Content uploaded by Vanessa Bauer
Author content
All content in this area was uploaded by Vanessa Bauer on Jan 06, 2020
Content may be subject to copyright.
P r e d i c t i v e P o l i c i n g i n G e r m a n y
Opportunities and challenges of data-analytical forecasting
technology in order to prevent crime.
MASTER’S THESIS
Empirical Method: Qualitative Analysis
Master’s Program
“Management, Communication & IT”
Management Center Innsbruck
Thesis Advisor: Dr. Christian Ploder
Author: Vanessa Katharina Bauer
Date: 9 September 2019
Abstract
This Master Thesis introduces theoretical fundamentals of Predictive Policing tools
used in German police institutions such as Hot-Spot techniques, Near-Repeat
approaches, Risk-terrain Analysis and Concentric-Zone Model. In times of Big Data,
police work has also changed and the usage of forecasting technologies in order to
prevent crime does not only varies state-wide in definitions but also in its’ application.
Therefore, objectives and appliances are described in general. Additionally, a
chronological transformation is established in order to compare lineages in Germany
with those in the USA. Since Predictive Policing polarises, the research question deals
with potential opportunities and challenges police institutions and the society have to
deal with, when it comes to leveraging data-analytical forecasting technologies in order
to prevent crime. For the empirical part, the guideline-based expert interview is key and
conducted on the basis of hypotheses revealed in literature. Results of the expert
interviews (N= 15) were evaluated with an adapted category scheme of Meuser and
Nagel. Resulting, that Predictive Policing is more than a buzzword. Approaches will
definitely manifest itself not only as a supportive tool but also as a ground base for
German police investigations. However, underlying technologies do not include
neutralization software to mitigate erroneous results. As a consequence, the term is in
the public eye with political and ethnological debates. For further results, a long-term
survey would be necessary; collecting and analysing data from similar urban areas
under the same complex conditions, which would require immense financial
investments as well as extensions of fundamental rights.
Acknowledgement
Personal thanks applies to the scientific co-worker and criminologist Simon Egbert, who
on the one hand has taken time for a personal interview to present the current state of art
and his published projects, which also concern the topic Predictive Policing. He has
already accompanied various scientific projects in the field of Predictive Policing
especially in German authorities and taught future policing at the Institute for
Criminological Social Research in Hamburg. Thanks, are also owed to Andreas
Vachenauer and Stefan Heisinger. Both have been working for the police for many years
in senior service level and also shared their experience during a personal interview,
discussing aspects of preventive police work and current methodologies. Furthermore, I
would also like to thank Ulrich Mayr who work as Commissioner for Domestic Violence
in the Swabia police district and shared professional insights as well as associated
technical changes from a different perspective.
I
Table of contents
LIST OF FIGURES………………………………………………………………….. III
LIST OF TABLES………………………………………………………………........ III
LIST OF ABBREVIATION…………………………………………………………. IV
1 INTRODUCTION………………………………………………………………….. 1
1.1 Criminal prosecution in times of Big Data…………………………………......... 1
1.2 Environmental circumstances and status quo of research area…………………... 3
1.3 Motivation driving the research question………………………………………... 6
1.4 Thesis structure……………………………………………………………........... 7
2 TEORECTICAL BACKGROUND……………………………………………...... 8
2.1 Terminology of Predictive Policing and related buzzwords……………….......... 9
2.2 Objectives and appliance of Predictive Policing………………………………… 11
2.3 Policing nowadays and its chronological transformation………………………... 15
2.4 Underlying theories and techniques ...……………………………………........... 16
2.4.1 Hot-Spot techniques as part of crime mapping …..……………………........ 18
2.4.2 Near-Repeat approaches……………………………………………………. 21
2.4.3 Risk-Terrain Analysis.....…………………………………………………... 24
2.5 Lineages in Germany compared to the USA…………………………….............. 25
3 EMPIRICAL WORK……………………………………………………................ 28
3.1 Guided expert interviews as an instrument of data acquisition…………………. 29
3.2 Qualitative implementation and setting………………………………………….. 30
3.3 Participants and Recruitment…………………………………………………….. 33
3.4 Hypothesis and evaluation methodology………………………………………… 36
II
4 DISCUSSION: OPPORTUNITIES AND CHALLENGES……………………... 42
4.1 Interpretation of Results……………………………………………………......... 42
4.2 Answer of the Research Question...……………………………………………... 46
4.2.1 Opportunities of applying Predictive Policing……………………………… 47
4.2.2 Challenges of applying Predictive Policing………………………………… 49
5 FINAL REMARKS………………………………………………………………… 53
5.1 Conclusion……………………………………………………………………….. 53
5.2 Limitations and further research…………………………………………………. 56
ACADEMIC REFERENCES……………………………………………………….. 58
NON-ACADEMIC REFERNCES………………………………………………….. 67
AFFIDAVIT………………………………………………………………………….. 69
TABLE OF APPENDIX……………………………………………...……………… 70
APPENDIX…………………………………………………………………………… 71
III
List of Figures
Figure 1: Percentage of total criminal offences in Germany in 2018……………….. 4
Figure 2: Criminal activities in Germany from 2015 to 2018………………………. 5
Figure 3: The Prediction-Led Policing Business Process…………………………… 13
Figure 4: Predictive Policing process from a methodological perspective………….. 14
Figure 5: Hot-spot and heat-maps coordinated by grids…………………...………... 19
Figure 6: Concentric-Zone model by Ernest Burgess………………………………. 20
Figure 7: Arranged excel file and QGIS settings……………………………………. 22
Figure 8: Final Near-Repeat result…………………………………………………... 23
Figure 9: Predictive Policing in Germany…………………………………………... 27
Figure 10: Interview evaluation process…………………………………………….. 40
Figure 11: Conclusive facts and figures……………………………………………… 56
List of Tables
Table 1: Law enforcement use of predictive technologies…………………………….17
Table 2: Hypothesis and assigned subcategories……………………………………... 37
Table 3: Examples of the second evaluation process step……………………………. 40
Table 4: Examples of the third evaluation process step………………………………. 41
Table 5: Examples of the fourth evaluation process step……………………………... 41
IV
List of Abbreviations
BKA
German Federal Criminal Investigation Office
ETL
Extraction–transformation–loading
IBM
International Business Machines
LKA
German State Criminal Investigation Offices
PI
Police Institution
PP
Predictive Policing
PRECOBS
Precrime Observation System
QGIS
QuatumGIS
RTM
Risk Terrain Modelling
SC
Society
1
1 Introduction
Since Predictive Policing is a highly topical issue especially in Germany, the introduction
provides a basic understanding of criminal prosecution in times of Big Data. Furthermore,
the purpose of thesis and the status quo of research areas as well as clear problem
definitions are explicitly addressed.
1.1 Criminal prosecution in times of Big Data
At least after the successful British science fiction series Black Mirror by Charlie
Brooker, the immense influence of Big Data and Artificial Intelligence on the media
society and its digital future has become visibly tangible and got shaken in many areas. It
can be seen, that some episodes have terrifying prophetic character (Hand, 2018),
especially when it comes to gathering and evaluating data in order to predict offences and
to automate decision-making processes. Those datasets administer social-media behavior
of citizens but also supplement information, which are collected in public space via video
cameras. Due to the enormous data growth and its variety, new possibilities and methods
in range of data science and thus also in the security sector arises. While the principle of
statistically evaluating offenders’ data in order to leverage preventive approaches is not
entirely unprecedented; but predictive approaches in order to forecast offences, pose a
new challenge. Nevertheless, the potential of merging different technologies such as Big
Data Analytics or Machine Learning has not been fully exploited in every sector.
Especially in the area of crime or crisis prevention, the application of such systems could
be used even more efficiently. Although the topic of pre-crime analysis has attracted more
attention in the last six years, police forces or other police organizations still hardly use
intelligent prognose tools (Gorr & Harries, 2003) to utilize limited resources. In view of
this challenge, companies identified new ways and opportunities to adapt their already
developed applications according to recent use cases. Also, tech-giant International
Business Machines (IBM) as a global Information Technology (IT) service provider
(Gongla & Rizzuto, 2001) announced 2008, that campaigns like 'IBM: we build a smarter
planet' will also address issues such as fighting against criminal operations. In IBM's
matching commercial, a policeman analyzes data on an electronical pad while sitting in a
car meanwhile a burglar prepares during the same time his robbery. In the next scene the
policeman waits relaxed with a cafe in front of the supermarket, which the burglar has
2
chosen as his victim. When the burglar is about to put on his mask to rob the supermarket,
he sees the police waiting for him and greeting him. Frustrated, offender walks away, as
he was again to late. With this scenario, it becomes apparent how IT companies like IBM
imagine the future of law prosecution and the protection of citizens and assets. IBM
Smarter Public Safety solutions are designed to make optimum use of personnel and
resources while minimizing misses (Anand et al., 2011). Since already accumulated data
and reports exist in different police authorities worldwide, IBM's vision is to link them
with each other. Thereby both structured data such as prosecution or judicial data can be
integrated into the analysis as well as unstructured data such as officer notes or social
media activities (Anand et al., 2011). Hence, the term Predictive Policing (PP), which is
one of the keywords making policing easier and less prone to errors in the future, stands
as a synonym for prosecution in times of Big Data.
The conceptional changes and mission of PP is driven by the fact that civil servants and
responsible ones are confronted with an ever-increasing number of different types of
problems (Joh, 2014). These are based on the one hand on the rapidly changing
demographic and ethnological environment, in which officers have to cope with a wider
range of tasks as the rise of urbanization increases the crime rate in certain categories in
both developed and undeveloped countries (Ahmad Malik, 2016). In addition, budget cuts
due to holes in exchequer make it more difficult to work in a targeted manner. Especially,
less staff is provided for dealing with daily tasks. Moreover, a trend in the way how crime
is done in the digital and real world can be observed (Farwell & Rohozinski, 2011). For
instance, nowadays wars can be waged online, since hackers collect data about an
organization such as state institutions in order to hack into the network and then slowing
down operations or even turning off the power of an entire nation. Because more and
more criminal organizations use highly technical means to commit illegal acts (Arquilla
& Ronfeldt, 1993) it is also important for the state to revolutionize police investigation
methods.
By collecting data about a countries’ citizen and geographical occurrences, the focus also
relies on minimizing disadvantageous decision making. Algorithm-based decision-
making should provide system advice and reinforce human advice (Joh, 2015). One of
the applied methods is Crime Mapping in combination with Data Mining techniques.
Hence, predictions can be made at which place, at what time, which perpetrator commits
which crime. A possible scenario could be, that an individual got arrested for a crime that
has not yet been committed. Because, computational criminal forecasts confirmed after
3
analyzing the individual data he or she will execute a criminal activity. Thus, system
advice origin could strongly influence the human advice origin and the instinctive
behavior of civil servants. According to the Max-Planck-Institute and its evaluation
results, Predictive Policing is particularly successful in Germany in the area of domestic
burglaries (Gerstner, 2017). However, having a closer look at the matter, it quickly
becomes apparent that Predictive Policing is not only a question of how to obtain and
evaluate data in a technical and ethical manner, but also a question of weighing the
consequences for innocent citizen who respect rights and moral claims. In addition, on
the one hand misgivings arise to what extent police forces currently have the funds to
integrate such technologies into their daily work and on the other hand how environmental
circumstances endorse these advancements. Since it is not unequivocal how society
envisages the prospective inclusion of Big Data as a part of police law enforcement
processes.
1.2 Environmental circumstances and status quo of research area
The goal of every police organization is to reduce the crime rate gradually in different
categories by using assistive technologies such as Predictive Policing (PP) software or
digital forensics which is mainly deployed for handling frauds in the range of Information
technologies or computer science (Garfinkel, 2010). But before employing various tools
to achieve the objectives, it is necessary to analyse crime groups and their annual statistics
in order to be able to assess whether the respective resources are deployed successfully.
Since the Master's Thesis focuses geographically on the development and use of
forecasting systems in Germany, only crime statistics of this area are relevant. On the one
hand, this decision stems from the fact that the global examination would go beyond the
scope of the study. Secondly, the limitation of one country allows a more detailed
elaboration as well as a greater similarity and thus comparability of knowledge. The
following graph shows crime shares of different crime categories out of a total of
5,555,520 cases which were recorded in Germany in 2018 (PKS, 2019).
4
Figure 1: Percentage of total criminal offences in Germany in 2018
(PKS, 2019, p. 7)
Admittedly, with 1,936,315 cases recorded, 34.9 percentage of the total number of crimes
committed in 2018 were thefts. Therefore, it is not surprising that PP is increasingly
applied in German federal states to prevent grand and petit larceny. In general, the overall
number of total criminal offences has declined. In detail, a total of 5,555,520 cases were
registered 2018, representing a decrease of -3.6% (PKS, 2019). It is difficult to judge
whether this is due to forecasting technologies such as Predpol or Precobs since different
environmental factors have to be taken into account. The measurement and evaluation of
these statistics depends, on the one hand, on the number of a population in the respective
country, which in Germany grew from 82,501,000 in 2005 to 82,792,351 in 2018. In
addition, the crime detection rate has risen to 57.7 per cent in 2018, which is a new peak
that may act as a deterrent for future ferrymen. On the other hand, Kuo et al. (2001) argues
that the vegetation, architectural style and ecosystem of a city can also influence the
criminal behaviour of its citizens. Meanwhile, Cohn et al. (2000) analysed that weekday
times as well as weather data are crucial predictors of property crimes.
In addition, the level of poverty, the severity of tourism, the presence of the police, the
unemployment rate as well as sociological aspects have an impact on the number of
crimes in a country (Howsen et al., 1987). It becomes evident, that the performance
measurements, which serves as a basis for the evaluation of opportunities and challenges
of data-analytical forecasting technology, depends on a myriad of environmental
circumstances. Therefore, research and analysis encompass four correlating players
Other criminal
offences
19,50%
Crimes against health
and life
0,1%
Sex offences
1,1%
Petit larceny
19,50%
Grand larceny
15,40%
Fraud
15,1%
Criminal property
damage
10,1%
Criminal assault
10%
Drug offences
6,3%
Immigration law
violations
2,9%
5
operating in the immediate ecosystem of Predictive Policing (PP) and influencing the
human advice origin. This involves, on the one hand, society in general, which should be
willing to accept these new technologies and feed systems with data. Furthermore, police
institutions need to devote time and resources in order to leverage from individual
forecasting innovations. In addition, police officers who are working with those
applications have to act responsibly with sensible data. Moreover, external tech-
companies as major stakeholders, provide the technical framework to integrate such
systems into existing structures and deliver the best possible outcome. Last but not least,
the government also play a decisive role within the ecosystem since they provide on the
one hand the legal framework, allowing police institutions to link and analyse collected
dat. On the other hand, the government guarantees citizens a certain transparency
regarding usage and collection of data.
Although the ecosystem, as mentioned at the beginning of the section, has experienced
positive developments, there are still critical categories, which can be defanged with
foreseeable police work. The following diagram shows the increase in crime between
2015 and 2018 in the categories of murder, weapons misuse and illegal heroin trafficking.
Figure 2: Criminal activities in Germany from 2015 to 2018
(Own illustration based on PKS 2015 - 2018)
It can be seen that criminal offences against the Weapons Act increased by +5.5% in 2018
compared to the previous years and that murder, manslaughter and killing on demand also
2.116
30.004
2480
2.418
34.443
3013
2.379
38.001
3420
2.471
40.104
4002
0
5.000
10.000
15.000
20.000
25.000
30.000
35.000
40.000
45.000
Murder, manslaughter and killing on
demand Offences against the Arms Act Illegal trafficking and smuggling of
cocaine
2015 2016 2017 2018
6
increased by +3.9% (PKS, 2019). An exponential increase could also be observed in drug-
related offences. In addition, an increase in the number of offences of at least +6.0% could
be calculated in the following categories: Resistance to and assault on state authority,
distribution of pornographic literature and drug offences in general (PKS, 2019).
The figures are based on current output statistics of the respective individual data records
available in the State Criminal Investigation Offices (LKA) and the Federal Criminal
Investigation Office (BKA). However, available figures and reports only reflect the so-
called ‘bright field’, which represents the criminality recorded by the police (PKS, 2019).
The number of unreported cases is estimated to be many times higher, especially with
regard to economic offences or corruption (PKS, 2019). These figures are intended to
prove that experts have to consider environmental and external influences when it comes
to Predictive Policing performance measurements. In this respect, positive evidence is
difficult to substitute (Santos, 2019). But still, the rise of criminality in various types of
crime indicate the necessity to invest in new technologies.
1.3 Motivation driving the research question
The motivation for writing the Master Thesis about the present topic stems from the fact
that it is highly current and has not yet been thoroughly studied. Preventing crime and
thus ensuring a safe environment is an important field of research in our society and
should be guaranteed with problem-oriented policing. Since there are varying
considerations and application measures of PP according to different country side
frameworks, the Thesis provides an overview about technical functioning and practical
appliance within Germany. Therefore, content provides on the one hand added value for
lecturers and students in the field of Public Security Management and related studies or
police officers in the upper grade of the civil service. On the other hand, it serves to
educate citizens about how far the technologies have progressed in this area and to what
extent this will influence the lives of citizens in the future. Many police departments
worldwide test software-based forecasting technologies according to their relevance in
practice. Forecasting systems work with data sets about already registered crime
activities. Those datasets are then complemented with socio-spatial, calendar and
meteorological data. Since the amount of collected and analyzed data increases day by
day, the question arises as to what extent Machine Learning and Artificial Intelligence
will influence the human advice origin to predict and prevent crime. The political and
7
judicial expectation regarding this automated data processing, promises to be capable of
preventing potential dangers and risks (Egbert & Krasmann, 2019). However, precise
analysis of the consequences for social impacts has not yet been systematically
researched.
PP, as described above, is applied in Germany to identify urban areas where burglaries or
vehicle thefts are most likely to occur. Predictive Policing, as it is currently used in
Germany, therefore does not constitute an encroachment on civil rights (Knobloch, 2018,
pp. 3-5). The various limited systems in action mainly process location-based
information, not personal data. However, this has led to the deduction of the research
question, as more and more countries are also focusing on personal analysis (Kaufmann
et al., 2018). The concrete research question of the project is therefore: ‘What are potential
opportunities and dangers for German police institutions and the society to leverage data-
analytical forecasting technologies in order to prevent crime?’ The aim of the Master's
Thesis is to determine how the introduction of technologies for predicting criminal
offences affects police practice (Egbert & Krasmann, 2019). In addition, it will be
researched to what extent PP and the underlying technologies change both the police's
and society's understanding of crime and danger. In a nutshell, the Master’s Thesis is
driven by the motivation, to provide replies to the following problem statements. The
problem statement base on the fact that with increasing population, criminal activities in
certain areas may rise. Therefore, it is indispensable to evaluate opportunities and
challenges of new prediction methods in order to be able to implement them in a
purposeful way. Furthermore, it is necessary to elaborate PP which is often perceived as
a black box in both police precincts and society.
1.4 Thesis structure
First of all, the introduction provides an overview of the topic by describing practical
examples from law enforcement in times of Big Data. After that, the environmental
conditions and focus as well as the research question and problem statement are explained
in more detail. Before specific opportunities and risks of PP are considered, chapter two
elaborates a conceptual and theoretical delimitation. In order to bring all readers to the
same level, the theoretical background with its definitions guarantee a basic knowledge.
This part of the Thesis will mainly reflect the informative purpose. In particular, the
terminologies such as Hot-Spot techniques or Risk-Terrain Analysis. In addition, not only
8
the goals of Predictive Policing are explained, but also other definitional refinements such
as Big Data are given. In addition, the current police work and its chronological changes
will be presented by taking reference to international application. Regarding this, an
outline of the international and national status quo is given for subsequent analysis.
Afterwards, the Master Thesis will illuminate the crime forecasting process and the
respective data generation methods according to its technical background (Ahishakiye et
al, 2017). Thereby the different categories in which data can be collected, such as
predicting time versus predicting the offender, should be differentiated in order to
guarantee a holistic picture.
In particular, the social embedding and evaluation of the criminological findings plays a
decisive role for the acceptance of further models. On the one hand it is necessary to
discuss already existing data collection and processing systems and possibly modify
them. On the other hand, the appropriate employee training in order to be able to leverage
an active external communication, should be eroded (Knobloch, 2018). To substantiate
statements, a qualitative study and expert interviews are carried out. The empirical data
collection first examines content of the expert interviews (Gobbens, 2010), whereas the
section ‘discussion’ interprets the results with regard to chances and dangers of described
technologies. Thereby, questions also deal with the practical appliance of Predictive
Policing methods, to what extent they are already used and how further developments
might decrease crime or might change natural behavior (Moses & Chan, 2018). This
section also provides explicit answers to the research question. At the end some future
scenarios and concluding remarks are mentioned.
2 Theoretical Background
The theory provides basis for generating qualitative research surveys with appropriate
hypotheses. On the one hand, the terminology, transformation, goals and applications of
PP are explained, but also basic theories and concepts such as Near-Repeat approaches
are elucidated. Afterwards it was necessary to compare current methods used in Germany
with those of the USA. Since Predictive Policing has its’ origins there.
9
2.1 Terminology of Predictive Policing and related buzzwords
Buzzwords such as Cloud Computing, Predictive Analytics, Information Centric Net-
working, Internet of Things and Big Data, have already impacted the futuristic paradigms
as well as the human horizons in the range of health, demographic change, sustainable
agriculture and mobility but also in the range of societal security (Atzori & Morabito,
2017, p. 1). Looking at the definitions of those buzzwords, it becomes apparent, that all
have a common denominator. These technologies are based on the exchange, analysis or
storage of data and information, which are not only acquired from man-to-machine but
also from machine-to-machine. For instance, Lupton (2014, p. 7) defines Internet of
Things as a ‘tool in which digitized everyday objects or smart things are able to connect
to the internet and with each other and exchange information without human intervention,
allowing for joined-up networks across a wide range of objects, databases and digital
platforms.’ Meanwhile predictive analytics deals with empirical methods to generate data
predictions based on analysed information as well as to assist in creating practically useful
models by leveraging explanatory modelling (Shmueli & Koppius, 2011, p. 1). As PP is
an interface of these various technologies and has developed from the fusion of different
approaches, it is difficult to assign the term to a topic.
However, Big Data and predictive analytics methods are particularly relevant in this
context. Since the data sets, which are collected for example via social and information
driven networking represents, similar to the already mentioned buzzwords, the genesis of
PP. The increased performance and complexity of computer systems and hardware
(Kümmerle, 1970, p. 342) as well as user-friendly software and exponentially growing
data volumes have intensified the human consciousness about predictive modelling.
These exponentially growing amounts of data, also known as data floods, are summarized
under the umbrella term Big Data and built ground for PP. But the term has become
ubiquitous (Ward & Baker, 2013, p. 1). The usage of the term varies according to a wide
range of settings, thereby its meanings are still blurred (Mauro et al., 2015, p. 1). Mauro
et al. have reviewed both, the existing definitions of Big Data as well as related research
priorities. They concluded that the core characteristics of Big Data can be summarized by
the following definition: ‘Big data represents the Information assets characterized by such
a High Volume, Velocity and Variety to require specific Technology and Analytical
Methods for its transformation into Value’ (Mauro et al., 2015, p. 7). This abundance of
information assets provides ground for various initiatives, but especially for forecasting
10
future behavior and occurrence. According to Weiser (1991, p. 66–75) the most profound
technologies are those that disappear as they weave themselves into the fabrics of
everyday life until they are indistinguishable from it (Weiser, 1991, p. 66–75).
Nowadays, the term Big Data has proven itself in the conjunction with predictive analyses
and is used on a daily basis in commercial and private companies as well as in state
institutions. Therefore, it is likely that in times of Big Data the future of police work is
envisioned in PP. Predictive policing in general, implies forecasting technologies, which
are able to predict the places and times of future criminal activities and preventing them
ad-hoc. Perry et al. (2013, p. 16) defines PP as an application of analytical techniques -
particularly quantitative techniques - to identify likely targets for police intervention and
prevent crime or solve past crimes by making statistical predictions.’ Thereby the
expression ‘prediction’ signifies a probability proposition and is therefore not defined as
a fact (Belina, 2016, p. 4). According to Perry et al. (2013) linguistic differences might
exist between the terms Predictive Policing and crime forecasting systems, but in practice
the same technical process is meant. Based on this, the Master Thesis would also use
these terms interchangeably. The generic term of both systems is Event Forecasting,
which is according to Armstrong (2001), defined as follows: ‘forecasting is often
confused with planning. Planning concerns what the world should look like, while
forecasting is about what it will look like.’ In detail, forecasting systems utilize past data
as a source to provide reasoned estimates that are predictable to identify the direction of
prospective events.
All terms, focus on similar outcomes – to reduce the risk of a certain event, activity or
reaction (Armstrong, 2001). However, for Egbert (2019, p.1) keywords such as ‘Big Data
Policing’ or ‘Algorithmic Policing’ are misleading in regards to PP. The idea, that police
institutions use digital technologies and sophisticated data mining systems in order to
combat crime seems to be a futuristic approach for society. However, practical
approaches indicate that these forecasting tools leverage with a more conventional
appliance than scientific references and narrative paradigms states. Ultimately, these
systems build on existing concepts and criminological insights, such as Rational-Choice-
Theory, Problem-Oriented Policing, Location-Based Policing or environmental
criminology (Egbert, 2019). Innovation is guaranteed on the one hand by the provision of
societal records and on the other hand by technically combining these datasets. For
instance, making sense of scaled crime-based data by linking databases to open up new
analytical approaches (Kitchin, 2014, p. 100) such as algorithmic-mediated analysis,
11
seems especially ground-breaking for German police institution. From Egbert's (2019)
point of view, the term PP can be divided into two layers since it is a cross-sectional
strategy with multidimensional processes. The first layer collects data and leverages input
techniques. It is important to consider the extent to which these techniques can really be
used in daily police tasks in terms of resources and personnel. The second level refers to
the generation of crime predictions by algorithmically mediated data analysis. Thus, PP
is an application of data analysis technologies used by the police to generate usable
predictions about scenarios and spatiotemporal conditions (Egbert, 2019, p. 3).
Concluding, the definition of Morgan in Craig D. Uchida's National Institute Justice
report (2009, p. 2) provides ground for an overarching terminology by covering discussed
characteristics. Predictive Policing refers to any policing strategy or tactic that makes
usage of Big Data. Those datasets are processed via analytical and quantitative
techniques. The aim is on the one hand to inform forward-thinking and crime prevention
in the range of scenarios and spatiotemporal conditions. On the other hand to predict
criminal offences throughout probability and criminal patterns in order to use existing
resources to an optimum.
2.2 Objectives and appliance of Predictive Policing
To ensure that these new approaches gain acceptance, evolve over time and are
implemented by policemen in different levels, results must be tangible (Perry et al., 2019,
p. 3). According to Završnik (2017, p. 1) the objective of PP is ‘to issue crime forecasts
in the same way as the Weather Service issues storm alerts’ and thereby to disrupt the
‘production cycle’ of crime. The opportunity of automated justice is to vaporize biases,
heuristics and to confine fundamentally value-based decisions to ‘clean and pure’
mathematical reason (Završnik, 2017, p. 1). The use of intelligent prediction tools, if
implemented and trained wise can provide various benefits. The analytical function
develops a variety of intelligent products to assist investigators in detecting, predicting
and solving criminal investigation. Thereby, prosecution is based on collected data, which
is displayed in well-arranged tables, charts, maps or other visuals. Aiming to support on
the one hand adjudication of trials and on the other hand to support decision-management
of chief executives or agency’s mission (Ioimo, 2018, p. 6). Law enforcement officers
can benefit from tactical and strategic recommendations. Such reports include crime hot-
spots, crime bulletins and summaries, view crime trends, possible threats, vulnerability
12
or provides risk assessment analysis (Ioimo, 2018, p. 7). Computerized databases as a
ground for PP, organize information and fosters meaningful relationships with other law
enforcement staffs by allowing them to quickly obtain information and assisting in
multijurisdictional cases. Since Predictive Policing tools are adapted to respective legal
conditions, results are compliant with local, state, tribal and federal laws and
regulations (Ioimo, 2018, p. 7). In addition, by leveraging underlaying software, existing
resources can be utilized in a more efficient way and resources can be saved in the long
run. This does not necessarily mean personnel savings, but rather the appropriate
employment of police officers. For instance, PP systems can process large amounts of
data in a short period of time. In the meantime, police officers have more capacity for
other activities such as street patrol. In this way, the use of new forecasting technologies
can simplify investigative efforts and, in the best case, crime statistics can be reduced
through appropriate prevention tactics. Therefore, an added value can be achieved in the
following sectors: police personnel management such as professional deployment and
recruitment, police budgets management such as measuring the costs of overtime and
other expenditures, offender monitoring, city or neighborhood planning i.e. design of
spaces, police security resource allocation or infrastructure protection (Craig D. Uchida's
National, 2009, p. 6). For instance, the LKA of the federal state North Rhine-Westphalia,
identified during the implementation phase of the PP project called SKALA, following
objectives for German police institutions: ‘its purpose is a strategic and target-oriented
police work, which detects emerging hotspots at an early stage on the basis of known,
crime-relevant determinants. The aim is to achieve a resource-conscious deployment of
police forces and a reduction in the frequency of crime’ (Landeskriminalamt NRW, 2018,
p. 10). This motive is predominant in the German federal states, although the total crime
rate in Germany is declining (Knobloch, 2018, p. 10). In practise, different types of
prediction are distinguished. On a superordinate level prediction between space and time
or persons can be differentiated (Krasmann & Egbert, 2019, pp. 12) and combined in
appliance. The data collection thus is determined in space and time related data or person
related data. Location and time related forecasts include methods for predicting crimes,
these approaches are used for predicting places and times with an increased risk of a
specific criminal activity (Perry et al., 2013, p. xiv).
The most common forecasting technology in the range of criminal activities are Hot-Spot
methods since police departments mainly work with location-based data. Here, ‘crime
analysts prepare maps of crimes that have already occurred and those maps are used to
13
deploy officers and to identify areas in need of intervention’ (Groff & La Vigne, 2002,
pp. 34). Methods for predicting offenders, perpetrators’ identities and potential crime
victims are approaches, which are more based on person related datasets. In detail,
methods for predicting offenders aims to identify individuals, who commit a crime in the
near future. Meanwhile, predicting perpetrators’ identities focus on profiles that
accurately match likely offenders with specific past crimes. Tools which forecast
potential crime victims are able to identify groups or individuals who are more likely to
become victims of offender (Perry et al., 2013, p. xiv). Although the forecasting
capabilities can be classified into different categories, further implementation and
appliance processes are similar. Perry et al. (2013) summarizes that the process of PP can
be presented in a classic four-step comprehensive business cycle (Perry et al., 2013). As
can be seen in figure three. First two steps deal with the collection and analysis of crime,
incident, and offender data, which requires data fusion. For detailed prediction, the data
will be analysed according to the individual police operations and departments.
Figure 3: The Prediction-Led Policing Business Process
(Perry et al. in corporation with RAND, 2013, p. xviii)
Data
Collection
Analysis
Criminal
response
Police
Operation
Situational
Awareness
Increase resources in areas at greater risk
Conduct crime-specific interventions
Address specific locations and factors
driving crime risk
Provide tailored
information to
all levels
Generic
Crime specific
Problem
specific
14
The third step implies performances of police institutions to act against the predicted
criminal event or even to solve old crimes. Possible interventions such as generic, crime-
specific and problem specific aspects are classified according to their complexity.
According to Perry et al. (2013, p. xviii), complicated interventions require more
resources such as personnel but achieve better, more goal-oriented results. In order to
successfully carry out missions, managers should not only discuss the critical part of
preventive analysis but also provide information that fills the need for situational
awareness among officers and staff (Perry et al., 2013, p. xviii). Building on this, the
fourth step of the cycle can be completed. Each intervention leads to a criminal response
which in the best case minimizes the risk or prevents the crime. Here, a short-term
feedback and assessment is considered by guaranteeing, that the interventions are being
carried out correctly and there are no apparent issues. In order to benefit from Predictive
Policing in long-term, it is necessary to reprocess the newly generated data after each
operation, which in turn leads to changing environmental conditions. Even though the
Prediction-Led Policing Business Process by Perry et al. is intended to constitute a holistic
approach, Bode et al. (2017, pp. 1-2) argue that this illustration does not meet the
methodological requirements as it is applied in Germany. The following figure depicts
the process from a police perspective (Bode et al., 2017, p.2) and is adapted by scientists
(Seidensticker, 2017, p.296).
Figure 4: Predictive Policing process from a methodological perspective
(Bode et al. 2017, p. 2; Seidensticker, 2017, p. 296)
Whether the usage of Predictive Policing tools is beneficial in the respective context or
not, depends on the resources. Therefore, it is necessary to start with a delinquency
analysis and to distinguish what application makes sense (Seidensticker, 2017, p. 296).
At the end, the executors give feedback whether the use of the Predictive Policing tool
was meaningful in this case. Thereby, deviations within different federal states are always
1. Data
•Collection
•Selection
•Processing
2. Statistics
•Modelling
3. Machine
Learning
•Forecast
calculation
4. Visualization
•Forecast
presentation
5.Measurements
& action
•Forecast
exploitation
by policemen
6. Formal evaluation, feedback and assessment by police department
Delinquency analysis
15
possible. The first step initiates the review and selection of data records as well as the
collection and processing of datasets. Thereby, spatial and temporal consolidation are
central. Already recorded police-related data can be combined with non-police-related
data such as weather or temperature changes. For this purpose, it is important to
geographically reference data in order to guarantee a uniform, machine-processable
dataset, which is ground for further analysis. During the second step, a concrete statistical
model, such as a regression or decision tree (Box et al. 2015, p. 305), is created depending
on the available data. The third phase involves analysing the data based on the selected
probability model. For example, the forecast calculation points out, which offence with
an increased risk will take place in which area. Within the fourth step results of the
prognosis are presented graphically for police officers and investigators. This can be done
with proper dashboards on smartphones or tablets in order to utilize them ad-hoc.
Depending on the current scenario, the data can be used as a basis for decision
management. Thereby policemen can implement accurate prevention measures. The last
step describes the performance measurement of the applied modelling, memorizes
lessons-learned and verifies the plausibility of intermediate results (Bode et al., 2017, p.
3). Compared to the Prediction-Led Policing process by Perry et al., the adaption of Bode
et al. includes a continuous evaluation after each step and leverage the feedback-culture.
2.3 Policing nowadays and its chronological transformation
Due to business consolidation and marketplace fragmentation the nature of police
organization has fundamentally changed. Even though changes can be seen in stuffing or
digitalization, physical approaches and objectives have remained steady. Preventive
policing and the assurance of social security by tailored interventions according to the
context, has always been a domain for police institutions. Between 1990 and nowadays
there have been various policy changes within German police prevention programmes.
The transformation is triggered by the anticipable adaptation phenomenon of social
transition but also by devastating happenings such as assassinations (Behr, 2016, p. 1).
Particularly in the 1990s, attempts were made to improve the relationship between police
as an organisation and the population. Thereby citizens were considered as customers and
social skills were decisive (Behr, 2016, p. 2). Especially, terms such as smooth policing
or community policing are coined. Aiming for a holistic, service ecosystem where police
acts as a buffer between citizens, stakeholders and the government.
16
In the early 2000s, the term smart policing was used, which not only links police
operations from a technical perspective but also leverage knowledge from other countries.
Smart policing initiatives and evidence-based, data-driven criminal-prosecution-tactics
built ground for Predictive Policing (Coldren, 2013, p. 275). However, such systems have
gained popularity for the first time in 2010. The first accompanying scientific evaluation
of PP as a test operation was conducted in Germany between 2015 and 2017 (Knobloch,
2018, p. 5). Thereby, new opportunities and challenges for stakeholders and police
organisations emerged in order use existing systems more effectively and benefit from
exponentially surging data volumes. In particular, there is a shift from post-crime to pre-
crime analysis (Wall, 2010, p. 22). These significant shifts ‘have occurred including
major reforms in public policing, and a substantial expansion of the private security
industry’ (Jones, 2002, p.1). These revolutions are not only due to the technical
circumstances but also the society demands change. Already, US-American psychologist
Abraham Maslow analysed that the need for security is the second most important value
for human being (Stum, 2001). To meet these societal needs, the police on the one hand
strengthened internal employee skills by implementing advanced study possibilities and
on the other hand increasingly cooperate with external IT enterprises. Since experts
operating in the free market economy are driven by different motivational factors and
benefit from the competitive business environment. As a result, talents in the free market
economy are used to work with current software and hardware systems, which makes
potential employers even more attractive. Since the police work is generally accused of
being too static and outdated (Marks, 2000, p. 1) the technical know-how of external
organizations has to be implicated. In turn, police processes are becoming more
transparent for society, which at the same time corresponds to the approaches of
community policing. In order to avoid the danger of being dependent on external
companies, internal personnel should be trained together with external staff. In
conclusion, Predictive Policing is not an entirely novel approach of policing, it is rather a
combination of prior police practices such as Crime Mapping and Geo-Policing, Quality-
of-Life Enforcement methods or Intelligence-Led Policing (Egbert, 2018, p. 10).
2.4 Underlying theories and techniques
For various use cases, different techniques can be used to facilitate criminal investigation
proceedings. These various approaches are not mutually exclusive, and can be combined
17
for a more detailed outcome. Before processing the data on the basis of the selected
theory, they must be available in systemic order. Since approaches are differentiated into
spatial and person predictions as well as into predicting crimes, predicting offenders,
predicting perpetrator identities and predicting crime victims, the following table
provides an overview of individual instructions.
Person-based Predictive Policing
measures
Spatial- and time-based Predictive
Policing measures
Predicting
offenders
Predicting
crime
victims
Predicting perpetrator
identities
Predicting crimes and
areas
• Regression
and
clustering
• Heat List
techniques
• Strategic-
subject-list
• Crime-
mapping
tools
• Data
mining
techniques
• SPSS
Modeler
and SPSS
Statistics
• Computer-assisted
queries
• Statistical modeling
• Geographic profiling
tools (e.g. Hot-spot
analysis)
• Analysis of sensor
databases
• Crime mapping (Hot-
Spot techniques)
• Near-Repeat
approaches
• Risk-terrain analysis
• Regression,
classification, and
clustering models
Table 1: Law enforcement use of predictive technologies
(Own illustration based on Perry et al, 2013, pp. 10-14;
Egbert & Krasmann, 2019, pp. 12-15)
In detail, the table reviews, which approaches are already in use in order to statistically
predict potential offenders, victims and to forecast in which area, when and what criminal
activity could happen. These software solutions and tools are based on different
psychological theories. Particularly relevant in this branch are Routine-Activity Theory,
Rational-Choice Theory and Broken-Window Theory. The Rational-Choice approach
states that all human actions are driven by desires and needs. In order to achieve this state
of affairs, the perpetrator has to weigh up the costs and benefits. Which means that the
higher the personal benefit, the more the perpetrator is willing to dare and sacrifice (Hill,
2002, pp. 29). Meanwhile, the Broken-Window Theory states that there is a causal link
between physical and social disorder functions. Therefore, it is more likely that criminal
18
activities occur in decayed and disordered areas then in neat neighborhoods (Ren et al.,
2019, p. 1). The Routine-Activity Theory is based on the assumption that offenders act
rationally and deliberately and thus follows a pattern. The precondition is, on the one
hand, a motivated offender and an appropriate, unprotected target or victim (Santos, 2015,
pp. 108-109). In order to determine the identity of possible offenders, both place and time
as well as personal data are required and thus specialists combine methods. Even though
person-related proceedings are becoming more and more attractive, civil servants in
Germany concentrate on spatial proceedings (Egbert & Krasmann, 2019, p. 12) and
mainly employ them in the area of burglary. The main justification behind their decision
is, that these procedures are promising as they not only process individual data sets but
also search for patterns. This indicates that crime follows a pattern in certain areas such
as planned burglary. Since Near-Repeat, Hot-Spot Analysis and Risk-Terrain Analysis
are the most frequently used techniques in Germany, the consideration is limited to these
methods. In addition, both person-related and spatial predictions can be realized with
Near-Repeat approaches. The explanation of more theories such as Heat-List techniques,
Strategic-Subject-List, computer-assisted queries or statistical modeling would go
beyond the scope of the master thesis and are therefore skipped.
2.4.1 Hot-Spot techniques as part of crime mapping
Crime mapping is a generic term used in criminology to describe the compilation,
visualization of spatial crime patterns. Based on this, crime cartographies can be drawn
according to the respective city (Paulsen et al., 2009). In order to calculate such crime
maps, geo-information systems are employed which do not indicate plotting of crimes,
but serve as a tool for processing collected spatial data. Collecting data refers to the
assumption ‘that crime will likely occur, where crime has already occurred. Thereby, the
past is prologue’ (Perry et al., 2013, p. 19). Crime mapping mainly refers to linking crime
scenes and perpetrators on a map by means of geographical information and spatial-
temporal coordinates (Hadamitzky, 2015, pp. 9-13). In this context, attempts are made to
trace past crimes in order to find the perpetrator or the victim.
Since 2015, crime mapping in Germany has also been used to predict potential crime
scenes and areas with a high crime density. Crime mapping is most frequently used in the
areas of street robbery, burglary, vehicle crime or community borders. These predictive
crime mapping methods are known in police jargon as Hot-Spot techniques. For Eck et
19
al. (2005, p.3), ‘hot spot is an area that has a greater than average number of criminal or
disorder events, or an area where people have a higher than average risk of victimization’.
Hot-Spot Analysis helps the police to identify areas of high criminality, to predict the
types of crime, which might be committed and suggest prevention tactics (Eck et al, 2005,
p. iii). Recent developments indicate, that approaches differ on the level, the hot spot size
and the geographic area of crime (Levergood et al., 2000, p. 2). In detail, place
predications, which forecast a crime at a specific coordinate, differ in visualization from
street, area or repeat victim prognoses (Eck et al, 2005, p. iii). Experts use different
methods according to the questions and objectives. For instance, the question 'where are
drugs sold?' refers to the identification of specific drug trafficking locations or street
segments where drug traffickers and customers routinely meet. While 'what is the market
for drugs' as second query, sounds similar, but focus on the origin of the customers (Eck
et al, 2005, p. 1). Then, these results are visualized on city maps. In particular, streets with
high crime probabilities are marked with lines. While a whole residential area behaves as
a polygon and the forecast of a certain event is marked as a point. In addition, hazardous
areas can also be displayed using heat maps. In practice, policemen can use up-to-date
heat maps in order to decide in which areas patrolling is more efficient. The following
image on the left shows a red polygon, which encircle a German neighborhood, where
burglaries are very likely to happen in the near future. While the graph on the right
represents a 'heat-map' of Portland, subdivided into grids. Red dots prove that there has
been a high density of crimes and therefore individual measures have to be taken.
Figure 5: Hot-spot and heat-maps coordinated by grids
(Landeskriminalamt NRW, 2018, pp. 4; Dantec, 2016)
Particularly relevant when using crime mapping is the inclusion of the Concentrated-Zone
Model, which visualizes the distribution of social groups in urban areas (Seidensticker,
2017, p. 293). This model states that a town can be scaled in rings starting with its center.
20
Figure 6: Concentric-Zone model by Ernest Burgess
(Own illustration based on Burgess, 1925)
As a result, the social status of the inhabitants changes according to the location. Families
who reach a higher socio-economic standard, leave the central zones and tend to move to
the outer areas of the city. While workers remain in the inner zone. Even though these
assumptions are already considered outdated, since there are now several centers
belonging to a city and the growth of cities is causing a constant shift in resident
structures, this can be of criminological interest. Based on this, special measures and
resources can be carried out in this area and the algorithm can be adapted. Burgess
confirmed that crime in the vicinity of centers, especially in the second zone, the so-called
'transition zone', reaches its highest level (Burgess, 2008).
Furthermore, it is important to distinguish between short- and long-term risk reduction.
For instance, George O. Mohler criticizes that the following methods, which are already
used in the USA, only provide short-term added value: ‘basic crime prediction or
forecasting techniques include: Crime counts, pin maps depicting past crime locations,
and crime hotspot maps. However, these methods have generally proven unsatisfactory
because they fail to take into account both long-term spatial variation in risk as well as
short term elevation in risk following crime in a systematic way.’ He argues that crime
hotspot maps in particular attempt to quantify the contagious spread of crime after past
events. However, they cannot estimate the probability of future background-events,
1. Central Business District (Zone 1):
- Highest land value
- Earns maximum economic returns
2. Transition Zone (Zone II):
- Mixed residential and commercial use
- Abandonment buildings
3. Inner City/ Working Class (Zone III):
- Single family tenements
4. Residential Zone (Zone IV):
- Single family tenements with yards
- Better residential area
5. Commuter Zone (Zone V):
- Peripheral area & high-income groups
21
which are the first events for triggering crime clusters. Additionally, Hot-Spot maps
usually assume that short-term crime activities may continue in the future without linking
to background events in order to find hidden pattern. Also, Gorr, W. and Harries, R.
(2003) highlighted that ‘conventional forecasting methods are of little use in predicting
the behavior of individual serial criminals’, but particularly Rule-based Expert Systems
or Data Mining technics can be useful for predicting the whereabouts of serial criminals
(Rossmo, 1999). Especially geographic profiling, which is described as a sub-discipline
of Environmental Criminology, investigates how crime occurrence is influenced by
opportunity. In summary, criminal events must be understood as several dependent
junction points in which law, offender, victim described as a goal, meet at a certain time and
location (Laukkanen, 2007).
2.4.2 Near-Repeat approaches
The algorithms behind Near Repeat approaches are used all over the world and are mainly
employed within the software PRECOBS (Precrime Observation System). Also, these
systems require predictable patterns in human origin in order to deliver reliable outcomes
(Haberman & Ratcliffe, 2012, p. 2). The idea of Near Repeat phenomenon is that previous
crime events leads to a heightened risk of victimization for spatially proximate targets,
places or victims (Johnson & Bowers, 2004, pp. 237–255). Bernasco (2008, p. 414)
discovered after several interviews with criminals that burglars break into a property
where they have already broken into before with a probability of 31 to 76 percent. But
also, nearby areas are of danger for a certain time frame.
Near repeats comprise multiple criminal acts, while Near-Repeat Pairs constitute of two
offenses. This does not exclude that one offence may be part of several Near-Repeat Pairs
(Schweer, 2015). The more Near-Repeat Pairs can be found, the better the prediction and
qualitative evaluation of a geographical area is. According to Haberman & Ratcliffe
(2012, p. 2) ‘the near repeat phenomenon therefore highlights crime clusters in space and
time’. Cluster are predefined by polygons, which allow the prediction of spatial and
temporal events without a fixed spatial reference (Seidensticker, 2017, p. 296). Polygons
are a prerequisite for not creating artificial geographical boundaries between near events.
Even if the Near-Repeat approach is not entirely based on spatial restrictions, it becomes
clear that the approach of Hot-Spot methods can be linked. Thus, the following polygons
must be defined in advance: The proximity of two criminal offences, the area unit when
22
events cease to spatially correlate, in which geographical boundaries police measures
should be done or how results and spaces are presented (Seidensticker, 2017, p. 296).
Recent studies could also identify two hypotheses in respect to Near-Repeat approaches.
Specialist differentiate between the ‘boost’ hypothesis as an offender-based perspective,
which suggests that successful past victimization boosts the likelihood of future
victimization (Haberman & Ratcliffe, 2012, p. 3). In other words, the same offender or
colleagues might return to the concerned neighborhood to implement the lessons learned
and capitalize opportunities in order to commit the crime even more efficiently or faster.
Meanwhile the ‘flag’ hypothesis states that an area is attractive when conditions are
optimal and the risk factors are low. Haberman & Ratcliffe (2012, p. 3) summarized, that
‘crime will be concentrated among those targets regardless of whether it is the work of
the same offender’ or not.
For the technical realization of Near-Repeat strategies, the open source program
QuatumGIS (QGIS) can be applied and is due to its simple usage widespread. In this use
case, randomized burglary data from England is prepared for further analysis. In the first
step, the collected data must be extracted, transformed and loaded (ETL) into the
respective open source. The ETL process is often used in data warehousing and describes
the copying of data from one or more sources to a target system that summarizes the data
and displays it in a clear and concise way (El-Sappagh et al. 2011, p. 1). Initially, the data
records from past burglaries are extracted to generate homogeneous data sets. During the
transformation, the data is cleaned up and finally turned into meaningful purpose. In this
case, the coordinates of past burglaries and temporal triggers such as time and weeks are
sorted in Excel and then uploaded to QGIS.
Figure 7: Arranged excel file and QGIS settings
(Own illustration based on University of Leeds, 2003)
23
Even though the Near Repeat approach was mainly evolved for burglaries, the process
works for every type of crime pattern. Afterwards practitioner define on the one hand the
layer name such as ‘burglary in Leeds’ and the geometry usually as ‘point coordination’.
On the other hand, it is necessary to specify the x and y columns, which gives information
about point coordination. After defining the range of the geo-data, for further predictions,
it is useful to set a proper time frame. As offenders usually turn back to the target within
the next four weeks, limiting the results to all criminal activities which happened the last
three weeks makes the result more accurate (University of Leeds, 2003). For detailed
results it is possible to add time period in order to make aware of daily, risky hours. By
setting time filters such as week one, two or three the results of the first search as well as
the new filtered data is displayed in the map. Therefore, it makes sense to transparent the
original results in the layer window in order to review relevant spots filtered by the time
periods on the map. Thereby, spots are small dots or points on the city map representing
the criminal event to a certain time. The last step, includes drawing the buffers around
each event. The buffers are colored depending on the weeks: all offences which happened
the last week are blue, yellow buffers represent burglaries two weeks ago and red three
weeks ago. Once all points are entered a city map can be deposited in order to leverage
police interventions within this district. The following figure shows circles with different
colors and radius, based on past offences.
Figure 8: Final Near-Repeat result
(University of Leeds, 2003)
24
The size of the circles depends on the definitions police institutions made before analysis.
For example, the police of Lower-Saxony defined the radius of the analyzed circles from
100 meters increments up to a diameter of two kilometers (Seidensticker, 2017, p. 297).
The map highlights neighbourhoods where burglaries occurred in every week. If the red,
blue and yellow circles overlap, then this area testifies a higher risk of criminal offence
also within the next weeks (University of Leeds, 2003). In practice, the Near-Repeat
prediction methodology builds ground for the German prediction prognosis instrument
Precobs (Leese, 2019, p. 62). The software is designed by the Oberhausen Institute for
Pattern-Based Prediction and sold globally. Here, among its other features, certain trigger
criteria can be specified for various crime attributes. Each trigger filter contains a list of
characteristics of the corresponding types. If an element of the list is detected during an
offence, the trigger is considered to be positive. A trigger offence consists of various
criteria, i.e. time of offence, prey and modus operandi (Schweer, 2015). But detractor fear
at the same time that the anonymous data collection might be supplemented by person-
related triggers and data. Moreover, since only police registered crimes and patterns are
sought (Biermann, 2015), the software risks narrowing the view to certain places.
2.4.3 Risk-Terrain analysis
In comparison to the other models, Risk-Terrain Modelling (RTM) offers the possibility
to make an analysis that is less influenced by previous events and more driven by a
dynamic interaction between social, physical and behavioural factors at different places.
The main objective is to find a way, how these variables can be combined in order to
identify interaction patterns (Kennedy et al., 2011, p. 342). This approach allows testing
different risk factors regarding accuracy by means of statistically valid selection
processes. Findings reveal risk concentration and clusters with the goal to determine on
the one hand what led to the problems and on the other hand to predict the type of future
crime and direct intervention.
In addition, this approach is also used to enhance resilience and extensibility in low-risk
areas (Kennedy et al., 2011, p. 343). The technical implementation is as follows: First, a
meta-analysis and other empirical methods as well as literature research are used to assess
all risk determinants related to a particular outcome. Each risk factor is then
operationalised and recorded on a map. Essentially, the RTM signifies the intensity,
abstinence and presence of the individual risk factors. Each factor is represented by a
25
separate risk map layer (Kennedy et al., 2011, p. 343). At the end, all map layers are
combined in order to create one holistic risk map. Each location is allocated to a
composite risk value that takes into account all factors associated with the crime result. If
the risk value is higher, there is a greater probability that offences will happen in that area.
In summary, the result of the RTM is a map showing which areas have a high risk for
future criminality (Kennedy et al., 2011, p. 343). This is not only due to police statistics
showing past crimes, but also because the environmental conditions are so favourable
there, that crimes may be committed tomorrow. As a result, violent crimes such as
shooting patterns can be predicted in addition to burglaries (Ferguson, 2011, p.283). By
combining interpersonal risk factors, past crimes and environmental triggers, RTM has
expanded the field of predictive analysis. According to Ferguson (2011, p. 283), future
admission incidents can be better predicted than with hotspot analysis. Aspects or events
that appear random at first glance can be exposed in patterns.
2.5 Lineages in Germany compared to the USA
Despite a certain delay to the origin country USA, Predictive Policing is now also
establishing itself in the German-speaking area; even if the use is defined differently
depending on legal restrictions. The first PP systems according to the described
characteristics, took place in Hollywood in 2011 within the project ‘Foothill Division’
(Mohler, 2015, p. 1403). This led to the development of the tool Predpol, which is
widespread and has been adapted. Originally ‘Foothill Division’ was grounded to
predict earthquakes in California at an early stage. In the meantime, the algorithm has
been expanded and adapted for German purposes as well as for commercial products or
imitators; one of the imitated products is PreCobs, which is mainly used in southern
Germany and Switzerland (Knobloch, 2018, p. 11). However, the introduction of the
systems has also given rise to certain difficulties. Especially in the USA, the potential for
ethnic discrimination is highly estimated since law enforcement, base to a certain extent
on maintained databases and past crime analysis. It is assumed that discrimination against
Afro-American and Latin American populations will continue to intensify (Knobloch,
2018, p. 11). Therefore, its impact may cause courts to reconsider the current approaches
according to reasonable suspicion. Concerns base on the fact, that this technology could
be applied in a manipulatively and discriminatory manner (Ferguson, 2011, p. 313). Since
personal data such as Social-Media behavior, camera shots or purchasing habits of
26
individual consumers are already included in the US rating. Thereby, special actions can
be interpreted negatively even if they are not illegal. Depending on which theory or
individual police assessment the profile is evaluated. Thus, historical crime data
constitute the basis for future police strategies (Knobloch, 2018, p. 11). In America,
personal data is already combined with location data even though somebody is not
classified as a threat. This leads to an over-representation of police forces in poorer areas,
which can lead to the detection of more illegal actions. Thus, theories, stereotypes and
assumptions may harden.
Angwin & Larson (2016) senior reporter at ProPublica voiced concerns regarding biased
Machine Learning and questioned ‘if it is possible to create a formula that is equally
predictive for all races without disparities in who suffers the harm of incorrect
predictions.’ The most prominent case of discrimination in connection with PP systems
took place in the USA. It was determined that the so-called Compas-Software
systematically discriminates black people by allowing judges to calculate how high the
risk is, that offenders will relapse. These calculations serve as a basis for the later
judgement of the offenders (Angwin & Larson, 2016). Hence, it is important for civil
society to assure functional transparency of the algorithms applied, particularly for
commercial Predictive Policing systems. In order to solve concerns Hustedt (2016, p. 12)
suggest, that governments need to consider on the one hand individual cultural
background and on the other hand how technological, managerial and legal factors impact
the functioning of investigations. Police officers and system developer need to realize that
Predpol or PreCobs are no ‘psychic crystal balls’ or neutral aids, since they mostly rely
on biased data (Hustedt, 2016, p. 13).
In Germany, six federal states are already working with both their own creations and
external solutions to prevent mainly burglary. It is worth mentioning here that in Germany
no individual personal data is yet compiled in connection with Predictive Policing. This
is based not only on the privacy protection regulations but also on the legal grey area
which makes it difficult to measure effectiveness. Personal data may only be collected if
there exists a judicial decision based on evidence that the person to be monitored is
actually a peril. The following hierarchy clearly presents the current situation and use of
PP solutions in Germany compared to other countries.
27
Figure 9: Predictive Policing in Germany
(Own hierarchy based on Knobloch, 2017, p. 13; Egbert & Krasmann, 2019, pp. 27-35)
The chart reveals that Germany is currently concentrating on location-based data and
analyses, which does not rule out that other countries such as the USA do not even use
Predictive
Policing
Spatial- and
time-based
data
Near-Repeat
approach
Commercial
Software
(External
consultancy)
Bavaria works with
Precobs / Precobs
Enterprise
Baden-Württemberg
works with Precobs but
also include meterological
and socio-ecological data
and theories i.e.
Concentric-Zone model
model
USA works with Predpol
Software base on
excisting theories
but is adapted
according to
individual
context
Berlin works
with KrimPro
based on
Precobs
Hot-Spot
Modelling
Lower Saxony
works with
PreMap
Criminologically
more complex
approaches with
larger particular data
sets
Northrhine-Westphalia
works with SKALA and
SPSS Modeler from IBM
Hesse works with KLB-
operative
Risk-Terrain
Modelling
Hesse works with
HessenDATA similar to
Radar-ITE (Risk Analysis for
Terrorist Attacks)
Person-based
data
Strategic-subject list,
Data mining techniques,
Regression or Computer-
assisted queries
For instance
UK, USA or
China
28
location- and person-based data. In summary, Hesse rely on the one hand on Risk-Terrain
Analysis as well on the Near-Repeat modelling. In practice, they developed their own
solutions inspired by Predpol. Every morning, the situation of the previous day is
computer-assisted processed, analysed and displayed on a map with the help of the
‘KLBoperativ’. On the basis of the results, a deployment plan for the next 24 hours is
drawn up. In order to be able to deploy police officers as effectively as possible. Based
on an improved crime analysis, the emergency forces in Hesse can be deployed very
specifically in designated focal point regions at relevant times (Hessen Ministerium,
2016). While HessenDATA, similar to Radar-ITE, represents Islamist terrorism as a rule-
based analysis of potentially destructive perpetrators to assess the acute risk
(Bundeskriminalamt, 2017).
North Rhine-Westphalia has also developed an individual system based on the SPSS
Modeler, which was used within the SKALA project from 2015 to 2017. However, the
federal states of Baden-Württemberg and Bavaria rely on the commercial version
Precobs. But Bavaria has strengthened systems and now employs Precobs Enterprise,
which allows a multidimensional data analysis that does not only consider predictions
(Egbert & Krasman, 2019, p. 35). The LKA in Berlin has already been working with
‘KrimPro’ since 2016, which is considered to be particularly versatile. Nearly 60 percent
of the predicted housing collapses in 2018 actually occurred (Dinger, 2019) and could be
prevented. Last but not least, Lower Saxony focuses on the application of Hot-Spot
modelling and leveraged Predictive Policing Mobile Analytics for Police also called
PreMap. Here, ad hoc data is analysed on mobile devices such as tablets, enabling risk
estimations of criminal acts and suggesting possible preventive measures.
3 Empirical work
The following section describes on the one hand why expert interviews were the most
suitable method for the empirical survey and on the other hand how respective experts
were acquired as well as how setting was constructed. In addition, this section evaluates
which hypotheses result from the literature and which evaluation method is particularly
relevant.
29
3.1 Guided expert interviews as an instrument of data acquisition
For the respective Research Question, it made sense from the very beginning to conduct
a qualitative study. The aspects to be examined relate mainly to sociological reviews
(Cohen & Felson, 1979) and technological developments. Therefore, the chosen method
is an expert interview, since it is a matter of exploring opinions and new concepts rather
than figures or value ranges. The Research Question, 'what are potential opportunities and
challenges for German police institutions and the society to leverage data-analytical
forecasting technologies in order to prevent crime?’, mainly addresses the qualitative
level. Because quantitative aspects, such as the number of PP software products used in
Germany, are not suitable for answering the Research Question. Therefore, inductive
findings can lead to exploratory insights as well as to plausibility studies of theoretical
cause-effect relationships (Skopp, 2002, pp. 1-3).
In order to reveal possibilities but also the challenges of PP for police officers and the
society, it is necessary to define suitable experts who can refer to these aspects and
represent an opinion. These qualitative statements and information can only be answered
by a small group of people. Meanwhile, an online survey with large samples and statistical
analyses with large random tests makes no sense for the frame of the undelaying thesis.
Rather, it is about the collection of in-depth correlations as well as combination between
location-related and personal-related data and how they are actually used in practice. In
order to refer to the already explained theoretical part of the work, it is important that
selected experts have exclusive insider knowledge and are usually members of a certain
organisation, institution or opinion-forming body. Therefore, it was important to note that
experts can identify processes, terminologies, projects or decision-making structures in
the field of PP within their institution or immediate environment. The respective expert
group is explained more detailed within the next section.
In literature, expert interviews either focus on actors who work as functional elites and
set up implicit and explicit rules or represent decision-making processes, which were
experienced in reality (Liebold & Trinczek, 2009, p. 35). Since there is a content limit for
this Master's thesis, the single guideline-based expert interview should not exceed one
hour if possible. Here, a thematically structured handbook is drawn up in advance. The
aim is to motivate the actors to present their work by means of narrative-generating
questions. To ensure both a focus on content and a self-running narrative. Above all,
sufficient space is guaranteed for free narrative passages, which can represent police
30
related use cases. Therefore, the social research can be described as an 'open procedure'
and thus corresponds to the basic assumptions of the interpretative paradigm (Liebold &
Trinczek, 2009, pp. 36-38). In detail, the video surveillance cameras at the train station
in Berlin may function as social reality and experts can elaborate those statements with
evidences.
According to Bogner et al. (2002, p. 2) conducting expert interviews can serve as a time-
saving data gathering processes, particularly if the focus group are treated as a
crystallization point, where they are interviewed as surrogates for a wider circle of
players. However, this was not the situation for this use case. Some experts consulted in
advance that their statements were not valid for the entire target group but should be
assessed individually. More detailed developments and results will follow in the next
sections. The decisive advantage of a qualitative study especially for this Master Thesis
topic is to leverage a theory- and hypothesis-generating character (Lettau, 1996). In terms
of methodology, the study is based on technography, which deals with the hypothetic
interaction between man and technology. Accordingly, the project is decidedly linked to
the sociology of technology, but also draws its theoretical impulses from the law
perspective as well as from algorithms and data studies (Krasmann, 2018). Thus, it allows
subjective and objective argumentation (Mey et al., 2011) in order to engender possible
synergies in knowledge management. The interview is structured as a semi-standardised,
problem-centred guideline interview (Flick et al., 1995). Jüttemann et al. (1985) argue
that the researcher only proceeds in a theoretically structured manner when it comes to
the evaluation, but is only an unconditional data retriever during data collecting. Experts
should therefore at best serve as bearer of knowledge (Meuser & Nagel, 2009, pp. 17-42).
3.2 Qualitative implementation and setting
In this case, the single interview is useful since the access to the social environment and
the working conditions of police officers in the upper grade is very difficult and much is
kept under lock and key. In detail, both closed and open questions are asked to experts
within a single interview, who in turn can reproduce their experiences and assumptions
within a physical interview moderated by the thesis author. Thereby, the author hopes to
shorten complex and time-consuming observation processes (Bogner et al., 2002). To
avoid suggestively influencing the expert's opinions with explicit hypotheses, the
construct interview rather poses questions in and around the problem environment of PP.
31
The result at the end of the interview would be implicit diagnostic hypotheses, subjective
constructs and action strategies. Possible diagnostic hypotheses can be; ‘biased data shape
these systems’ or ‘Predictive Policing tools used nowadays are more successful then
methods used ten years ago’. All interview partners do not have to speak English. The
interview will be translated in the follow-up. Between 12 and 15 interviews are necessary
in order to ensure that each persona group includes at least 6 individual interviews. For
this purpose, a comprehensive guideline was drawn up in advance. In detail, the guide is
structured as follows: First, the Research Question is split into its literal parts in order to
be able to define hypotheses. A total of eight hypotheses were divided into three distinct
areas. Firstly, the advantages and challenges of PP in Germany. Secondly, hypotheses
were evaluated around the ‘Theoretical background’. This is followed by a detailed
inspection of the necessary experts, who should at best take reference to all hypotheses.
The data collection is then available in form of texts. Texts are defined as a structured,
non-random arrangement of symbols, which can also include images (Helfferich, 2019,
p. 669).
Depending on the Research Question, both groups of experts, 'police institutions' and
'society', were then defined with characteristics in order to create an explicit circle of
experts. The fourth organisational point then contains a script with possible interview
questions and sub-questions if the expert cannot explicitly answer the first question. This
allows the interviewer who in this case is the author of the Master Thesis to go deeper
and keep the conversation going. In this case, the interviewer also assumes an expert role
and can thus create an authentic, interactive communication that also creates a pleasant
interview situation for the surveyed expert. Eight main questions consist of two
introductory questions on personal experiences and understanding in order to bring both
communicators up to the same level. Then there are five core questions which become
more complex and critical the closer one gets to the end of the interview questionnaire.
The aim here is also to ‘discover the unknown’. The last question contains a prognosis
for the future as well as a desired outlook. All questions were derived from the literature
and the pre-defined hypotheses. The interview questionnaire for both expert fields can be
found in the appendix (p. 71).
The timeframe of all expert interviews was in total seven weeks with two weeks of buffer
for illness or absence. For this Master Thesis, 15 interviews were conducted in total, not
only in person but also via telephone. It became apparent that telephone interviews in
particular were more time-saving in terms of travel time, but challenging regarding
32
agreements and technical implementation. Before expert interviews were executed, a Pre-
test was made. To substantiate statements, a qualitative study is carried out. The
respective results may highlight possible challenges and criticisms of those practices. On
the one hand, the structure of the survey was evaluated, the formulation and sequence of
the questions were checked, the timeframe was fixed, the extent and type of answer
failures were noted, the technical framework such as the audio equipment was reviewed
as well as the atmosphere for the participants and the choice of location. In addition, the
interview dynamics were subsequently assessed in order to increase validity of the data.
It became obvious that a comparative interpretation of the orderliness of the topics is
necessary for both the census situation and the evaluation of the data material. During the
Pre-test it was remarkable that there were too many sub-questions. As a result, the first
interview lasted over an hour. Therefore, some questions were deleted or merged.
In addition, the choice of the selected cafe was not particularly appropriate, as there were
too many disruptive factors. Subsequently, the interview locations were usually
determined by the interviewer at home or in the office of the expert. For the audio
recordings open source products were chosen at the beginning, but there were usually
limitations especially for test versions. Therefore, the smartphone with the app ACR was
used as a recording device for telephone conversations and the voice memo function for
personal interviews. There were hardly any changes to the question structure or content
since during the Pre-test nearly every question could be answered by the expert. Before
the interviews started, a privacy statement was signed by each expert, with the agreement
that the meeting process will be recorded and afterwards interpreted. If experts were too
far away for a personal meeting, an appointment was made for a telephone interview, here
the ethics assessment was then sent by e-mail and signed with PDF reader.
On average, the interviews lasted between 20 and 45 minutes, depending on how detailed
the experts were able to answer the questions. In general, there were no negative
experiences or interruptions. Also, during the interviews there were no surprising
challenges except that some experts did not want to express themselves regarding certain
questions. Because on the one hand they do not have any profound knowledge in this area
and on the other hand they were not allowed to answer them for professional reasonings.
However, these unanswered questions had no influence on the findings or results
regarding the Research Question.
33
3.3 Participants and Recruitment
According to Flick (1995), a qualitative approach enables the generation of theory as a
research goal by dealing with an expert field. This expert field do not have to be
homogeneous. Although they should have numerous similarities, finding valuable
interview participants is therefore more likely. As already described, the first task was to
draw up the guidelines for the interview and to generate different hypotheses based on
the literature. In order to define which personas are relevant for this Master's thesis and
who of them in the best case is able to add new categories to the already defined
hypotheses. Since the Research Question intersects two levels - the opportunities and
challenges for police officers as well as for society - it makes sense to analyse two expert
fields. For this purpose, members of German police institutions on the one hand and on
the other hand the target group representing society were suitable. The professional status
is rather of secondary interest, since experience-based information is more important
(Flick, 2016).
In detail, the target group named as police institutions should provide the following
expertise and characteristics: Police institutions comprise civil servants or employees who
are either members of the Federal Police, the Federal Criminal Police Office, the former
Federal Border Police, the 16 state police forces or 18 independent police forces in
Germany (Groß, 2008). Since mainly civil servants at the LKA or BKA use Predictive
Policing strategies (Knobloch, 2018), civil servants in the upper service were given
preferential treatment. Due to a lack of access to the respective experts, as some were not
allowed to give detailed comments about this topic, the expert field was extended to
students or graduates enrolled in police related study programs such as police and security
management. These include graduates and students of criminology (Egbert, 2018),
violent research, police science, criminal justice, governance and police science since the
topic of PP or similar solutions are discussed in these courses. In addition, experts from
this target group must be able to comment on the topic of preventive and predictive police
work and related terms. Terms such as Big Data or Predictive Policing can be assigned
and in the best case, they have already worked with such or similar systems or attended
further training courses in this area to report about status quo.
Since the Master Thesis is limited to the application of Predictive Policing strategies in
Germany, experts should also live in Germany or have worked there for a longer period
of time. Age and gender are not relevant as the focus is on the profound knowledge of
34
each individual expert. A similar definition exists for the 'society' test group. People in
this expert group can form an individual opinion on the subject of preventive and
predictive police work, since they have already dealt with factual contexts before the
interview. Keywords such as Predictive Policing, forecasting tools, Crime Mapping,
clustering, crime analysis, risk factors, risk terrain analysis, predicted hot spots, predictive
analytics or Geographic Information Systems are not completely alien. Due to legal
simplifications, this expert group was limited to the age range above 18 years because
there is a large pool of potential experts in comparison to the first target group.
Additionally, relevant personas respectively to the second expert group are born either in
the age of Digital Natives or Digital Immigrants. Since both generations have to deal in
their immediate private and working environment with digital media related topics. Even
though Digital Natives grew up with computers and the Internet, the Digital Immigrants
had to learn it and extended over times their skill set. According to Bennett et al. (2008,
pp. 1-2) people born roughly between 1980 and 1994 has been characterised as the Net
Generation or Digital Natives because of their familiarity and reliance on information and
communications technology. Meanwhile Digital Immigrants are those who were not born
in the digital world but have, at some later point in their lives, become triggered by this
world and adopted many aspects of the new technologies. Thereby, Digital Immigrants
were able to adapt their environment accordingly (Prensky, 2001, p. 2). These limitations
of the second expert group are supportive, as both generations can mostly demonstrate a
certain understanding of the functioning of the described technologies.
The potential experts of both expert fields were spotted through Social-Media, internet
websites and gatekeepers or the world of mouth. As soon as a contact person was
acquired, it was a matter of recruiting this person. Contact was made by telephone or
email, which often required two to three telephone calls to discuss the situation in detail.
In doing so, the aim and content of the research were presented and transparency about
facts and theoretical basics were given right from the start. In addition, on the one hand
phone calls were made with the press department of municipal police agencies and on the
other hand with the personnel department of police institutions in order to narrow down
the search for experts. There was no financial reward as this was a student project as part
of a final Thesis. Results will of course be shared with the experts after the end of the
project. For the second target group ‘society’, reference could be made to friends and
fellow students who themselves deal with similar topics for their Master's Thesis.
35
In practice, a total of 15 experts were recruited, including seven experts in the category
‘Police Institutions’ and eight personas who are accountable for ‘society’. In detail,
officials from the LKA, BKA, the dog squadron, chief police superintendents of domestic
violence, bachelor graduates of Police Management, lawyers, technical assistants at
Fujitsu or master students of Management, Communication and IT were recruited. The
recruited experts for police institutions (PI) can make a relevant contribution to the work
with their following expertise. On the one hand, four of the seven PI experts have worked
with tools such as Precobs or other similar forecasting technologies. Another expert is
responsible for purchasing such systems and can therefore comment on the external
cooperation as well as the necessary must-haves of the software. At the same time,
another expert has learned the expertise and skills during studies and has already written
similar papers. Thus, knows the current state of research in Germany and can provide
added value by summarizing methods, which are actively in use. This is especially useful
as different police institutions also work with other tools and the holistic view of the topic
would have been missing. Another expert of the PI group persuaded with the following
characteristics: profound knowledge of the subject, having obtained a doctorate in this
field, already leading both practical and scientific projects, studying criminology,
interviewing experts about the subject and understanding data processing technologies.
This paragraph describes in more detail the characteristics and expertise of the selected
SC specialists and why they were of particular interest. On the one hand a profound
knowledge about the described buzzwords like Prediction-led forecasting technics, Big
Data or Data Mining was necessary to understand the context of the interview questions.
Therefore, fellow students were suitable for this purpose because they have already
discussed these contents in class or just wrote their Master Thesis about related topics.
Researchers understand the importance of qualitative research in order to achieve the
desired results within a given timeframe. For example, an expert who writes the Thesis
about 'Smart Home technologies' has been able to identify synergies in order to map
existing knowledge structures in respect to involved risks (Cox et al., 2003). Another
expert from this target group has a legal background and was thus able to shed light on a
new perspective. In this expert group, it was particularly important to achieve a high
degree of variation in the background in order to realise possible new trigger points. In
general, necessary character trades were similar to those of PI: technical know-how
regarding data processing, the ability to come up with an opinion as well as a problem
36
solution regarding Predictive Policing, analytical and fast logical thinking skills as well
as the ability to articulate oneself in a proper way. However, these experts have not yet
worked actively with the underlying technologies but this was also not a necessary
requirement.
Since some experts wanted to remain anonymous, it is necessary to anonymise all experts
within a pseudonymisation table in order to remain uniform (Bayardo & Agrawal, 2005).
The definition for pseudonymisation according to the Federal Data Protection Act
(Neubauer et al., 2010) is as follows: 'Pseudonymisation is the substitution of the name
and other identification features by a mark for the purpose of excluding or substantially
complicating the determination of the person affected.' In other words, this means that the
experts and their names are made unrecognisable with the help of an anonymisation table
and a suitable numerical code. In contrast to anonymisation, pseudonymisation keeps
references to different data sets that have been pseudonymised in the same way. It is
therefore decisive that a combination of person and data is still possible. A
pseudonymisation of the content is not necessary, since no explicit places of residence,
date of birth or other identification characteristics were discovered during the interviews.
In this context, it should be mentioned once again that the opinions of the experts,
especially those of the first expert group, do not correspond to the views of an entire
police authority, but rather refer to individual aspects.
3.4 Hypothesis and evaluation methodology
The aim of the study is not testing already existing theories, but to intensify patterns for
specific subject areas. In this respect, induction-logical steps play a greater role in gaining
knowledge. In the final procedure of induction, general hypotheses are concluded on the
basis of concrete expert experience (Lettau, 1996). All expert interviews were protocolled
and recorded. Subsequently, these interview protocols were transcribed and the answers
assigned to the individual hypotheses as well as subcategories. It should be mentioned
that the area of application to be analysed was Germany and therefore the expert group
mostly preferred the German language as communication language during the dialogue.
This was agreed in advance by telephone or email. The main argumentation of the experts
was that most of them feel more comfortable discussing issues of the highly complex
topic if they can use their mother tongue. It should therefore be mentioned that the results
and evaluations of the interviews were translated by the author from German to English.
37
The hypotheses form the basis for the interview questions and revolve around the field of
the research question. The following paragraph summarizes and enunciates the
hypotheses, which evolved after theory inspection.
Hypothesis
Category
1. Predictive Policing is a well-known
term among police officers and the
society.
Expert knowledge
2. When it comes to the definition of
Predictive Policing, there is
inequality in the explanation of terms
and their application in practice.
Definition
3. Predictive Policing Software is
mainly developed externally and
implemented internally
Development and Implementation
4. In Germany, only location-based data
are analyzed for Predictive Policing
approaches.
Data usage
5. Larger data sets mean more detailed
and better results.
Data quality and quantity
6. Predictive Policing will result in
faster and more efficient police work
Opportunities for police institutions and the
society
7. Crime follows a certain pattern and
might influence the human origin
Human origin
8. Civil servants and citizens trust the
systems as long as they are not
ground for decision management.
Trust level
9. Predictive policing reinforces
discrimination
Challenges for police institutions and the
society (ethical and technical challenges)
10. Trends and investments show that
German authorities will also
increasingly work with the
underlying technologies.
Future outlook
Table 2: Hypothesis and assigned subcategories
The first hypothesis confirms how current the subject is and seeks to explore the current
state of knowledge in society as well as of police institutions. Also, Gerstner (2018)
describes in his introduction that several pilot projects have already taken place in
Germany. Since some podcasts, short films or documentaries (Wormer, 2018) about the
phenomenon are already online, it can be assumed that a certain part of society at least
those interested in digital trends and developments know the term. According to the
definitional analysis by Mauro et al. (2015, p.1), it was unambiguous that the term
Predictive Policing is blurred in demarcation as well as in application. On the one hand,
this is based on the fact that there are different development approaches and goals to be
38
fulfilled. Furthermore, the market for police forecasting software has grown strongly in
the meantime, and there are even various freeware programs. However, since the IT group
IBM is the market leader for predictive-based police work (Monroy, 2015) and most IT
enterprises of this size pay attractive remuneration and thus have better access to IT
specialists (Bhasin et al., 2002), it is assumed that such programs are increasingly
designed externally and then implemented individually by the police authorities. The
fourth hypothesis is based on Knobloch's paper (2017) in which it is stated that personal
data in connection with Predictive Policing is not yet analyzed.
In the main part of the interview, the hypotheses and questions were differentiated into
advantages and challenges. Previous literature analyses have shown that police work can
be made more efficient through the Predpol software or other applications (Perry et al.,
2013). This means that preventive measures can be developed more expeditiously
(Hardyns & Rummens, 2018) and complex data sets can be analyzed more quickly. The
assumption is that this will not only benefit police institutions but also society. However,
the underlying algorithms are based on theories and proven patterns. Therefore, the
seventh hypothesis is that crime must follow a pattern for an optimal predictive policing
strategy (Hill, 2002). According to Brantingham, the inventor of the Predictive Policing
software Predpol, usually criminal characteristics can also be described as a 'physical
process'. In order to be capable of using such systems wisely, it is essential that the level
of trust exists on behalf of society and the police. Since the systems have been in use in
Germany for more than 5 years (Egbert & Krasmann, 2019), the eighth hypothesis is
based on the fact that trust has already been generated. Since the LKA and BKA would
not work with a software that has not achieved a certain success so far. Furthermore, there
are strict data protection guidelines in Germany and no personal data has been analysed
up to now. In general, such systems are faithful to the local regulations. Therefore, it is
assumed that society will trust such systems as long as they are not the grounds for
decision-making (Ioimo, 2018). Having already hypothesized the opportunities, the
literature research also revealed some technical and ethical challenges, such as increased
discrimination (Ferguson, 2011; Angwin & Larson, 2016). The last hypothesis address
future developments and trends that may emerge in comparison to the US or other planned
projects. These assumptions are on the one hand based on Perry et al. (2013), who have
identified potentials that could lead to future ideas. Jones (2002) also argues that the
influence of the private industry advocates a radical change in police management. More
and more investments will be made in this area in order to provide a secure future for
39
society. In general, the aim of the expert interviews is to confirm or deny these
hypotheses, but also that in the best-case new hypotheses and subcategories can be
formed.
The Meuser and Nagel category scheme was chosen as the evaluation methodology. The
process of categorization implies on the one hand a subsumption of parts under a general
concept of validity and on the other hand a reconstruction of this concept in reality
(Meuser & Nagel, 2009, pp. 461-464). In this reconstructive procedure, correlations of
meaning are linked to typologies and theories. Thereby, addition and pragmatic
coexistence are predominated (Meuser & Nagel, 2009, pp. 463-465). The aim of the
evaluation is to extract the similarities in the statements of the experts. The challenge with
semi-open interviews is that each interview is unique. Therefore, an explicit method of
evaluation should be envisaged. A prerequisite for comparability of the findings is
achieved by pre-defined hypotheses and subcategories. Later, the experts' statements
were assigned to the hypotheses. Hence, the process evaluation of Meuser and Nagel is
rudimentary suitable, since the content of each questionnaire is regarded as central and
individual. Nevertheless, it is also worthwhile to work out what is common and typical
between the two expert groups and to reduce the amount of data material. The Meuser
and Nagel analysis usually consists of five steps, but for this use case an adapted
evaluation procedure is used. Since two expert groups were selected, hypotheses were
already prepared in advance and an intensive literature analysis (Auer-Srnka, 2009) of
more than 100 sources in the theoretical field was examined. According to Meuser and
Nagel, each transcription is followed by a paraphrase (Meuser & Nagel, 2009, pp. 475-
478). However, this is omitted as a separated step because the literature has already been
examined in order to limit the field of hypotheses and to develop a deeper understanding
of the research question. The following flowchart provides an overview of the steps that
were taken to evaluate the interviews.
40
Figure 10: Interview evaluation process
(Own process based on Meuser & Nagel, 2009)
The paraphrase is normally used to reproduce the text in chronological order, but the
experts have followed the guidelines precisely, which means that all expert answers are
already in a chronological and meaningful context. Paraphrasing the transcribed text
happens therefore in combination with the second step. Thereby, the hypothesises are
assigned to the categories, which are already subsumed in Table two. Within the second
step the structured interview questions are assigned to the categories built out of the
hypothesises in order to organize later on expert paraphrases in a proper way. The
following table presents two examples of the second evaluation process step. Interview
questions as well as categories and paraphrases of experts are translated from German to
English.
Interview Questions according to the
Hypothesis (translated from German to
English)
Category
The topic of Predictive Policing cannot be seen in
isolation from the debate on discrimination. What could
be potential risk factors and how could they be
counteracted?
Challenges for police institutions and the society
An essential assumption of the 'Routine-Activity
Theory' or the 'Crime Pattern Theory' is that crime
follows a certain pattern. According to Brantingham, the
inventor of the Predictive Policing software Predpol,
criminal properties can be described as a 'physical
process'. Can you confirm such a statement on basis of
your professional experience?
Human origin
Table 3: Examples of the second evaluation process step
By reviewing the third step of figure nine the material is condensed by the experts’ ID
number. Thereby it is key to filter and organize only core statements and relevant passages
1. Hypothesises
and its
categories
•Key questions of
the interview
build on the
literature and the
10 main
hypothesises
•Categories are
formed out of
hypothesises
2. Categories
of the
Interview
questions
•Translating
interview questions
•Organizing
questions to
headings and
categories
3. Organizing
statements to
headings and
categories
•Adding the experts' IDs
to the table of process
step 2
•Filtering paraphrases
•Only assigning core
statememts to headings
and categories
4. Merging the
results
•Final evaluation
consists of
summarising
results of step 3
41
of the quotes from the already transcribed interview to headings and categories. These
headings give an overview of the core statements of the experts. For instance, the
crystallized statements are then classified and sorted according to the hypothesis and
categories; such as the quote of F8273 in table four can be assigned to ‘Expert
Knowledge’ and can be headed to ‘Touchpoints in connection with Big Data and Data
Analytics’.
Interview
Questions
according to the
Hypothesis
(translated from
German to
English)
Category
ID Number
Paraphrase
(translated from
German to English)
Headings
Could you please
describe when and
in which context
you heard about the
topic Predictive
Policing for the
first time?
Expert knowledge
F8273
▪ I became aware of
Predictive Policing
especially in
connection with Big
Data and Data
Analytics around
2015
▪ Touchpoints in
connection with
Big Data and
Data Analytics
(2015)
Table 4: Examples of the third evaluation process step
The results of step three of the two different expert groups are recorded in different tables.
The last step presented in figure nine deals with a thematic comparison of the different
expert statements quoted by referring to the respective headings. Thereby new
subcategories are built such as displayed in table five.
Category
Combined headings from both expert groups
Summary and iterations
Expert
knowledge
Society:
▪ In connection with Big Data and Data Analytics
(2015)
▪ Documentary in the TV
▪ Developed similar software
▪ Newspaper
▪ Documentary
▪ Professional crossroads
▪ First touch points between 2000
and 2018
▪ Studies and private research
▪ Interfaces with other technological
terminologies
Police Institution:
▪ Preventive vs. predictive
▪ Heard about Predictive Policing during university
lecture - home burglary series in 2017
Table 5: Examples of the fourth evaluation process step
42
The similarities but also differences of the target group are differentiated and subsumed.
The last step deals with the theoretical generalization of category formation, in which the
results of the different tables are interpreted by means of answering the research question
and hypothesis.
4 Discussion: opportunities and challenges
The following unit refers to the research question, presents the results and interprets them.
The aim is to explore the opportunities and risks for the society and police institutions of
applying Predictive Policing methods. The respective similarities and differences that
resulted from the interviews with the two different expert groups will be discussed.
Results are divided into two different sections. The answer to the research question
summarizes the experts’ responses and appropriate headings about opportunities and
challenges. While the subsection 'Interpretation of results' relates and interprets comments
on the other hypotheses.
4.1 Interpretation of results
In the final assessment phase, the headings resulting out of the fifteen expert interviews
were summarized in a common table in order to identify possible similarities and new
aspects. The following evaluation and discussion are based on tables, which can be
viewed in the appendix (Interview guideline, 2019, p. 71).
The first question of the expert interview referred to the hypothesis that the term
Predictive Policing is familiar among both society and police institutions (PI). The first
points of contact with the topic were between 2000 and 2018. It emerged that the expert
group of society (SC) does not know the term by the police press or similar trade unions,
but rather through documentary films, private Internet research and literature as well as
study subjects. In addition, it could be elaborated that some experts from both PI and SC
were able to show professional crossings and interfaces to other technological
terminologies. For example, the expert RB728 of the SC group works in the insurance
business and was able to devise interfaces from professional experience: ‘We had a series
of break-ins in a certain neighbourhood in 2016. The district continues to be my preferred
burglary area - this gave me the opportunity to focus more on the topic of risk
management. Accordingly, I sensitize our customers to invest in prevention methods such
Smart Home devices’ (RB728, personal communication, July 20, 2019). Those devices
43
make customers aware of dangerous months, areas or other conspicuous appearances. It
can be interpreted that in order to exploit the full potential of Predictive Policing, a fusion
of different technologies and sub-areas is required. With Smart Home devices, data can
be collected and analysed ad hoc, which can then be transmitted to the police in order to
identify new behaviour patterns and update data sets.
Meanwhile the subject was scarcely mentioned during the training period or studies of
the target group PI, it can be concluded on the one hand that terminologies are very topical
and not used before 2000 or that technologies during that time did not exist. For instance,
H8264 argued that during the training phase ‘such programs had a different name and
referred to purely graphical situational images and analyses’ (H8264, personal
communication, June 28, 2019). Implying that the term Predictive Policing is rather
fashionable and was adopted from the American area. However, this does not indicate
that the underlying methods, techniques and analytical procedures were not applied
before but had a different wording and definition. The hypothesis can thus be confirmed,
but it is important to mention that the selected experts do not represent a sample and are
not representative of society as a whole or every German police institution. All selected
experts knew the term and were able to put it into a suitable context.
The responses concerning definitional background of Predictive Policing varied among
both individual experts and expert groups. However, the following matching
characteristics could be identified: IT supportive police work, crime prevention,
intelligent data analysis tool, individual measures, crime prediction, probability
calculation and disruptive technology. A detailed definition was provided by E7284,
‘Predictive Policing is the application of analytical, digital methods to generate
operational forecasts regarding the origins, times and locations of future offences’
(E7284, personal communication, July 4, 2019). While S5744 assumed that ‘the term is
used to describe an analysis of personal and police data in order to predict future crimes,
for example by finding out patterns in cases of burglary or analysing areas with a high
crime rate’ (S5744, personal communication, June 18, 2019). Regarding the third
hypothesis that 'Predictive Policing Software is mainly developed externally and
implemented internally' the following interpretations and results could be extracted. An
external cooperation already takes place. However, the framework and the guidelines are
created internally by the executive, for example the LKA and BKA, and the legislative
restricts the use of data in order to secure the personal rights of citizens. While experts of
the PI Group see external software development as advantageous, as it gives better access
44
to IT services, talents and their knowhow, the other expert group disagrees. SC experts
prefers internal development and implementation to guarantee data safety. Since external
developments involve risk factors such as data manipulation, data breaches, misuse of
access rights or hacker attacks.
Additionally, police institution might get dependent on the external service provider and
the data exchange takes more time. The expert F8273 is concerned that, ‘the data is very
sensitive and the system should be developed internally to prevent possible manipulation.
In this way it can also be avoided that only the external developers know the function of
the algorithm and misuse it’ (F8273, personal communication, June 15, 2019).
Nevertheless, H0163 argues against it: ‘If the software is developed externally or
internally does not matter for me as the internal control processes and data storage
locations are of particular relevance. I guess that the requirements are specified by the
internal authorities, so I see only little room for manipulation’ (H0163, personal
communication, July 9, 2019). In summary, it can be affirmed that the PI expert group
has a better insight into internal developments and therefore the hypothesis can be
confirmed. Nowadays the police cannot program everything themselves and rely on
external cooperation for this purpose. As soon as software has been developed, it can be
adapted internally to the federal state.
Furthermore, it was necessary to verify if no personal data is analysed in Germany as it
is noted in literature. According to PI personal, time as well as location-based data is
collected and evaluated in practice. But more or less for preventive police work instead
of predictive. For instance, in the field of domestic violence or burglaries, police officers
work with local and personal data to make individual and forecasted measures. Here SC
would not adapt or assimilate their natural human origin or habits since information is
secured by data protections rules. Some experts encountered that ‘personally, I feel more
confident to know that police men are using all means and resources to achieve their
goals’ (H0163, personal communication, July 9, 2019).
But in general, collecting personalized data is an ongoing debate, which has to be judged
depending on the case. According to the statements it can be interpreted om the one hand
that police officers in the upper level already involve personal data especially if
conspicuous person is already defined as hazardous. On the other hand, the society does
not consider the analysis of personal data necessary at the moment. However, this always
depends on the sense of security among the population. If more terrorist attacks will
happen in the future, renewed and stricter methods of analysis will be needed. In addition,
45
every new technical introduction requires an acceptance period, which means that there
may be an initial rebirth, but in the long run citizens will get used to it. In realizing
predictive policing, data must first be collected or existing data records must be processed
in order to recognize a pattern at the end. From this train of thought the following
hypothesis has developed: 'larger data sets mean more detailed and better results.' Nine
out of fifteen experts affirmed the statement and only four refused it. The majority of
experts agreed that for Predictive Policing the quantity is even more relevant then quality.
One the one hand larger data sets are necessary for accurate algorithm calculations and
other hand for statistically relevant results the quantity is essential.
In this context experts concerned the more data sets and information is collected, the
higher the chance that false or irrelevant data is also included. Accordingly, B6253
argued: ‘Data is evaluated according to mathematical principles, the more information,
the more data for evaluation, the more wrong data sets - but also more concrete results
can be presented to society. The question is, if the wrong data sets stand in a reasonable
relation to a concrete hit?’ (B6253, personal communication, June 23, 2019). As a
counterargument it was mentioned that as more data is processed, clustering and other
methods require more time and computing power. Resulting, that even if the quantity of
data is of greater interest, the data sets should still be purposeful, maintained and verified
with a quality software in order to work effectively as well as time saving. Referring to
the hypothesis, that crime follows a certain pattern and might influence the human origin;
experts’ opinions was dichotomous. Ones reasoned, that patterns can be identified on the
one hand for first-offender and for successful criminals, who might repeat their actions
and on the other hand for planned criminality such as car theft, drug or gun abuse and
gang crime.
Thereby, Routine-Activity Theory is a ground methodology for Predictive Policing.
Admittedly, D2640 stated that, ‘not every single field of crime may follow its own pattern.
There are also crime fields in which no patterns can be identified, for example killings.
Since certain crime groups are driven by conditions, emotions or affects. I would rather
advocate, specializing in certain individual fields of crime in which there are proven
patterns’ (D2640, personal communication, July 10, 2019). Additionally, B6253
substantiated, that the 'Routine-Activity-Theory' or other approaches does not provide an
explanation of crime patterns since they are too simple in its conception and ignores
contextual or environmental facts. Thus, it is entangled in tautologies (B6253, personal
communication, June 23, 2019). In summary, this means that underlying patterns and
46
theories do not automatically determine and predict the human origin. To some extent, a
basic pattern can be identified in certain crimes, but the behaviour could take a different
course in the smallest context change, such as weather or financial wealth. The next point
relates to the trust level among police officers and society. Six out of fifteen experts
definitely do not trust the systems since technology is not yet mature or error-free, the
algorithm is a black box and trustworthiness depends on aim as well as the already noted
successes factors. Meanwhile officials trust the system as the final decision is based on
human origin. The hypothesis is confirmed by J9352: ‘One hundred percent trust is
always gained through knowledge one has gained him or herself. However, officials also
have a high level of trust regarding their used programs since the last instance in deciding
whether the program has delivered a meaningful result, is always a human official’
(J9352, personal communication, June 21, 2019).
In conclusion, more tests, experiences and reports are necessary for an optimal trust base.
Therefore, the hypothesis that civil servants and citizens trust the systems as long as they
are not ground for decision management, can be confirmed. Last but not least, the
following future scenarios for Predictive Policing were interpreted by the experts. Ones
stated that Heat-maps are used more frequently, police institutions will work with better
data quality and person-related data sets, the frequency of controls and camera appearance
at public places will increase and cross functional technologies will be merged and
expanded. Furthermore, changes in this range depends on the peace-index of a country,
which means. If the peace index might rise the next years it indicates that this country has
become more dangerous and Predictive Policing might be implemented in all federal
states. On the other hand, if Germany continues to be one of the most peaceful countries
in the world, investments in this area may not be justifiable, and data protection rules will
be handled even more strictly.
4.2 Answer of the Research Question
In this section, explicit reference is made to the research question: ‘what are potential
opportunities and challenges for police institutions and the society to leverage data-
analytical forecasting technologies in order to prevent crime?’ At the beginning, the
possibilities and challenges faced by the society were dealt separately from those faced
by the police. However, during the evaluation it was established that initial thoughts are
not meaningful. As there are tight links between the two target groups and possibilities
47
and challenges of the two target groups correlate. In detail, opportunities which develops
during PI expert interviews, represent also advantages for the society. Therefore, only
opportunities and challenges in general have been separated in this section.
In summary, after the literature research and the qualitative survey the following results
could be recorded. Even if police institutions have to face some technical challenges such
as data processing or correlation versus context as well as ethical challenges such as
discrimination rates, the early use of Predictive Policing is recommended. To be able to
gather empirical values and Know-how in order to deliver a less error prone and
predictive police work and investigations in the long run. Flawless and efficient police
investigations not only provide an advantage for executives but also for society, as
resources such as taxpayers' money can be saved over the long term and the crime rate
drops.
4.2.1 Opportunities of applying Predictive Policing
From the discussions with the various expert groups, an answer to the following Research
Question could be evolved: ‘What are potential opportunities and challenges for police
institutions and the society to leverage data-analytical forecasting technologies in order
to prevent crime?’ In general, it can be stated that during evaluation phase especially
regarding challenges experts’ opinions varied. Thereby, it was possible to identify new
aspects beside of ethical and technical challenges. While hardly any new aspects in sense
of possibilities or advantages of PP were elaborated. Already existing assumptions and
hypotheses regarding this point of view were confirmed. Both members of the target
group PI and SC repeated several times in conversations that the underlying technologies
mainly produce advantages when used as facilitation tools. Senior officers confirmed
during interviews that programs such as RADA-ITE or Predpol are currently used only
as assistance, but decisions are not made by the program itself. Tools provide
recommendations for further actions, which can then be accepted, adapted or even
rejected by the responsible commissioners. It should be mentioned that the results of
these automatic evaluations probably already influence the human advice origin because
human officials cannot free themselves from experiences. As soon as statistical successes
of these programs are proven, the basis of trust is intensified. Instead of a supportive
instrument, PP might become a basic requirement for any predictive as well as present
determination. Used as a supportive instrument, PP offers the following possibilities for
48
the German market. With preventive analyses, personnel and other resources such as
workplaces, equipment or budget, which, according to the experts, are only available to
a limited extent can be used more purposefully and thus more effectively. According to
D8236 Predictive Policing leverages ‘on the one hand, detection of stereotypical crimes,
which are then prevented in advance’. Thereby similar forms of crime activities or
measures can be implemented more effectively and this could have a positive effect on
budgeting’ (D8236, personal communication, June 18, 2019).
This approach has also been confirmed by the literature, since the information is accurate
in terms of time and location. In addition, it represents important decision-making aids
regarding planning of patrol trips and operations, as well as in investigative work.
Thereby, tax money can be saved on a long-term. In general, PP has definitely established
itself as necessary device for crime prevention. This is because an increased police
presence, can reduce the ‘opportunity structures’ for criminal offences and prevent a
criminal offence from arising (Jordan et al., 2017, p. 59). This also supports the hypothesis
that the use of PP reduces the criminality rate. Additionally, those ‘modernization’
approaches intensify the image of the static and low paced police office environment.
Residents get the feeling that they are living in a state that cares for its citizens and is
committed to an increased sense of security. Because modern technology guarantees a
reduction in crime (H0163, personal communication, June 7, 2019). Those technological
innovations enable better police performance through visualizable quantifications
(B6253, personal communication, June 23, 2019). This advantageous data visualization
allows police officers who are responsible for executive actions to understand the current
situation more easily. At a glance on the smartphone, it becomes clear which area or
which group of people needs to be observed. Resulting, that respective algorithms can be
used to monitor criminal gangs or to identify and protect potential victims in advance
(F8273, personal communication, June 15, 2019). Therefore, the sixth hypothesis
‘Predictive Policing will result in faster and more efficient police work’ can be vindicated.
In summary, it is acknowledged that using Predpol or Precobs for analysing large amounts
of data facilitates quicker identification of criminological patterns, accelerates data
exploration, and counteracts criminal patterns at an early stage. Expert RB728 mentioned
in this context, that opportunities can be realised leveraging Big Data in order to ensure
that people do not become criminals: ‘People can change through help such as special
care or schooling, if this is recorded and dealt with a preventively character, the criminal
49
pattern can be counteracted’ beforehand (RB728, personal communication, July 20,
2019). In addition, knowledge management within the police is intensified by external
experts and innovative communication channels. These tools connect and disseminate
information between the federal states, both fast-track and intelligent. Above all, several
experts have confirmed that the use of Predpol or similar tools supports crime prevention
in the following areas: Terrorist attacks for example on popular holidays such as New
Year's Eve., billing fraud, gang crime or burglaries. This insight was substantiated not
only by literature (Gerstner, 2018) but also by V0627; ‘forecasts are particularly helpful
in cases of control offences and planned crime such as car theft, drug or gun abuse and
gang crime but everything that deals with emotions is not predictable’ (V0627, personal
communication, July 7, 2019). However, this does not exclude the possibility that in the
future software will also be used for drug abuse or other criminal offences - test attempts
are already discussed in practice (Egbert, 2019).
According to the PI expert group, such tools are also used as decision-making assistance
and lead to smart priority selection. Commissioner ratified from duty, that, ‘an advantage
could be, that such systems filter priorities and at a certain stage analyses prevention
measures individually to the use case’ (H8264, personal communication, June 28, 2019).
This can have a positive effect on the rate of discrimination, as it is no longer just based
on experience or feelings of executives. A new point of view also came up by B8350,
since such systems can above all demonstrate a high level of plausibility in everyday life.
Meaning, ‘that results of risk areas are immediately helpful and applicable in the present,
thus guaranteeing a high level of everyday plausibility’ (B6253, personal communication,
June 23, 2019). Lastly, according to expert B6253 and E7284, the fear of data misuse or
racial profiling is over-founded, at least at the present juncture. Since the data protection
regulations in Germany treat the privacy of society very strictly and an abuse of sensitive
data is rather unlikely.
4.2.2 Challenges of applying Predictive Policing
Discussions during the interviews focused not only on benefits and opportunities, but also
on risk factors and challenges of technical and ethical origin for both society and the
police. In particular, one argument was interpreted differently by several experts.
According to V0627, PP tools only provide added value during investigation proceedings
since ‘programs determine on the basis of probabilities, if it is necessary to observe a
50
person by trawling ones` privacy, which in the long run would likely erode the
presumption of innocence. Therefore, I would use it as an aid kit, but not yet classify it
as predictive measurement’ (V0627, personal communication, July 7, 2019). A valuable,
technical innovation should also be able to stand alone. In detail, this means that, on the
one hand, trust in these systems as a basis for decision-making has not been optimised
due to the rare success evaluations and studies. On the other hand, human or technical
failure in the range of data mistreatment has serious consequences.
Confirmed by B6253 and S5744 there might be too many interferences of numerous
influencing factors, which makes it impossible to review effectiveness. Since social
frameworks are too elapsed to be solved by technical means. Criminals might be arrested,
but probably more innocent people are targeted by the authorities, even though they are
not guilty (S5744, personal communication, June 18, 2019). Furthermore, systems
described in practice show technical bounds, which are also the reason for the curtailed
usage. So far, different correlation theories have mainly been used in the evaluation,
which implies that the outlier and the context behind the crimes are neglected. According
to K7239, ‘the most challenging aspect about Predictive Policing is, if problems are
systematized’ and the software draw certain profiling conclusions based on false
assumptions. This in turn can lead to circuits, ‘because increased attention is paid to these
certain people or areas and increased controls are carried out’. Therefore, more controls
will lead to more hits (K7239, personal communication, July 14, 2019). Thus, results are
falsified and do not reflect reality. Also, Egbert (2019) confirmed in literature as well as
during the expert interview that, ‘it must be clearly stated that the police database is not
neutral and if the software generates forecasts based on this data, these forecasts are of
course not neutral either - new approaches and algorithms are needed here’ (E7284,
personal communication, July 4, 2019). Additionally, B6253 elaborated that this type of
crime analysis is based on recognized data, which represents the brightfield. However,
there is a lack of including the darkfield. Meaning, certain crimes or figures are not even
recorded, which completely distorts the success of statistics (B6253, personal
communication, June 23, 2019). These concerns have also been confirmed by literature
(Perry et al., 2013).
Making reference to the ninth hypothesis, which states that PP reinforces discrimination
- this can only be confirmed with restrictions. Since such a statement only applies, if
personal data is collected actively and used in this context. Although the PI expert group
mentioned that personal data is used for certain investigations, it is not used in connection
51
with forecast-based policing. In comparison, F8273 predicated, those systems can further
strengthen prejudices against minorities and discriminate certain social groups.
Moreover, it should be mentioned here, that both expert groups vary widely in the
definition of PP, which has been already elaborated in 4.1. The main reason why this
aspect finds relevance in this context, derives from the fact that different interpretations
regarding the use of such tools can lead to communication difficulties. According to
V0627, in Germany, police units use different modules and theories in their respective
federal states, which makes it difficult to compare empirical values and guarantee a
common optimal use across the states. Thus, also knowledge bottlenecks can develop
with various applications. Such risk factors could be avoided by establishing a uniform
assessment and analysis process, guidelines or codes of conducts throughout the federal
states.
Possible solution for mentioned challenge can be individual neutralisation software,
which also takes outliers into account (E7284, personal communication, July 4, 2019).
However, for such an improvement, the German data protection ordinance curbs results
and developments. For this reason, it is particularly challenging for German residents and
police institutions to find a balance between security and personal freedom (J9352,
personal communication, June 21, 2019). Furthermore, after evaluating experts’ answers
referring to the third hypothesis about ‘external developments’ within this business, new
challenging risks emerged. It can be interpreted, the more complicated technical
applications or demands raised by society and the legal systems are, the more
cooperation’s with external companies has to be initiated. As a result, police might lose
on the one hand their sense of human advising or gut feeling as well as their own talents
in the long run. Thereby, institutions get too dependent on external developments (H8264,
personal communication, June 28, 2019). Thus, results depend not only on data
management, but also on the state of the art as well as donors investing in further
evolution (RB728, personal communication, July 20, 2019).
Since the acquisition and licensing of such systems can be very expensive at the
beginning, the training of the employees and the know-how are time-intensive. Referring
to F827: ‘I also think that issues like third party intervention, denial of service attacks and
data security could be a debate especially by relying on external providers. Therefore, it
is also important to train employees who will ultimately analyse the results’ (F8273,
personal communication, June 15, 2019). Hence, the benefits depend on the objectives
52
and definitions of the respective police institution. If successes and savings have to be
realized within the next six months, the implementation of Predictive Policing presents
an expense factor.
Last but not least, decentralised administration of data also represents a risk. As it is
necessary to collect a large number of information in order to justify high-quality results
and patterns. Usually, institutions prefer decentralised data management when it comes
to working with a huge amount of data. Especially challenging is not only collecting but
also evaluation relevant data at the very beginning (H8264, personal communication,
June 28, 2019). As long as no studies have yet been elaborated in order to determine
which data really counts (Perry et al., 2013), data might be collected on the off chance.
These big datasets probably have to be stored decentral. Thereby, processing is very time-
consuming, which is detrimental for efficient police investigations.
Another point for further considerations is, that criminal patterns and human behaviour
can change according to contextual movements and the knowledge they might gain about
Predictive Policing tools (V0627, personal communication, July 7, 2019). It is a challenge
to handle the balance between the degree of transparency. Meaning, that what information
can be publicized and which process steps are only communicated internally. As soon as
it becomes notorious for the society what data is collected for what reason, potential
perpetrators can adapt and manipulate the system. At this juncture J9352 intervened:
‘people want to continue to be able to move anonymously in public. For example, critics
see an encroachment on the fundamental right to general freedom of action. The great
challenge for the future will be to find an optimal balance between public safety and
personal rights’ (J9352, personal communication, June 21, 2019). This would include
disadvantages for society as well as for the police, because perpetrators adapt everyday
habits according to the systems requirements and disguise oneself as 'innocent' people. In
this case, the police have to constantly change their methods and set up even stricter data
regulations and controls.
In addition, during the interviews with SC, at least four experts expressed concerns about
the responsibilities of such systems. In this regard, F8273 commented as follows: ‘it
should also be clarified which factors with which weighting influence the result of the
system and how to deal with matches or mismatches’ (F8273, personal communication,
June 15, 2019). The question is whether the responsible police unit, the state or the
software manufacturer must defend themselves in court in the event of mismatches and
wrong decisions. It is therefore indispensable that codes of conduct should deal with the
53
fears and concerns raised by society. Misuse will cause irreparable damage for both
society and the police and worsen the relationship as well as trust to police institutions.
Finally, experts specified issues of fluctuating talent and human resource management.
Since the underlaying technology in the best case should be more efficient than human
commissar. As a result, on the one hand employees are made redundant and tax money is
saved, but on the other hand professional fields of the policemen are transformed. Mainly
IT specialists are required who can work with applied software. Such developments might
have a challenging impact on the traditional police work (B6253, personal
communication, June 23, 2019).
5 Final remarks
In the last section, final lines of reasoning and limitations arising during empirical
research are summarized. In addition, it is outlined how these limitations can be omitted
in further research objectives, which however have not been extended further as this
would go beyond the scope of the work.
5.1 Conclusion
It is expected that police departments nationwide have to struggle with budget freezes and
deep cuts even though crime rates in certain areas are raising. Because of this, it is
obviously that there is a greater emphasis on attempting to prevent crime before it will
happen by investing valuable resources in new forecasting technologies and citizen-
oriented policing (Feltes, 2013). In order to stabilize the public safety force, detailed
analysis within already registered higher crime risk areas will be needed. Therefore, it is
to be expected that crime hotspot maps will still be the most widely used tool for the
quantification of future crime risk and present key elements in hotspot policing (George,
2015).
However, there are also some limitations for researcher and practitioners. For instance,
Perry et al. (2013) argue, that ‘Predictive Policing methods are not a crystal ball: they
cannot foretell the future. They can only identify people at increased risk of crime’.
Moreover, Perry et al. (2013) expressed some concerns about the collection and use of
data. Since the privacy and civil liberties of each individual can be attacked if necessary.
Thus, it will be an additional task for police services to create guidelines for ‘policing the
54
police’ (Heffernan, 2019). In detail, defining any legal precedent, to conduct regular
audits, developing public relation strategies as well as increasing engagement and
understanding of society will be even more important. In order to dispel the possible
concerns of citizens, the police should accentuate proactive strategies by making work
processes to a certain degree transparent. Thereby, police forces can build a strong
relationship between them and their communities in order to solve crime problems
together (Perry et al., 2013). Another explanatory aspect is the lack of definitional
precision, especially in Germany. As various technologies, which are associated with
Predictive Policing diluted the term.
While in Germany mainly preventive police work with heat maps or crime mapping is
applied in practice, experts ask themselves to what extent this is actually renewal.
Especially the prediction without Artificial Intelligence is thoughtful regarding the right
to be considered as an innovation (Braga & Weisburd, 2019). Several police institutions
affirm the use and pose that they already work with underlying technologies. In this
context, the question arises what is seen as preventive or predictive police work. Also,
Meijer & Wessels (2019) conclude ‘that the current thrust of Predictive Policing is based
on convincing arguments and anecdotal evidence rather than on systematic empirical
research’.
At the moment, usually only two or three attributes are linked and an analysis takes place
on two levels such as time and place or crime and time. While PP, according to literature,
automatically should link several relevant attributes and in best case adds personal data.
By leveraging Artificial Intelligence, attributes can be linked even though literary given
connections are not indicated at the first glance. Therefore, it is also important to adapt
the already existing theories to the present time. Because theories like the Concentric-
Model are hardly applicable nowadays due to demographic changes.
According to the expert interviews as well as literature, Predictive Policing is mainly used
for burglary or theft. Therefore, future challenges may be working prognosis-based in
other criminal groups such as murder or drug abuse. As Predictive Policing has only been
used in practice for a limited number of pattern-based and calculable criminal acts, the
question arises why emotional offences cannot be foreseen. Since context and the
environment in which a crime occurs also change with its actors involved. Thus,
emotional decisions reflect the reaction towards an action, which may also be pattern-
based to a certain extent. Imaging the eating habits, personal Netflix favourites or the
choice of an airplane seat seems to be an unconscious event, but can correlate with future
55
criminal occurrences. As a result, new approaches need to be identified, especially in the
context of discrimination debates. Resulting, that further neutralisation software would
be of assistance. However, due to strict data protection regulations regarding personal
data, a fairly slow technical development can be observed in Germany. Whether this is
advantageous or not depends on individual objectives and is a matter of interpretation.
However, since an upsurge can be noted in certain areas of criminal activities, it can be
expected that test attempts with camera shots like those at the train station in Berlin will
become rule. Above all, to counteract terrorism in the long term. In general, the aim is to
identify correlations between data sets and improving the crime-based patterns after each
event in order to minimize step by step prediction errors and to leverage economies-of-
scale. Therefore, Prediction-based police work is definitely more like a trend, even if the
term Predictive Policing seems more like a buzzword or black box, because it is used in
different forms. In the long run, the use of PP is not only worthwhile in terms of more
effective and efficient police work, but can also have a positive effect on budgeting.
Over the next years, Predictive Policing will certainly not be able to achieve the same
status in Germany as in the USA. There is the fear of data violations and feeling of being
a 'glass man'. In addition, the peace index in Germany is healthy compared to other
countries. Therefore, society is not willing to give up freedom when considering the pros
and cons. But still, Predictive Policing on the one hand will merge with other technologies
and on the other hand will prove itself beyond a supportive tool. However, the human
advice origin will not be replaced by the machine advice origin as the human agent on the
spot is always an executive agent who spontaneously adapts strategies according to the
situation. In order to guarantee a holistic picture, the following infographic summarizes
important facts and figures about Predictive Policing and its deployment in Germany.
56
Figure 11: Conclusive facts and figures
As can be seen in the chart, according to the PKS, the burglary offences in Germany have
dropped nearly by half, occasionally also due to the increased use of proper PP
methodologies.
5.2 Limitations and further Research
In summary, it can be stated that the Research Question does not aim for a final polarized
statement regarding the application of Predictive Policing. In detail, it is not disputed
whether the use of such systems is good or bad for German institutions and society.
Rather, it is about revealing all possibilities and challenges allowing readers to build their
own judgements. A final statement is not possible in the current state of research, since
technologies in this area are still in their infancy and hardly any long-term studies exist.
In addition, the term challenges and opportunities were deliberately chosen instead of
disadvantages and advantages. Since opportunities mean, that only under certain
conditions and circumstances an advantageous result can be achieved, which is also
transmissible to Predictive Policing. Disadvantages would mean that something has
57
unpleasant effects. Again, the term challenges seemed more appropriate. Since the error-
free use of PP seems to be a complex but intriguing task, which can be accomplished by
leveraging different technological and human cooperation. Limitations appeared
primarily in the selection of experts, especially those of the expert group SC as some of
them where recruited from family, friends, employees and fellow students. This does not
falsify the results, as experts also provide necessary expertise and characteristics. But in
general, for scientifically relevant results, seven experts do not correspond to the 'law of
large numbers'. The study thus provides a valuable point of reference for future inquiries.
But for precise analysis of possibilities and challenges, studies should be carried out in
larger areas and with a larger group of specialists. However, this was not possible as it
would go beyond both the time and financial scope of the research. Moreover, it is
important to mention that recent tangible outcomes in the use of Predictive Policing are
mostly facts and statistics produced by external providers. Thus, the results are not highly
representative, as they are supposed to advertise underlaying products such as Precobs or
Predpol.
For a detailed investigation of Predictive Policing systems in Germany, experts from each
German federal state should be interviewed, as different processing methods are used in
each federal state. In detail, relevant experts should have been working with the tools for
many years and need to observe the state of development in an ongoing process. In
addition, future research projects should focus on the definitions and codes of conduct of
the systems. Since many police institutions have so far lacked Predictive Policing
guidelines and user interfaces that are applicable across locations. In particular, it would
make sense to organise nationwide workshops or webinars for relevant stakeholders to
discuss best practices and uses cases.
58
Academic References
Ahishakiye, E., Omulo, E., Taremwa, D. & Niyonzima, I., (2017). Crime Prediction
Using Decision Tree (J48) Classification Algorithm (Volume 06 – Issue 03). Uganda:
International Journal of Computer and Information Technology.
Ahmad Malik, A. (2016). Urbanization and Crime: A Relational Analysis, Vol. 21, Issue
1, Ver. IV (Jan. 2016) PP 6. New Delhi, India: IOSR Journal of Humanities and Social
Science IOSR. Retrieved from
https://pdfs.semanticscholar.org/a090/6d4d7fb65f135b6c28d6dcba9a2bfed52759.pdf
AIP Conference Proceedings (Ed.) (2015). AIP Conference Proceedings: AIP Publishing
LLC.
Anand, P., Cleverley, M., Kerr, W., Marzolini, F., Reade, M. & Russo, S. (2011). Smarter
Cities Series: Understanding the IBM Approach to Public Safety (No. REDP-4738-
00). New York. Retrieved from
https://www.aeiciberseguridad.es/descargas/categoria6/6193257.pdf
Andresen, M. A., & Farrell, G. (2015). The Criminal Act. London: Palgrave Macmillan
UK. https://doi.org/10.1057/9781137391322
Armstrong, J. S. (Ed.). (2004). International series in operations research & management
science. Principles of forecasting: A handbook for researchers and practitioners (4.
printing). Boston, Mass.: Kluwer.
Arquilla, J., & Ronfeldt, D. (1993). Cyberwar is coming! Comparative Strategy, 12(2),
141–165. https://doi.org/10.1080/01495939308402915
Atzori, L., Iera, A., & Morabito, G. (2017). Understanding the Internet of Things:
definition, potentials, and societal role of a fast-evolving paradigm. Ad Hoc Networks,
56, 122–140. https://doi.org/10.1016/j.adhoc.2016.12.004
Auer-Srnka, K. J. (2009). Qualitative und kombinierte Methoden in der
wissenschaftlichen Marketingforschung: Theoretische Betrachtung und
Literaturanalyse [Qualitative and combined methods in scientific marketing research:
Theoretical consideration and literature analysis], 48(1-2), 7–20.
https://doi.org/10.1007/s12642-009-0002-2
Baur, N., & Blasius, J. (Eds.). (2019). Handbuch Methoden der empirischen
Sozialforschung [Handbook Methods of Empirical Social Research]. Wiesbaden:
Springer Fachmedien Wiesbaden. https://doi.org/10.1007/978-3-658-21308-4
Bayardo, R. J., & Agrawal, R. (2005). Data Privacy through Optimal k-Anonymization.
In 21st International Conference on Data Engineering (ICDE'05) (pp. 217–228).
IEEE. https://doi.org/10.1109/ICDE.2005.42
Belina, B. (2016). Predictive Policing. Monatsschrift für Kriminologie und
Strafrechtsreform [Journal of Criminology an Penal Reform], 99(2), 201.
https://doi.org/10.1515/mkr-2016-0202
Bennett, S., Maton, K., & Kervin, L. (2008). The ‘digital natives’ debate: A critical review
of the evidence. British Journal of Educational Technology, 39(5), 775–786.
https://doi.org/10.1111/j.1467-8535.2007.00793.x
59
Bennett Moses, L., & Chan, J. (2018). Algorithmic prediction in policing: assumptions,
evaluation, and accountability. Policing and Society, 28(7), 806–822.
https://doi.org/10.1080/10439463.2016.1253695
Behr, R. (2016). Präventionsstrategie(n) der deutschen Polizei im Wandel der letzten 25
Jahre [Prevention strategy(s) of the German police in the change of the last 25 years].
In Gewalt Prävention Information, 1. Auflage, pp. 99–109, original work published
2016.
Bernasco, W. (2008). Them Again? European Journal of Criminology, 5(4), 411–431.
https://doi.org/10.1177/1477370808095124
Bhasin, B. B., & Low Kim Cheng, P. (2002). The fight for global talent: new directions,
new competitors – a case study on Singapore. Career Development International, 7(2),
109–114. https://doi.org/10.1108/13620430210421632
Bode, F., Stoffel F. & Keim, D. (2017). Variabilität und Validität von Qualitätsmetriken
im Bereich von Predictive Policing [Variability and validity of quality metrics in
predictive policing]. In Konstanzer Online-Publikations-System (KOPS)
(Kriminologischen Forschungsstelle im LKA NRW, pp. 1–14). Retrieved from
https://d-nb.info/1129599035/34
Bogner, A., Littig, B., & Menz, W. (2002). Das Experteninterview [Expert interview].
Wiesbaden: VS Verlag für Sozialwissenschaften. https://doi.org/10.1007/978-3-322-
93270-9
Bogner, A., Littig, B., & Menz, W. (Eds.). (2009). Interviewing Experts. London:
Palgrave Macmillan UK. https://doi.org/10.1057/9780230244276
Bogner, A., Littig, B., & Menz, W. (2009). Introduction: Expert Interviews - An
Introduction to a New Methodological Debate. In A. Bogner, B. Littig, & W. Menz
(Eds.), Interviewing Experts (pp. 1–13). London: Palgrave Macmillan UK.
https://doi.org/10.1057/9780230244276_1
Box, G. E. P., Jenkins, G. M., Reinsel, G. C., & Ljung, G. M. (2016). Time series analysis:
Forecasting and control (Fifth edition). Wiley Series in Probability and Statistics.
Hoboken, New Jersey: John Wiley & Sons Inc. Retrieved from
http://search.ebscohost.com/login.aspx?direct=true&scope=site&db=nlebk&AN=106
1322
Braga, A., & Weisburd, D. (2019). Conclusion: Police Innovation and the Future of
Policing. In D. Weisburd & A. A. Braga (Eds.), Police Innovation (Vol. 8, pp. 544–
563). Cambridge University Press. https://doi.org/10.1017/9781108278423.024
Bundeskriminalamt (2017). Presse Information: Neues Instrument zur Risikobewertung
von potentiellen Gewaltstraftaten [Press release: new instrument for risk evaluations
of potential offences]. Retrieved from
https://www.bka.de/DE/Presse/Listenseite_Pressemitteilungen/2017/Presse2017/170
202_Radar.html
Burgess, E. W. (2008). The Growth of the City: An Introduction to a Research Project.
In J. M. Marzluff, E. Shulenberger, W. Endlicher, M. Alberti, G. Bradley, C. Ryan, C.
60
(Eds.), Urban Ecology (pp. 71–78). Boston, MA: Springer US.
https://doi.org/10.1007/978-0-387-73412-5_5
Hustedt, C. (2016). A Public Value Perspective on Predictive Policing in the US: A
systematic case study analysis of the impact delivered through the Atlanta Police
Department’s predictive policing project.
https://doi.org/10.13140/RG.2.2.14410.62404
Cohn, E. G., & Rotton, J. (2000). Weather, seasonal trends and property crimes in
Minneapolis 1987–1988. A moderator-variable time-series analysis of routine
activities. Journal of Environmental Psychology, 20(3), 257–272.
https://doi.org/10.1006/jevp.1999.0157
Cohen, L. E., & Felson, M. (1979). Social Change and Crime Rate Trends: A Routine
Activity Approach. American Sociological Review, 44(4), 588.
https://doi.org/10.2307/2094589
Coldren, J. R., Huntoon, A., & Medaris, M. (2013). Introducing Smart Policing:
Foundations, Principles, and Practice. Police Quarterly, 16(3), 275–286.
https://doi.org/10.1177/1098611113497042
Cox, P., Niewohner, J., Pidgeon, N., Gerrard, S., Fischhoff, B., & Riley, D. (2003). The
Use of Mental Models in Chemical Risk Protection: Developing a Generic Workplace
Methodology. Risk Analysis, 23(2), 311–324. https://doi.org/10.1111/1539-
6924.00311
Craig D. Uchida (2009). A National Discussion on Predictive Policing: Defining Our
Terms and Mapping Successful Implementation Strategies. In U.S. Department of
Justice, Nov. 18-20, 2009 (NCJ 230404, pp. 4–6). Retrieved from
https://www.ncjrs.gov/pdffiles1/nij/grants/230404.pdf
Eck, J., Chainey, S., G. Cameron, J., Leitner, M., & E. Wilson, R. (2005). Mapping
Crime: Understanding Hot Spots. Research report of the National Institute of Justice.
Retrieved from http://discovery.ucl.ac.uk/11291/1/11291.pdf
Egbert, S. (2018). Predictive Policing und das Polizieren der Zukunft [Predictive
Policing, police work in the future] (Vortrag). Universität Hamburg, Hamburg.
Retrieved from http://www.surveillance-studies.org/blog/wp-
content/uploads/20180109_Folien_SimonEgbert_PredictivePolicing.pdf
Egbert, S. (2019). Predictive Policing and the Platformization of Police Work.
Surveillance & Society, 17(1/2), 83–88. https://doi.org/10.24908/ss.v17i1/2.12920
Egbert, S., & Krasmann, S. (2019). Predictive policing: not yet, but soon preemptive?
Policing and Society, 48(1), 1–15. https://doi.org/10.1080/10439463.2019.1611821
El-Sappagh, S., Hendawi, A., & El Bastawissy, A. H. (2011). A proposed model for data
warehouse ETL processes. Journal of King Saud University - Computer and
Information Sciences, 23(2), 91–104. https://doi.org/10.1016/j.jksuci.2011.05.005
Farwell, J. P., & Rohozinski, R. (2011). Stuxnet and the Future of Cyber War. Survival,
53(1), 23–40. https://doi.org/10.1080/00396338.2011.555586
61
Feltes, T. (2013). Bürgernahe Polizeiarbeit in Deutschland [Citzen-focused policing in
Germany]. In OSZE-Jahrbuch 2013 (pp. 241–252). Nomos.
https://doi.org/10.5771/9783845259680-241
Ferguson, A. G. (2011). Predictive Policing and the Future of Reasonable Suspicion.
SSRN Electronic Journal. Advance online publication.
https://doi.org/10.2139/ssrn.1965226
Flick, U., Kardorff, E. v., & Keupp, H. (1995). Handbuch qualitative Sozialforschung:
Grundlagen, Konzepte, Methoden und Anwendungen (2. Auflage). Grundlagen
Psychologie [Handbook of qualitative social research: basics, concepts, methods and
applications (2nd edition). Fundamentales of Psychology].
Weinheim: Beltz. Retrieved from http://www.content-
select.com/index.php?id=bib_view&ean=9783621278355
Flick, U. (1995): Stations of the qualitative research process. In: Flick, U. &
Kardorff, E. v.
Garfinkel, S. L. (2010). Digital forensics research: The next 10 years. Digital
Investigation, 7, S64-S73. https://doi.org/10.1016/j.diin.2010.05.009
George, O. M. (2015), San Jose, CA. Subject to any disclaimer, the term of this patent is
extended or adjusted under 35 U.S.C. 154(b) by 224 days. US 8,949,164 B1.
Gerstner, D. (2017). Predictive Policing als Instrument zur Prävention von
Wohnungseinbruchdiebstahl [Predictive Policing as an instrument for crime
prevention] (1. Auflage 2017). Research brief: Vol. 50. Freiburg im Breisgau: Max-
Planck-Institut für ausländisches und internationales Strafrecht.
Gerstner, D. (2018). Predictive Policing in the Context of Residential Burglary: An
Empirical Illustration on the Basis of a Pilot Project in Baden-Württemberg, Germany.
European Journal for Security Research, 3(2), 115–138.
https://doi.org/10.1007/s41125-018-0033-0
Geva, S., Trotman, A., Bruza, P., Clarke, C. L.A., & Järvelin, K. (Eds.) 2014. Proceedings
of the 37th international ACM SIGIR conference on Research & development in
information retrieval - SIGIR '14. New York, New York, USA: ACM Press.
Gongla, P., & Rizzuto, C. R. (2001). Evolving communities of practice: IBM Global
Services experience. IBM Systems Journal, 40(4), 842–862.
https://doi.org/10.1147/sj.404.0842
Gorr, W., & Harries, R. (2003). Introduction to crime forecasting. International Journal
of Forecasting, 19(4), 551–555. https://doi.org/10.1016/S0169-2070(03)00089-X
Gobbens, R. J. J., Luijkx, K. G., Wijnen-Sponselee, M. T., & Schols, J. (2010). In search
of an integral conceptual definition of frailty: Opinions of experts. Journal of the
American Medical Directors Association, 11(5), 338–343.
https://doi.org/10.1016/j.jamda.2009.09.015
Groff, E., & La Vigne, N. (2002). Forecasting the future of predictive crime mapping.
Crime Prevention Studies, 13, pp. 29-57.
62
Groß, H. (Ed.). (2008). Handbuch der Polizeien Deutschlands [Handbook, police work
in Germany] (1. Aufl.). Wiesbaden: VS Verlag für Sozialwissenschaften. Retrieved
from http://www.loc.gov/catdir/enhancements/fy1406/2008386193-d.html
Haberman, C. P., & Ratcliffe, J. H. (2012). The Predictive Policing Challenges of Near
Repeat Armed Street Robberies. The British Journal of Criminology, 6(2), 151–166.
https://doi.org/10.1093/police/pas012
Hadamitzky, G. (2015). Crime Mapping: Digitale Kriminalitätskartierung in Zeiten der
Risikogesellschaft. Schriftenreihe Polizei & Wissenschaft [Digital crime mapping in
times of the risk society. Series Police & Science]. Frankfurt am Main: Verlag für
Polizeiwissenschaft.
Hand, R. (2018). How Black Mirror combines a disturbing future with a familiar past. In
conversation: Academic rigour, journalistic flair.
Hardyns, W., & Rummens, A. (2018). Predictive Policing as a New Tool for Law
Enforcement? Recent Developments and Challenges. European Journal on Criminal
Policy and Research, 24(3), 201–218. https://doi.org/10.1007/s10610-017-9361-2
Heffernan, W. C. (2019). Policing the Police. In W. C. Heffernan (Ed.), Rights and
Wrongs (pp. 95–109). Cham: Springer International Publishing.
https://doi.org/10.1007/978-3-030-12782-4_6
Helfferich, C. (2019). Leitfaden- und Experteninterviews. In N. Baur & J. Blasius (Eds.),
Handbuch Methoden der empirischen Sozialforschung [Guideline and expert
interviews., Handbook Methods of Empirical Social Research] (pp. 669–686).
Wiesbaden: Springer Fachmedien Wiesbaden. https://doi.org/10.1007/978-3-658-
21308-4_44
Hill, P. B. (2002). Rational-Choice-Theorie Einsichten. Themen der Soziologie [Rational-
Choice-Theory insights. Topics of sociology]. s.l.: transcript Verlag. Retrieved from
http://www.content-select.com/index.php?id=bib_view&ean=9783839400302
Howsen, R. M., & Jarrell, S. B. (1987). Some determinants of property crime: economic
factors influence criminal behavior but cannot completely explain the syndrome.
American Journal of Economics and Sociology, 46(4), 445–457.
https://doi.org/10.1111/j.1536-7150.1987.tb01992.x
Ioimo, R. (2018). Crime analysis and crime mapping. In R. Ioimo (Ed.), Introduction to
Criminal Justice Information Systems (pp. 181–206). CRC Press.
https://doi.org/10.4324/9781315368504-8
Ioimo, R. (Ed.). (2018). Introduction to Criminal Justice Information Systems: CRC
Press. https://doi.org/10.4324/9781315368504
Joh, E. (2014). Policing by Numbers: Big Data and the Fourth Amendment (89 Wash. L.
Rev. 35). University of California, School of Law, California. Retrieved from SSRN:
https://ssrn.com/abstract=2403028
Joh, E. (2015). The New Surveillance Discretion: Automated Suspicion, Big Data, and
Policing (Harvard Law & Policy Review). Retrieved from SSRN:
https://ssrn.com/abstract=2680266
63
Johnson, S. D., & Bowers, K. J. (2004). The Burglary as Clue to the Future. European
Journal of Criminology, 1(2), 237–255. https://doi.org/10.1177/1477370804041252
Jones, T. (2002). The Transformation of Policing? Understanding Current Trends in
Policing Systems. The British Journal of Criminology, 42(1), 129–146.
https://doi.org/10.1093/bjc/42.1.129
Jordan, P., Pietruska, F., & Siemer, J. (2017). Geoinformation & Visualisierung: Pionier
und Wegbereiter eines neuen Verständnisses von Kartographie und Geoinformatik:
Festschrift anlässlich der Emeritierung von [Geoinformation & Visualization: Pioneer
and pioneer of a new understanding of cartography and geoinformatics] by Prof. Dr.
Hartmut Asche, March 2017. Potsdamer Geographische Praxis: Vol. 12. Potsdam:
Universitätsverlag Potsdam. Retrieved from https://publishup.uni-potsdam.de/opus4-
ubp/files/1007/pgp12.pdf
Jüttemann, G. (Ed.). (1985). Qualitative Forschung in der Psychologie: Grundfragen,
Verfahrensweisen, Anwendungsfelder [Qualitative research in psychology: basic
questions, procedures, fields of application]. Weinheim: Beltz.
Kaufmann, M., Egbert, S., & Leese, M. (2018). Predictive Policing and the Politics of
Patterns. The British Journal of Criminology, 48, 3. https://doi.org/10.1093/bjc/azy060
Kennedy, L. W., Caplan, J. M., & Piza, E. (2011). Risk Clusters, Hotspots, and Spatial
Intelligence: Risk Terrain Modeling as an Algorithm for Police Resource Allocation
Strategies. Journal of Quantitative Criminology, 27(3), 339–362.
https://doi.org/10.1007/s10940-010-9126-2
Keupp, H., Rosenstiel, L. v., Wolff, St. (Ed.): Handbook of <