What is the cause of crime? One can find answers in many places: in Genesis, in the plays of Sophocles and Shakespeare, in Dostoyevsky’s Crime and Punishment and Brothers Karamazov. The many answers include impiety, appetite, rage, and even the lure of transgression itself. For many, religious and poetic insights still offer the most compelling explanations. In the eighteenth-century, Enlightenment scholars and intellectuals began seeking a different kind of answer—one rooted in systematic inquiry and rational study. Eventually these inquiries crystallized as the field of criminology, a distinct branch of social science, interwoven within the history of sociology.
Within the field of criminology today the dominant answer to “What causes people to commit crime?” includes three elements: blocked opportunities, delinquent peers, and delinquent labels. That’s the story leading criminology textbooks and theorists tell. Oddly, however, the supporting evidence adduced by criminologists falls far short. Most of it is marred by a deep systematic error that confuses cause and effect.
This doesn’t mean that the Enlightenment-inspired search for social scientific explanations of criminality was misguided. To the contrary, we can learn some important things from the empirical and statistical study of crime. But honest twenty-first century criminology will have to review the “blocked opportunities, delinquent peers, delinquent labels” explanations with renewed skepticism. Criminology would do better to reexamine the views of its founders, who believed that criminal behavior was a function of individual choice.
Rational Choice Roots of Criminology in the Nineteenth Century
Criminology in the nineteenth century could hardly be characterized as a “sociological” discipline. Due largely to the influence of Cesare Becaria (1738–1794) and his “classical” school of thought, the prevailing view of the criminal was that of a rational and hedonistic individual who possessed a will free to choose between criminality and conformity. Discussion seldom focused on societal limitations on the choices the individual could make. Indeed, there was an insistence that an unwavering scale of exact punishments for equal acts be administered without any consideration to the individual involved or any special circumstances he may have faced.1 The early criminologists were a far cry from the kinder, gentler ones we hear from now.
In the first days of the discipline it was hoped that punishment—swift, certain, and severe—would deter people from making the wrong choice of criminality over conformity. If internal mechanisms taking the form of moral (generally religious) objections to crime proved insufficient, it was the government’s job to frighten people into making the right choice.
When, later in the nineteenth century, the emphasis in criminology shifted from a classical view to a “positive” view—from criminal behavior as freely chosen to criminal behavior as caused—the discipline remained rooted in the belief that a criminal was primarily to be understood as a predator from whom society needed to protect itself.
Cesare Lombroso (1835–1909) led the transformation of the discipline in the wake of Darwin by asserting that criminals were less “evolved” than their non-criminal counterparts. While willing to regard crime in a more deterministic way, Lombroso left no room for a discussion of societal limitations on the human potential to conform to the law.
Perhaps by relieving them of the need to fund costly reforms, the concept of the “born criminal” appealed particularly to the upper classes. However, by focusing on readily identifiable physical characteristics—those evidencing a more “primitive” stage of evolutionary development—Lombroso’s work also appealed to every class. Who wouldn’t want to be able to identify some external physical characteristic associated with a crime such as rape or murder? If in fact correct, this theory made the task of avoiding the dangerous predator much simpler. Just keep your eyes peeled and run from those possessing criminal traits.
The notion of rehabilitation was therefore rendered irrelevant—or at least less relevant—in the minds of those who perceived, and preferred, an opportunity to avoid criminal victimization. The preference is easy to understand.
Similar to views concerning physical appearance, theories relating to mental deficiency became popular just before the end of the nineteenth century. Richard Dugdale’s 1875 family study, The Jukes: A Study in Crime, Pauperism, Disease and Heredity, provides one example of the effort to link low IQ to crime.2 After finding six members of a family in a New York jail, Dugdale researched the family’s history over a period of two hundred years. He suggested that the family was “degenerate” and prone to imbecility and criminality. Dugdale’s discussion of the link between intelligence and criminality did much to reinforce the notion that charity was actually bad for society. The absence of social programs would lower the chances for “degenerates” lacking the intelligence to make good choices to survive and reproduce. According to this idea, efforts to produce a “Great Society” would actually create a “Retrograde Society.”
“Survival of the fittest” may have been little more than a rationalization for those who did not want to bear the burden of rehabilitation on behalf of “society.” Regardless, this viewpoint starkly contrasted with twentieth-century theories, which regarded society rather than the individual as the cause of criminal conduct.
“Sociological” Criminology of the Early Twentieth Century
Most influential theories of the twentieth century took a more innocent view of human nature, which resulted in the reluctance to hold an individual responsible for his criminal conduct. To the extent that the individual had “appetites” for crime, they were considered to be “culturally induced” as opposed to “natural” tendencies
According to Robert K. Merton, social structure imposed limitations on the individual’s ability to satisfy those appetites.3 This tension, or “strain” as it became known, in turn resulted in crime. Thus, the concentration of crime in the lower classes was regarded less as a function of low IQ and other individual deficiencies and more due to the absence of legitimate opportunities for achieving goals, such as personal wealth, in lower segments of society.
Just a few years after Merton presented his strain theory, Edwin Sutherland published his “differential association” theory.4 Sutherland adopted a similar view of human nature—that criminal tendencies are not inborn—and his theory suggested that both the motivation and the techniques used to commit crime were transmitted to the “good” individual by the “bad” society. How the sum of “good” individuals equals a “bad” society is nowhere explained by differential association theory—or any other theory, for that matter.
Since all people are exposed to pro-delinquent and anti-delinquent influences, delinquency is explained by Sutherland as “an excess of definitions favorable to violation of law over definitions unfavorable to violation of law.”5 Later known as the “principle of differential association,” this was a fancy way of saying that delinquency occurred when bad influences outweighed good influences. Obviously, this principle characterizes crime as a perfectly normal response to an abnormal environment, and diverges from the nineteenth-century conception of the criminal as an individual who makes bad decisions as a result of irrationality or psychological abnormality.
“Labeling” theory emerged in the 1930s by reclassifying yet another presumed consequence of crime as a cause. Edwin Lemert would eventually paint a more detailed portrait of labeling theory than some of its earliest proponents like Frank Tannenbaum. In the process of expanding upon that earlier work, Lemert would make a crucial distinction between “primary” and “secondary deviance”—the latter term encompassing what is perhaps the principal contribution of the labeling perspective.6
For Lemert, virtually everyone would—at one time or another—commit crime as a result of biological, psychological, or social factors. Crime resulting from one of these broader causes was dubbed “primary deviance.” It was assumed, rather safely, that among those committing acts of “primary deviance” only a subset would be caught. When caught, these individuals would likely be subjected to a greater degree of scrutiny than their counterparts whose crimes go undiscovered.
According to labeling theorists, the process of getting caught in an act of primary deviance and being subjected to “formal labeling” (probation, prison, etc.) or “informal labeling” (ridicule, ostracism, etc.) had further consequences. Among the possible consequences: lower self-esteem, identification with the criminal label, and exclusion from certain segments of society. Any subsequent deviance beyond the original cause (primary deviance) was identified as “secondary deviance.”
Taken together, these twentieth-century theories about the origins of crime called for greater social intervention than nineteenth-century explanations. If “bad” society could create crime, then social engineering could undo the damage. Sociologists and criminologists would be well-suited for the task, if indeed they possessed a better understanding of the “root causes” of crime.
Consider the key differences in the classical view, the positive view, and the sociological view of crime causation:
Classical View: The decision to commit crime has consequences that include blocked opportunities, ostracism from conformists, and the attachment of a delinquent label.
Positive View: Genetic factors including, but not limited to, low IQ affect the decision to commit crime. Such decisions have consequences that include blocked opportunities, ostracism from conformists, and the attachment of a delinquent label.
Sociological View: Sociological factors including blocked opportunities, association with delinquents, and the attachment of a delinquent label exert a causal effect on crime.
Naturally, any discipline that wishes to call itself scientific must be willing to collect data that helps reconcile these very different explanations of crime causation. In the late twentieth century, criminologists would spend a lot of time—and no small amount of government funds—doing just that.
Quantitative Criminology in the Late Twentieth Century
With the explosion of journals publishing sociological and criminological research in the twentieth century, social scientists found many outlets to produce a wealth of literature on the topic of crime causation. Most of this research would take the form of self-report questionnaires administered by unofficial (non-government) entities. These self-report questionnaires simply asked respondents to report how much crime they had committed, usually in the previous year. While official (government collected) sources were established in the twentieth century, they simply were not able to answer—indeed, they did not even seek to—complex questions about crime causation.
Although self-report questionnaires were developed in the 1940s, they were not widely used to address questions of crime and delinquency causation until the 1960s. Even then, self-report questionnaires suffered from serious flaws, including an undue emphasis on trivial forms of crime and delinquency and a tendency to focus on simplistic “yes/no” response options when inquiring about particular criminal acts committed during the past year. For example, a question might read: “Yes or No, in the last year, did you strike a member of your family in anger?”
One can easily imagine how focusing on trivial items—especially with a measure that assigns a maximum outcome for a single offense—could result in unrealistically high scores. It follows that this could also reduce the amount of variation between different respondents. Such a result is problematic for those wishing to study, for example, the relationship between social class and crime. The class/crime issue is especially well-suited for analyses employing the self-report method. But any class differences in crime go undetected if flaws in the instrument hide the true extent of class variations.
Changes in the nature of self-report measures during the 1970s significantly attenuated these difficulties. By (typically) employing five-item ordinal dependent variables—and also by focusing on more serious forms of predatory crime—criminologists were now able to address previously inaccessible issues. But the discipline still had one major issue left to resolve. This involved the continued use of cross-sectional studies, which survey subjects only once, rather than longitudinal studies, which include follow-up surveys. These longitudinal studies, rather than mere correlational analysis, are crucial to assessments of true cause and effect.
The fundamental difference between nineteenth-century and twentieth-century views of crime causation rests on whether certain bad social outcomes (blocked opportunities, labeling, association with delinquents) precede or follow criminal involvement. Hence, an honest assessment of the accuracy of either perspective requires the use of longitudinal data that measure both crime and key theoretical variables at more than one time interval.
For example, subjects may be asked “How many of your friends smoke marijuana?” as one measure of the independent variable of delinquent peer associations. They can be asked “How often in the last year did you smoke marijuana?” as one measure of the dependent variable of juvenile drug use. But this does not put the researcher in a position to evaluate cause and effect. One can only speak in terms of correlation.
The problem is resolved when employing a longitudinal study, which re-interviews subjects a year later. In such a design, one can simply use the independent measure (peer drug use) from the first survey and the dependent measure (respondent drug use) from the second survey. Here, cause clearly precedes effect.
Researchers in the field are sometimes able, at least somewhat, to control for the lack of longitudinal data by using “age-stratified inquires.” For example, they may ask fourteen-year-old boys about their career aspirations. They may then ask forty-year-old men about their actual accomplishments. But such studies merely note contemporaneous events and posit one as the cause of the other. The “longitudinal panel study”—which poses the same questions to the same (panel of) subjects over time—allows for statistical control. For example, one can measure for levels of delinquency prior to and after measuring the dependent variable.
Researchers cannot control for everything because they cannot measure everything, but they are in a better position to make causal arguments by measuring and controlling for some potentially spurious variables. If, on the other hand, modern criminologists simply ignored the issue of causal ordering of key variables they would be violating standards of professional competence. If modern criminologists simply pretended their theoretical variables predicted, without measuring them prior to, delinquency, they would be violating standards of professional ethics.
Ronald Akers and John Cochran published “Adolescent Marijuana Use: A Test of Three Different Theories of Deviant Behavior,” a 1985 study which raised potential ethical questions concerning practices common in the discipline of criminology by the end of the twentieth century.7 After warning readers that researchers must address the “usual caveats” of using cross-sectional research to make causal arguments, they disregard their own advice in a way that allows them to assert that data supporting a nineteenth-century view of crime causation instead supports a twentieth-century view.
In an effort to compare the relative predictive value of “social learning” theory (versus “social control” theory and strain theory), the Akers and Cochran study reports that social learning theory explained a remarkable 68 percent of the variance in self-reported marijuana use. This was much higher than the modest level of variance explained by social control theory and the very low level of variance explained by strain theory.
Akers and Cochran asserted that factors such as the modeling of marijuana use by one’s peers led to a higher level of self-reported marijuana use. Unfortunately, the level of marijuana use by peers was measured in the present tense. Self-reported marijuana use was assessed by asking subjects how often they smoked the drug over the past year. The researchers’ reasoning was backwards. The data, properly interpreted, should lead to the opposite conclusion—that those who smoke marijuana are more likely in the future to associate with those who also smoke marijuana. In other words, certain consequences attach to the decision to smoke marijuana. Among the consequences is an unwillingness of conformists to the law to associate with non-conformists—in this case, marijuana smokers.
So, there really wasn’t anything remarkable about the predictive efficacy of social learning variables in the Akers and Cochran study. These variables were actually predicting the past, which is hardly a scholarly achievement. Anyone can do it.
The Akers and Cochran study also misinterprets the relatively low level of variance explained by the model employing strain variables. Their observation that strain theory was unsupported by the data was meant to convey this assertion: perceptions of blocked educational and occupational opportunities do not lead to marijuana use. However, these perceptions were also measured in the present tense while self-reported delinquency (acts subjects reported they committed with no indication that they were actually caught committing them) was measured over the course of the preceding year. Hence, the only conclusion warranted is: youths who are not caught smoking marijuana do not expect their conduct to result in blocked educational opportunities or blocked occupational opportunities.
While the Akers and Cochran study is nearly twenty-five years old, more recent studies fall prey to similar errors of interpretation. For example, in a 2007 article Ieva Cechaviciute and Dianna T. Kenny claim that “techniques of neutralization”—or rationalization for criminal conduct—are strong predictors of self-reported delinquency.8 They also incorporate present-tense labeling measures when asking respondents whether they agree with the statement “Most people think I am a delinquent.” This is done in conjunction with a self-report scale measuring delinquency in the past tense.
Despite the fact that they measure their dependent variable before crucial independent variables, Cechaviciute and Kenny wrongly report that their results are consistent with labeling theory, which suggests that being labeled delinquent is a cause, not a consequence, of delinquency. The authors also incorrectly posit that certain neutralization factors are predictors of self-reported delinquency. Again, these predictors follow rather than precede self-reported delinquency measures. Finally, they urge that “caution must be observed in interpreting the findings of (their) correlational study.”9 They claim that it is quite possible, for example, that prior delinquency could facilitate the development of perceived delinquent labels, rather than vice versa.
But Cechaviciute and Kenny understate their case. More than mere possibility, that prior delinquency facilitates the development of perceived delinquent labels is what the data in their study actually demonstrated—conclusively and in direct contrast with their published interpretation. Their conclusion that labeling affected delinquency is not a possibility given the reversed measurement of key variables.
Unfortunately, sociology and criminology journals are filled with so many studies making the same fundamental error that a comprehensive survey of such fatally flawed literature is impracticable. However, studies using similar theoretical variables, but which also employ longitudinal measures, do exist. It is worthwhile to examine a couple.
Jón Gunnar Bernburg, Marvin D. Krohn, and Craig Rivera conducted a longitudinal test of delinquency using variables from both labeling and differential association theories.10 The study found that juvenile justice intervention is associated with greater subsequent involvement in delinquent networks. Such networks include membership in delinquent gangs.
Perhaps more important, this longitudinal study shows that increases in subsequent gang involvement and involvement in informal delinquent networks mediate the effects of formal labeling on delinquency. My own research, conducted prior to this study, demonstrates that informal labeling effects on delinquency are mediated by subsequent delinquent peer associations.11 In other words, when longitudinal—not cross-sectional—studies are considered, formal labeling does not seem to be a direct cause of delinquency. Nor does informal labeling seem to be a direct cause of delinquency.
There is also some longitudinal evidence that suggests that engaging in delinquency makes one more likely to spend time socializing with other delinquents in the future. Using longitudinal data, Cesar Rebellon conducted a study of male delinquents.12 Results indicated that subsequent increases in socializing with other delinquents are due not to their own desire to socialize with delinquents, but instead to the fact that only delinquents are willing to socialize with them.
It is interesting to observe the way that differential association/social learning variables perform in the presence of other theoretical variables—both in cross-sectional and longitudinal studies. Clearly, studies like the Akers and Cochran demonstrate that strain variables lose considerable predictable efficacy in conjunction with such theoretical variables. In the more sophisticated longitudinal studies mentioned earlier, labeling variables—both formal and informal—lose their power when differential association/social learning variables are added to the equation.
The paucity of longitudinal research in criminology journals speaks volumes about the academic and ethical climate of the discipline in the twentieth century. It also clearly points to the future direction of criminology.
Quantitative Criminology in the Twenty-First Century
The discipline of criminology desperately needs to take a new path in the twenty-first century, largely because it continues to tolerate the use of the same obvious methodological errors while deeming itself “progressive.” This is surely not progress. No discipline can hope to move forward by testing its theories backwards.
By the mid-1980s—when Akers and Cochran published their study on social learning and marijuana use, for example—criminology had developed sophisticated measures of key independent and dependent variables. But criminologists continued to make these three fundamental errors, listed from least to most egregious, which rendered their research suspect:
Conduct cross-sectional studies in order to test cause/effect relationships. This is certainly understandable, given that longitudinal studies are so much more expensive and time consuming to conduct. The pressure to publish, especially among untenured researchers, makes this the least harmful of the three errors.
Warn that the cross-sectional nature of a study merely causes ambiguities when interpreting cause and effect. It does not. While it may appear that measuring the dependent and independent variables at the same point in time produces ambiguities, the problem goes deeper. Given that the dependent variables are always measured in the past tense—respondents cannot be asked whether they are presently committing a crime or plan to in the future—the “effect,” crime, actually precedes the cause. Since this is impossible, it cannot be an effect. It must be a cause. There is nothing complicated about this problem and ignoring it will not make it disappear. Regrettably, most researchers end up doing just that in their studies, which is why this error causes greater concern.
Claim falsely that results confirm a sociological view of crime causation. When the direction of the independent and dependent variables actually reverses, the researcher risks confirming the classical perspective while claiming to confirm the sociological perspective. This error causes the greatest concern because it points to the researcher’s methodological incompetence or intellectual dishonesty.
More than twenty years separate the publication of the Akers and Cochran and the other cross-sectional study examined in this piece, that by Cechaviciute and Kenny. Both make the same three fundamental errors described above. The reason for this is simple: similar examples are simply too numerous to review. In other words, these egregious errors have become institutional norms within the discipline of criminology.
One can get a sense of just how embedded these norms are by looking at the texts used to teach criminology to undergraduate students. For example, Criminology, by Piers Bierne and James W. Messerschmidt, employs thirty-five scholarly references in its presentation and evaluation of strain, labeling, and differential association—the three sociological theories discussed in this article.13 Predictably, only six, or 17 percent, of those thirty-five references actually use data to test the theories. When they do, the studies using cross-sectional analysis outnumber the studies using longitudinal analysis by a ratio of five to one.
This is a very serious issue. Most college students who take a course in criminology will take only one. Texts like the Bierne and Messerschmidt spend an inordinate amount of time on theories with decidedly leftist policy implications—as opposed to classical theories with very different policy implications. Due to methodological error, they also present studies that disproportionately measure theories in a way that cannot confirm sociological theories but can only potentially confirm classical theories. Despite this, Bierne and Messerschmidt do not even hint that when routinely measuring a theory backwards one runs the risk of confirming a different theory.
And that is precisely why a change is so desperately needed—one that must take the form of an exclusive focus on longitudinal studies when exploring issues of crime causation. Because such studies are more expensive and time-consuming it means that criminology will produce less research. But this is not a disadvantage if it means that criminologists will be producing less shoddy research.
We have already gotten a glimpse of what longitudinal studies might tell us about the relative predictive power of various sociological theories. But even before that research is examined, studies like the one by Akers and Cochran show just how weak strain theory is relative to differential association/social learning theories. The handful of longitudinal studies suggests that formal labeling and informal labeling theories also fare poorly in relation to differential association/social learning theories.
Sooner or later, criminologists must get down to the serious business of expanding the literature based on legitimate longitudinal analysis. When the dust settles, differential association/social learning theories may perhaps be the only sociological theories left standing. But that will only mean we have answered the question of how criminal behavior is transmitted, not how it is initiated.
Criminologists may eventually discover that the end of the road leads back to the beginning—to an individual choice to commit crime. Perhaps that explains why they choose to travel so slowly. Clearly, their desire to advance progressivism is preventing academic progress.