Overview
Q: Please give us a two minute summary of your research.
A: Sure Ron, the project’s titled "Careers in Crime and Substance Use". It’s a secondary data analysis of a
program conducted in the mid 1970s, in which they tried to give jobs to unemployed addicts, offenders, and dropouts. What I’ve tried to do is study what, if any, are the effects of the experiments on both drug use and crime, and then, specifically, I wanted to answer whether drugs are a cause of crime or whether the two seem to go hand in hand and they’re both just manifestations of the same underlying thing.
Research Questions
Q: What theory or theories, if any, helped you formulate the research questions?
A: Well, I think within criminology Merten’s [?] theory was a starting point and theories of, I guess I’d say choice among socially structured alternatives. Rational choice theory and deterrence theory in crime suggest that the greater the costs of crime, the less one is likely to engage in it. Beyond that, with the analysis of drug use and illegal earnings, we were trying to really merge theories of legitimate attainment with theories of deviant attainment and considering human capital, criminal capital, social capital as predictors of deviant behavior and there drug use took the role as consumption. Those were some of the starting points that we used to get going.
Design Sampling
Q: When you planned this study, what were your research design considerations?
A: Well, as a secondary analysis we had to decide if these data would be appropriate for what we wanted to do or whether we would have to seek either primary data collection or an alternative. So, I guess the biggest consideration was whether to analyze the experiment and to have a standard and to really take advantage of the randomized design or whether to do more of an observational study and we’ve tried to do both.
Q: Are you pleased that you decided to do both?
A: Oh yes, I think…I’d written a piece on causality and crime in which I, the longer I thought about it the more difficult I thought it would be to do observational work, or more challenging to overcome selection processes into levels of our independent variables predicting our dependent variables. So, I thought, well, experiments are at least one plausible way to gain some control over that, but on the other hand, we can’t randomly assign heroine habits, so there really is a need for good observational research.
Q: Could you give us some details about the original study? How many cases there were, what period of time, how the experimental assignment was made and so on.
A: Yeah, I can try and I’m sure there’s others who can do it as well. It was a large scale social program, it was a hundred million dollar intervention, there were ex-addicts, who were referred from drug treatment and social service agencies, criminal offenders who were referred from probation and parole offices, criminal justice service agencies, there were youth dropouts, half of who had an official arrest history, there was also an AFDC sample in which they did not collect crime data, so I haven’t analyzed that, but among the three, sort of deviant populations, there were about five thousand cases, I believe eleven cities, and they were given the opportunity to participate in a subsidized job program that would last between twelve and eighteen months and then they were followed up for up to thirty-six months and there was a randomized assignment process, there was a program staff and an evaluation staff at each side. They were given jobs in crews of eight to ten people, largely service oriented jobs, some construction and the evaluation was taken place in the mid to late 1970s so it was, we have some statistical techniques now that were not available at that time so I think we can do a more fine-grained analysis today.
Measurement & Data Collection
Q: Was there a control group, so some were not assigned to the…
A: That’s right. Fifty percent. So, you’d walk in the door, and you’d go left or right, which is very difficult to accomplish in social services, because of course, we want to be able to provide the best training or experience for everyone who walks through the door. On the other hand, it’s a very powerful design because then you have a real comparison group who would have participated had they not been assigned been assigned with the control condition.
Q: The people that were in the jobs program were paid a reasonable wage, but the people in the control group, were they paid for their time?
A: No, I don’t believe so, I’ll have to check if there was a token interview fee, but those in the job group were given either minimum wage or slightly below market wage jobs so there was a concern that they would be competing with private sector employees.
Q: And the subjects in the jobs group apparently were more likely to stop using drugs or were less likely to start using drugs than the control group, is that correct?
A: Well, among the addict group, when we looked at time until people would use cocaine or heroin, there really wasn’t a program effect, that is, the jobs didn’t help people stay off the drugs. There was a crime effect though, in which those who had the jobs were less likely to reoffend, so the program seemed to decrease crime, but not drugs use.
Q: How do you account for that?
A: Well, I think something the original investigators suggested, that I believe as well, is that there may be an economic mechanism as work, and that is that once addicts have their economic means met, even through marginal employment, they’re less likely to be offending, but the drug use, the affecting drug use, is a much more complicated process and that those, the job program just didn’t touch that.
Q: Now, in this project that you just recently completed, what did you do to ensure a quality measurement?
A: Well, I hope we’ve taken some steps to ensure it, but measurement is a great difficulty here; since the program was in the 1970s, one of the obstacles we had was that we did not have the urine analysis for drug testing that is currently available and used, so we had to rely on self-reported drug use. Alternatively, with crime, we did have self-reported arrest measures and self-reported measures of illegal earnings. With the crime, there were some pains taken at several of the sites to do reverse record checks, in which you’d go to the police department and look up how many people who said they’d been arrested were arrested and how many who said they hadn’t been were not and there was some underrepording of the incidents or the frequency of arrest, but whether one was arrested or not was rather accurately reported. And that’s also been the case with test/retest of self-reported substance use in other studies. Those are concerns that we continue to worry about, but that we try to bring secondary evidence to bear on it wherever possible.
Q: And then you also used this individual case analysis for examining these questions?
A: Yes, that’s right. We started off where we literally tried to compile case histories in which we arrayed every person’s monthly activities over the, up to a three year period. And then, to see well, what did crime go with drug use for people and then analytically we tried to [?] all four thousand nine hundred of these cases and their time lines to look at within person changes, to see whether, indeed, people were using more on average or people were stealing more when they were using more and answer those sorts of questions.
Data Analysis
Q: Could you describe the statistical models you used for your analysis?
A: Well, essentially, the technique is called a fixed-effects model, which is, I understand, more common in economics than in sociology, but it’s variant of change score for [?] different type models, in which there’s an individual effect, so that every person’s activities, or what one is doing in a month is deviated from what that person is usually doing in that month and it’s a way to try to net out stable within person differences. We also worry about that in crime, is that there’s so many things that we leave unmeasured—ambition, intelligence, family background—that, at least if we could look within person for change, that we know that our measured factors are going to…we’ll be able to isolate those effects a little bit better. There’s still unmeasured things that are changing, but it’s a big step in the right direction.
Q: All the subjects were surveyed every month for three years and you looked at each month, statistically, you looked at each month and compared it to the previous month on other variables; was the month your unit of analysis or was the person the unit of analysis?
A: The person-month, the "person hyphen month", is how we would say it. I have to be careful, ideally we would have had data collected at each month, but they were really collected at nine-month intervals and the people were asked to recall the previous nine months activities. So, we suspect that some of the activities were estimated, some were, so the unit is the person-month. As I said, one way we could analyze that is by aggregating all of those observations, person-month observations, but to do that we would be assuming we had a lot more data than we had, because we’d have to net out those persons’ specific effects, otherwise we’d be artificially deflating our standards errors and all sorts of statistical problems.
Q: I’m trying to get a sense of the implication of your procedure, effectively, would it be true to say that your modeling, your fixed-effects model, predicted for each individual, or in other words you estimated an equation for each individual and then you aggregated across all the individuals, so you have both the within and the between individual effects separate, is that correct?
A: Well, I think of it more as one very lengthy linear equation that has an intercept term for every person. And that would be, if I’m predicting illegal earnings, that would be one’s mean illegal earnings in that month, and that intercept would only come into the equation for that person’s specific observations. The other coefficient, drug use for example, would refer to the average increase across person’s, I guess, on their within person gains in illegal earnings.
Q: What were the main conclusions of this analysis?
A: The first one was, the more theoretically interesting conclusion perhaps, is that the same sort of predictors of legal attainment, legal earnings, seem to apply to illegal earnings as well. For one thing, when we modeled the human capital or criminal capital, we fit age and experience terms and their squares and we found the same sort of [?] linear relationships that are observed in studies of legal attainment, that is, diminishing returns to age and experience on the criminal market as well. And there were some other similarities in those analyses to legal earnings, but the main, perhaps, the largest finding statistically was the effect of cocaine and heroin use on illegal earnings among this population. We estimated between four hundred to seven hundred dollars per month additional illegal earnings associated when one is using drugs then when one is not using drugs. This just dwarfed every other effect in our model and so that, I think that’s what jumps off the page both statistically and also perhaps policy-wise.
Q: And this was true whether or not they were in the experimental jobs program?
A: That’s right. The jobs program didn’t appear to affect the drugs use to a dramatic degree.
Q: How about family things, like living with spouse or someone?
A: That was something that emerged as well, that while one is cohabiting, or living with a spouse or partner, that one is much less likely to be earning money illegally, one’s illegal earnings decrease somewhere between a hundred and two hundred dollars less per month while one is living with a partner. It’s difficult, we of course lagged these effects, but it is difficult to sort out causality of whether one gets booted because one has resumed criminal activity or whether one is actively conforming in response to the informal social control [?] that a spouse provides.
Q: Or perhaps the spouse is earning illegal earnings as well.
A: That’s correct and we, unfortunately, have too little individual level data on the spouses to make the determinations. If one’s living with a deviant spouse, it’s certainly possible.
Interpretation & Dissemination
Q: You found that the effect within persons was different from the effect across persons. Could you tell us when that effect occurred and what it means?
A: Some things that we think of as being powerful predictors of crime seem to wash out in the within person analysis, for example, the effects of perceived risk, on average, people who perceive greater risks, or associate greater risks with criminal activity are less deviant, but within person fluctuations in those risks did not predict the amount one would earn illegally, to a much lesser degree. Similarly, the perceived frequency of illegal opportunities, so if one sees how often you have an opportunity to earn money illegally. Those who see frequent opportunities tend to earn more than those who don’t see frequent opportunities. On the other hand, within persons, those fluctuations do not predict illegal earnings, so in general, the across person picture is a bit different than the within person picture.
Q: What’s more important for a sociologist or social scientist—the within person or the across person?
A: It’s an interesting question, I think looking across persons allows for what I call statistical discrimination in which we can identify persons on a basis of their group means, who are more likely to be committing crime and I think, in the paper we say that might be useful for, I suppose, law enforcement profiling. In the within person picture, you identify the things that seem to change when criminal behavior changes and that might be more useful for those who are involved in actually trying to rehabilitate offenders, who are doing hands on programming. Now, sorting out causality is quite difficult here, but I think the within person picture gives us a better, gives us a more leverage to making causal inferences than the across person. I think they’re both sociologically interesting for descriptive purposes as well as inferential ones.
Q: It sounds like this is an important study because you were able to look at both kinds of effects and because in most studies it’s not possible to look at the within person because they don’t have the time series or longitudinal data.
A: Right. I think it’s an important advance, we have to be cautious, some things we’re not sure that even though that in the world we conceive of them as continua, that we’re only measuring them in very crude, discrete increments. Perceived risks, we might be bumping up against ceiling effect, that could be a very sensitive indicator that’s predictive and I have a five point ordinal scale that I’m using to measure it. But, I think, that said, yes it is potentially very powerful; the data requirements are fairly stringent, but I think it does give us a view that we wouldn’t have had otherwise.
Q: What are your primary audiences for this research?
A: There’s two: that’s the social scientific community and by that, I hope I’m speaking to sociologists as well as to psychologists, economists, all who study crime. The other primary audience is people, like those at the National Institute of Justice, who funded the research, who are involved in policy making and program design and evaluation.
Q: Now that you’ve completed this project, what’s the next logical step and are you going to do it?
A: I haven’t quite completed, we have the final report to write and some papers to submit, but the next logical step analytically is that even though we’ve made some strives in analyzing these process, there’s quite a bit of disaggregation to do and one of the things we want to test is the generality of our model of earnings attainment and of crime by looking at specific offending and offense types. Are those who earn their money from the drug trade, do we observe the same predictors as those who commit robbery and burglary? My research assistant is currently trying to sort the offense types into conceptually meaningful, theoretically meaningful, subtypes and then we have some hypothesis about which predictors will be especially strong for certain types of crime.
Q: What does your research imply in terms of the legalization of drug controversy?
A: Well, it’s a difficult question to answer, it’s something we’ve given some thought to. The very large effects of drug use, specifically cocaine and heroin use, on illegal earnings suggests that first that, I think there’s an economic incentive to providing treatment or anything that’s been shown to be effective in reducing drug use. I’m a little nervous on speaking on legalization, partly because we don’t whether, I think legalization may indeed reduce the cost of drugs, which may in turn reduce the amount or cost of the crime associated with drug use. On the other hand, it’s unclear whether new users would be induced to take up drug use in a legalized environment, so we’re very cautious in advocating the policy. On the other hand, I think, programs such as methadone maintenance for serious opiate users would be—I shouldn’t say encouraged by these results—but these results would probably support those sorts of programs.
Q: Are there any other policy implications of your study that you haven’t mentioned?
A: Well, I think that we talked about a little bit was [?] the marriage effect, the idea we said before the cameras went on that half a million people are released from prison each year and we know very little about them and I think we’ve talked about the economic support and the effects of employment, but I think there’s also room for family interventions and for family support services that may play a very important causal role in [?], or cessation from crime, and so I think some of the within person analysis suggests that indeed, if we can keep families together, if we can keep particularly fathers in that breadwinner role, that it does engender less crime.