Offending, employment and benefits – Emerging findings from data

Offending, employment and
Offending
benefits – emerging findings from
the MoJ / DWP / HMRC data
share
RSS Event – 21 February 2012
M li
Melissa
C
Cox – Statistician
St ti ti i in
i M
MoJ
J
Nick Murphy - Economic Adviser in DWP
This presentation will cover:
- Background to the data-share
- Approval for the data-share
- Data matching methodology and results
- Introduction to the linked data
- Cleaning
g and validating
g the matched data.
- Initial findings from analysing linked data
- Benefits of data-share: supporting policy development
- Next steps
2
Glossary
W ’ll ttry tto avoid
We’ll
id acronyms, b
butt jjustt iin case…..:
CJS – Criminal Justice System
y
JSA – Jobseeker’s Allowance (key out-of-work benefit)
ESA - Employment and Support Allowance
IB – Incapacity Benefit
IS – Income Support
P45 – P45 employment spell
NBD – National Benefit Database
PNC – Police National Computer
DPA – Data Protection Act
MoU – Memorandum of Understanding
PIA – Privacy Impact Assessment
3
Background to the data share
Joint analytical data sharing project between DWP and MoJ
which shares administrative data between MoJ and DWP /
HMRC on offending, benefits and P45 employment to:
• improve
i
th
the ((very thi
thin)) evidence
id
b
base on offender
ff d employment
l
t
and benefit outcomes;
• get a better understanding of the links between re-offending,
employment and benefits; and
• use that evidence to develop effective policies to reduce reoffending
ff di and
d welfare
lf
d
dependency.
d
4
Coverage of data share
MoJ data on offending (primarily Police National Computer (PNC) data)
p y
linked to DWP / HMRC data on benefits and P45 employment.
3.6 million offenders who have at least one caution/conviction between
2000 and 2010,
2010 and have at least one P45 employment or benefit spell
spell.
The data covers all offending, benefit and P45 employment spells over
thi ti
this
time – including
i l di ttypes off offences
ff
and
d sentences,
t
types
t
off benefits
b
fit
claimed.
The employment data included in the data-share is derived from P45
forms sent to HMRC by employers. P45 employment spells do not
usually record employment paid at levels below tax thresholds, or selfemployment or cash-in-hand informal economy work but should provide
a useful proxy of employment.
No information included on earnings,
g type
yp of employment.
p y
5
Our two year data sharing journey
Nov 2009
Nov 2010
Nov 2011
Data Sharing approval
Data
matching
Data
prep
Analysing linked
data
6
Approval for the data-share– why?
•Full consideration was given to relevant legal and ethical issues before
a decision was taken to proceed
proceed.
•Data sharing needs to be:
-
lawful,
fair,
jjustified and
proportionate
•Approval needed by MoJ,
MoJ DWP,
DWP HMRC and PIAP (data owners)
•Agreement reached and full approval for a one-off data-share was
given
i
iin D
December
b 2010
2010.
7
Legal basis
MoJ, DWP and HMRC lawyers agreed legal basis for data
sharing:
Section 14
14, 4 (c) of the Offender Management Act 2007
Which enables disclosure of information for offender
management purposes - including the development or
assessment of policies relating to matters connected with the
management of offenders.
To avoid multiple flows of data, DWP will act as Data Processor
for HMRC and process and transfer HMRC data that DWP holds
under other gateways.
Condition that shared data can be used for analytical purposes
only.
8
Ethical approval – safeguards
• DWP did matching
t hi - more proportionate
ti
t
• Only shared a limited number of variables - justification for
each
• Excluded offenders aged under 16
• Anonymised at earliest opportunity and personal data deleted
• Shared data has restricted access and securely stored
• Approval from Ethics Committee
• Approval from operational security to do secure data transfers
- personal delivery
• Bound byy an Memorandum of Understanding
g and Privacy
y
Impact Assessment
•Joint moderation group set up to ensure compliance
9
Challenges in getting approval
- Agreeing
A
i on llegall power
- Building relationships across Departments and data owners – multi
multidisciplinary and cross-Government team
- Change in personnel through project
- Persevering with the necessary paperwork
- Dependent
p
on successful feasibility
y study
y
- DWP Ethics Committee
10
Method of data transfer
Data transfer 1
MoJ
DWP
Dataset 1 – MoJ data
Personal identifiers of 4.2 million offenders
from Police National Computer extract – all
adults who have offended from 2000
DWP successfully
matched
PNC d
data
t tto
Master
Data transfer 2
DWP
MoJ
Dataset 2 – matched data
- Personal identifiers removed.
removed
- Agreed DWP / HMRC variables added.
Index and P45
according to
agreed matching
algorithm
- MoJ and DWP anonymous identifiers
retained.
Data transfer 3
MoJ
11
DWP
Dataset 3 – anonymised dataset
MoJ added on agreed MoJ variables to
Dataset 2 (removed DWP/HMRC variables).
Dataset 1 destroyed
by DWP once data
successfully
matched
Data matching - Variables
No unique identifier in common between MoJ and DWP / HMRC
data so data matching techniques needed.
Identified common variables across administrative data sources:
Forename
Surname
Date of birth
Gender
P t d (full
Postcode
(f ll postcode
t d hi
history)
t )
•Also included alias names
Reference ID (MoJ, DWP)
12
Data matching - methodology
Matching rules developed using common variables (including
initial of forename and fuzzyy matching
g on names to g
get best
match)
37 step
t matching
t hi algorithm
l ith originally
i i ll agreed
d using
i a scoring
i
system and combinations of at least 3 of the 5 variables
((forename,, surname,, date of birth,, postcode,
p
, gender).
g
)
Ranging from:
Exact matching on all 5 variables
3 out of 5 variables: surname, date of birth, gender
13
Final matching algorithm
• Quality assurance process to try to minimise matching errors:
• False positives: an identified but incorrect match
• False negatives: an unidentified but correct match
• Concentrated
C
t t d on minimising
i i i i ffalse
l positives
iti
even if thi
this llostt some
additional true matches.
• QA process:
- Match data using all 37 rules
- Sample
p of matches found at each stage
g
- Manually examine personal details
- If estimate more than 5% false positive – rule abandoned
• Algorithm simplified/improved from 37 to 20 rules – following quality
assurance
14
Final matching process
MOJ
data
DWP data
(Master
Index))
20 matching rule
algorithm:
-1st matching
rule – match or
no match?
Matching algorithm
Matched
data – 66%
Unmatched
data – 34%
HMRC
data (P45)
Matching algorithm
86%
matc
h
rate
15
Matched data
– 20%
Unmatched
data – 14%
- Unmatched
records then
used 2nd
matching rule
etc
etc…….
Data matching results
• 86% match rate: 3.66 m offenders matched to DWP /HMRC data
•Much
M hb
better
tt th
than expected
t d - match
t h rate
t off feasibility
f
ibilit study
t d was jjustt over 70%
• Quality of matches:
- 40% of matches = exact match on 5 personal identifiers
- Over 75% of matches was an exact match on all 5 variables, or exact
match on 4 variables (all excluding postcode)
•Representativeness of matched data:
- distributions of key variables between the matched data, un-matched data,
and the total.
- only
l reall diff
differences are ethnicity
th i it ((slight
li ht under
d representation
t ti ffor ethnic
th i
minority groups) and disposal category (higher proportion of cautions in the unmatched data)
Expertise of developing and running matching algorithm
16
Introduction to matched data
Criminal Justice System information (MoJ data):
- basic offence details (date of offence and offence type);
- basic
b i d
details
t il off sentence
t
received;
i d spellll off prison
i
and
d probation
b ti
where known;
Benefit,
B
fit P45 employment
l
t and
d programme iinformation
f
ti (DWP/HMRC
data):
- benefit spells (start and end dates, benefit type);
- P45 employment
l
t spells
ll ((start
t t and
d end
dd
dates);
t )
-programme spells (start and end dates, programme type)
- date of death
- ICD code (identifies type of illness for incapacity benefit claim)
-geographic level data; and
- necessary
y variables to use data including
g extract dates,, details of
match strength, anonymous identifiers and so on.
- Data matched internally for further analysis e.g. to look at specific
interactions with the benefit or criminal justice systems etc.
17
Introduction to the data
Combined spells dataset has approx. 40 million rows;
3.6 m offenders
7 6 m non-custodial
7.6
t di l sentences
t
1 0 m custodial sentences
1.0
13.8 m benefit spells
p
18.8 m employment spells
2.2 m programme spells
18
Cleaning and validating the data
Considerable time spent cleaning and understanding data due to scale
and
d complexity
l i off matched
h dd
data.
Statistical QA procedures applied to protect integrity of matched data
including;
- removal of duplicated entries,
- checks for completeness, and
- cleansing of inconsistent data based on business intelligence.
•Worked closely together:
• Set up
p and maintained an issues log
g
• Weekly video conference meetings
19
Any questions or comments so far?
15 mins
20
Initial findings
Initial findings
f
from
f
analysing linked data to support policy
development in specific areas and are intended to demonstrate
the potential of the improved evidence base.
Published in November:
http://www.justice.gov.uk/publications/statistics-and-data/adhoc/index.htm
These statistics do not imply causality between benefit or
employment status and proven offending
offending.
21
Initial findings – visualising data
Example 1 – Person 1338
Male, Born mid 1960s
July 2006 sentenced to 4 years prison for “violence
violence against the person”
person
P45
Employment
Prison
22
Initial findings – visualising data
Example 2 – Person 534
Male, Born mid 1970s
Sentenced several times for theft
theft, once for burglary
burglary, once for violence
and many times for other indictable offences
Basic Skills
ESA
Emp. Zones
Incap. ben
Income sup
JSA
Non-custodial spell
Prison
23
Initial findings – visualising data
Example 3 – person 1773
Male, born early 1980s
Sentences for (in order) theft
theft, robbery
robbery, summary offences excl.
excl
motoring and violence.
P45
Employment
JSA
Non-custodial spell
Prison
24
Initial findings–descriptive statistics
Th proportions
The
i
off offenders
ff d
with
i h each
h type off record
d are:
86 % have at least one P45 employment spell;
76 % have at least one DWP benefit spell; and
28 % have at least one DWP programme spell.
26 % of all 4.9 million out-of-work benefits being claimed on 1
December 2010 in England and Wales were claimed by offenders in
the data-share.
- 33 % of JSA claims were by offenders.
- 20%+ of Incapacity Benefit, Employment and Support Allowance or
Income Support claims
25
Benefit and P45 employment status around
time of sentence
Offenders sentenced during year ending 30 November 2010
Proportion of offenders claiming
benefits or in P45 employment at
some point in the month before
sentence
Cl i i
Claiming
benefits
b
fi (any
(
benefits)
b
fi )
54%
4%
Claiming out of work benefits
51%
Jobseeker's
Jobseeker
s Allowance
24%
Incapacity benefits
13%
Employment and Support Allowance
Income Support
9%
14%
Other benefits
3%
N b
No
benefits
fit claimed
l i d
46%
In P45 Employment
33%
26
Benefit and P45 employment status around
time of sentence – by sentence type
Benefit and P45 employment status for offenders in the month
before sentence by sentence type for offenders in the matched data
who were sentenced in the year ending 30 November 2010 and recorded on the PNC
Proportion of offenders
Any out-ofwork
benefits
All disposals
Caut o
Caution
1
Fine
Community Sentence
Suspended Sentence Order
Immediate Custody
Discharges (Absolute / Conditional)
Other
In P45
employment
51%
33%
40%
0%
47%
57%
57%
51%
64%
60%
46%
6%
39%
29%
30%
13%
27%
24%
Offenders sentenced
to a caution less likely
to be claiming
g
benefits, and more
likely to be in P45
employment
Only 13% of offenders sentenced
to immediate custody are in P45
employment in month before
sentence
1. Care should be taken with the analysis on fines. The PNC data largely covers 'recordable' offences where the coverage
of fines in the matched data only includes fines that are given for the more serious summary offences. The PNC includes
less than a fifth of all fines given by the courts so these findings must not be interpreted as representative of all fines.
27
Benefit, P45 employment and prison
status for all prisoners released in 2008
• 47% claiming out-of-work benefits 2 years after release from prison
•15 % in P45 employment
• 11% back in prison
Proportion of offenders
s
60%
50%
JSA,ESA,IB,IS
P455
PRISON
Any NBD Benefit
40%
30%
20%
10%
0%
0
13
26
39
52
65
78
91
104
117
Number of weeks since release from prison
28
130
143
156
Benefit status for all prisoners released in 2008 –
by benefit type
35%
P
Proportio
on of offen
nders
30%
25%
20%
15%
10%
5%
0%
0
26
JSA
IS
SDA
52
P45
ICA
WB
78
PRISON
AA
PC
104
130
IB
RP
PIB
Number of weeks since release from prison
29
156
ESA
DLA
BB
Whether ex-prisoner had any benefit, P45 or
prison spell in weeks following release in 2008
Cumulative proportion:
• 75% of offenders made at least one claim to an out-of-work benefit within 2 years of release from prison
• 29% started at least one P45 employment spell
• 46% had at least one prison spell at some point in 2 years following release
Proporrtion of offfenders
80%
70%
60%
50%
40%
30%
%
20%
JSA+IB/ESA+IS
10%
EMP
PRN
0%
0
30
13
26
39
52
65
78
91
Number of weeks since release from prison
104
Whether ex-prisoner claimed certain benefits in
weeks following release in 2008 – by benefit type
Cumulative proportion:
• 60% of offenders made at least one claim to Jobseeker’s Allowance within 2 years of release from prison
• Just under 30% made at least one Incapacity Benefit or Employment and Support Allowance claim
• Around 15% made at least one Income Support claim
70%
JSA
IB/ESA
Proportio
on of offe
enders
60%
IS
50%
40%
30%
20%
10%
0%
0
31
13
26
39
52
65
78
Number of weeks since release from prison
91
104
JSA survival rate comparison
DWP: May 2011
%S
Still Claim
ming JSA
• Ex-prisoners seem to perform pretty much the same as the average JSA
claimant – onlyy 10% of claims last for a year.
y
But.......
JSA
Prisoners
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%
0
13
26
39
52
Duration (weeks)
32
JSA survival rate comparison
DWP: May 2011
....ex-prisoners spend almost 40% more time on benefits in the
3 years following a new JSA claim than the JSA average
Whether on benefits in weeks following new JSA claim
Proporrtion of cllaimants
100%
Average JSA
Ex-Prisoner
80%
60%
40%
20%
0%
0
13
26
39
52
65
78
91
104
weeks since JSA claim start
33
117
130
143
156
Analysis informing Policy (1):
Prisoner Work Programme development
Two Key Findings:
1. Offenders spend more time on benefits from “day 1” of a JSA claim
th th
than
the average JSA claimant
l i
t spends
d ffrom th
the “12 month”
th” point.
i t
2. A little over 30,000 offenders leave prison each year and claim JSA
within 13 weeks.
Time on benefit
over 2 years
Time in prison
over 2 years
Maximum time on
benefit or in
prison
JSA25+
66%
1%
67%
JSA E
Early
l A
Access
83%
2%
85%
JSA Prison Leaver
59%
10%
69%
34
Analysis informing Policy (2):
The Policy
From March 2012 all offenders leaving prison and claiming JSA
within 13 weeks will be mandated to the Work Programme.
Extensive reforms are taking place within the benefit system to
enable prisoners to start the JSA claim process prior to release.
- Minimise “prisoner finance gap” and
- Maximise
M i i employment
l
t supportt
35
Some Other Findings in brief
- 44% of 500,000 Community Care Grant (CCG) applications made in England and Wales in
2009/2010 were made by people on the DWP/MoJ dataset.
- This includes 15% that were recorded on the dataset as having been in prison at some
point.
-9% of 3 million Disability Living Allowance claims by offenders
- 1% off 11.6
11 6 million
illi R
Retirement
i
P
Pension
i claims
l i
b
by offenders
ff d
-Over the three year period, it is estimated that, per individual, the ex-prisoner population
receive an extra £1,500 ((or 38 per
p cent more)) in benefits than the average
g JSA claimant.
- Response to August public disorder:
-35 % of adults were claiming an out-of-work benefit at the time of the August 2011 public
disorder (compared to 12 per cent of the working age population in England)
England).
-45 % of all offenders who were sentenced for an indictable offence in 2010 were claiming
benefits.
36
Benefits of shared data - uses
Breaking the cycle: rehabilitation of offenders and reducing
welfare dependency
Analysis from linked data already playing important role in
helping DWP and MoJ to produce better evaluations, monitoring
information on interventions and in targeting resources and
developing implementation plans (such
(
as on the Work
Programme extension to Prison Leavers claiming JSA).
37
Press coverage
FURY AS JUNKIES GET £1BN BENEFITS – Express, May 2011
g handed to convicted
“A further £162million a yyear is being
criminals who go straight on to jobless benefits after they are
released from prison.”
THIRD OF UNEMPLOYED ARE CRIMINALS – Telegraph, Mail,
Metro + others
others, December 2011
TOO SICK TO WORK BUT NOT TOO SICK TO RIOT: 1 in 8
defendants were on incapacity or disability benefits – Mail,
October 2011
38
Next steps
• Continue to analyse shared data, e.g:
- Evaluate the links between employment and re-offending
- Link shared data to other datasets (ensuring compliance with MoU)
- Grateful for your ideas after we’ve finished presentation!
• We intend to move to an ongoing data share (pending approval
being given) given the value from the shared data
• Exploring data shares with other Government Departments to
support policy development and evaluations
• Keen to exploit
p
the shared data as much as p
possible.
Considering ways to provide anonymised access to data –
possibly through a Datalab – we will keep you posted!
39
• Questions and discussion
40