The Micro Domination
Despite the development of important sociological explanations of crime (e.g., Braithwaite 1989; Burgess and Akers 1968; Cloward and Ohlin 1960; Cohen 1955; Merton 1938; Messner and Rosenfeld 1994; Shaw and McKay 1942; Sutherland 1945) and the recent influential work of criminologist Robert Sampson (1986, 2000, 2002; Sampson and Groves 1989; Sampson and Laub 1993, 1994; Sampson, Morenoff, and Gannon-Rawley 2002; Sampson and Rauden- bush 1999), the discipline of criminology is largely dominated by microlevel research and explanations. We define microlevel explanations of crime as those that focus on individuals and individual explanations of crime.
How much does the microlevel approach dominate criminology? For example, in 2013, the journal Criminology—arguably one of the most influential journals in the discipline—published 27 articles. Seventy-five percent of those articles studied individuals such as boys in gangs, offenders, middle school students, men of color, Finnish males between the ages of 20 and 30 years old, inmates, juvenile delinquents, police, and individuals with an arrest history. Only a few studies departed from these microlevel studies to examine such things as a count of hate crimes, days of rioting, or the number of violent crimes in a hotspot. Neighborhood-fevel studies of crime only appeared in 2 of the 27 studies in the journal. In short, contemporary criminology remains dominated by microlevel research and explanations of crime that explain individual “choices."
The domination of microlevel research is somewhat of a puzzle from an empirical perspective. There is little doubt that microlevel explanations have logical appeal, and the idea that criminals are different than those who do not violate the law is an assumption that has broad acceptance. That assumption is questionable with regard to the distribution of crime, however, since we know from self-report studies that participation in crime is widespread among the public, suggesting perhaps that crime is influenced as much by social structure as it is by variability across individuals. Less widespread is the form of crime that would identify an individual as a career criminal. That form of criminality may require individual-level explanation. At the same time, depending on how rare that kind of behavior is, it will be more difficult to predict. Nevertheless, if one examines the content of criminological literature and approaches its findings objectively, one could say that empirically, microlevel explanations of crime (there are also macrolevel explanations of crime to which this observation also applies) tend to produce weak results.
By “weak results" we mean two things. First, the combined impact of the relevant explanatory variables is weak insofar as these variables typically explain a small amount of variation in crime across individuals. Criminologists often measure what amount of crime they explain using various goodness-of-fit measures (Hagquist and Stenbeck 1998). These goodness-of-fit measures include R-squared, adjusted R-squared, and various pseudo-R-squared measures. Goodness of fit is useful for illustrating the general fit of a model to a set of data. And one should expect researchers to provide this information as part of the assessment of the appropriateness of the model from which conclusions will be drawn. While we take the position that a true model exists, we recognize that this position is debated (Berry 1993, 30-33).
Second, by weak results, we also mean that the results for individual explanatory variables in the model are impacted by the underestimated nature of the models—that is, models that criminologists produce tend to have goodness-of-fit statistics that indicate the models are not correctly specified and depart significantly from the “true model” (Achen 1990). We interpret these results as suggesting that that there is a strong likelihood that the model suffers from omitted variable bias (i.e., a problem of specification error), meaning that important factors that explain crime are missing from the model. One such factor may have to do with the definition of crime and forces that impact the political construction of crime. Since the definition of crime is a social construction that varies over time and place, individual-level variables that are used to predict crime are not necessarily likely to produce strong empirical results. When variables that would improve the fit of the model are omitted (i.e., those that explain the construction of law and its content), the empirical results for the individual variables in the model are likely to be improperly estimated. This raises questions about the empirical validity of the results of many empirical tests in the criminological literature. Thus we are concerned with the interaction of a weak-fitting model and its impact on the estimates derived for any given independent variables in the model. Instead, we argue that criminology is built upon theoretical models that are determined by “the best-fitting model” so that explanations of crime are based on significant variables included within a model that is assumed to contain all the important variables necessary to predict an outcome so that no important explanatory variables that may alter the model are omitted (Berry 1993).
When the equations used to predict crime perform poorly as a result of misspecification, researchers ought to employ sufficient caution when interpreting the empirical results. We examine this problem in greater depth later in this work. While some researchers might suggest that weak results are typical of criminological research that predicts the causes of crime, the larger question is not whether these results are usual, but whether they ought to consistently be accepted as empirically and scientifically valid indicators of the utility of the causal explanation being examined regardless of their normalcy. And as of yet, criminologists have not agreed on some standard that ought to be met empirically and instead leave that determination up to individual readers, researchers, and peer reviews to determine.
The defense of weak empirical results at the microlevel suggests that these kinds of results are typical within the criminological literature. By treating these results as acceptable, researchers consent that weak to moderate goodness-of-fit statistics are the best that can be expected in criminological research (R-squared, for example, rarely exceeds 0.5 and it is often around 0.1). This response implies that there is no need to consider that extant empirical results may indicate that these kinds of causal explanations are weak and that they are not, therefore, the best way to explain crime.
Having reviewed our position on the measurement and standards for the measurement of statistical effects, these concerns can now be related back to the subject of this book: the definition of crime. It is possible that weak results for individual-level explanations of crime may have something to do with the nature of the traditional definition of crime and how that relates to testing individual-level theories of crime. Criminologists tend to address weak results produced by their studies by investigating new methods of analysis in the hope that those methods will change the outcome. This is unlikely, since a new statistical approach is based on the same set of probability relationship found throughout the data, and a new method of analysis may not reveal new relationships between the variables and is more simply an alternative means of assessing those relationships. This is not to say that a new method of analysis never works in the way they are expected to work. The potentially larger issue here is that the relationship between microlevel causes of crime and crime can originate with the definition and measurement of crime itself. The new methods address analytic procedures and not the problem of crime measurement and it is for this reason that the new method may yield a similar outcome.
There is also a logical problem hiding behind the typical empirical results, the nature of the explanations employed to explain crime at the microlevel, and the definition and measure of crime. This significant logical problem has to do with the inconsistencies between the concept and measurement of crime and whether the legal measurement and concept of crime is logically consistent with the effort to explain crime at the microlevel. That logical inconsistency is this: the legal definition of crime is a political construction and is influenced by structural processes while individual-level explanations of crime are not typically political and do not include variables that explain how political processes shape crime. Thus we can say that in effect criminology looks for individual-level behavioral explanations of crime to explain a socially and politically generated outcome. Ignored in these microlevel explanations are those variables that may influence the making and application of law and how that process influences the distribution of crime.