Predicting Crime: The review of Research



6.1 Summary and Analysis: Methodological Design

This section summarizes the research and analytical methodologies used to develop crime trend series and make predictions about the future of crime. This analysis includes a discussion of the strengths and weaknesses of the research methods, findings, and predictions themselves.

Predicting the future of crime is growing in popularity, as can be seen when one compares the amount of literature dedicated to the future of crime that was published during the 1990s with that available in previous decades. However, despite the continual refinement of the increasingly scientific forecasting methods, it is likely to continue to prove extremely difficult to generate accurate, long-term forecasts of the nature and scope of crime.

Although all uncertainty cannot be removed, it is still possible to systematically formulate a range of possibilities using established methods and analytical tools. The tools of the disciplined futurists, according to Cole (1995), are a sound methodology, a sense of history and theory, knowledge of key factual data, and the ability to examine crime in the contexts of broader social, political, technological, and economic trends. The data sources and methods used to guide forecasting include crime statistics; surveys of experts, practitioners, and the general public; literature reviews; scenario writing; and statistical (time series) models that extrapolate crime trends into the future.

Following the adage “the best predictor of the future is the past” is a common method of developing crime forecasts: extrapolating historical and contemporary trends into the future. Mathematical models that describe the behaviour of observed past values can be used to forecast future crime trends by projecting a time series analysis of crime trends into the future. In general, the source of quantitative time-series forecasting is police- and victim-reported crime statistics. Modelling consists of describing the causal sequence of variables and the prediction of their interactions. Any predictive model endeavours to show a relationship between certain independent (predictor) variables and a dependent variable (i.e., the criterion to be predicted).

It is universally acknowledged that simply to extrapolate past trends is an unsatisfactory predictor of the future; to maximize the validity of crime predictions, time series models must take into consideration the broader context within which crime exists (Britt, 1995). In other words, these models must account for the broader social, economic, political, and technological factors that produce crime. As such, to ensure greater accuracy, these models must identify and predict the scope and nature of a number of factors that will influence crime and victimization in the future. Indeed, many of the time series models identified in this research are based on correlating crime trends with trends in major predictive variables, most commonly the strength of the economy and demographics, particularly the size of the crime-prone population.

The principal strength of quantitative models, in relation to qualitative forecasts, is that descriptions of future crime rates are much more specific and precise, although not necessarily more accurate. By extrapolating crime trends and adjusting them according to influencing variables, predictions can be made with greater empirical validity than by relying exclusively on qualitative methods.

The most significant weakness of the predictive aspects of these models is the difficulty of identifying, anticipating, and factoring in the impact of the variables expected to influence future crime trends. The quantitative models identified in this research did not, and probably could not, factor in the vast array of variables that may influence crime; most of the time series models used to predict future crime rates only took into consideration one or two key influential (demographic and macro-economic) variables. While there is substantial evidence of the impact that these variables have on the crime rate, a key reason they are used in crime forecasts is that their trends can also be quantitatively documented, and hence, forecasted into the future. What is absent from these mathematical models is the integration of influential variables that are more difficult to quantify through historical time series data such as technology, life-style changes, criminal justice responses, and the level of private security and crime prevention efforts undertaken by the public.

Finally, quantitative models suffer from the inability to anticipate unpredictable future events, including unforeseen technological advances, economic vicissitudes, social trends, and advances made in law enforcement and private security technology. The problem inherent in correlating crime with influencing factors is that the future movement of these extraneous factors must also be subject to tenuous predictions in order to gauge how their future direction will influence crime. Despite these problems, an extrapolation of past trends is still informative as it describes some of the underlying pressures on crime, and with careful interpretation, can provide a useful baseline for developing forecasts (Dhiri et al., 1999). Ideally, time series forecasting should develop a range of predictions based on various optional models of how extraneous variables will move in the future and the impact of each of these on crime.

The predictions reviewed in this research offer various forecasts that range from a decrease to an increase in crime. They are generally based on differing assumptions and the margins of error inherent in mathematical modeling. For example, the mathematical model of Dhiri et al. (1999) offers substantially differing crime predictions. At one extreme lies their prediction that the number of recorded burglaries and thefts in 1999 and 2000 will increase by approximately 40 percent when compared to 1997. Alternatively, they project that burglary might fall below the 1997 level. Abrahamse (1996) projects homicide arrest rates in California until 2021 using pessimistic, nominal, and optimistic assumptions. Under the pessimistic assumption, by 2021 homicide arrest rates will nearly double the 1994 rate; under the nominal assumption, homicide arrest rates will be about 28 percent higher in 2021 than in 1994; and under the optimistic assumption, homicide arrest rates in 2021 will be about 14 percent below 1994 levels.

Crime trend analysis and statistical modelling may provide one means of estimating future crime rates. However, there is a need for a rational identification of new targets for criminal activity that keeps pace with changes in social mores and technological innovation. This type of prediction is best accomplished through qualitative methods. The principal weakness of qualitative models is that they are often restricted to generalities; any attempt at predictions of actual crime rates simply involves educated guesses by experts.

Below are some of the qualitative methods that have been used to forecast crime:

Environmental Scanning
In the context of developing predictions, environmental scanning represents a systematic effort to identify future developments (i.e., trends or events) that could plausibly occur over the time horizon of interest and whose occurrence could alter a particular environment in important ways. Such developments may come from a number of domains, including economic conditions, demographic shifts, government policies and enforcement resources, international events, social attitudes, technological advances, and so on. In essence, the scanning process is concerned with identifying future conditions that could emerge and how they affect the phenomenon that is the focus of the forecast. Future researchers are particularly interested in identifying those developments-whether of low or high likelihood of materialization-that are capable of producing the most significant changes in the character of the issue that is the subject of the forecast. Two methods are most frequently used in environmental scanning: reviewing and synthesizing the literature in disciplines relevant to the issue at hand and gathering the opinions of experts through techniques such as Delphi Groups (Cole, 1995, 7).
Nominal Group and Delphi Technique
Surveying the opinions and judgments of experts about future developments has been used in environmental scanning, as well as for providing direct predictions about the future. Various techniques, including questionnaires, telephone conferences, and face-to-face group meetings, can be utilized to capture expert opinions and encourage discussion and consensus. The Delphi process is frequently used for this purpose. Delphi-which takes its name from the ancient Greek oracle-is a technique by which a panel of experts is convened to examine and debate the probable impacts of a series of possible future developments. The Foresight program (2000b) relied heavily on the responses to a Delphi questionnaire that was designed to solicit information examining the links between future technology and criminal activity. Eighty experts in a number of relevant fields-including law enforcement, insurance, loss adjustment, academia, science, and the computer industry-were asked to complete two questionnaires about the probability of future innovations eliciting criminal activity. For example, experts were asked: “What is the probability of some companies defrauding customers whilst selling goods over the Internet? - Definitely Not, Unlikely, Likely, Very Likely, Definitely.” Interviews were also conducted with selected experts to obtain more detailed information throughout the Delphi exercise.
Scenario Writing
Scenario writing attempts to describe how current conditions may evolve in the future. It is a mechanism through which the influences of possible future developments, identified by the scanning process, on the issue of interest can be examined. Many different outcomes are feasible, and good scenario writing strives to identify the range of possible conditions that might emerge, given the variety of forces and events deemed feasible. Scenarios are not forecasts per se; they are descriptions and portrayals of events and trends as they could evolve. Different scenarios can be developed to illustrate the consequences of varying assumptions about the trends and events that will occur and their timing and impacts. Crime forecasts, for example, might involve constructing several scenarios that differ in their assumptions about birthrates, population, economic conditions, technological innovation, etc. When a set of scenarios is prepared, each treats the same variables, but the resulting outcomes will vary according to the dynamic interactions that are formulated. Changing one or several key assumptions can, in turn, generate different sets of scenarios. For example, the assumed economic growth might be low in one scenario, moderate in another, and high in a third (Cole, 1995, 7-11). Each scenario will result, potentially, in a different prediction of future crime rates.

In sum, there are a number of different quantitative and qualitative methods that can be used to develop crime forecasts. One of the significant gaps in the methodological designs of the studies reviewed for this report is the lack of a combined use of quantitative and qualitative analysis. Most, if not all of the studies, used quantitative modeling or qualitative research, but not both. This is a significant weakness, because, as the above analysis insinuates, quantitative and qualitative approaches are quite complementary. While a quantitative analysis is useful for extrapolating crime trends, qualitative methods, such as environmental scanning or scenario writing, are useful in identifying variables that will influence crime rates. This process includes the identification of different potential movements of these variables, which can then be used to develop a range of options for quantitative modeling.

6.2 Accuracy of Past Crime Predictions

Of course, any prediction of future events will suffer from inevitable inaccuracies. This is no different with respect to crime, and a review of past predictions of future crime rates does reveal problems with accuracy.

The results of the time series model developed by Fox (1978) indicate a general reduction in the upward trend in crime rates during the 1980s and an increase during the 1990s. Fox also predicted that the violent crime rate would decline in the 1980s before increasing once again in the 1990s. Fox accurately predicted the levelling out of the crime rate in the 1980s, but was incorrect in his predictions of an increase in the 1990s.

The time series models developed by Field (1990; 1998), which correlate macro-economic expansion and consumer expenditures with a growth in property crime in the U.K., anticipates an increase in crime rates in 1999 through 2003. However, this rise is not borne out by either police- or victim-reported crime rates for 1999 and 2000, both of which indicate that crime continued to decline in the U.K. during these years, while the economy expanded.

The problems with developing accurate crime forecasts are also reflected in greatly divergent predictions, despite the use of the same data and analytical models. For example, while the same data and statistical modeling procedures were used by Dhiri et al. (1999) and Deadman (2000), the former predicted a rise in the U.K. crime rate while the latter predicted a decline. This difference stemmed not from the data used for the predictions, but from the use of different analytical techniques.

6.3 International Comparative Analysis

In the context of an international comparative analysis, with Canada at the core of this comparison, the most significant gap in the field of crime forecasting is the lack of research that has been conducted in this country.

As would be expected, much of the crime forecasting has come from the United States and Great Britain, and to a lesser extent, Australia. In recent years, the British government has been the most active of these in directly funding research into crime trend forecasting. This has led to financing of at least three studies by the Home Office using econometric modelling (Field, 1990; Field, 1998; Dhiri et al., 1999). In addition, the British government created within the Department of Trade and Industry, the Foresight Directorate to anticipate future trends, including trends in crime and crime prevention. The initiatives undertaken by these two separate government departments are complementary, at least from a methodological perspective, through the use of quantitative and qualitative methods respectively. These governmental initiatives are joined by private sector efforts-in particular by the British Insurers Association-that have also funded original research to develop crime predictions. Accordingly, of the countries studied for this report, the Great Britain has the most impressive body of knowledge on the future of crime. Particularly promising is the emphasis placed by the Home Office on the continual refinement of time series models that are used to correlate crime trends with the movement of predictor variables. These models are impressive for their identification of factors influencing crime, the use of rigorous modelling techniques, and their ability to develop a range of crime predictions based on different future movements of the predictor variables. The Foresight Programme is also impressive for its ambitious attempts to map the future, its widespread consultations with experts and the public, and with respect to crime specifically, its attempts to simultaneously examine both the future of crime and crime control.

In contrast, this research did not identify any Canadian studies in the past decade that involved forecasting crime into the 21st Century. In fact, this study turned up little Canadian research from the past three decades that attempts to construct any predictive models of crime. This void stems in part from the smaller criminological community in Canada, relative to Great Britain or the U.S., as well as minimal government demand or funding for this type of research. In contrast, the Australian Government, and the Australian Institute of Criminology in particular, has funded research into the future of crime in that country.

Although largely beyond the scope of this paper, most of the attempts at criminal justice forecasting in Canada have been developed around the future of policing (Leighton & Normandeau, 1990; Rossmo & Saville, 1991; Bayley, 1991; RCMP, 1998; Police Futures Group[1]) and future prison populations in federal correctional facilities (Canadian Corrections Service, 1982). While it makes sense that predictions of the future of policing or prison populations would be premised on future crime trends and rates, none of the above-cited organizations or publications seem to take these factors into consideration.

6.4 Replication of Foreign Research in Canada

This section conjectures about whether foreign organizations and analytical models examined in this review could be replicated in Canada. Particular emphasis is placed on determining if there is available data in this country, as this will largely dictate the ability to conduct such research (although this is not to discount other important determinants, such as adequate research funding or existing expertise).

Data for quantitative modelling of future crime trends can be demarcated into two categories: basic crime statistics and statistical data on predictor variables, in particular economic output, consumer spending, and demographic characteristics of the Canadian population (with an emphasis on the size of certain age populations).

A cursory knowledge of Canadian data sources indicates that there are no real obstacles to replicating in this country the organizations and models developed elsewhere. Basic quantitative data exists on most property and violent criminal incidents. In particular, within Statistics Canada, there exists the expertise, resources, methodologies, and more than 40 years of experience in conducting national quantitative estimates of criminal incidents, which include national and historical police-reported crime statistics as well as the results of national victimization surveys. Relying on over four decades of uniform crime reporting, Statistics Canada also conducts historical time series analyses, the results of which are published on an annual basis. The extent of information on criminal incidents contained in police-reported crime statistics has been greatly improved with the implementation of the UCR2 survey, an incident-based reporting system.[2] Like most other countries however, there is a substantial lack of national quantitative data on other significant (organized) criminal incidents such as fraud, money laundering, drug trafficking, etc.

Statistics Canada also collects data on “prime environmental factors” that influence crime and which can be correlated with crime-related data to conduct time series analyses, forecasts, and impact assessments. Relevant statistics include the unemployment rate for young males Labour Force Survey); Gross Domestic Product (National Accounts); the number of individuals, families and children with incomes below low-income cut-offs (Census, Survey of Consumer Finances); and age group populations (Census).

However, while Statistics Canada has collected and analyzed quantitative data in historical time series analysis, and despite the fact that this agency also collects data on relevant environmental factors, this study did not identify any attempts by Statistics Canada or other organizations or individuals to use this data to project future crime trends. Regardless of the inherent problems with such crime forecasting, this void should be considered a significant weakness in applied research and analysis of crime in this country. There has been little concerted effort by the criminal justice researchers or the academic community to anticipate future crime trends, let alone develop policies and programs to minimize the impact of future crime. This lack of foresight will continue to hamper the ability of the criminal justice system and society at large to counter the increasingly sophisticated, technological, and organized nature of crime that will take place in the future.

Date modified: