Tayebi, M.A., Glasser, U., & Andresen, M.A. (2020). Future applications of big data in environmental criminology. In B. Leclerc & J. Cale (Eds.), Criminology at the Edge: Big Data (pp. 40 – 53). New York, NY: Routledge.
The advent of big data in the social sciences is beginning to change the research landscape both in terms of what type of research questions we can ask and how we answer those research questions. The field of environmental criminology is well-suited to have future applications of big data because of the types of data used in this research, but also because of the increased number of dimensions in the analyses because of the considerations of both space and time. In this chapter, we consider some of the research that has begun to investigate crime patterns using big data, but also where this research needs to go and some considerations on how to move forward. Of particular salience for environmental criminology is the nature of the questions asked in research, how those questions are answered, and with whom environmental criminologists must collaborate to be able to take advantage of this relatively new form of data.
Keywords: big data; environmental criminology; data quality; computational criminology
Big data are increasingly becoming a topic in research in both the physical and the social sciences. Though its definition has been a moving target with increasingly powerful computers, big data is generally defined as extremely large or complex (multiple dimensions) data sets that need to be analyzed using methods outside of the traditional data-processing applications commonly used. These data have significantly increased the possibilities for analysis of new topics but also presenting new opportunities to re-visit only questions in new ways.
Big data are increasingly being used in criminology and, specifically environmental criminology. Environmental criminology is primarily concerned with the analysis of crime through a spatial and temporal lens. Topics can range from crime prevention to policing to theoretical testing, but spatial and temporal methods are at the heart of much of this research. These added dimensions of space and time alone could turn “small data” into big data, but there is also a trend of environmental criminology using smaller and smaller units of analysis that leads to a greater number of observations in the data sets environmental criminologists use (Weisburd et al., 2009).
In this chapter we discuss some of the potential uses of big data and its associated computational methods for the future of environmental criminology. There are aspects of research in which big data has already been used. We discuss some of this research, but also new directions. We conclude with a brief discussion of the considerations that must be taken into account with the use of this new data source in future research.
Making better small data with big data
Though big data is often used an analyzed as a whole, we would also argue that pieces of big data can be used to make the small data we use in criminology better. For example, some research within environmental criminology has used satellite data, social media data, and mobile phone data to measure where people are in order to better identify risky places (Andresen, 2011; Malleson & Andresen, 2015a, 2015b, 2016). Because such data may better represent where people are through the geo-referencing of social media and mobile phone networks, they can be useful in identifying where people are, on average (rather than census data), or where they are on particular days of the week, or times within a given day. Given that there are hundreds of millions of tweets every day alone, specialized statistical and computing techniques are necessary to be able to analyze these data. However, most often, researchers in environmental criminology are interested in particular places. These places could be the size of cities, but also the size of a street segment (Andresen et al., 2017a, 2017b; Hodgkinson et al., 2016).
Within these contexts of cartographically larger places, social media or mobile phone information can be obtained to better identify where people are and when they are there. This leads to the better identification of risk for different places and those places at different times of the day. This is what has been done, but the potential for such analyses has only begun to be utilized. There are a number of confidentiality and ethical implications for these types of analyses (to be discussed further below), but the potential for understanding crime patterns (spatially and temporally, among others) is tremendous with the availability of big data.
For example, knowing the age and gender of people, how long they spend at particular places, and their spending habits, can all be instructive for understanding why offending and victimization occurs here and there, now and not then. These data can be identified from larger big data datasets for smaller areas and then aggregated to protect confidentiality. Such data could then be used in traditional (statistical) analyses to better measure theoretical constructs, testing theory in ways would simply could not do before. But the presence and availability of big data should not only be used to revisit old methods in new ways, but open the door for new types of analyses and pushing the field of environmental criminology forward.
Big data, data mining, and machine learning in environmental criminology
One of the ways in which environmental criminology may move forward with big data is through the use of modern computing techniques such as data mining and machine learning. Generally speaking, data mining involves the identification of patterns in big data that are not feasible, or extremely costly in terms of time and/or money; machine learning is one method within data mining that uses algorithms and statistical modeling without explicit instructions. In other words, these techniques rely less on theory guiding what we think should be in the model than on inference itself. As such, all the data available can be used to test theoretical models and policy implementation, rather than a relatively small set of data our theories tell us should matter.
For example, studies that use census data for predictor variables rarely use more than 10-20 predictor variables when there are hundreds of such variables available through most censuses. Data mining can be used for inferential testing of all these variables and their interactions that is simply not tractable using conventional techniques. Moreover, with the availability of crime and police data at relatively fine spatial resolutions for many police agencies across many countries (sometimes available over many years), data mining techniques can be used to analyze crime patterns at multiple scales (spatially and temporally), across multiple crime types, across multiple policing services, using predictors that have not been able to be incorporated before because of data, software, and user limitations. The implications of such data and methods are significant for the testing of our theories across many different contexts with similarly defined data and the evaluation of larger scale policies that may work in some places but not others.
Another potential with the use of big data and data mining is the testing of theories that have simply not been possible in the past. For example, the theoretical construct of the geometry of crime requires activity patterns for offenders and non-offenders, alike to be able to test the predictions of this theory. These activity patterns need to be for both deviant and non-deviant activities and measured over time in order to know where and when people travel to their activity nodes and the pathways between them. These data could be identified through mobile phone activity, for example, and then cross checked with victimization and offending data. Hypotheses and propositions that have never been fully tested in the past can be tested using big data and its corresponding methods.
Predictive analysis and big data
Turning to more predictive analyses that are already being undertaken but need further development, policymakers inevitably face enormous challenges deploying notoriously scarce resources ever more efficiently to apprehend criminals, disrupt criminal networks and effectively deter crime by investing in crime reduction and prevention strategies. While data collection from different sources, data preparation, and information sharing pose difficult tasks, the big challenge for law enforcement agencies is analyzing and extracting knowledge from their large collection of crime data. Applying data-driven approaches on such data can provide a scientific foundation for developing effective crime reduction and prevention strategies through analysis of offenders' spatial decision making and their social standing. The main idea behind crime prediction techniques is that crime is not random but happens in a patterned manner. In the crime data mining process, the goal is to understand criminal behaviours and extract criminal patterns in order to predict crime and take steps to prevent it.
The rapid evolution of data science, employing techniques and theories drawn from broad areas such as machine learning and data mining, through availability of massive computational power increasingly influences our daily lives. Data are collected, modeled and analyzed to uncover the patterns of human behaviour and help with predicting social trends. This is changing the way we think about business, politics, education, health, and data science innovations will undoubtedly continue in the years to come. One particular area that has seen limited growth in accepting and using these powerful tools is public safety. This is somewhat surprising given the important role that predictive analytics can play in public safety.
New methodologies emerging in data science can advance crime analysis to the next level through the use of big data and move from tracking patterns of crime to predicting those patterns. This has led to a new paradigm of crime analysis, called predictive policing. Predictive policing uses data science and big data to identify potential targets for criminal activity with the goal of crime prevention. Successful predictive policing results in more proactive policing and less reactive policing.
Specifically, “predictive policing refers to any policing strategy or tactic that develops and uses information and advanced analysis to inform forward-thinking crime prevention” (Uchida, 2012), which involves multiple disciplines to form the rules and develop the models. Given that research strongly supports that crime is not random but rather occurs in patterns, the goal of predictive policing methods is to extract crime patterns from historical data at both macro and micro scales as a basis for prediction and prevention of future crimes. This approach uses data-driven tools and big data that benefit from data mining and machine learning techniques for predicting crime locations and temporal characteristics of criminal behaviour.
Predictive analysis for policing can be divided into four classes:
Predicting offenders. The goal is predicting future offenders using the history of individuals such as features of their living environment and behavioural patterns.
Predicting victims. This is about identifying individuals who more likely than others may become victims and predicting risky situations for potential victims.
Predicting crime locations. This task aims at predicting the location of future crimes at individual and aggregate level. An important aspect of crime is the geographic location that crime happens. Every neighborhood provides some condition in which criminal behaviour takes place, but crime distribution in city neighborhoods is not even. Understanding the spatial patterns of crime is essential for law enforcement agencies to design efficient crime reduction and prevention policies. Although mining spatial patterns of crime data in the aggregate level took special attention in the criminology literature, there is not that much work about crime spatial patterns for individual offenders.
Predicting criminal collaborations. Predicting likely future collaboration between offenders and the type of associated crime using social network analysis. We focus our discussion on this latter topic area.
Co-offending network analysis
Social networks represent relationships among social entities. Normally, such relationships can be represented as a network. Examples include interactions between members of a group (such as family, friends, or neighbors) or economic relationships between businesses. Social networks are important in many respects. Social influence may motivate someone to buy a product, to commit a crime, and any other decision can be interpreted and modeled under a social network structure. Spread of diseases such as AIDS infection, the diffusion of information and word of mouth also strongly depend on the topology of social networks.
Social network analysis (SNA) focuses on structural aspects of networks to detect and interpret the patterns of social entities. SNA essentially takes a network with nodes and edges and finds distinguished properties of the network through formal analysis. Data mining is the process of finding patterns and knowledge hidden in large databases. Data mining methods are increasingly being applied to social networks, and there is substantial overlap and synergy with SNA. These methods are particularly important for the combinations and permutations of possible social ties or links when using large data sets (big data). For example, even a criminal event data set with only a few thousand criminal events leads to crime linkage pairs in the millions that are difficult to analyze when considering traditional data analyses.
New techniques for the analysis and mining of social networks are developed for a broad range of domains, including health and criminology. These methods can be categorized depending on the level of granularity at which the network is analyzed 1) methods that determine properties of the social network as a whole; 2) methods that discover important subnetworks; 3) methods that analyze individual network nodes; and 4) methods that characterize network evolution. In the following, we list the primary tasks of SNA:
Centrality analysis aims at determining more important actors of a social network so as to understand their prestige, importance or influence in a network;
Community detection methods identify groups of actors that are more densely connected among each other than with the rest of the network;
Information diffusion studies the flow of information through networks and proposes abstract models of that diffusion such as the Independent Cascade model;
Link prediction aims at predicting for a given social network how its structure evolves over time, that is, which new links are likely to form; and
Generative models are probabilistic models that simulate the topology, temporal dynamics and patterns of large real-world networks.
Co-offending networks. Criminal organizational systems differ in terms of their scope, form and content. They can be a simple co-offending looking for opportunistic crimes, or a complex organized crime group involved in serious crimes. They can be formed based on one-time partnership for committing a crime, or their existence can have continuity over time and across different crime types. For example, in a criminal organization system interaction among actors can be initiated from family, friendship, ethnic, or other possible ties.
A co-offending network is a network of offenders who have committed crimes together. With increasing attention to SNA, law enforcement and intelligence agencies have come to realize the importance of detailed knowledge about co-offending networks. Groups and organizations that engage in conspiracies, terrorist activities, and crimes such as drug trafficking typically do this in a concealed fashion, trying to hide their illegal activities. In analyzing such activities, investigations do not only focus on individual suspects but also examine criminal groups and illegal organization and their behaviour.
Thus, it is important to identify co-offending networks in data resources readily available to investigators, such as police arrest data and court data, and study them using social network analysis methods. In turn, social network analysis can provide useful information about individuals as well. For example, investigators could determine who are key players, and subject them to closer inspection. In general, knowledge about co-offending network structures provides a basis for law enforcement agencies to make strategic or tactical decisions.
Co-offending network analysis in practice
Co-offending networks analysis contributes to predictive policing by detecting hidden links and predicting potential links among offenders. In this section, we introduce important applications of co-offending networks analysis in predictive policing that are covered in this research.
Organized crime group detection. Organized crime is a major international concern. Organized crime groups produce disproportionate harm to societies, and an increasing volume of violence is related to their activities. Because the primary aim of organized crime groups is gaining material benefit they try to access to resources that can be profitably exploited. In terms of economic-related crimes (e.g., credit and debit card fraud) organized crime costs tax payers billions of dollars per year.
Understanding the structure of organized crime groups and the factors that impact on it is crucial to combat organized crime. There are several possible perspectives how to define the structure of organized crime groups, but recent criminological studies are increasingly focusing on using social network analysis for this purpose. The idea of using social network analysis is that links between offenders and subgroups of an organized crime group are critical determinant of the performance and sustainability of organized crime groups.
Confronted with a bewildering diversity of characteristics referred to in existing definitions of organized crime and criminal organizations, the conceptual model itself appears not clearly rendered in the literature. Striving for a definition that is general and open, a potential source is the criminal code, although this depends on a specific country. For instance, a baseline definition of criminal organization is provided by the Criminal Code of Canada:
“In Canada a criminal organization is a group, however organized that: (a) is composed of three or more persons in or outside Canada; and (b) has as one of its main purposes or main activities the facilitation or commission of one or more serious offences, that, if committed, would likely result in the direct or indirect receipt of a material benefit, including a financial benefit, by the group or by any one of the persons who constitute the group. The definition further specifies that it excludes a group of three or more persons that has formed randomly for the immediate commission of a single offence. Section 467.1(1) of the Criminal Code of Canada.”
Looking for a quantitative definition, in an attempt to measure organized crime, van der Heijden (1996) proposes a number of common characteristics:
Collaboration of more than two people;
Commission of serious criminal offences (suspected);
Determined by the pursuit of profit and/or power;
Each having their own appointed tasks;
For a prolonged or indefinite period of time;
Using some form of discipline and control;
Operating across borders;
Using violence or other means suitable for intimidation;
Using commercial or businesslike structures;
Engaged in money laundering;
Exerting influence on politics, the media, public administration, judicial authorities, or economy.
According to van der Heijden (1996), for any criminal group to be categorized as organized crime it needs to have at least six of the above characteristics, where items 1, 2, and 3 are obligatory, thus adding three more characteristics. A major study in the Netherlands (Fijnaut, 1998) mentions great variations in collaborative forms of organized crime and concludes that “the frameworks need not necessarily exhibit the hierarchical structure or meticulous division of labor often attributed to mafia syndicates. Intersections of social networks with a rudimentary division of labor have also been included as groups in the sub-report on the role of Dutch criminal groups, where they are referred to as cliques.
An impressive collection of definitions of organized crime specific for various countries, comprising more than individual 150 entries in total, has been gathered by von Lampe (2019). In addition, this collection also includes comments on how to define organized crime, and definitions by prominent individuals and government agencies, for instance, such as the Federal Bureau of Investigation (FBI). Not included though are definitions of the term `organized crime group'. Given the abstract nature and informal language of these definitions, it is not clear at all how and to what extent one may utilize this resource for defining organized crime in precise computational and/or mathematical terms. The use of big data and its corresponding analytic techniques come into play here because various forms of law enforcement data can be used to show how offenders interact from the ground up and identify potentially organized behavior (using all of the available data), rather than imposing a definition a priori and then searching for interactions that are consistent with that definition.
Organized crime group detection using community detection. In most cases, existing definitions in the literature on organized crime concentrate on three essential perspectives for characterizing the nature of this form of crime (von Lampe, 2019). First, organized crime is primarily about crime, such that organized crime is seen as a specific type of criminal activity that has certain specific characteristics (continuity, in contrast to irregular criminal behaviour). Second, organized crime is related to the concentration of power, either in economic or in political structures of the society. Third, the emphasis is on being organized. That is, the important aspect of organized crime is on how offenders are connected to each other more than what they do.
Based on the third view, we can formalize central aspects of criminal networks in a coherent and consistent formal framework to provide a precise semantic foundation that is consistent with criminological research, social network analysis, and law enforcement operations. Such work would aim at bridging the conceptual gap between data level, mining level, and interpretation level, and is intended for developing advanced computational methods for analyzing co-offending networks to detect and extract organized crime structures and how they evolve over time in order to assist law enforcement and intelligence agencies in their investigations.
Community detection in social networks has attracted considerable interest and many definitions of the concept of community have been proposed. In social science studies, social networks are considered as basis of social behaviours and activities. Studies of different social networks show that community structure influences information transfer, communication, and cooperation. Sense of community is generally defined as a feeling that members of a group matter to one another and to the group, and a common belief that members' needs will be satisfied through their commitment to be together. The nature of organized crime groups, however, is different from other types of groups such as friendship or co-authorship groups. Organized crime groups are usually well established with group membership being defined explicitly and strictly. Unlike friendship or co-authorship communities, offender groups as well are characterized by member relationships that are more systematic and organized to achieve material benefit from committing crime. Therefore, detecting organized crime groups calls for a stricter definition of community. Given the large volumes of law enforcement data, analyzing and extracting knowledge from their data using data-driven approaches such as machine learning, as discussed above, can help define what matters for organized crime groups that is different from other forms of groups to better identify their presence.
On this note, and based on fundamental discussions in the criminology literature, one can summarize the important characteristics of organized crime groups as: 1) these groups have at least three members and can be categorized as centralized or distributed or hierarchical groups, but the focus is on offender groups for which the density of their intra-group collaborations is higher than the density of intergroup collaborations; 2) organized crime groups are characterized by a distribution of roles and different degrees of agency amongst individuals, where groups can overlap and may have common members; 3) these groups commit serious crimes with the perspective of gaining material benefit; and 4) their activity is more continuous compared to regular offender groups. Using big data analytics, these characteristics of organized crime groups can be searched and then tested with regard to their validity on a grand scale with all possible combinations. Moreover, predictive analytics can then be used to identify the potential emergence of these characteristics in real time.
Therefore, for the purpose of organized crime group detection, in each time snapshot of a co-offending network the following tasks are carried out in consecutive steps: 1) discover offender groups in the current network; 2) compute the activity and criminality of these groups in the time period between the current network and the previous network based on the offences that were committed by their members; 3) assess the material benefit associated with each of the offences considered in Step 2; 4) identify those groups that qualify as possible criminal organizations; 5) update the groups evolutionary trace for the current time period. It should be clear at this point that sophisticated data mining techniques become critical with the increasingly complexity of such analyses. For example, rather than relying on traditional policing methods to identify organized crime groups, police calls for service data with co-offender identifiers can be used to identify clusters of individuals who may then be assigned to these more formal group relationships. But before such analyses can take place, identifying all co-offences is critical.
Co-offence prediction. Co-offence prediction is defined as a link prediction problem for co-offending networks. In the context of suspect investigation, law enforcement can more precisely focus their efforts based on probable relationships in criminal networks that have previously not observed. Traditional suspect investigation methods use partial knowledge discovered from crime scenes to identify potential suspects. Co-offending networks analysis provides a complementary approach to traditional criminal profiling methods insofar as it can contribute to investigation in cases where multiple offenders committed a crime, but only a subset of offenders are charged. Therefore, link prediction is an important aspect of social network analysis with the aim to better understand the network structure. Link prediction methods can be used to extract missing information, identify hidden interactions, and evaluate network evolution mechanisms. For example, missing one key player in a criminal network (who may or may not be directly related to each crime in a law enforcement database) could bring together or solidify an existing network structure.
Contrary to other social networks, the concealment of identities and activities of actors is a central characteristic of co-offending networks. Still, the network topology is a primary source of information for co-offence prediction. Moreover, there are two other major information sources: environmental activity and criminal activity. Offenders who are spatially close tend to be socially close, because this increases the chance of meeting each other and forming new criminal collaborations (Tayebi et al., 2012). Further, common criminal experience (with the same type of offences, for example) also affects co-offending behaviour (Weerman, 2003).
Several studies now show that supervised link prediction approaches outperform unsupervised methods (Lichtenwalter et al., 2010) who use only topological features (Liben-Nowell & Kleinberg, 2007). In contrast to unsupervised methods, supervised learning methods can overcome the class imbalance problem, an issue that arises when the ratio of groupings of data (co-offending or not, for example) is not close to one-to-one (Lichtenwalter et al., 2010). Exploiting the geographic information provided by location-based social networks services, some recently proposed link prediction methods consider spatial characteristics of users (Scellato et al., 2011; Wang et al., 2011). Scellato and colleagues (2011) used information about places visited by users, in addition to their social network features, to define prediction spaces that reduce the class imbalance ratio and improve the prediction of people to people and people to places.
Co-offending networks are spatially embedded in a manner similar to location-based social networks. However, the environmental effects on the formation of co-offence links and, accordingly, our approach in defining offenders' spatial closeness are different from those in location-based social networks (Scellato et al., 2011; Wang et al., 2011; Zhang et al., 2012; Cho et al., 2011). Tayebi et al. (2014) proposed a framework that builds on criminological theories (Brantingham & Brantingham, 1981; Brantingham et al., 2017; McGloin et al., 2008; Morselli, 2009; Rossmo, 2000; Sutherland & Cressie, 1947) and, considering the available information on offenders, distinguishes three different criminal cooperation opportunities: socially-related, geographically-related and experience-related. In this work, the authors studied the co-offence prediction problem in each of these prediction spaces separately, achieving two goals. First, the heavy class imbalance between positive (existing links) and negative samples (non-existing links) is the main challenge of the link prediction problem (Lichtenwalter et al., 2010). The restriction of the training and test data to the different prediction spaces reduces the class imbalance ratio significantly, while keeping about half of the positive samples (co-offences). Second, the prediction spaces enhance the understanding of co-offence patterns in different criminal cooperation opportunities.
They define the prediction features in four different categories (social, geographic, geo-social, and similarity), and evaluate their prediction strength both individually and as a set. Social features indicate social closeness of offenders based on their position in a co-offending network. Geographic features show spatial proximity of offenders based on their residential locations and the location of offences they have committed. Geo-social features combine social and geographic characteristics of offenders. Finally, similarity features capture homophily-based characteristics of offenders, such as age and gender. Evaluating features individually and also as a set shows that the geo-social features we define outperform other features.
The co-offence prediction framework proposed aims at advancing the state-of-the-art in crime data mining by making the following contributions: 1) defining co-offence prediction spaces to reduce the class imbalance; 2) introducing novel prediction features for co-offence prediction; and, 3) experimentally evaluating the proposed approach on large real-world crime data. Some of the main findings in this research include results that show that social and geographic features have important implications for the evaluation of networks. First, repeated exposure to individuals is a strong predictor compared to the social features present with common friends in the network. This implies that the chance of criminal collaboration increases more with the opportunity to commit crimes than with trust or transitivity in the co-offending network. Crime location distance is a better predictor of co-offending than home location distance, meaning that being criminally active in close proximity may lead to new criminal collaboration. Second, geo-social features are better co-offence predictors than geographic or social features alone. This result implies that researchers and practitioners need to focus more on combined patterns in environmental and social features to enhance crime reduction and prevention. This also shows the role of environmental criminology in the context of co-offending. Third, experimental results show that, although there is variability in the performance of different classifiers (e.g. the area under the receiver operating characteristic (ROC) curves.), the probability of predicting a co-offence for similarity-related offenders is higher than for socially or geographically-related co-offenders.
Taken together, this information can then be used in the context of co-offending network disruption. Actors of a social network can be categorized based on their relations in the network. Actors in the same category may take similar roles within an organization, community or whole network. These roles are usually dependent on the network structure and the actors' position in the network. For instance, actors who are located in the central positions of a social network may be detected as key players in that network. Actors who are connected to many other actors may be viewed as socially active players, and actors who are frequently observed by other actors may be identified as popular players.
In terms of co-offending network disruption, the goal is finding a set of players whose removal creates a network with the least possible cohesion. In the other words, their removal maximally destabilizes the network. This task is critical in the co-offending network analysis where removing the key players may sabotage the network and decrease the aggregate crime rate. Given the large volume of law enforcement data, big data and its corresponding analytic techniques can be used to identify these players, at times using co-offence prediction techniques, and then simulating the impact of their removal to better understand where scarce law enforcement resources may be best allocated.
Directions and issues moving forward for big data in environmental criminology
As stated above, thousands of criminal events can have millions of crime pairs, so tens of thousands of criminal events (not a large data set in the grand scheme of things) can have hundreds of millions of crime pairs. As such, what may be traditionally considered “small data” becomes big data through the addition of inherent complexity within the small data (new dimensions are created and added), requiring the application of data mining techniques to make the analysis tractable. Therefore, it is important to note that the use of big data, or the transformation of small data to big data, has implications for social science (e.g. environmental criminology) research. There are two primary considerations in the future of big data in environmental criminology: 1) the need for specialized knowledge; and, 2) ethics in the use of big data. Each is discussed in turn.
The need for specialized knowledge is not a limitation or cautionary aspect of big data, but a pragmatic concern. Though there are social scientists (environmental criminologists) with the necessary computing skills to undertake the computational methods necessary for the analysis of big data, most often it will be necessary for environmental criminologists to collaborate with other scholars from fields such as computing science and statistics. At face value, this is a simple task, but it is important to recognize that different disciplines have different research and publication cultures. Some differences involve where research is published, author order, and the number of authors, among many others. These aforementioned differences, for example, need to be known up front and accounted for when publications are “counted” for salary and promotion decisions. One of the most pertinent publication cultural differences between the social sciences and computing science is publishing in traditional journals and conference proceedings. In the social sciences, conferences proceedings are viewed as a lesser publication form, deemed to not have the same rigor and value as an article in a peer-reviewed journal. However, conference proceedings in computing science undergo similar levels of peer-review as traditional journals in the social sciences in terms of the peer-review process and rejection rates. It can be difficult to assess academic outlets across disciplines, but if interdisciplinary research is to move forward, cultural differences such as these, along with many others, must be figured out and accounted for to ensure fairness in measuring academic output that is truly interdisciplinary.
With regard to ethics and big data, privacy concerns increase in magnitude almost as quickly as the sizes of big data data sets themselves. Environmental criminologists, in particular, analyze sensitive data sets with regard to criminal behaviour or victimization. Even without personal identifiers, the same methods used to analyze big data can be used to create data linkages across different data sets, placing personal identification at risk. For example, with the large volumes of social media data that may have information regarding when and where someone was as well as what they were doing (or what happened to them), criminal offending or victimization may be able to be connected to specific individuals. With larger and more data sets, this probabilistic matching becomes more of a concern. This is only one example but shows the importance of asking new questions of not just if we can do something, but if we even should.
Despite these considerations for the need of specialized knowledge and the ethical concerns, the potential for the continued and advanced use of big data in environmental criminology is great. What has been done has made significant advances in the field and what will be done in the future is expected to alter the ways in which we think about environmental criminology as a discipline.
Andresen, M.A. (2011). The ambient population and crime analysis. Professional
Geographer, 63(2), 193 - 212.
Andresen, M.A., Curman, A.S.N., & Linning, S.J. (2017). The trajectories of crime at places:
Understanding the patterns of disaggregated crime types. Journal of Quantitative
Criminology, 33(3), 427 - 449.
Andresen, M.A., Linning, S.J., & Malleson, N. (2017). Crime at places and spatial
concentrations: Exploring the spatial stability of property crime in Vancouver BC, 2003-
2013. Journal of Quantitative Criminology, 33(2), 255 - 275.
Brantingham, P.J., & Brantingham, P.L. [Eds.] (1981). Environmental criminology. Beverly
Hills, CA: Sage Publications.
Brantingham, P.J., Brantingham, P.L., & Andresen, M.A. (2017). The geometry of crime and
crime pattern theory. In R. Wortley & M. Townsley (Eds.), Environmental
criminology and crime analysis (2nd ed.) (pp. 98 - 115). New York, NY: Routledge.
Cho, E., Myers, S.A., & Leskovec, J. (2011). Friendship and mobility: User movement in
location-based social networks. Proceedings, 17th ACM SIGKDD International
Conference on Knowledge Discovery and Data Mining, 1082-1090.
Fijnaut, C., Bovenkerk, F., Bruinsma, G., & van de Bunt, H. (1998). Organized crime in the
Netherlands. The Hague, Netherlands: Kluwer Law International.
Hodgkinson, T., Andresen, M.A., & Farrell, G. (2016). The decline and locational shift of
automotive theft: A local level analysis. Journal of Criminal Justice, 44(1), 49 - 57.
Liben-Nowell, D., & Kleinberg, J. (2007). The link-prediction problem for social networks.
Journal of the Association for Information Science and Technology, 58(7), 1019-1031.
Lichtenwalter, R.N., Lussier, J.T., & Chawla, N.V. (2010). New perspectives and methods in
link prediction. Proceedings, 16th ACM SIGKDD International Conference on
Knowledge Discovery and Data Mining, 243-252.
McGloin, J.M., Sullivan, C.J., Piquero, A.R., & Bacon, S. (2008). Investigating the stability of
co‐offending and co‐offenders among a sample of youthful offenders. Criminology,
Malleson, N., & Andresen, M.A. (2015a). Spatio-temporal crime hotspots and the ambient
population. Crime Science, 4, Article 10.
Malleson, N., & Andresen, M.A. (2015b). The impact of using social media data in crime rate
calculations: Shifting hot spots and changing spatial patterns. Cartography and
Geographic Information Science, 42(2), 112 - 121.
Malleson, N., & Andresen, M.A. (2016). Exploring the impact of ambient population
measures on London crime hotspots. Journal of Criminal Justice, 46, 52 - 63.
Morselli, C. (2009). Inside criminal networks. New York, NY: Springer.
Rossmo, D.K. (2000). Geographic profiling. Boca Raton, FL: CRC Press.
Scellato, S., Noulas, A., & Mascolo, C. (2011). Exploiting place features in link prediction on
location-based social networks. Proceedings, 17th ACM SIGKDD International
Conference on Knowledge Discovery and Data Mining, 1046-1054.
Sutherland, E.H., & Cressey, D.R. (1947). Principles of criminology (4th ed.). Chicago, IL: J.B.
Tayebi, M.A., Frank, R., & Glässer, U. (2012). Understanding the link between social and
spatial distance in the crime world. Proceedings, 20th ACM SIGSPATIAL
International Conference on Advances in Geographic Information Systems, 550–553.
Tayebi, M.A., Ester, M., Glässer, U., & Brantingham, P.L. (2014). Spatially embedded co-
offence prediction using supervised learning. Proceedings, 20th ACM SIGKDD
International Conference on Knowledge Discovery and Data Mining, 1789-1798.
Uchida, C. (2012). A national discussion on predictive policing: Defining our terms and
mapping successful implementation strategies. Washington, DC: National Institute of
van der Heijden, T. (1996). Measuring organized crime in Western Europe. Policing in
Central and Eastern Europe: Comparing first hand knowledge with experience from
the West. Slovenia: College of Police and Security Studies.
von Lampe, K. (2019, 08 February). Definitions of organized crime. Retrieved from:
Wang, D., Pedreschi, D., Song, C., Giannotti, F., & Barabasi, A-L. (2011). Human mobility,
social ties, and link prediction. Proceedings, 17th ACM SIGKDD International
Conference on Knowledge Discovery and Data Mining, 1100-1108.
Weerman, F.M. (2003). Co-offending as social exchange: Explaining characteristics of co-
offending. British Journal of Criminology, 43(2), 398-416.
Weisburd, D., Bruinsma, G.J.N., & Bernasco, W. (2009). Units of analysis in geographic
criminology: Historical development, critical issues, and open questions. In D. Weisburd,
W. Bernasco and G. J. N. Bruinsma (eds), Putting crime in its place: Units of analysis in
geographic criminology (pp. 3 – 31). New York, NY: Springer.
Zhang, C, Shou, L, Chen, K., Chen, G., & Bei, Y. (2012). Evaluating geo-social influence in
location-based social networks. Proceedings, 21st ACM International Conference on
Information and Knowledge Management, 1442-1451.