1
Improving victimisation
estimates from the Crime
Survey for England and
Wales (CSEW)
Joe Traynor
24 January 2019
CSEW survey manager
Centre for Crime and Justice
We’ve changed how repeat
victimisation is estimated
• This includes a small change to the survey weights.
• Repeat victimisation is defined as the same thing,
done under the same circumstances, probably by the
same people, against the same victim.
2
We’ve replaced the cap of 5 from
the old methodology
• This was a limit on the number of repeat incidents
included in the estimates.
• We’ve replaced this limit with the 98th percentile value.
• We’ve adjusted the weights used on the survey to
better suit the inclusion of these higher counts.
3
There has been no impact on the
long-term picture of total crime
• However, the number of incidents for total CSEW
crime are slightly higher across the entire time series
than previously published.
• Since the year to March 2002, the average increase in
total CSEW crime (excluding fraud and computer
misuse) was 2.8%.
4
The increase is primarily seen in
violent offences
• Since the year to March 2002 CSEW, estimates of
violence have increased between 6.4% and 31.6%
than previously published. This is due to repeat
incidents being more common in violent offences.
• For most crime types, the estimated number of
incidents is unaffected.
5
The number of victims of crime is
almost identical
• The new cap doesn’t impact the number of victims.
• However, the small change to the survey weights had
a marginal effect on all crime survey estimates.
• For example, for the year to March 2018 CSEW, the
estimated number of victims of violent crime increased
by 0.4%.
6
Published data
• Estimates using this new methodology are published for
the first time in the Crime in England and Wales: year
ending September 2018 release on 24 January 2019.
• Users should not use releases published before January
2019 for estimates on the number of incidents from the
crime survey.
7
8
Why we’ve changed our
method
We interview one adult from
each household selected for
the Crime Survey for England
and Wales (CSEW) sample.
We ask them whether they
have been a victim of any
crime in the last 12 months.
9
Measuring incidents of crime
Sometimes a respondent reports being a victim of
multiple crimes that are each isolated incidents differing
in nature. These incidents are counted separately.
10
Fraud
1 incident
+
Theft
1 incident
+
Violence
1 incident
+
Violence 2
1 incident
= 4
incidents
Series of incidents
Occasionally, we find people have been victims of the
same thing, done under the same circumstances, probably
by the same people. We call these a series of incidents.
11
Fraud
1 incident
+
Theft
1 incident
+
Violence
1 incident
+
Violence 2
3 incidents
= 6
incidents
To gain estimates of the actual amount of crime in
England and Wales, we multiply respondents’ answers by
around 1,300
We survey around 1 in
every 1,300 people
12
Total incidents
experienced by
respondent
1 + 1 + 1 + 3 = 6
This respondent
represents 1,300 people so
we multiply their answers
by around 1,300
Fraud: 1 incident
+
Theft: 1 incident
+
Violence A: 1 incident
+
Violence B: 3 incidents
Total incidents counted
1,300 x 6 = 7,800
Weight
Out of around 1,300
adults in the population
we interviewed this adult
in this household.
Incident count Weight x incidents
13
Occasionally people report being a victim of a large
number of repeat incidents. The number of such victims
included in the survey, varies from year to year. This can
lead to volatility in the data.
Volatility in the data
14
Total incidents
experienced by
respondent
1 + 1 + 1 + 95 = 98
Weight
Out of around 1,300
adults in the population
we interviewed this adult
in this household.
This respondent
represents 1,300 people so
we multiply their answers
by around 1,300
Violence B - 97 incidents
Total incidents counted
1,300 x 98 =
127,400
Fraud: 1 incident
+
Theft: 1 incident
+
Violence A: 1 incident
+
Violence B: 95 incidents
Incident count Weight x incidents
15
The number of incidents can change
dramatically between years
This makes it difficult to
determine trends
16
When the survey was set up in 1981, it was decided to
cap the number of incidents that can be counted within a
series. This was set at a maximum of 5.
Removing this volatility
17
Total incidents
experienced by
respondent
1 + 1 + 1 + 95 = 98
Weight
Out of around 1,300
adults in the population
we interviewed this adult
in this household.
This respondent
represents 1,300 people so
we multiply their answers
by around 1,300
Violence B - 97 incidents
Total incidents counted
1,300 x 98 =
127,400
Fraud: 1 incident
+
Theft: 1 incident
+
Violence A: 1 incident
+
Violence B: 95 incidents
Incident count Weight x incidents
18
Total incidents
experienced by
respondent
1 + 1 + 1 + 5 = 8
Weight
Out of around 1,300
adults in the population
we interviewed this adult
in this household.
This respondent
represents 1,300 people so
we multiply their answers
by around 1,300
Violence B - 97 incidents
Total incidents counted
1,300 x 8 =
10,400
Fraud: 1 incident
+
Theft: 1 incident
+
Violence A: 1 incident
+
Violence B: 5 offences
Incident count with cap Weight x incidents
19
Capping the number of incidents
reduces volatility
This makes it possible
to discern trends over
time
20
However…
• We now know that the arbitrary cap of 5 disregards a
lot of incidents, especially acts of violence.
• Victims of violence are more likely to be repeat victims.
21
22
0
20
40
60
80
100
Violence Theft Offences Criminal damage Robbery
Percentageofincidents
INCLUDED EXCLUDED
40% of violent
incidents were
excluded in this
CSEW survey
year compared
to only 9% of
theft offences.
Proportion of CSEW incidents excluded by crime type,
example taken from the year ending March 2017
23
between handling volatility in the estimates and not
disregarding a lot of incidents
A balance was needed
To find this balance we…
• commissioned an independent review
• ran a public consultation
• sought advice from the National Statistician’s Crime
Statistics Advisory Committee
• published our decisions in response to the consultation
in November 2016
24
We decided to…
• remove the arbitrary limit of 5
• move to the 98th percentile value where this value is
greater than 5
• adjust the weights used on the survey to better suit the
inclusion of these higher counts
• publish the uncapped data with caveats to their use as
they are subject to considerable volatility
25
Calculating the 98th percentile
• Order the victims (across 3 years) by the number of
repeat incidents they have experienced.
• Find the victim who is at the 98th percentile, that is,
below which 98% of victims are found.
• Use the number of repeat incidents experienced by
that victim.
26
27
We now include a much higher volume of violent incidents
reported by respondents
What does this mean in
practice?
28
0 10 20 30 40 50
Jan '95 to
Dec '95
Apr '07 to
Mar '08
Apr '16 to
Mar '17
Apr '17 to
Mar '18
Percentage of incidents
Proportion excluded
with 98th percentile
Proportion excluded
with Cap of 5
Proportion of violent incidents excluded by methodology for a
selection of survey years
The overall picture hasn’t changed, the number of
incidents has increased across the whole series
What’s the impact on
total crime?
29
30
0
5,000
10,000
15,000
20,000
Jan '81
to Dec
'81
Jan '87
to Dec
'87
Jan '93
to Dec
'93
Jan '97
to Dec
'97
Apr '01
to Mar
'02
Apr '03
to Mar
'04
Apr '05
to Mar
'06
Apr '07
to Mar
'08
Apr '09
to Mar
'10
Apr '11
to Mar
'12
Apr '13
to Mar
'14
Apr '15
to Mar
'16
Apr '17
to Mar
'18
Numberofincidents(thousands)
Capped at 5
98th percentile
Total CSEW crime excluding fraud and computer misuse by old
and new methodology, year ending December 1981 to year
ending March 2018
The overall picture is broadly similar, the number of
incidents has increased across the whole series
What is the impact on
violent crime?
31
32
0
500
1,000
1,500
2,000
2,500
3,000
3,500
4,000
4,500
5,000
Jan '81
to Dec
'81
Jan '87
to Dec
'87
Jan '93
to Dec
'93
Jan '97
to Dec
'97
Apr '01
to Mar
'02
Apr '03
to Mar
'04
Apr '05
to Mar
'06
Apr '07
to Mar
'08
Apr '09
to Mar
'10
Apr '11
to Mar
'12
Apr '13
to Mar
'14
Apr '15
to Mar
'16
Apr '17
to Mar
'18
Numberofincidents(thousands)
Capped at 5
98th percentile
This rise in violence that we see here
is a genuine reflection of sustained
higher volumes of repeat incidents
Being reported by respondents
around this time-period.
Across financial years, new violence estimates will be between
6% to 32% higher than previously published estimates
What’s next
• We plan to publish microdata containing:
• new crime category variables based on the 98th percentile
• new crime category variables with the removal of the cap
altogether for specialist users to conduct their own
analyses
• We are also planning a new strand of research on the
extent and characteristics of repeat victimisation.
33
In summary: how have we changed the way we treat CSEW
data back to 1981?
For some crimes
where repeat
victimisation
appears more
frequently in the
untreated data, the
cap of 5 was too
stringent.
We have decided
to use the 98th
percentile count
for each of these
crime types
instead.
No change
Maximum number of
incidents in a series
remains at 5 where
98th percentile is
below this level.
Otherwise maximum
number of incidents
reflects the 98th
percentile count of
number of incidents
for that crime type.
• Burglary
• Other Household Theft
• Personal Theft Offences
• Vehicle Crime
• Bicycle Theft
• Robbery
• Fraud
• Computer Misuse
Violence
New maximum values
from 1981 to 2018
range between 8 - 20.
Criminal damage
For some years between
1981 and 2018 a maximum
value of 9 will be applied.
Weights
The inclusion of
more count data
from violence
has meant that
we have had to
pay more
attention to the
impact of
extreme
weights and
alter they way in
which we treat
these cases.
34

Improving victimisation estimates from the Crime Survey for England and Wales

  • 1.
    1 Improving victimisation estimates fromthe Crime Survey for England and Wales (CSEW) Joe Traynor 24 January 2019 CSEW survey manager Centre for Crime and Justice
  • 2.
    We’ve changed howrepeat victimisation is estimated • This includes a small change to the survey weights. • Repeat victimisation is defined as the same thing, done under the same circumstances, probably by the same people, against the same victim. 2
  • 3.
    We’ve replaced thecap of 5 from the old methodology • This was a limit on the number of repeat incidents included in the estimates. • We’ve replaced this limit with the 98th percentile value. • We’ve adjusted the weights used on the survey to better suit the inclusion of these higher counts. 3
  • 4.
    There has beenno impact on the long-term picture of total crime • However, the number of incidents for total CSEW crime are slightly higher across the entire time series than previously published. • Since the year to March 2002, the average increase in total CSEW crime (excluding fraud and computer misuse) was 2.8%. 4
  • 5.
    The increase isprimarily seen in violent offences • Since the year to March 2002 CSEW, estimates of violence have increased between 6.4% and 31.6% than previously published. This is due to repeat incidents being more common in violent offences. • For most crime types, the estimated number of incidents is unaffected. 5
  • 6.
    The number ofvictims of crime is almost identical • The new cap doesn’t impact the number of victims. • However, the small change to the survey weights had a marginal effect on all crime survey estimates. • For example, for the year to March 2018 CSEW, the estimated number of victims of violent crime increased by 0.4%. 6
  • 7.
    Published data • Estimatesusing this new methodology are published for the first time in the Crime in England and Wales: year ending September 2018 release on 24 January 2019. • Users should not use releases published before January 2019 for estimates on the number of incidents from the crime survey. 7
  • 8.
  • 9.
    We interview oneadult from each household selected for the Crime Survey for England and Wales (CSEW) sample. We ask them whether they have been a victim of any crime in the last 12 months. 9
  • 10.
    Measuring incidents ofcrime Sometimes a respondent reports being a victim of multiple crimes that are each isolated incidents differing in nature. These incidents are counted separately. 10 Fraud 1 incident + Theft 1 incident + Violence 1 incident + Violence 2 1 incident = 4 incidents
  • 11.
    Series of incidents Occasionally,we find people have been victims of the same thing, done under the same circumstances, probably by the same people. We call these a series of incidents. 11 Fraud 1 incident + Theft 1 incident + Violence 1 incident + Violence 2 3 incidents = 6 incidents
  • 12.
    To gain estimatesof the actual amount of crime in England and Wales, we multiply respondents’ answers by around 1,300 We survey around 1 in every 1,300 people 12
  • 13.
    Total incidents experienced by respondent 1+ 1 + 1 + 3 = 6 This respondent represents 1,300 people so we multiply their answers by around 1,300 Fraud: 1 incident + Theft: 1 incident + Violence A: 1 incident + Violence B: 3 incidents Total incidents counted 1,300 x 6 = 7,800 Weight Out of around 1,300 adults in the population we interviewed this adult in this household. Incident count Weight x incidents 13
  • 14.
    Occasionally people reportbeing a victim of a large number of repeat incidents. The number of such victims included in the survey, varies from year to year. This can lead to volatility in the data. Volatility in the data 14
  • 15.
    Total incidents experienced by respondent 1+ 1 + 1 + 95 = 98 Weight Out of around 1,300 adults in the population we interviewed this adult in this household. This respondent represents 1,300 people so we multiply their answers by around 1,300 Violence B - 97 incidents Total incidents counted 1,300 x 98 = 127,400 Fraud: 1 incident + Theft: 1 incident + Violence A: 1 incident + Violence B: 95 incidents Incident count Weight x incidents 15
  • 16.
    The number ofincidents can change dramatically between years This makes it difficult to determine trends 16
  • 17.
    When the surveywas set up in 1981, it was decided to cap the number of incidents that can be counted within a series. This was set at a maximum of 5. Removing this volatility 17
  • 18.
    Total incidents experienced by respondent 1+ 1 + 1 + 95 = 98 Weight Out of around 1,300 adults in the population we interviewed this adult in this household. This respondent represents 1,300 people so we multiply their answers by around 1,300 Violence B - 97 incidents Total incidents counted 1,300 x 98 = 127,400 Fraud: 1 incident + Theft: 1 incident + Violence A: 1 incident + Violence B: 95 incidents Incident count Weight x incidents 18
  • 19.
    Total incidents experienced by respondent 1+ 1 + 1 + 5 = 8 Weight Out of around 1,300 adults in the population we interviewed this adult in this household. This respondent represents 1,300 people so we multiply their answers by around 1,300 Violence B - 97 incidents Total incidents counted 1,300 x 8 = 10,400 Fraud: 1 incident + Theft: 1 incident + Violence A: 1 incident + Violence B: 5 offences Incident count with cap Weight x incidents 19
  • 20.
    Capping the numberof incidents reduces volatility This makes it possible to discern trends over time 20
  • 21.
    However… • We nowknow that the arbitrary cap of 5 disregards a lot of incidents, especially acts of violence. • Victims of violence are more likely to be repeat victims. 21
  • 22.
    22 0 20 40 60 80 100 Violence Theft OffencesCriminal damage Robbery Percentageofincidents INCLUDED EXCLUDED 40% of violent incidents were excluded in this CSEW survey year compared to only 9% of theft offences. Proportion of CSEW incidents excluded by crime type, example taken from the year ending March 2017
  • 23.
    23 between handling volatilityin the estimates and not disregarding a lot of incidents A balance was needed
  • 24.
    To find thisbalance we… • commissioned an independent review • ran a public consultation • sought advice from the National Statistician’s Crime Statistics Advisory Committee • published our decisions in response to the consultation in November 2016 24
  • 25.
    We decided to… •remove the arbitrary limit of 5 • move to the 98th percentile value where this value is greater than 5 • adjust the weights used on the survey to better suit the inclusion of these higher counts • publish the uncapped data with caveats to their use as they are subject to considerable volatility 25
  • 26.
    Calculating the 98thpercentile • Order the victims (across 3 years) by the number of repeat incidents they have experienced. • Find the victim who is at the 98th percentile, that is, below which 98% of victims are found. • Use the number of repeat incidents experienced by that victim. 26
  • 27.
    27 We now includea much higher volume of violent incidents reported by respondents What does this mean in practice?
  • 28.
    28 0 10 2030 40 50 Jan '95 to Dec '95 Apr '07 to Mar '08 Apr '16 to Mar '17 Apr '17 to Mar '18 Percentage of incidents Proportion excluded with 98th percentile Proportion excluded with Cap of 5 Proportion of violent incidents excluded by methodology for a selection of survey years
  • 29.
    The overall picturehasn’t changed, the number of incidents has increased across the whole series What’s the impact on total crime? 29
  • 30.
    30 0 5,000 10,000 15,000 20,000 Jan '81 to Dec '81 Jan'87 to Dec '87 Jan '93 to Dec '93 Jan '97 to Dec '97 Apr '01 to Mar '02 Apr '03 to Mar '04 Apr '05 to Mar '06 Apr '07 to Mar '08 Apr '09 to Mar '10 Apr '11 to Mar '12 Apr '13 to Mar '14 Apr '15 to Mar '16 Apr '17 to Mar '18 Numberofincidents(thousands) Capped at 5 98th percentile Total CSEW crime excluding fraud and computer misuse by old and new methodology, year ending December 1981 to year ending March 2018
  • 31.
    The overall pictureis broadly similar, the number of incidents has increased across the whole series What is the impact on violent crime? 31
  • 32.
    32 0 500 1,000 1,500 2,000 2,500 3,000 3,500 4,000 4,500 5,000 Jan '81 to Dec '81 Jan'87 to Dec '87 Jan '93 to Dec '93 Jan '97 to Dec '97 Apr '01 to Mar '02 Apr '03 to Mar '04 Apr '05 to Mar '06 Apr '07 to Mar '08 Apr '09 to Mar '10 Apr '11 to Mar '12 Apr '13 to Mar '14 Apr '15 to Mar '16 Apr '17 to Mar '18 Numberofincidents(thousands) Capped at 5 98th percentile This rise in violence that we see here is a genuine reflection of sustained higher volumes of repeat incidents Being reported by respondents around this time-period. Across financial years, new violence estimates will be between 6% to 32% higher than previously published estimates
  • 33.
    What’s next • Weplan to publish microdata containing: • new crime category variables based on the 98th percentile • new crime category variables with the removal of the cap altogether for specialist users to conduct their own analyses • We are also planning a new strand of research on the extent and characteristics of repeat victimisation. 33
  • 34.
    In summary: howhave we changed the way we treat CSEW data back to 1981? For some crimes where repeat victimisation appears more frequently in the untreated data, the cap of 5 was too stringent. We have decided to use the 98th percentile count for each of these crime types instead. No change Maximum number of incidents in a series remains at 5 where 98th percentile is below this level. Otherwise maximum number of incidents reflects the 98th percentile count of number of incidents for that crime type. • Burglary • Other Household Theft • Personal Theft Offences • Vehicle Crime • Bicycle Theft • Robbery • Fraud • Computer Misuse Violence New maximum values from 1981 to 2018 range between 8 - 20. Criminal damage For some years between 1981 and 2018 a maximum value of 9 will be applied. Weights The inclusion of more count data from violence has meant that we have had to pay more attention to the impact of extreme weights and alter they way in which we treat these cases. 34