Adding Census Geographical Detail into

1

Adding Census Geographical Detail into the
British Crime Survey for Modelling Crime
Charatdao Kongmuang Naresuan University,
Thailand Graham Clarke and Andrew
Evans University of Leeds, UK
2
Background

• Crime and risk of victimisation
• Unevenly distributed across population, time and
  space
• Varies dramatically with demographic,
  socio-economic and area characteristics
• Crimes at the small area scale often do not match
  expectations based on national averages

3
Background (cont.)

• Although, the British Crime Survey (BCS) provides
  rich information about levels of crime and crime
  victimisation, it cannot be used to explain crime
  victimisation for small geographical units (not
  currently released below the national level)

4
Solution

• Attach the information from the BCS to the more
  geographically disaggregated census data using
  spatial microsimulation technique.

5
What is Microsimulation?

• A methodology aimed at building large-scale
  datasets on individual units such as persons,
  households or firms and can be used to simulate
  the effect of changes in policy or other changes
  on these micro units

6
What is Spatial Microsimulation?

• A microsimulation that takes space into account
• It contains geographical information that can be
  used to investigate the policy impacts

7
SimCrime Model

• A static spatial microsimulation model designed
  to estimate the likelihood of being a victim of
  crime and crime rates at the small area level in
  Leeds.

8
SimCrime Model (cont.)
Combines individual microdata from the British
Crime Survey (BCS) with spatially small area
aggregated census data to create synthetic
microdata estimates for output areas (OAs) in
Leeds using a Combinatorial Optimisation
Simulated Annealing method.
9
SimCrime Model Specification

• The synthetic microdata dataset was generated at
  the census output area for Leeds with the use of
  Simulated Annealing-Based Reweighting Program.
• 514,523 individuals aged 16-74 living in
  households found in Leeds in the UK 2001 Census
  were recreated

10
Simulated Annealing Based Reweighting
Program(generate a population microdata dataset
at the Output Area Level)

• Implemented in Java.
• The process involves selecting the combination of
  individuals from the BCS microdata which best
  fits the known constraints in the selected small
  areas (of the 2001 UK Census).
• The process is repeated with the aim of gradually
  improving fit between the observed data and the
  selected combination of individual from the BCS.

11
Data for generating synthetic micro-population

• Census Area Statistics of the 2001 UK Census
• Constrained Tables
• Number of total population in small-areas
• Microdata from the 2001 British Crime Survey

12
Census Area Statistics (CAS)

• Equivalent to the Small Area Statistics (SAS) of
  the 1971, 1981, and 1991 Censuses.
• Available for geographical levels down to output
  area (OA), the smallest unit of the 2001 Census.

Note Each OA contains approximately 290 persons
or 125 households
13
Variables related to crime
Category Indicator High Propensity
Demographic Characteristics of Offender Age Sex Marital Status Family Status Family Size Young adult Male Single Broken Home , divorce (weak family life) Large
Socio-Economic Status of Offender Income Employment status Education Deprivation Low income Unemployed Less High level of deprivation
Household Characteristics Density of living Tenure Substandard Rented
Victim Characteristics Age Sex Ethnicity Lifestyle Tenure Young adult Male Minority Group Away home Rented, not owner occupied
Neighbourhood types and characteristics Urbanisation Population Density Proximity High High Inner city, proximity to disadvantage areas
14
Constrained Tables

• CS004 Age by Sex and Living Arrangements (16
  categories)
• CS047 National Statistic-Socioeconomic
  Classification by Tenure
• (18 categories)
• CS061 Tenure and Car or Van Availability by
• Economic Activity (24
  categories)

15
SimCrime Constrained Variables Categories
Age Aged 16-24 Aged 25-34 Aged 35-49 Aged 50-74
Sex Male Female
Living Arrangement Couple Not couple
Economic Activity Employed Unemployed Inactive Full-time Student
Tenure Type Owned Rented
Car or Van availability No Car One Car Two or more car
Socio-economic Classification Higher Managerial and professional occupations Lower Managerial and professional occupations Intermediate occupations Small employers and own account workers Lower supervisory and technical occupations Semi-routine occupations Routine occupations Never worked and long-term unemployed Not classified
16
Discrepancies in census counts between tables
Source 2001 Census Area Statistics Note Each
cell shows the number of people aged 16-74 living
in households
17

• The constraint tables should be adjusted to
  minimise discrepancies between the total
  populations in small areas.

18
Constraint Tables Adjustment

• Total number of people in the small areas
  (GroupNumber)
• Each table cell

19
Constraint Tables Adjustment (cont.)

• Number of people in each cell
• Number of people from the constraint table x
  GroupNumber
• Total Sum for each area

20
(No Transcript)
21
What can we get?

• The adjustment method ensures the constraint
  tables are more consistent or at least can be
  guaranteed to produce the smallest discrepancy.

22
The British Crime Survey

• One of the largest social research surveys
  conducted in England and Wales (Sample 40,000
  households)
• A victimisation survey (whether or not reported
  to the police)
• Covers a wide range of topics (1,642 variables)
• The BCS can now provide limited information at
  the police force area level, but NOT for smaller
  geographical units.

23
Microdata (The 2001 BCS)
1,642 variables with 32,824 records
24
The Program

• The microdata filtering process
• Goes through the entire micro-database and checks
  whether an individual fits into each column of
  constraining tables for the current area.
• Simulated Annealing process
• Searches for the best combinations of individuals
  based on the result of the filtering process.

25
Output from the Simulated Annealing Based
Reweighting Program

• Synthetic Population A list of individuals which
  contains the demographic and socio-economic
  characteristics (crime variables from the BCS are
  attached).
• Error Report Provides information on the
  difference between distributions of constrained
  table and synthetic microdata at the output area
  level.

26
Error Report
The absolute differences between estimated
expected counts
27
Distribution of female single, widow, or divorce
aged 25-49 living in rented house
28
(No Transcript)
29
Distribution of high class households, owner
occupier having at least 1 car
30
Evaluation of Synthetic Microdata

• Evaluate in terms of their match to the
  constraint tables from the census at the output
  area level.

31
Evaluation of Synthetic Microdata (cont.)

• The measure of difference between distributions
  of constrained table and synthetic microdata is
  the Total Absolute Error (TAE)
• The sum of absolute differences between estimated
  and observed counts.
• To compare across the tables Standardised
  Absolute Error (SAE)
• TAE / Total expected count

32
SAE of 0 or perfect fit 1,318 output
areas Note The number in the bracket show
number of output area for each SAE group.
There are 2,439 output areas in Leeds.
Source SimCrime
33
Spatial distribution of SAE for all constraints
at output area level
SAE of 0 or perfect fit 1,212 output
areas Note The number in the bracket show
number of output area for each SAE group.
There are 2,439 output areas in Leeds.
Source SimCrime
34
Modelling Crime

• Each individual in the BCS has crime variables
  associated with them, the microsimulation allows
  us to make small area estimates victims of crime
  and high-risk areas.
• Assume that if the synthetic population have the
  same characteristics as the population from the
  BCS, they will have the same propensity to be a
  victim of crime.

35
Estimated victim rate per 1,000 households by
ward of burglary dwelling in Leeds
36
Conclusion

• SimCrime effectively adds geography to the
  British Crime Survey
• The spatial aspect of the data make it possible
  to do analysis at different spatial scales.
• Demonstrated a method to minimise discrepancies
  between the totals of the constraint tables

37
Conclusion (cont.)

• The spatial microsimulation has enabled the
  modelling of crime victimisation at small area
  levels. Before this the smallest area of
  modelling crime in the UK was at the police force
  area level.

38
More information

• Modelling Crime A Spatial Microsimulation
  Approach
• (Completed PhD thesis)
• http//www.geog.leeds.ac.uk/people/old/c.kongmu
  ang/
• SimCrime A Spatial Microsimulation for Crime in
  Leeds (Working Paper 06/1) http//www.geog.leeds.
  ac.uk/wpapers/index.html
• Email charatdao_at_gmail.com
• Dept. Natural Resources and Environment
• Fac. of Agriculture, Natural Resources and
  Environment
• Naresuan University, Muang, Phitsanulok, 65000,
  THAILAND

39

• Thank you