Visualising Crime in Space
and Time
Chris Brunsdon, Jon Corcoran,
Gary Higgs
University of Glamorgan
Introduction and Context
zCertain types of crime tend to happen in
certain places
zCertain types of crime tend to happen and
certain times
zSpace and Time are linked – ie
disturbances may happen in city centres
on Friday evenings
Space-Time Patterns in Crime
zRoutine Activity Theory (Cohen & Felson)
{“Criminal offences are related to the nature of
everyday patterns of social interaction”
zTime Geography (Hagerstrand)
{“... concerned not so much with the long time
spans of historical research but with daily,
weekly and seasonal rhythms within human
behaviour over space”
Are all the patterns obvious?
zNo – although the theory predicts that
such patterns exist, they are silent on
exactly when and where hotspots will
appear.
zFelson’s (2001) work suggests daily rises
and falls in incidents vary geographically –
so knowing the daily ‘cycle’ in one area
does not imply it will be the same in
another area
Aims of the talk
{Given a need to investigate space-time crime
patterns
{Consider three approaches to visualisation
z‘Comapping’
zIsosurface Visualisation
zAnimation
{Compare the three approaches
Test Data Set
z Reported disturbances (1999-2001)
z City Centre and surrounding areas
z Each report is time-tagged – at the time it was
reported
z Each report also grid-referenced
z Each report also date-stamped
z Data was processed using kernel density
estimation (risk surface)
{Geographical bandwidth = 500m
{Temporal bandwidth = 2h30m
Approach #1 : Comaps
z coplots (Clevelend, 1993), modified to comaps
(Brunsdon, 2001)
z Makes use of Tufte’s idea of ‘small multiples’
z Small multiples of diagrams allow visual
comparisons within eyespans. They are also
good for showing time series, or the evolution of
systems.
z c.f. Snapshot temporal analysis (ie Ratcliffe &
McCullagh,1998)
Comap for Time of Day
Characteristics of comaps
z They divide maps up according to time of day
(the co-variate – hence co-map)
z But divisions overlap in time
{Allows for some uncertainty
{Avoids patterns that are artefacts of the cut-off times
(rather like kernel smoothing does)
z Divisions also chosen to base each diagram on
the same number of data points
{Roughly same degree of random variability in each map
{Make comparisons more plausible
z Here the technique is modified to allow for
‘circularity’ of time-of-day
Comap for Time of Day and Day of Week
Approach #2: Isosurfaces
zAgain it is helpful to consider ‘snapshots’
in time
zThis time in terms of contours on a kernel
density (ie hotspot) map
Isosurfaces – extending risk map in time
Consider the contours ‘floating’ in
spacetime
Join them up – obtain a 2D surface in 3D
space
Finally – fill out the wire frame to get the
surface…
Isosurfaces in practice
zOpenDX – IBM open source software for
scientific visualization
zAnalyst needs to be able to interact with
the isosurface – and see associated risk
maps for any given time-slice (snaphot)
zDemo of OpenDx
Approach #3: Animation
z Show changing space time patterns as risk
surfaces, but animate them throughout the day.
z Key issue with animation: unlike comaps and
isosurfaces, it isn’t synoptic – ie does not show
all whole day in a single frame
z Viewers may well have forgotten early frames
towards the end of the movie
z Proposed solution the ‘lap-timer’ button – cf LCD
stopwatches
The ‘stopwatch’ – really a movie player –
but see ‘lap’ button
Result of pressing ‘lap’ button – another
kind of ‘snapshot’?
Can also examine weekly patterns...
Mon
Tue
Wed
Thu
Fri
Sat
Sun
Three Methods – Which is best ?
zAll do different jobs
zPublic vs Private usage
{Comaps perhaps good public displays
{Isosurfaces/Animations more interactive, better
for an individual analyst to interrogate for
patterns
Isosurfaces vs Animations
z Isosurfaces
z For
{ Show whole day’s pattern
{ Allow a large degree of
interaction
z Against
{ Mathematically abstract
{ Require some
familiarisation with
underlying idea
z Animations
z For
{ Intuitive
{ Can link in with a laptimer (or snapshot)
approach
z Against
{ Don’t show whole day at a
glance
{ Can forget earlier frames
while viewing later ones –
not so good for the ‘big
picture’
Further Issues
zNeed for a user-based evaluation?
{Which kind of method best for which kind of
user or audience?
zUncertainty in data
{In some cases, time of event not well known
{Aoristic crime analysis in a space-time
visualization framework?
zUse with analysis techniques other than
kernel density estimation
Practical Issues – Review Strategies and
Resource Allocation
zLinking these visualisations to other data
{Times pubs/clubs close etc.
{Compare with information about geographical
movements of police officers over the span of a
day
zComparison
{Spacetime patterns for ordinary Saturday vs
match day
{Seasonal variations