientist. One is tenaciousness. Janet Lauritsen focuses on a topic and does not let go until she figures it out. That is rare in this age of fleeting tweets, email, let's figure out something quickly and move on to the next topic. She is one of the most tenacious scientists I've ever met.

The second trait is courage. When we embrace evidence-based practice, we cannot cherry-pick. We cannot say, “Oh, these findings fit my comfort zone; let's run with them. Oh, these scientific findings make me nervous; let's not run with those.” You need to be courageous to be a good scientist. As I tell my students all the time, science is not for the timid. Janet is tenacious, courageous, as well as having all the analytic abilities of being a first-rate scientist.

With that, please join me in welcoming Dr. Janet Lauritsen

Janet Lauritsen: I'd like to thank the National Institute of Justice for inviting me to be here today and also for sponsoring some of my work on victimization over the past 20 years or so. I really appreciate the fact that you've all come out to hear me speak today about repeat violence. I'd also like to acknowledge the support that some of my current research is getting from the National Science Foundation, and especially from BJS, who has been kind to me throughout the years as I worked to solve various different problems related to victimization.

I must note the standard Department of Justice disclaimer that the work that I've done here reflects my own opinions and not that of any one — any other agency that sponsored this research.

I'm here today to talk about the phenomenon which has come to be termed “repeat victimization” and more specifically about how critical attention to this experience may help us understand and reduce violence. As John mentioned, I first began studying this issue long ago in a project with him and Robert Sampson, in which we examined victimization patterns among youth using longitudinal data and discovered, as several others had before, that one of the strongest predictors of whether a youth was a victim of crime in any given year was whether they had also experienced victimization in the prior year.

In a subsequent paper, my colleague, Kenna Quinet, and I described in more detail the extent of repeat victimization among this same sample of youth. We developed models to try to understand why prior victimization was such a consistent predictor of future risk and how this information might be useful for theory and practice.

More recently, I've been working with BJS to improve our understanding in estimates of series or high-rate repeated victimizations, and define ways to develop annual estimates of violent victimization that distinguish the number of victims in the population from the number of incidents they experienced, so we can learn, for instance, whether the general decline in violence since the mid 1990s included a decrease in the numbers of persons experiencing revictimization, as well as in the number of incidents.

What I've learned from this research over the years is that repeat victimization is an important component of many forms of violence, but at the same time it's one of the more challenging issues to assess, understand and prevent.

Today, what I'm going to discuss is what we know about repeat victimization and about the prospects and challenges for developing a better understanding of the issue in order to reduce violence. I'll begin with a brief overview of the research in the area that includes some of the work here in the U.S. and abroad. My focus will be on repeat violence.

I then examine a series of empirical questions suggested by prior research, using data from the National Crime [Victimization] Survey, or NCVS, that will allow us to look at the levels and trends in repeat violence at a national level. One of the criticisms of the annual NCVS reports on levels of violence in the U.S. is that the aggregate victimization rates don't distinguish the number of victims from the number of incidents they experience, and so we simply don't know if we're paying adequate attention to repeat victimization.

This is a research problem that I've been working with BJS to solve, so I'll present new findings from my research using those data that examine the extent to which repeat victimization plays a role in our national rates of violence, and how that may have changed over time.

I'll also present findings that show how repeat victimization affects the levels and trends in some of the types of violence known to most often include repeat events, in particular, intimate-partner violence against women and school-related violence against youth. An examination of repeat victimization for these specific types may produce findings that differ from one another and from the overall violence patterns.

Finally, I'll use the survey data to show how victims experienced with violence in one time period are associated with a risk in a subsequent time period, which will tell us how well some types of victimization predict future risk. This can be done with the NCVS data because it's designed to interview persons every six months for a maximum of seven interviews. Longitudinal data collection on victimization with a large representative sample of the population is very rare. Exploiting this aspect of the data can tell us things about repeat victimization that we can't learn from cross-sectional interviews that speak with persons just once. We can determine, for example, the likelihood that a victim of violence in one six-month time period will go on to additional violence in the next six months, controlling for some of the factors that we know are known to be associated with risk. Today I'll be looking at victimization both within six-month periods and across a one-year period.

After I present the new findings, I'll discuss what we know about the mechanisms and processes underlying repeat violence and what we've learned from various programs that have been developed to try to reduce some types of repeat victimization. I'm sure this audience knows that efforts to reduce different forms of violence have been met with varying degrees of success. A consideration of program results, along with findings about the extent of repeat victimization can help us discover what it is we need to know to better understand and reduce rates of violence.

Attention to the phenomenon of repeat victimization is certainly not new. More than 30 years ago some of the earlier analyses of victim data from the National Crime Survey, which was the precursor to the NCVS, revealed that when persons were asked to report about their violent victimization over the past six months, some victims reported experiencing one incident in a disproportionate share of total incidents. Analyses that took advantage of the fact that the survey interviewed respondents every six months also showed that previous victimization was correlated with future risk.

In fact, the issue of repeat victimization had been discovered earlier during a research and development period for the survey when the pretests revealed that some victims had experienced so many incidents within a six-month period, that they had difficulty recalling the details of each of the events. For those persons, victimization was no simply an event, but better described as a chronic condition, at least within a specific timeframe.

Two themes, both somewhat cautious in their implications, emerged from the early NCS research. First, that further study of repeat victimization might help us better understand the theoretical processes underlying victimization risk, and second, that the measure of repeated violence is fraught with special challenges.

Similar patterns were observed in the data from the then-new British Crime Survey. Using a 12-month recall period, this survey also showed that a relatively small proportion of respondents experienced a disproportionate share of incidents. Also, because their sampling design allowed researchers to study victimization in different areas of the country, it was discovered that high-crime areas appeared to have higher rates of violence because they had more persons experiencing repeated victimization. The focus that emerged from this British research was decidedly pragmatic in nature, arguing that if past victimization is a strong predictor of future risk, then crime prevention resources should be directed at those most recently victimized in order to produce the most notable reductions in crime and the most efficient use of resources.

Underlying these claims are assumptions that efforts targeted at recent victims will be more effective than broader prevention efforts to reduce crime and that once known victims are identified, we know what to do to reduce their risk.

In support of this idea where studies of residential burglary in the U.K. and elsewhere, which showed that burglary victimization at one time point was a strong predictor of future burglary, and also that the probability of a repeat event was highest in the period immediately following the first victimization; that is, within the first 30 days. This suggested that if preventive measures were to be most effective, they should occur immediately after the incident.

One of the first crime prevention programs to test this idea was the Kirkholt Burglary Prevention Project, a broad-based effort specifically targeted to areas with high rates of burglary and to recently burglarized households. Their program used a cocoon approach, which involved a combination of target hardening and target removal actions, additional police attention and patrols, and notification and engagement of neighbors in the immediate area.

The results of the program showed notable declines in burglary in the area and thus inspired optimism in the value of paying particular attention to other forms of repeat victimization.

Since the 1990s, the phenomenon of repeat victimization has been widely recognized as an important issue in violence. Other sources of data, such as calls for service to police departments, emergency room visits, and women's shelter data suggest that oftentimes police officers are called back to the same address, that victims of violence revisit emergency rooms, and that women return to shelters. Special purpose surveys, such as those designed to measure intimate-partner violence, sexual victimization, youth violence, and child abuse and neglect also indicate the same general patterns of repeated incidence of violence.

When the type of crime under consideration is repeat violent victimization, evidence of specific programs producing large reductions in revictimization is difficult to find. Why is this the case? If repeat violence has been so often observed, why hasn't it been easy to develop victimization reduction programs as successful for violence as the Kirkholt Burglary Project?

One argument is that we simply do not know enough about the levels and trends in repeat victimization and that without this information, we haven't had sufficient empirical basis from which to develop programs. For example, Farrell, Tseloni and Pease argue in a 2005 article that the failure of the NCVS to be used to study repeat victimization has hampered the development of violence reduction programs. They argue that the violent victimization rate routinely published is misleading and may have misguided crime policy because it fails to provide information about repeat victimization. They summarize their point there by saying that if the true extent of repeat victimization in the U.S. were known, it's possible that more effort would be put into strategic thinking as to how the prevention of repeat victimization could be developed there and elsewhere.

I'll begin with their challenge. To address the issue, I'll use the NCVS data to generate estimates of the levels in trends in violence and repeat violence in the U.S. using data for the period 1993 to 2010. But because varying definitions of repeat victimization have been used in the literature, I must first note how I am defining this concept.

So, what is repeat victimization? There are several ways to define it, and no single way is inherently superior to any other. The definition must be tailored to the theoretical or programmatic concerns of the research. The definitions that have been used have typically varied according to the types of victimization included, the timeframe under consideration, and the relationship between the victim and the offender.

A more restrictive use of the term might be to define a repeat victim as someone who experiences the same crime type within a short time period at the hands of the same offender. An example would be a woman who is assaulted by her boyfriend and then assaulted a second time four weeks later by him. After the second assault, she would be considered a victim of repeat violence. A less restrictive definition might also define her as a repeat victim if the second assault occurs say 10 months later or three years later or even if it's committed by a different man.

Take the case of a teenage boy who is being bullied at school. His first victimization of the school year may have been an assault by an older boy, but the second incident may be that he was robbed later in the year by two boys who came to view him as an easy mark because of what they learned from their peers about the first incident. In this instance, we might consider him to be a repeat victim even though he experienced different crime types by different offenders. Even if that later robbery was committed during the following school year, we may still want to define him as a victim of repeat violence if we believe that the subsequent incident was somehow linked to the occurrence of the first.

Next, consider the case of a college student who is assaulted at a campus party, who then experiences a second assault six months later by a stranger in a bar while on spring break. We might have second thoughts about defining him as a victim of repeat violence because it may not be as immediately obvious how the two incidents might be linked. Clearly, he's experienced more than one incident in a relatively short one-year time period, but this case seems perhaps distinct from that of the woman or the teenage boy because the second incident does not involve an offender who could've known about the victim's vulnerability from the first incident. Here, though, the second incident may have been preventable if we knew, for example, that this student often drank too much and in so doing became more vulnerable as a target. In that instance, we might think that a change in his drinking behavior may reduce his risk.

What if we can't determine an underlying common correlate of the two incidents because the student doesn't drink? How then might the second incident have been prevented by knowledge of the first? This exercise of thinking through the various dimensions of when we might want to define someone as a repeat victim tells us that there's not a simple or obvious answer and that we need to think carefully about the purpose of the concept itself. This is important, and I'll return to this issue later after I show the extent of repeat violence in the U.S. using several definitions.

That said, it should also be noted that the wider the definition used in terms of type of crime, victim-offender relationship, and length of time under consideration, the higher the prevalence estimates a repeat victimization will be. Higher or broader estimates may not be the most useful estimates. For these analyses, I'll use the definition of repeat victimization that falls somewhere in between more and least restrictive.

I define someone as a repeat victim of violence if, in a six-month period, they've experienced more than one incident of attempted or completed rape, sexual assault, robbery or aggravated or simple assault. This definition is wide with respect to types of violence and victim-offender relationship, but more narrow with respect to timeframe.

However, because arguments about reducing repeat victimization emphasize the period immediately following the event, this definition seems like a reasonable place to start. The definition also makes it less difficult to estimate repeat victimization because the NCVS asks persons about their experiences with violence using a six-month recall period.

For those of you who may not be very familiar with the NCVS, let me briefly describe it for you. It's a household-based, self-report survey designed to estimate victimization in the United States. Conducted by the Census Bureau for the Bureau of Justice Statistics, the survey uses a nationally represented sampling frame, and in-person and telephone interviews are conducted with all persons ages 12 and older in each sampled household.

Persons living in places such as college dorms and shelters are included in the sample; however, persons living in institutions, such as juvenile detention facilities and those who are homeless, are not included. Households selected to participate are interviewed every six months for a period of 3 1/2 years. Victimization experiences are measured through a series of cues and common-language questions and are coded into crime types following the interview.

Some of the key strengths of the data are its high response rates, which recently have been around 85 percent to 90 percent; its large sample size of roughly 165,000 interviews in 2010; a stable methodology; the use of what's known as “bounded interviewing,” which prevents double counting of incidents; and the fact that the complete series of data have been collected for almost 40 years.

The last major change to the survey occurred in 1992 and these changes resulted in increases in victimization rates, particularly for crimes of sexual violence, violence committed by non-strangers and intimate partners, and for incidents not typically reported to the police.

Of course the data are not without error and limitations. The disadvantages include the fact that it excludes persons under age 12, so it's impossible to use the data to estimate victimization such as child abuse and neglect. As noted above, it also excludes some persons, such as those living on the streets and in institutional settings. Research suggests that these persons are at higher risk for violent victimization. The trends and the patterns that I talk about today cannot be generalized to these groups.

To estimate the extent of repeat violence in the U.S. over time, it's helpful to briefly note how the standard victimization rate is estimated with the NCVS and how it can be decomposed to produce a repeat victimization rate. Generally speaking, the standard victimization rate that's published in the annual reports — that's the expression on the left-hand size of the slide — is estimated by counting the number of victimizations that all persons interviewed during the year report that they've experienced, and dividing that number by the number of persons interviewed. It's a bit more complicated. Sampling weights are used to take sample design into account so that the data can be used to produce estimates of the number of victimizations and rates that are representative of the population ages 12 and above.

This victimization rate is the number that some repeat victimization researchers argue is misleading because it tells us the average risk in the population and not how many victims there are or how many incidents they each experience. The data can be used to do this, to produce a prevalence rate and a concentration rate. The prevalence rate tells us what proportion of the population experiences one or more incidents in a particular time period. The concentration rate tells us how many incidents the average victim experiences. By decomposing the rates, we can then define any victim who experiences more than one victimization in a specified time period as a repeat victim.

I'll start by showing the victimization rates for the first and second six-month periods of each year for the period 1993 to 2010, along with the corresponding prevalence rates for those periods. I'll be happy to go into statistical detail for anybody who's interested after the talk. I do want to quickly note that in all these trends, the series incidents are included, and they're estimated using the new BJS counting rule.

The trend lines shown here suggest that there have been decreases over time in both the violent victimization rate and in the prevalence rate. In 2010, for example, the victimization rate was about 10. That is about 10 incidents per 1,000 population in both the first and second half of the year, compared to a rate of about 40 in the early part of the series. The prevalence rates, which are based on the number of victims, were about 7 per 1,000 in each half of 2010, compared to about 22 in 1993.

You might also notice that the gap between the two trend lines appears to have closed some over time, which can only occur if the average number of incidents per victim is declined.

If we also show the average number of incidents per victim, we can see that this does appear to be the case. The axis corresponding to this number appears on the right side of the figure. There has been a small decline in the average number of incidents per victim, from about 1.7 incidents per victim in the early to mid 1990s to about 1.3 or 1.4 in '09 and '10. Those are six-month averages.

The general conclusions we can draw from the figure is that there have been declines in risk for violence according to both the victimization rate based on the number of incidents, and according to the prevalence rate based on the number of victims. What these decomposed rates don't show us is what percent of all violent crime victims experience more than one incident, or how that proportion may have changed over time.

This next figure shows the percent of victims that were repeat victims of violence and the prevalence rate for repeat victimization. Looking first at the blue line with its corresponding axis on the left, we see that over these years, about 10 percent to 15 percent of violent crime victims were repeat victims in the six-month period. The average is about 13 percent. We also see that there is no clear trend over time in this estimate. Even though this percentage remained fairly stable over the years, we need to be careful not to conclude that there's been no change in the risk for repeat victimization. In fact, the red line with its axis on the right, shows that the prevalence rate for repeat violence did decline during the 1990s, though that decline leveled off and even increased during a few of the years around 2006 and 2007, before returning to levels of the early 2000s. Thus, person's risks for repeat violence have declined over time as the overall victimization rate has declined.

We can also use these percentage estimates to conclude that most victims of violence do not experience a subsequent incident within a six-month time period. if the percent of victims who are repeat averages 13 percent, this means that on average that about 87 percent of victims do not experience another incident in that period. This suggest that if the purpose of a victim-based program is designed solely to prevent future violence of any form and to do so in a relatively short six-month timeframe, the delivery of that program to all victims might not be cost effective because the program will have been delivered to a large number of victims who are likely to have not needed it. This is a highly qualified statement.

What happens when we consider special types of violent crime? I mentioned earlier that I've been working with BJS to study how to reliably estimate high-rate repeat victimization incidents, which are referred to as series incidents in the NCVS publications. In a BJS report that was released last week, we show that the typical forms of high frequency repeat victimization have not changed much over time, even thought the rates of those crimes have declined.

Series victimizations, which are the kinds of violence that persons often have difficulty detailing for interviewers, are primarily incidents of intimate-partner violence, youth violence at school, and victimizations that occur while at work or on duty, typically in professions that involve policing and security, medical personnel, and social work. Over the 1993 to 2010 period, we found that about two-thirds of the violence that victims report as series victimizations are one of these three types. Thus, it makes sense to look at these types of violence separately from the overall rate. In the interest of time here today, I'll focus just on the levels of intimate-partner violence and violence among youth at school.

Let's first look at intimate partner violence. Note that the axes for the rates here appear on the left side of the figure while the axis for the average number of incidents for the victim appears on the right. The trends shown here suggest that there were some decreases during the 1990s in both the victimization rate and in the prevalence rate, with a general leveling off in both rates during the 2000s. It appears that the average woman's risk for intimate-partner violence has remained fairly steady over the past ten years, but that there was a decline over the longer period.

In 2010, the six-month victimization rate was about 2 1/2 per 1,000 females, compared to rates of about 8 [per 1,000 females] in the earlier part of the series. The prevalence rates that are based on the number of victims were about 1 1/2 per 1,000 in 2010, compared to about 3 1/2 or 4 in the early 1990s. Also, you may recall from the earlier figure that the average number of incidents in a six-month period for victims of any form of violence ranged from about 1.7 to 1.3 [per 1,000].

Here we see that the average number of incidents for victims of intimate partner violence in a six-month period is higher, averaging about 2 per six-month period.

Now let's take a look at the percent of female intimate partner violence victims that were repeat victims in a six-month period and the prevalence rate for repeat intimate partner violence. Looking first at the red line, we see that the prevalence rate of repeat intimate partner violence among women is, generally speaking, lower in the most recent decade, compared to the decade of the 1990s. But the blue line tells us that the percentage of victims that experience repeated intimate partner violence has ranged from about 15 percent to 30 percent in a six-month period, with an average of roughly 25 percent over the years, and no clear trend over time. Thus, although women's risk for repeat intimate partner violence has declined over time, this is because fewer numbers of women are experiencing intimate partner violence now compared to the early 1990s, and not because the repeated nature of that violence has changed over time.

Though the fact that intimate partner violence is especially likely to be repeated is probably not news to this audience. What we do learn here that we didn't know before is that this proportion of repeatedness for women suffering from intimate partner violence is not much less now than it was almost 20 years ago. This comparatively higher level of repeatedness suggests that preventions of this type of violence, assuming a successful program, will be more cost-effective than a broader prevention program targeted at all victims of violence, because a greater proportion of intimate partner violence victims are repeat victims.

Finally, let's take a look at youth violence at school. Like the previous trends, the data here suggests that there were decreases during the 1990s in both the victimization rate and in the prevalence rate of youth violence at school. But for this form of violence, we see less of the leveling off in the rates during the 2000s than we saw in intimate partner violence and for violence overall. Instead, we see some additional declines in the rates during the 2000s.

In 2010, that six-month victimization rate was about 6 incidents per 1,000 youth ages 12 to 17, compared to a rate of roughly 34 in the early part of the series. The prevalence rates were about 5 per 1,000 in each half of 2010, compared to about 22 in 1993. So we've seen fairly steady declines in violent victimization at school.

When we look at the repeated violence at school, we see that the prevalence rate of repeat victimization, the red line, is generally lower in the most recent decade, compared to the decade of the 1990s, and the youth are now at lower risk for repeated victimization at school than they were in the previous decade.

The blue line tells us that the percent of victims that experience repeat violence has ranged from about 10 percent to 27 percent over time with an average of roughly 19 percent, and there's no clear trend in this percentage over time. This proportion of repeatedness is somewhat lower than was found among women suffering from intimate partner violence. So, from this figure, we learn that the repeated nature of violence at school is not different now, compared to earlier years. And like intimate partner violence, the proportion of repeated victimization suggests that prevention programs targeted at victims of this type of violence — assuming a successful program — should be more cost-effective than a broader prevention program targeted at all victims of violence.

I realize I've just shown you quite a few trends. Here's a summary of what we've seen so far in these trends. For all three categories of violence, we've seen declines in the victimization rate, based on the number of incidents; the prevalence rate, based on the number of victims; and the repeat victimization rate; based on proportion of the population experiencing two or more incidents in a six-month period. The declines in intimate partner violence have been somewhat smaller than those in other categories. Although the declines are certainly noteworthy, we saw that with the exception of youth violence at school, the declines were much smaller during the 2000s than they were during the 1990s.

Even thought the various rates of violence have declined, there were only small changes in the percent of victims with two or more incidents and in the average number of incidents per victim. So, while risk is declined, the repeated nature of victimization today is much the same as it was in the early 1990s.

Earlier I noted that the definition of repeat victimization is flexible and should be tailored to a specific research question. One of the limitations of population-based research in this area is that repeat victimization has been assessed to the extent possible with data gathered cross-sectionally. The analyses I've just shown treat the NCVS data as a series of six-month cross sections or snapshots of risk. For some purposes, this might not be what we want, because although we can describe what proportion of victims experience two or more incidents within a specific period, we can't say much about longer periods of victimization or how risk across the periods might be linked.

There are ways to try to estimate longer risks with cross-sectional data, but they typically require asking subjects to recreate the details of their victimization histories over long periods of time. These types of data have unknown levels of error that can be consequential if one is trying to sort out the time sequencing of events and trying to determine why persons experience different victimizations at different periods in their lives. These are the types of issues that longitudinal data can help begin to address.

In the next part of the analysis I'll present some preliminary findings about the extent of repeat victimization in a one-year period. These are preliminary findings because they're only looking at person's experiences in two adjacent six-month periods of the years 2008 and 2009. The goal is to be able to look at persons for as long as they are in the sample; that is, for up to three years, and to do so for the full 1993 to 2010 period. Ideally, researchers would like to be able to follow persons across the live course, but the NCVS data can't be used to do this because persons aren't followed for extended periods of time. My analyses will not be able to estimate long-term vulnerabilities, such as those that exist from childhood into adolescence or from adolescence into adulthood.

This next analysis will then estimate the likelihood that a victim of violence in one six-month period will go on to experience additional violence of the same type in the next six-month period, and the strength of that relationship between victimization at time one and two. Briefly, what I've done here is used the NCVS to analyze the experience of persons who were interviewed twice in 2008 and interviewed twice in 2009. The estimates will present the average findings across the two years.

For each type of violence, I first ask what percent of victims at time one were also victimized at time two. I'll then examine how strong that relationship is between the two time periods after controlling for some basic demographic factors. First, let's look at what percent of victims at time one went on to be revictimized at time two.

What we see in this table is that about 6 percent of victims of violence in the first part of the year were also victims in the second part of the year. Like the earlier findings, this indicates that the majority of violent crime victims in one six-month period are not revictimized in the next six-month period. The table also shows that this proportion is higher for women suffering from intimate partner violence. About 12 percent of women suffering from intimate partner violence in one time period go on to experience additional incidence in the next time period, which is a revictimization rate, about twice that found for violence overall. Like the earlier analyses, this finding highlights the higher degree of repeatedness in intimate partner violence, compared to other forms.

The percent of youth victims of violence at school who go on to experience additional violence in the next six months is somewhat lower than was found for intimate partner violence, but higher than for violence overall. Again, this is similar to the patterns within the six-month period shown earlier. About 8 percent of youth victimized at school at time one report that they were victims again during the next six-month period.

Now let's examine the strength of the relationship between victimization at times one and two. Researchers who try to determine risk and protective factors for victimization at the individual level of analysis have looked at various characteristics, such as a person's demographic attributes; family, neighborhood and school characteristics; their lifestyle and peer associations; as well as many other factors. Though this literature contains literally thousands of studies that reveal many consistent and significant predictors, as a whole it suggests that even the best set of predictors do not do a great job of predicting who will be victimized within, say, a one-year period. In research terms, most of these models have low R-squared statistics, which means that they cannot account for a large portion of individual variation and risk.

Like any statistically rare event, like predicting whether a person will have a heart attack or an auto accident in the next year, victimization is difficult to predict in the short term. I mention this because whenever we're trying to predict whether an individual that will experience an event that is, in statistical terms, uncommon, and known risk factor are relatively weak, the best predictor will often be whether the experience the same incident in a recent time period. Let's first look at violence generally defined.

Here we see that in 2008 and 2009, the likelihood that a victim of violence at time one will experience another victimization at time two, is about eight times greater than it is for someone who was not a victim at time one. Even though we see that most victims of violence are not revictimized, the fact that a person was a victim in the first period is a strong predictor of their risk in the next.

Here we see how strongly intimate partner violence at time one is associated with intimate partner violence at time two. Even with some controls for demographic factors, we find that in 2008 and 2009, women who experienced intimate partner violence at time one are roughly 75 times more likely to report additional violence at time two than are women who did not experience that violence at time one. Clearly, recent victimization is a very large predictor of future risk, far surpassing in magnitude any other known risk factor that's in the literature. To some extent, we expected to see a stronger relationship than found for violence overall because we're already aware that intimate partner violence has a greater degree of repeatedness and because it's statistically more rare than the broader definition of violence. But we can see here that it is certainly much stronger than was found for the case of violence overall. Now let's look at the relationship for youth at school.

The risk differential shows that, with controls for some demographic factors, 12- to 17-year-old victims of violence at school at time one are about 12 times more likely to be victims of additional school violence in the next six months than are non-victims. The finding that this risk differential is much closer in magnitude to the overall violence risk estimate is, in part, related to the fact that youth violence at school is more common. What the full set of findings highlights is the unique repeated nature of intimate partner violence compared to these other forms.

Before I turn to a discussion of what these findings may mean for understanding risk and for practice, I want to urge caution about this particular set of estimates. It's very important that these patterns be assessed using additional years of data to determine how sensitive they are to the year selected for analysis. The confidence intervals around these relative risk estimates are also quite large, particularly for intimate partner violence. For this type of violence, the confidence interval suggests that the relative risk ratio may be as low as 25, rather than 75, as well as higher. Preliminary analysis of data from 1994, a period when intimate partner violence rates were higher overall, showed a ratio of about 20. So, we see how sensitive these estimates across the years can be, and how they're related to the overall base rate of the phenomenon.

What do such numbers actually mean? How should these indicators of repeat violence be used? To answer these questions, we need to learn more about the mechanisms or theoretical processes that underlie victimization risk.

The research on repeat victimization tells us that there are generally two ways to think about what the relationship between victimization at time one and a later period means. If we think about victimization as having a set of causes or predictors, we might depict that model to look something like the following.

Here we see a simple illustration showing that some set of factors that exist prior to the occurrence of victimization at time T are causally related to victimization. Broadly defined, these are factors that vary across individuals and are generally described in the literature as involving exposure to offenders, victim vulnerability, and target attractiveness, and the degree to which some form of guardianship that might prevent victimization from occurring is available. Identifying these factors is the goal of much research on victimization.

Take the case of the boy who experiences a violent assault at school. Compared to other boys, including those going to school elsewhere, he may be at greater risk because, for example, he's younger and smaller than his classmates, and he has greater physical and emotional vulnerabilities. The school he attends may have greater numbers of juvenile offenders and fewer numbers of teachers and other guardians in the hallways and the areas around the school.

Think about the fact that we see a strong relationship between victimization at time one and two. This may occur simply because the same factors are present at both periods. For the boy at school, then, there's a strong relationship between victimization and time one and two because the causal factors haven't changed. We would expect him to continue to remain at higher risk than other boys elsewhere.

In this scenario, the statistical estimate depicting the relationship between victimization at time one and two, those risk differentials, will be a reflection of the consistent presence of those background causal factors.

If this is what is going on, researchers will describe the relationship between victimization at time one and two as reflecting a process of consistent heterogeneity and risk factors or predictors. However, researchers have also examined whether some other processes may be occurring that help account for repeatedness. That is, the possibility that the occurrence of the first victimization somehow added to the individual's future risk.

The boy [who] was assaulted at school during the first half of the year and later robbed by different youth during the second half may have been robbed in part because these youth learned about the boy's vulnerability from their peers after the first assault. This in turn added to his risk, because he became known as a more vulnerable target to the other youth who are looking for someone to rob.

When repeat victimization occurs for this kind of reason, researchers describe it as reflecting a “state-dependent” or an “event-dependent” process. The background factors are likely to continue to remain relevant, but something about the outcome of the first event added to the victim's risk. That's why there's an additional arrow there [on the slide] from victimization at time one to time two.

Earlier, we saw estimates of the relationship between several forms of victimization at time one and two. To think about what those numbers mean, we have to think broadly about several potential processes, to see how they might operate for various types of violence.

If the revictimization of the boy at school reflects heterogeneity processes, then identifying those processes should suggest specific prevention practices. For example, if the issue is primarily a lack of sufficient guardianship at school, then increased supervision in places where potential victims and offenders are in contact with one another is called for. If the school's well supervised, and victimization remains high, it may be that it can only be reduced if youth who commit these acts have less contact with potential victims.

However, if it is the case that state-dependent processes are also occurring, increased guardianship and reduced exposure to the original offender may not necessarily reduce this victim's risk for repeat victimization.

If he's become known as a vulnerable target, then this label may continue to haunt him outside of the school in his neighborhood or other places where he might be in contact with his schoolmates.

The types of programs that have been developed to try to prevent youth violence in schools are typically school-based programs, because that's the institution that can be used to delivery programs to youth. To my knowledge, most efforts that explicitly attempt to reduce repeated victimization at school are primarily anti-bullying programs, a phenomenon that represents one form of repeat victimization and includes acts that are intended to intimidate, but which may not be illegal, such as insults, verbal attacks and harassment.

There also are many other school-based violent programs that are focused on reducing disruptive and aggressive behavior, but their assessments don't often include indicators of victimization or revictimization. Systematic reviews of bullying programs have shown that the most successful programs decrease this type of repeat victimization by an average of roughly 20 percent. Much can be learned from these reviews. But because bullying is often defined to exclude incidents between persons of similar physical and psychological strengths, it's not clear whether these programs will have effects on other types of repeat victimization.

Similarly, broader school violence programs have also been found to be successful at reducing aggressive behavior. If we want to know whether they also reduce repeat victimization, the answer is largely unknown. Successful programs that reduce offending at school are, by definition, also reducing youth's exposure to offender. To the extent the victimization is primarily driven by this factor, these exposure reduction programs should reduce repeat victimization as well.

What if a state-dependent process is occurring? If it's some form of labeling process, then exposure reduction in school may not reduce the likelihood of revictimization because other offenders, including those outside of the school, might have become aware of the victim's vulnerability. We don't know the extent to which this process may be occurring among teenagers at school, but related research on chronic victimization using children's play groups has suggested that children with fewer social interaction skills are more likely to be aggressed against or victimized. These children tend to react differently to aggressive episodes and over time, other children also choose that child to victimize.

The initial risk factor may have been the child's relative lack of social skills, but the revictimization was promoted by others' perceptions of the victim's vulnerability and larger group processes and dynamics. Attending only to the earlier risk factor, the relative lack of skills, may not have an immediate effect on the child's risk for future victimization, since others may not come to view the victim differently. Attending only to the labeling process will likely also be insufficient, since the initial risk factor's left unaffected.

Among adolescents, the factor that has been shown to be the strongest predictor of victimization, once prior victimization is taken into account, is involvement in offending. Youth who are victimized are more likely to offend than those who are nonvictims. The reverse is true as well: Youth who engage in offending are more likely to be victims than youth who do not. There's also evidence that these patterns are reciprocal over time and the victimization at time one is associated with increased risk at time two because some victims retaliate. Some do things like join gangs, adopt a code of the street's attitude, or obtain weapons in an attempt to reduce their vulnerability.

Youth often do these things because they believe they will be safer, but longitudinal research has shown that this is not the case. Clearly, these are much different state-dependent mechanisms than the labeling process I just described. Where there is evidence that this latter set of factors is occurring, attention must be paid to improving the ways in which youth might be responding to or coping with victimization and fear.

Although the earlier estimate told us that victims of youth violence at school are about 12 times more likely than nonvictims to experience another event, this number does not tell us what we should do to prevent future violence. It's only preliminary information. Basic research that uses longitudinal data to identify the processes underlying the phenomenon is necessary to inform the development of an effective prevention program.

What about intimate partner violence? How might thinking about heterogeneity and state-dependent processes help us understand and reduce this form? We saw earlier how strongly intimate partner violence at one time period is associated with victimization in the next, even though at the same time most of the victims will not report additional violence, at least during the subsequent six months. Compared to the amount of research on youth, there's relatively little literature that explicitly focuses on modeling heterogeneity and state-dependent processes that may be involved in intimate partner violence, because the necessary data following women over time is rare and difficult to obtain.

In general, longitudinal and nationally-representative data on adults is rare. Collecting such data is an especially daunting challenge when the topic is victimization because victims often move to escape violence, making it difficult to keep track of subjects for subsequent interviews. We have learned a great deal about intimate partner violence from clinical samples and from programs designed to reduce such violence that we can use to think about these issues.

In terms of heterogeneity processes, there is some research that suggests that in the U.S. some demographic attributes are significant correlates of this type of violence. For instance, younger women, unmarried and cohabitating women with children, and women living in highly disadvantaged neighborhoods are more likely to experience intimate partner violence than their counterparts. There's also some evidence that women living in households experiencing financial crises, and those who experience victimization earlier in their lives, are at higher risks. Longitudinal data from New Zealand found that young women who grew up in families with high degrees of conflict and harsh discipline practices were more likely to experience intimate partner violence, as were women who as teenagers engaged in aggressive delinquency, had substance abuse problems, contact with the police, and then dropped out of school.

Though the correlates I've mentioned have been found to be statistically significant, they can be interpreted in various ways. Although they're certainly not required for the occurrence of intimate partner violence, they're often instead used to suggest where the risk is greatest. In that sense, they're thought of as representative of heterogeneity processes.

We might try to categorize these factors into dimensions of exposure to offenders, target vulnerability, and availability of capable guardianship in order to see how they might operate to increase risk. As we try to think about these correlates in this way, we see that in contrast to thinking about youth violence in school, this exercise does not resolve itself as neatly. Like school violence, if intimate partner revictimization reflects heterogeneity processes, then identifying those processes should suggest specific practices. Unlike school violence, intimate partner violence may involve continuous exposure to the offender, especially if the partners live together. Because the violence typically takes place in private spaces, one cannot simply add guardianship and supervision to the home. There is no institution like the school that can offer this type of direct monitoring and protection from repeat victimization.

Efforts to reduce intimate partner violence have tried to use analogous mechanisms to reduce exposure, such as mandatory arrest policies, protection or restraining orders, and shelter availability. By themselves, these efforts have not had large effects because oftentimes one or both of the intimate partners want to continue the relationship. If the assaultive partner wants to do so, but the woman does not, this is a very dangerous situation, because constant guardianship is not a possibility. If both parties want to continue in the relationship, it can also be very dangerous, because even if the batterer is ordered to participate in a battering intervention program, most of these programs have been shown to have relatively weak effects.

One approach to reducing repeat incidence of intimate partner and family violence that was inspired by the discovery of repeat victimization has been referred to as “second-responder programs.”

They're called second-responder programs because after initial police response to a family violence incident, a second visit — typically including a police officer and a victim advocate or family violence specialist — would follow up with the victim to provide her with information about her legal rights and the available services in the area that she might need, such as shelters, counseling, and other forms of assistance. These programs were specifically targeted at reducing revictimization under the assumption that the victims are most amenable to help and assistance in the period immediately following an incident. In that sense, the programs are trying to create a new state-dependent effect to reduce the likelihood that a subsequent incident will occur.

The initial results for these programs were mixed. Some programs appeared to show decreases in victimization, while others did not. A few even suggested increases in risk. A set of these experimental programs was recently assessed in an NIJ-sponsored systematic review by Davis, Weisburd and Taylor. The researchers assessed 10 programs that met key methodological criteria, such as having a control group for comparison, and it included data on whether a new domestic violence incident occurred within the next six-month period, using both victim survey data and police reports.

Their systematic review concluded that the second-response interventions did not reduce the likelihood that the household would have another family violence incident. But the intervention did appear to increase victims' willingness to report incidents to the police, perhaps because the victims came to have more confidence in them due to their extra concern and [the] attention that the victims received.

The repeat victimization estimates I reported today may help us understand why these programs have had disappointing results. The NCVS data showed that the relationship between intimate partner violence at time one and time two is very strong in recent years, while for other types of violence the relationship's not quite as strong. Interventions that seek to reduce revictimization have to be powerful enough to overcome the influences of all the other risk factors that initially contributed to the prior victimization.

For intimate partner violence, this means that a successful program will have to break a stronger association between violence at time one and time two than would be the case, for example, for school violence programs. It appears that if that program cannot reduce exposure to the offender — the original source of the violence at time one — it's unlikely to have much effect on violence at time two.

What should we make of our additional finding that most of the victims at time one didn't report intimate partner violence at time two? Here, I think it's important to keep in mind that the second time period only captured six months. There's very little research that attempts to document the temporal patterns of intimate partner violence or any other form of victimization. It may be the case that such episodes cluster in short time periods and are followed by nonviolent periods, only to re-emerge perhaps again in clusters eight months or a year or two later.

I am unaware of any data that can be used to describe the typical patterning of incidents for a representative sample of women over long periods of time, such as 5 or 10 or more years. If they did exist, they'd provide useful information about what a reasonable follow-up period should be to determine the usefulness of a prevention program. For instance, the second-responder's program may have shown more positive results if the follow-up period was longer, or if there were repeated or incremental increases in the dose of the program, and the outcomes were evaluated over longer periods of time.

I noted what many of these intimate partner violence programs are trying to do is create a state-dependent process to break the association between violence at time one and two. There are other state-dependent processes to think about as well, and these include the consequences of violence for the victim.

We know from research that women are psychologically and emotionally damaged by these experiences and that many victims suffer from anxiety, depression, PTSD, and other symptoms that may further increase their risk for revictimization because the victim may not have the strength to defend or remove herself from a dangerous situation.

Victim-service providers, who do important work treating these symptoms sometimes do so under the expectation that these types of services will eventually reduce revictimization. In many instances, this may be the case, but in my opinion, this should not be the standard by which the funding for and availability of such services should be judged. Services related to mental health may, in some cases, play a role in reducing women's risks for revictimization, but like other types of medical care, they should be assessed according to their value from a mental health perspective, and not for their ability to lower rates for victimization.

So, my conclusions: I said at the beginning of my talk that repeat victimization is an important component of many forms of violence, but the early optimism, resulting from programmatic success with repeat burglary and other forms of property crime is more difficult to sustain when the focus shifts to preventing repeat violent victimization. This is not to say that violence reduction programs are ineffective in lowering risk for victimization, but that the effort to target recent victims to reduce revictimization can be very difficult, because it seems likely that the most important aspect of the program will be its ability to reduce victims' exposure to offenders. Unlike property, victims are active agents, and they may be unable or unwilling, for a variety of reasons, to participate in the process.

How much value is there then to focusing on repeat victimization, rather than on victimization more generally? I think there's certainly value in continuing to study the issue, in part because the estimates themselves of revictimization remind us that it will be more difficult to have an impact on some types of problems compared to others. Programs that result in small successes for a very difficult problem are likely to provide important information to build on. Different types of revictimization rates may also be useful for telling us when, where, and for what groups the experience of chronic victimization is most likely to be found, and where victim services and prevention efforts are needed.

These new repeat victimization rates may also tell us what we already know. If it's the case that revictimization proportions are primarily a function of crime type — for example, as in the case of intimate partner violence — the groups with the higher victimization rates will also likely be the same groups with the higher prevalence rates and the higher revictimization rates.

To be most useful, I think the definition of the concept will need to be carefully considered. It's not obvious to me what the length of time of the follow-up period should be to judge the success of a program. Shorter timeframes are often preferred for methodological reasons to ensure that the results are not contaminated by external factors and events, but it may be the case that it takes time, and perhaps repeated efforts, before victims are willing and able to heed the suggestions of an intervention.

I think that the primary value of the concept of repeat victimization lies in what these experiences can tell us about the processes underlying victimization risk more generally. This point was made long ago, and subsequent developments and longitudinal modeling have made it possible to assess how revictimization might be influenced by both heterogeneity and state-dependent processes. Much of the modeling of these processes has been limited to adolescents, due to the relative lack of victimization data that follows children and adults prospectively over various periods of the life course.

Continued research of this type with new data sources can help distinguish between distal and more proximate causes of violent victimization, and the extent to which they might be social, psychological, or perhaps even biological in origin. The findings from this research might also show that, in some instances, our prevention programs have done as much as is feasible, given a set of persistent underlying causes. When this is the case, reductions in revictimization will have to come from altering the broader, underlying set of factors that promote the violence in the first place.

Thank you.

Laub: Thank you, Janet, for that fascinating talk. Another trait of a scientist is I counted seven times “preliminary” and “caution.” So, with that, we're going to have a question-and-answer period.

Lauritsen: Great.

Akiva Liberman: Akiva Liberman, Urban Institute. Thank you very much. That was fascinating. I want to ask about the conceptual distinction between thinking of the repeat aspect of victimization as a general feature of victimization versus trying to differentiate those who are repeat victims from those who tend to be victimized only once perhaps. In that regard, I noted a couple of important things. One, when you mentioned risk factors for victimization, both for intimate partner violence and for youth violence at schools, they were general risk factors as opposed to risk factors that would discriminate those groups. I would imagine — just jumping off without any caution whatsoever — that in the case of intimate partner violence, where the prevalence rate of repeat victimization was something like 25 percent to 30 percent, that the risk factors will tend to mostly be the same, that it's more of a persistent feature of intimate partner violence that tends to have a repeated quality. But for other types of victimization that have lower rates of prevalence of repeat victimization, it would seem like it might be quite different. That's the general question. What do we know about that?

Lauritsen: I think I understand your general question. There have been other experiments of trying to use recently victimized persons and applying the same kind of logic to their situation as the second-responders programs did — some experiments that were run in New York — and they included burglary victims, assault victims, and robbery victims. What's interesting about those experiments, I think, is — to cut to the end — they also had little effect on repeat victimization of that type.

What the programs did, though, was increase the victim's knowledge about what risk factors might be, precautionary behaviors that they might take. And follow-up assessments showed that they retained that knowledge, but in the end it didn't help. What that suggests is this is not just a problem limited to one type of repeat victimization, but that the problem is very difficult. If you know somebody's at risk, and you can identify the source of that risk as exposure to offenders, if you can't do anything about that, it's hard to imagine what you could do then to reduce revictimization. I think in most cases you'll have to ultimately address the source of the revictimization in the first place. Situational factors, like more proximate to the occurrence of the event, are constantly changing and hard to measure. Is that…?

Liberman: Well, in mentioning those interventions, if I understand it, you're talking about interventions that take all victims as at greater risk for revictimization than other people, and therefore target resources. I'm not hearing anything about an attempt to further distinguish among first-time victims, those who are at greater risk for revictimization than others.

Lauritsen: Right. The attempt to do that, to distinguish first-time victims, is really difficult conceptually, right?. So if you take the case of intimate partner violence or youth violence at school, how far back do you go to get the so-called initial victimization? Do you go back to adolescence? Do you go back to childhood? Do you go back to infancy? There are arguments and there are correlations of those victimization patterns over time throughout the full life course, but there is no clear information about where we should stop and where we should start that assessment.

I think part of this is because there is no unifying life course data on victimization. The child victimization specialists focus on one set of factors, the adolescents something else, and the adults something else. Where would we go back to? How far? That's the challenge.

David Johnson: Hi. Great talk. My name's David Johnson. I'm with DREL Solutions. My question is about dissociating chronic victimization versus nonchronic. Thank you.

Lauritsen: That's a good question. I'm going to go back to the point I made at the beginning, and that is, for what purpose would it be most useful to distinguish between chronic victims, sporadic victims, occasional, and rarely victimized persons? If the process is research-based, and it's exploratory and descriptive, then we can try all sorts of possibilities, drawing the line at “you need this certain amount of incidents and in this particular time period to be called chronic” versus nonchronic.

You could do all sorts of research. I can do it any which way we would want to, but why do we want to do it? I think we need to think about that. We use the terms all the time, but why do we want to make those distinctions? Presumably it's because we think there is something different about those people. But what may be different about the most chronic victim and the one that's just moderately victimized is they have more of the initial, more of the causal factors in their situation, more of them. Not that there's anything different about their victimization, but that there's just more of the risk factors in their life.

So, I don't have an answer where we would draw that line. We can try.

Gam Rose: Thank you. Gam Rose, DEA. Our interest in your presentation is less direct in what we do than just hearing, but maybe get some ideas from you because we respect your research and want to hear from you, so we're here today, Ann Crutoff and I. My question is, you drew a nice distinction between the two sort of principal categories, one which is more based in in the interactiveness between revictimization and the likelihood of being victimized again — I'm sorry, previous victimization. One of the factors that I think might be interesting in that group, since you kind of despair about the guardianship, that's possible in that group —

Lauritsen: [Interposing] In some situations, yeah.

Rose: Because, for example, in a private domestic situation it's hard to control that. It would be interesting to study a factor — if you can get information on it — about the living arrangement and the number of people and the extended familiness — you know, different cultural living arrangements — that might affect that.

Lauritsen: I agree.

Rose: I don't know if you can do that.

Lauritsen: There's been a little bit of research on this, on whether things like family size —

Rose: [Interposing] Grandparents.

Lauritsen: Right, grandparents, other adults in the household, older children. But children are not good sources of guardianship.

Rose: That's true.

Lauritsen: That's enormous responsibility to put on them, and they're damaged by the experience as well. But I don't know of anything systematic that's been done with a representative sample of, say, families to see whether the presence of additional people might help, or whether they live close by to family members who are willing to intervene in a family situation. I think one of the differences between family violence compared to, say, what they were able to do in the burglary situation, is they're able to tell neighbors. They're able to say to the neighbors of someone whose house was recently burglarized, “This is what's happening in your area. We want you to be on the lookout for all these kinds of things.” This will help reduce your risk, as well, because obviously there's someone in the area.

When it comes to family violence, neighbors are not as a rule, willing to become involved. It seems to be a situation that communities are often afraid of those situations. Women who care about their neighbor are also afraid of the offender. It's really, I think, a wholly different problem.

I think one of the things that repeat victimization studies can tell us is that we really need to think about crime types, like intimate partner violence, rather than things like assault versus robbery versus different other types of crime. We really need to think about types like family violence as a whole, what Jim Lynch calls “domain-based violence,” and where it's occurring. That will help set the parameters about what other sources can be brought in to bear and what can't be. I think that's the direction, but I agree it would be really interesting to know, specifically about neighbors' willingness to be involved in assistance.

Joseph Heaps: My name is Joe Heaps. I'm here at NIJ. On the chart you showed, it seemed that there was an increase between '05 and '06?

Lauritsen: Yes. Part of that increase in '06 is methodological. There was a problem in 2006 in the sense that there was an addition of new sample to the data to increase sample size. That involved new interviewers and new cases. We know that people newly interviewed tend to report more than people who were interviewed for the second time. It appears that the interviewers were a bit more productive in their interviews. There also was in police data a slight uptake in violence in 2006, so it's not fully that, but partly that I think.

Laub: Janet, it looked like you did some modeling. Were you modeling cross interview repeats or where you modeling —

Lauritsen: [Interposing] Yes.

Laub: It would be really interesting to see if you broke the file into person months —

Lauritsen: [Interposing] If I could time it.

Laub: — within the reference period. I think Min Xie did that for her dissertation.

Lauritsen: Right. It would be interesting. You could begin to piece together three-year, perhaps, intervals of victimization patterns. That's what you're referring to?

Laub: Well, I think what it might allow you to do is, at least for when you take your state dependents and heterogeneity distinction, that some of the state dependence variables will vary. The NCVS reference period is broken into six months, so a person could have a victimization in the first month, another one in the second, another one in the third —

Lauritsen: [Interposing] — And then nothing.

Laub: — and then nothing. You would be able to know things like whether they reported it to the police or not.

Lauritsen: [Interposing] — Yes.

Laub: You wouldn't have…the heterogeneity variables would not change, but the state dependence variables would, and it would be very interesting to see whether the person who's victimized the first time, if she reports it to the police, that this doesn't go forward.

Lauritsen: [Interposing] — Right.

Laub: — And I think, then, if you did that it would be…also because the Brits, the stuff with the police data, is in rapid succession — mostly the burglary stuff — but within three days or four days or five days. You're going to miss that if you're going to interview to interview kind of thing.

Lauritsen: That would be nice. That would be great to have that data.

Laub: I think Min has that file.

Lauritsen: The one issue about that, though, is the series victimizations. Right? The series victimizations, we know roughly when they occur, but we don't know how they were distributed across the six months.

Laub: No, but you could use different models. You'd know if it was still going on or not.

Lauritsen: I agree. That would be a useful approach after the other years are done. I think, though, the statistical issues are really difficult. Part of the reason there hasn't been much progress on it is it's pretty challenging. It's not the simple thing you put your graduate student on, for example.

What I do also note — and I didn't get a chance to talk about it — is that violence has come down since the 1990s. The more rare it is, statistically speaking, the more difficult both the research problems become and the programmatic problems, I think. Because it may be the case that violence came down initially, because the kinds of things that would work when more women were exposed to violence essentially took care of the least difficult cases in the population, where those women were more amenable to programmatic interventions. When the rate becomes particularly low, it's likely that those women may be suffering from an overwhelming number of risk factors, and their cases may be more difficult to intervene with. Right? So, we don't know that answer to that question either, but the statistical issues are important for both documenting the individual's history and what might be going on there, but also in the aggregate level, figuring out how much of that is responsible for the decline, for example, that we saw in the '90s.

Laub: Janet, we're going to have to wrap up in a minute. I just want to make one point. I think in addition to the methodological challenges, I think we are also suffering here from a conceptual problem —

Lauritsen: [Interposing] — Yeah.

Laub: — in that we have divided the study of victimization from the study of offending. A key to Liberman's question was really about are the correlates of one-time offenders different from the correlates of career criminals?

Lauritsen: Right.

Laub: And there's a long body of research that we have because people did longitudinal studies starting in the '60s — causes and correlates, project to delinquency, so on and so forth. So I think my take-away from this is that from here on, every study of offending should include victimization…

Lauritsen [interposing]: Yes!

Laub: — including things like the National Crime Victimization Survey. But again, I think if we're really serious about understanding repeat victimization, we have to start in childhood and do these studies over the life course. We've had missed opportunities because we've divided into these two camps: These folks study victims, these folks study offenders, and never the twain shall meet. I think what your research really points out is how we need to move forward and be much more broader in our thinking.

Lauritsen [interposing]: I agree.

Laub: So, I'll get off my soapbox.

Lauritsen [interposing]: I agree.

Laub: Thank you for a wonderful talk.

Lauritsen: Thank you. Thank you.