Analytical mapping of registered criminal activities in Vilnius city.
Beconyte, Giedre ; Eismontaite, Agne ; Romanovas, Denis 等
1. Introduction
Distribution and concentration of crimes and delinquency have
always been important in understanding city life. Geographic approach
has been successfully applied since early 20th century in the U.S. when
first (non-digital) crime distribution maps were made. New research
methods have been developed (Boba 2005; Bruce 2008) and more intensely
applied in order to support or to disprove theories about
differentiation of criminal activities between city districts, such as
social disorganization theory explaining street crime levels by
characteristics of neighbourhood (Zhang, Peterson 2007). More intense
research into geography of crimes began in the second half of the 20th
century as computers and GIS technology allowed processing large volumes
of geographic data and efficient visualizations of the results (Maltz et
al. 2000). Traditionally, such research is still more popular in the
U.S. where stronger differentiation between city zones is observed and
processes of succession are much faster than in Central or Eastern
Europe.
The main purpose of the research on crime and delinquency
distribution conducted in Vilnius University in 2009-2012 is to reveal
the spatial pattern of overall distribution in Vilnius City and
distribution of different types of offences. We expected to observe a
spatial trend of change of crime rates in 2011. We did not have initial
hypothesis about the location and number of highest concentration areas,
though it had been anticipated that they would match neither the pattern
with crime rates spatially dispersing from the city centre outward nor
the opposite.
2. Data and data processing
Data on criminal activities in Vilnius City in 2010 and 2011 were
obtained from the registry of Vilnius County Police headquarters. The
dataset of 2010 did not include Grigiskes. The incident record typically
consists of street address, number of injured/fatalities, date and time
when information was submitted to police offices or by 112 (common
emergency telephone number). The address information had not been
geocoded and used for spatial analysis.
The incidents were initially classified into types but different
sets of type values had been used in 2010 and 2011. The authors grouped
several types of incidents into four large types: assaults, burglaries
and thefts, motor vehicle thefts and minor offences. These four types
together cover about 58.8 percent of all analysed incidents of each year
(Fig. 1). Due to their different character (thus different factors that
influence the spatial pattern) each of the four types was analysed
separately.
Geographic coordinates of the registered incidents were determined
using street address information. Vilnius city address database
containing 48674 address points was used as reference dataset.
Because number of registered incidents was very large, the
geocoding process had to be automated. Initially we attempted to employ
ArcGIS Geocoding tool but lack of proper address locators and a poor
documentation made it inefficient. Then Google Geocoding API was tested,
but it could not be used due to its limitations of use: Google Geocoding
API may only be used in conjunction with a Google map and query limit of
2500 location requests per day exists. In order to avoid such
complications a custom geocoding program was written in Python.
The original addresses where incidents had been registered were
already split into components: city name, street name and optionally
house number, all stored in separate table columns. Such structure
facilitated identification of the address components. Street names was
the most problematic part of the address data due to misspellings and
inconsistency of spelling of the compound names, particularly personal
names that in different records would or would not include the title,
first or second given name or abbreviation. They all had to be
transformed into the same form.
In order to produce better address matching results, reference
dataset was altered in this way: city type was separated from its name
(same was done with street name) and city name was reduced to its root.
Reference dataset was also populated by two new fields: city name
soundex field and street name soundex field. That was necessary to
assure that small misspelling error in address did not result in a false
match.
Then every address record was processed as follows:
1. Address information was standardized. If address city name
included city type, it was separated from the city name string and was
standardized, e.g., "m.", "miest.",
"miestas", "mieste" (different forms and
abbreviations of "town") into "miestas" (nominative
of "town"). The same was done with the street names. Address
strings were converted into lowercase.
2. Soundex representations of city name and street name were
calculated.
3. The reference dataset was queried for address soundex
representation and all matches were returned.
4. The best matching address out of the returned list was computed
using Levenshtein algorithm for measuring the amount of difference
between the strings (Levenshtein 1965) and corresponding coordinates
were returned.
5. If registered incident address contained house number and the
latter was found in the reference dataset, the coordinates of the
address point were returned. If registered incident address contained
house number, but such number was not in the reference dataset, linear
interpolation was made between two presumably closest addresses on the
same street. If registered incident address did not contain house
number, location of a random house of the matched city street was
returned.
[FIGURE 1 OMITTED]
About 99.4 percent of the 97912 and 117417 incidents that occurred
correspondingly in 2010 and 2011 were successfully located. 84.7 percent
of incidents were matched to address points using exact incident address
point co-ordinates, 5.0 percent of incidents were located approximately
by interpolating nearest address point coordinates along the street
where incident had occurred and 9.7 percent of incidents were located by
randomly choosing incident's street address. Only 1186 incidents
could not be even approximately located.
Four main reasons why the incidents could not be located are:
1. Too many misspells in the address field (e.g., "Vilniaus m.
UKMR", "Vilniaus m. Rodunioskelias", "Vilniaus m.
Ghelvonu.").
2. Other locator than address was used (e.g., "Vilniaus m.
centras", "Vilniaus m. s.b. Vyturys").
3. Vilnius city address database did not contain address points for
the incident street.
4. Matching algorithm resulted in a wrong match. The false matches
were identified manually by scanning the address matching log.
Some records have been located manually. Mainly unclassified
incidents with completely incorrect location information could not be
matched.
After additional filtering (some located incidents were outside
Vilnius City), the total of 97812 incidents of 2010 and 116997 incidents
of 2011 were used in further calculations.
3. Methods of calculation
The probability of density of incidents has been estimated using
spatial kernel density method based on the quadratic kernel function
(Silverman 1986: 76) and a corresponding ArcGIS tool that calculates the
density of point features in a neighbourhood around each cell of an
output raster (Chainey, Ractliffe 2005; Gibin et al. 2007). Whereas in a
simple point density calculation points that fall within the search area
are summed, and divided by the search area size to get density value for
each cell, kernel density also evaluates influence of the occurrences
registered in the neighbourhood. We used 1500 meter neighbourhood radius
and the 60x60 meter cell size for the output raster dataset for Vilnius
city and correspondingly 400 meter neighbourhood radius (with exception
of small districts of Naujamiestis and the Old Town where 200 meter
radius was used for better precision) and the 10x10 meter cell size for
individual districts. The points have been weighted depending on the
total number of incidents that were registered at the same address. The
output raster datasets of 2010 (Fig. 2) and 2011 (Fig. 3) were produced
and subtracted to identify the changes in the crime and delinquency
landscape of Vilnius in one year.
Location quotient technique (Brantingham, P. L., Brantingham, P. J.
1997; Harries 1999) has been applied to determine the relative weight of
four major types of incidents. Location quotient was calculated as an
index for comparing a district's share of a particular type of
criminal activity with the share of that same activity at the city
level. It allows evaluating the deviation of impact of a particular type
of criminal activity in a district. We applied formula:
[LQ.sub.ij] = [C.sub.ij]/[summation over (j=1 ... 21)] [C.sub.ij] /
[summation over (i=1 ... 4)] [C.sub.ij]/[summation over (i=1 ... 4, j=1
... 21)] [C.sub.ij], (1)
where [C.sub.ij] is a rate of criminal activity i in district j.
[FIGURE 2 OMITTED]
Location quotients were estimated for assaults, burglaries and
thefts, motor vehicle thefts and minor offences separately.
4. Incident maps
The initial density maps (Figs 1, 2) show distribution of
probability of criminal activity for all Vilnius City based on incident
data of correspondingly 2010 and 2011. After many experiments, relative
density values have been grouped into 8 classes (1 - low, 8 - high)
using natural (statistical) breaks method in order to achieve best
visual expression. It can easily be seen that crime rates grew in 2011
but the concentration areas retained their general shape and structure.
The same classification method was used for the larger scale
density maps of four major types of crimes in the central part of
Vilnius. Neither total density nor density by each type of crimes has
any characteristic distribution in peripheral districts and the relative
density value for those districts is "low", that can be
explained by much lower population density outside the central part of
the city.
There are two large concentration areas: the major one of the Old
Town and secondary one of the flat block residential districts with the
centre in Pasilaiciai. These areas roughly match the highest population
density areas of Vilnius city. The central part of Vilnius including
those two areas was analysed in more detail and by different types of
criminal activity. The results are shown on four larger scale maps
(Figs. 3-7).
The patterns of distribution of four different types of criminal
activities (Figs. 4-7) reveal their dependence on the population density
as general configuration of each crime concentration areas is always
similar to the area of high population density and to the general
incident density areas. Distribution of the most common types of
incidents, namely minor offences (such as hooliganism and other public
nuisances, Fig. 5) and burglaries and thefts (Fig. 7) is very similar to
the general pattern and to each other. However, assaults (Fig. 4) tend
to concentrate in several smaller areas different from highest
population density areas and motor vehicle thefts (Fig. 5) have strongly
different distribution pattern with highest concentration area in newer
residential districts of Fabijoniskes and Pasilaiciai that have
relatively lower crime rates.
[FIGURE 3 OMITTED]
Location quotient choropleth maps (Fig. 8) show prevalence of
particular type of criminal activity in each district independently from
the total number of registered incidents. Thus, assaults strongly
dominate in crime landscape of Paneriai and Naujoji Vilnia, robberies
and thefts in Verkiai and motor vehicle thefts in Pasilaiciai,
Fabijoniskes and Grigiskes.
5. Conclusions
Density analysis can be applied for analysis of criminal activities
and produces results that are visually very expressive. However, if
volumes of initial data are large and not completely consistent,
pre-processing such as geocoding and generalisation may be time
consuming.
The authors have not made any assumptions about dependency between
crime rates and specific population groups. Indeed, the spatial pattern
of crime rates in Vilnius does not seem to be related with specific
communities but rather concentrates around specific public areas, such
as shopping and entertainment centres.
The spatial pattern of crime rates in 2010-2011 showed stability
with significant concentration in the Old Town and some immediate
neighbourhoods and not so strong concentration in Central-Western
residential districts. Overall incident rate is roughly proportional to
population density but patterns of different major types of crimes
deviate from Vilnius population density pattern in different ways.
[FIGURE 4 OMITTED]
[FIGURE 5 OMITTED]
[FIGURE 6 OMITTED]
[FIGURE 7 OMITTED]
[FIGURE 8 OMITTED]
Density analysis can be efficiently complemented by other methods
of spatial analysis. In our research location quotient was estimated to
determine structural differences of criminal activity in different
districts of Vilnius. Burglaries and thefts relatively prevail in rather
dissimilar districts of Snipiskes, Verkiai and Zverynas. Assaults tend
to prevail in the areas most distant from the city centre (Rasos,
Grigiskes, Naujoji Vilnia) with exception of Pasilaiciai residential
district. Motor vehicle thefts concentrate almost solely around
Pasilaiciai.
The geocoded point data and visualizations of registered incidents
of 2010 and 2011 are available at Lithuanian Spatial Information Portal
www.geoportal.lt.
doi:10.3846/20296991.2012.755343
References
Boba, R. 2005. Crime Analysis and Crime Mapping. California: Sage
Publications. 320.
Brantingham, P. L.; Brantingham, P. J. 1997. Mapping crime for
analytic purposes: location quotients, counts, and rates, in D.
Weisburd, J. T. McEwen (Eds.). Crime Mapping and Crime Prevention. New
York: Criminal Justice Press, 263-288.
Bruce, C. V. 2008. Exploring Crime Analysis: Readings on Essential
Skills. Kansas: International Association of Crime Analyst, 342-363.
Chainey, S.; Ractliffe, J. 2005. GIS and Crime Mapping. Chichester:
John Wiley and Sons. 422.
Gibin, M.; Longley, P.; Atkinson, P. 2007. Kernel density
estimation and percent volume contours in general practice catchment
area analysis in urban areas, in Proceedings of Geographical Information
Science Research Conference (GISRUK), 11-13 April 2007, NUI Maynooth,
Ireland, 270-276 [online], [cited 26 November 2012]. Available from
Internet: http://ncg.nuim.ie/gisruk/materials/proceedings/
Harries, K. 1999. Mapping Crime: Principle and Practice. Washington
DC: U.S. Department of Justice, Office of Justice Program, National
Institute of Justice, Crime Mapping Research Center. 206.
Levenshtein, V. 1965. Binary codes capable of correcting deletions,
insertions, and reversals, Doklady Akademii Nauk SSSR 163(4): 845-848.
Maltz, M. D.; Gordon, A. C.; Friedman, W. 2000. Mapping Crime in
Its Community Setting: Event Geography Analysis. New York:
Springer-Verlag. 197.
Silverman, B. W. 1986. Density Estimation for Statistics and Data
Analysis. New York: Chapman & Hall. 22.
Zhang, H.; Peterson, M. P. 2007. A spatial analysis of
neighbourhood crime in Omaha, Nebraska using alternative measures of
crime rates, Internet Journal of Criminology 31 [online], [cited 26
November 2012]. Available from Internet:
http://www.internetjournalofcriminology.com/Zhang%20
Peterson%20-%20A%20SPATIAL%20ANALYSIS%20 OF%20NEIGHBOURHOOD%20CRIME.pdf
Giedre BECONYTE. Professor, Vilnius University, M. K. Ciurlionio g.
21, LT-03101 Vilnius, Lithuania. Ph +370 640 16583, Fax: +370 5 2398296,
e-mail: giedre.beconyte@gf.vu.lt.
A member of the Commission of Theoretical Cartography at the
International Cartographic Association, the Executive secretary of
Lithuanian cartographic society, has published more than 40 papers
addressing information engineering methods of cartography, actively
participated in the projects on national thematic cartography and in the
process of developing the infrastructure of Lithuanian spatial data.
Research interests: thematic cartography, system analysis and
engineering, graphic design and cartography.
Agne EISMONTAITE. Master of Science in Cartography at the JSC
"HNIT-Baltic", S. Konarskio g. 28A, LT-03127 Vilnius,
Lithuania. Ph + 370 5 215 0575, Fax +370 5 215 0576, e-mail:
agne.eismontaite@gmail.com.
A graduate with Magna Cum Laude from Vilnius University (2012)
actively participating in activities of Lithuanian cartographic society.
Research interests: social geography, thematic cartography, GIS.
Denis ROMANOVAS. MSc student at Vilnius Gediminas Technical
University, GIS programmer at the National Centre of Remote Sensing and
Geoinformatics "GIS-Centras", Seliu g. 66, LT-08109 Vilnius,
Lithuania. Ph +370 5 2724 741, Fax +370 5 3737 723, e-mail:
d.romanovas@gis-centras.lt.
Research interests: GIS analysis and application programming.
Giedre Beconyte (1), Agne Eismontaite (2), Denis Romanovas (3)
Centre for Cartography, Vilnius University, M. K. Ciurlionio g. 21,
LT-03101 Vilnius, Lithuania
E-mail: 1giedre.beconyte@gf.vu.lt (corresponding author)
Received 03 October 2012; accepted 12 December 2012