Home Office Police National Computer

1

• Home Office Police National Computer
• Adrian Shepherd and Lucy Cuppleditch,
• Reconviction Analysis, RDS-NOMS

2
Outline
Main topics

• What is HOPNC?
• What can it be used for and how have we used it?
• What have we learned?
• Practical example Proven re-offending rates
• Practical example Frequency of proven
  re-offending
• Practical example Local proven re-offending
  rates

3
What is HOPNC?

• Is a partial copy of the Police National
  Computer, containing
• Offender details (name, sex, ethnicity)
• Offence details for all recordable offences (what
  offence it was, where and when it was committed)
• Disposal details for all recordable offences
  (whether disposed of through caution, conviction)

Within RDS-NOMS, HOPNC was designed to supplement
the Offenders Index a court based system for
recording criminal histories. Outside of
RDS-NOMS, Peter Grove, Chris Kershaw, and others
were pivotal in securing this system.
As it contains named individual data from the
police, the database is kept on a stand-alone
server in a secure room in 2 MS.
4
What and how can PNC be used for

• What
• Ad-hoc analysis. E.g. Whats the most common
  names amongst offenders on the database. And less
  serious questions like the average number of
  TICs.
• Criminal histories. E.g. What are the
  characteristics of offenders in prison, of
  offenders who receive fines?
• Reconvictions Analysis. E.g. What happens after a
  community penalty or prison sentence?
• Cohort Studies. What are the offending behaviours
  of those born in a certain year?
• Monitoring and Performance Information. E.g. Are
  offenders being moved from community penalties to
  fines?
• How?
• All are delivered through the provision of
  extracts to RDS or other teams or through the
  output from SQL programmes

5
What are we doing with it?
Developing new estimates of risks of re-offending

• It emerged on Friday that Hanson had been let
  out of jail on licence despite an official
  assessment calculating that his chances of
  committing another violent offence were 91.
  Anything above 75 is considered high risk. BBC
  News, 19th December 2005
• current OGRS2 algorithms predict risk of
  reconvicted within two years for groups of
  offenders.
• work with Lancaster University and ODEAT team in
  the Home Office has resulted in a new measure of
  calculating all re-offending, not only of those
  on community penalties but also for those
  offenders who have been cautioned.
• the new measure is a more accurate indicator
  than older OGRS owing to the richer data held in
  the PNC.
• by moving from logistic regression to ordinal
  regression, we have both one and two year
  re-offending rates.

6
What are we doing with it?
Using PNC to get more accurate information about
reconviction generally
More than a third of criminals reoffended within
six months of ending their sentence and almost 50
per cent within a year. The Times, December 22nd.
PNC gets round the problem of pseudo-reconvictions
to give us an accurate record of months to first
re-offence. Allow more timely re-offending rates.
7
What have we learnt

• About the process
• Running a database like this is time consuming.
  It needs substantial efforts to keep it up to
  date in terms of offence mappings, updates,
  back-ups, and security arrangements. This takes
  at least 1 week per month.
• The learning curve for the organisation is about
  12 to 18 months and we are still on that curve.
  Each month we learn something new, or take a
  different approach to old problems.
• As its a SQL database, retrieving information
  requires writing code to query the database. Most
  SQL courses that are available concentrate on
  application development. As such, learning to use
  PNC requires an apprenticeship.
• Facts
• Lots. From the previous histories of the prison
  and probation population, re-offending from
  cautions, and so on.

8
Measuring Proven Re-offending
Measuring Proven Re-offending
9
Step 1 Measuring Actual Rates
Measuring Proven Re-offending

• Data on community sentence commencement and
  custodial release
• Match names on data-sets to PNC records using
  names, gender, date of birth, conviction date.
• Quarterly
• Check conviction date is within one week of PNC
  data

10
Defining Proven Re-offending
Measuring Proven Re-offending

• Re-offending is defined as someone committing an
  offence within two years of commencing a
  community order or being released from custody
  for which they were convicted regardless of when
  the conviction occurred

11
Measuring Proven Re-offending
Offence
Conviction
Two years
Index date
Offence must occur within two years of index
date, conviction can occur after.
12
Step 2 Generating Predicted Rates
Measuring Proven Re-offending

• They control for changes in the offender cohort
  in their underlying predisposition to re-offend
• How do they work? Predict re-offending in 2000
  (baseline cohort) using offender age and sex, and
  various aspects of offending history, and
  previous prison sentences in logistic regression.
  
• Then using coefficients in that regression and
  applying the same coefficients to subsequent
  cohorts can generate predicted scores on a
  like-for-like basis.

13
Measuring Proven Re-offending
Step 3 Combining the actual and predicted rates
14
Measuring the frequency of proven offending
Frequency of proven re-offending
15
Understanding the figures
Frequency of proven re-offending
Proven re-offenders 25,000

• Cohort 45,000

Proven re-offences 100,000
Desisters 20,000
Proven re-offences per cohort member 2.2
16
Frequency of proven re-offending
Distribution of data
17
Predicting frequency - Methodology
Frequency of proven re-offending

• OLS No, not normal
• Poisson No, mean ? variance
• Zero-inflated Poisson Still no, mean ? variance
• Negative binomial No
• Ordered logistic regression (Cameron and
  Trivendi, 1998) - Robust, flexible, simple.

18
Predicting frequency - Data
Frequency of proven re-offending

• Dichotomous vs. continuous (specificity)
• Proven re-offending
• Limited variables

19
Local reports on proven re-offending
Local reports on proven re-offending
20
How the data is created.
Local reports on proven re-offending
Probation caseload 250,000
Matched243,000 (97)
PNC 218,000
All offenders on the caseload those on licence
or community orders
Used to see if the offender has been proven to
re-offend
PNCID, name, date of birth, sex, date of
conviction
Report to be produced for each probation area
every quarter
Caseload taken every March, June, September and
December
21
The time period for the proven re-offending rate
Local reports on proven re-offending
22
Areas under investigation.
Local reports on proven re-offending

• How similar are the probation areas and can they
  be compared?
• What is the affect of the Criminal Justice
  System?
• How different are the caseloads over time and can
  they be compared?

23
Initial analysis How different are the probation
caseloads?
Local reports on proven re-offending
24
Initial analysis How different are the probation
caseloads?
Local reports on proven re-offending
25
Initial analysis How difference are the
caseloads over time?
Local reports on proven re-offending