Transcript Microsoft PowerPoint - NCRM EPrints Repository

Living near to burglars:
estimating the small area level
risk of burglary in
Cambridgeshire
Robert Haining
Department of Geography
University of Cambridge
ESRC Research Methods
Festival, Oxford, July 2010
Outline:
1. The nature of ecological analysis.
2. Geographical variation in numbers of burglaries
by Census Output Area (COA) in Cambridgeshire:
description and explanation.
3. The challenges presented by ecological analysis.
4. Concluding remarks.
1. The nature of ecological analysis.
Ecological: study of groups or aggregates using
data grouped by:
social class;
socio-economic status;
demographics (sex, age cohorts) ....
geography (and time).
Geography
scale
UK Census: Census Output Areas (COAs) – c. 220,000.
Design criteria for COAs:
- aggregations of postcodes;
- recommended to contain approx. 125
households;
- socially homogeneous based on housing
tenure and dwelling type.
(ONS website)
Police recorded crime database, data quality issues:
[1] Accuracy (geocoding; time of event)
[2] Completeness (e.g. offender not
known; not all offences reported;
extent of personal details)
[3] Consistency.
[4] Resolution (COA level).
Forms of ecological analysis:
Descriptive: maps (presentation
graphics; visualization tools);
graphical and numerical
summaries (e.g. hotspot
locations).
Confirmatory: model fitting for
parameter estimation and
hypothesis testing.
Why are ecological analyses of crime data useful and
important?
[a] Police are “territorial” and one aspect
of resource allocation is by geographical
area. PFAs, BCUs and beats/
neighbourhoods.
[b] Many theories about the location of
offences have an ecological level, but:
- what is the appropriate spatial
framework?
- what is relationship between the
appropriate framework and data
availability?
2. Geographical variation in numbers of
burglaries by COA in Cambridgeshire:
description and explanation
Map of burglary counts at COA level in
Cambridgeshire, 2002.
Conceptualising the problem: Burglary as
the outcome of a rational choice “two
stage process”.
Stage 1: Select area
Stage 2: Select target within the
chosen area.
Each stage involves a distinct set of
factors
Area selection factors:
(1) area attractiveness (reward):
likely gains from a burglary =>
affluent areas might be favoured over
less affluent and deprived areas.
also areas where households are
expected to have high value and easy to
steal goods.
(2) area opportunity (risk):
likelihood of not getting caught =>
areas with fewer formal and/or informal
“capable guardians” offer a greater
likelihood of success. (Routine activity
theory; Cohen and Felson, 1979.)
- affluent neighbourhoods where
residents are absent for extended
periods during a day/at weekends.
- areas with low levels of collective
efficacy (social cohesion +
willingness to act for common
good).
(3) area accessibility (familiarity + least
effort principle):
- areas which are known to the
offender perhaps because they are
near to where they live (or work
etc) but where (s)he will not
be recognized.
Summary:
At stage 1, for the motivated offender,
choice of area is a balance of risk against
reward whilst taking into account the effort
involved.
At stage 2, choice of target may be
opportunistic.
Questions:
[1] What is the statistical significance of each
of these three sets of factors and how far do they
help us to explain area differences in burglary rates.
{Modelling for the purpose of hypothesis testing}
[2] By how much, on average, do area level
rates of burglary increase for unit increases in each
of the different factors.
{Modelling for the purpose of parameter estimation}
Problem:
We are dependent on the UK National Census but there is often no clear
or unambiguous link between Census variables and the attractiveness or
opportunity factors.
However we are mainly interested in estimating the importance of the
accessibility factor whilst controlling for these two other factors.
We collected census data on 20 variables covering:
- household composition (e.g. Prop. lone parent households)
- living arrangements (e.g. Prop. single people in households)
- household tenure (e.g. Prop private rented)
- accommodation type (e.g. Prop. detached housing)
- population turnover
- social and ethnic composition (e.g. index of ethnicity)
n
access index (i) 
 f(d (i, j )) z(j)
j 1
u(i)
 1000
f(d(i,j)): a function of the distance, d(i,j), between
the centroids of COAs i and j.
z(j) denotes the number of burglaries committed by
residents of COA j in 2001.
u(i) denotes the number of dwellings in i.
After Bernasco and Luykx (2003).
Distance from centroid of COA where offender resides to centroid of COA where offence was
committed: (a) offenders resident in urban COAs (b) offenders resident in rural COAs (c)
Gamma functions fitted to the distances travelled by offenders resident in urban and rural
COAs: Dashed line: Gamma (1.12, 0.0005) for offenders resident in urban COAs. Solid line:
Gamma (0.7, 0.0003) for offenders resident in the rural COAs.
Histogram of burglary counts: Cambridgeshire COAs, 2002.
Mean: 2.28
Variance: 9.56
Burglary counts by COA: 2002
(1) Negative binomial GLM
log it [ p(i)]  log[N(i)]  β 0  β1 X1 (i)  .....  β k X k (i)
(2) Poisson model with spatial (S) and non-spatial (U) random effects
log[  (i)]  log[N(i)]  β 0  β1 X1 (i)  .....  β k X k (i)  S(i)  U(i)
Multiplicative change in the expected number of burglaries in a COA,
due to a unit increase in the corresponding variable: Cambridgeshire
COAs 2002.
Negative
binomial GLM
Poisson with spatial
random effects
Prop. detached hhlds [Att]
Prop. private rented hhlds [Att]
Prop. lone parent hhlds [Op]
Prop. Economically inactive
Prop. multi-person hhlds (not
students) [Op]
Prop. single people in hhlds [Op]
Access index quantile 3 [Ac]
Access index quantile 4 [Ac]
1.0026
1.0112
1.0286
1.0170
1.0363
1.0026
1.0070
1.0198
1.0124
1.0369
1.0116
1.4978
2.3560
1.0076
1.3324
1.8826
Peterboroughc
DIC
Psi [sd(S)/sd(U+S)]
Dispersion parameter (r)
Moran test (Pearson residuals)
1.2827
8493.03
1.4552
7863.35
0.670 (.577, .850)
2.160 (1.9, 2.47)
0.190 (prob=0.001)
Remarks:
1.The British Crime Survey (BCS) reveals the types of ACORN areas with
the highest burglary rates. These areas include:
(1) areas with council flats, high unemployment and persons living
alone (3.1 times the national average (x3.1)) or many lone
parent families (x 2.8).
(2) areas with furnished flats and bedsits housing young single
people (x2.4).
Many of these ACORN areas seem to be characterised by providing
“opportunity” rather than being “attractive”.
2. This study gives a Cambridgeshire county level perspective on these
factors:
(1) a similar emphasis but with increases in risk that appear modest
by comparison with findings from the BCS.
(2) Comparable increases in risk are found in the case of
Peterborough (x1.45) and when COAs are close to the homes
of motivated offenders (x1.88). A COA in Peterborough
close to concentrations of motivated offenders has a raised risk
of x2.74.
(3) Mapping the residual relative risk reveals high levels of risk in
north/northwest Cambridge unaccounted for by these factors.
3. Some challenges presented by ecological
analysis.
(a) modifiable areal units problem (MAUP):
(i) scale effect (different results at
different resolutions).
(ii) grouping effect (different results
from different aggregations)
The MAUP has implications for:
- mapping (pattern detection);
- hot spot detection;
- results of modelling (regression)
and correlation):
(b) Selection of “appropriate” spatial
units: neighbourhoods.
(c) Incompatible spatial frameworks.
(d) Areas with small populations –
populations tend to be more homogeneous
but statistics suffer from the small number
problem.
Areas with large populations –
statistics more robust (with smaller standard
errors) but populations tend to be more
heterogeneous
(e) Classical statistical analyses need to
contend with the problems created by
(inter-area) spatial autocorrelation:
(i) in dependent variable
(ii) in model residuals.
inference problems
4. Concluding remarks.
-Spatial ecological analyses have a place in
research into crime and disorder and are also
relevant to the way police forces operate.
- Spatial ecological analyses present a number
of challenges to data analysts.
- Spatial ecological analysis continues to be a
rapidly developing area of methodological
research that crime analysts ought to keep an
eye on.