Methodological and analytical issues in using the British Crime Survey to model the impact of respondent Ethnicity: Difficulties and solutions arising.

Methodological and analytical issues in using the British Crime Survey to model the impact of respondent Ethnicity:

Difficulties and solutions arising from an ESRC-UPTAP funded research project

Dr. Paula M. Kautt

BACKGROUND

Ethnicity and the BCS

The British Crime Survey• The British Crime Survey (BCS) is a face-to-face

victimisation survey utilising a nationally representative sample of adult, private household residents

• Since 1982, it has been used to examine a broad range of criminological questions– Yet, few analyses use it to examine variation in criminological

experiences between black and minority ethnic groups (BME) or the factors influencing those experiences

– Several methodological issues are associated with doing so• Design changes in 2001 means that data from previous

years are not comparable– Today’s discussion focuses on post-2001 BCS issues

BCS Sampling Strategy

• BCS data– Are collected year-round– Use a 12 month reference

point– Purposefully include roughly

1,000 cases per PFA (Police Force Area)

– Are sampled via Postcode Address File (PAF)

– Stratified by various factors– Include a non-White boost

sample

BCS Structure: Follow up Modules

• BCS Questionnaire– Core battery of questions that all participants answer– Four distinct follow-up modules

• Attitudes to the Police• Attitudes to the CJS• Crime Prevention • Ad hoc

– These are also sometimes divided into sub-modules• Each respondent answers questions from only one follow-up module

(and sub-module)– Which follow-up module is pre-assigned, preserving the sampling frame

DEFINING ETHNICITY

Ethnicity and the BCS

Ethnicity and Ways of Capturing It

• Ethnic/racial categories– Skin Colour– ‘Other’

• Religion– Subgroups?– Active?

• National/Geographic origin or Birth country– Can move from country to

country• Language

– Native/first language– Main language spoken at home– Fluency in dominant language

• Citizenship – Naturalised or Native?

• Immigration Status– Country

• EU, former colony• Regions

– Legal or illegal?• Length of time in country• Appearance

– Immutable characteristics– Mutable characteristics

• Cultural affiliation– Self-declaration

• Name• Ancestry/Family origin

– All of the above for ancestors

ANCESTRY

PhysicalAppearance

ETHNICIDENTITY

ReflectedAppearance

Group History

Components of Self-Identified Ethnic Identity (as modified from Hirschman & Perez, 2007)

Culture&

Language

Community Attachment &Participation

FamilySocialization

CollectiveMemory

Citizenship & Immigration Status

Religion & its Importance

Components of Ascribed Ethnic Identity

Ascribed Ethnicity

Skin colour and Physical features

Name

Dress

Accent

Religious Practice

Proficiency in English

Social Customs, Practices

The Sixteen BCS Categories of Ethnicity

• White - British • White - Irish • White - Other White

Background • Mixed - White and Black

Caribbean • Mixed - White and Black African • Mixed - White and Asian • Mixed - Any Other Mixed

Background • Asian or Asian British - Indian • Asian or Asian British -

Pakistani

• Asian or Asian British - Bangladeshi

• Asian or Asian British - Other Asian Background

• Black or Black British - Caribbean

• Black or Black British - African

• Black or Black British - Other Black Background

• Chinese • ‘Other’ Ethnic Group

BCS-Captured EthnicityBenefits• Same 16 categories as

the UK Census – Permits consistency

between these data

• Respondent classifies self by categories

• Improvement over capturing only national origin

• Latest version captures year of arrival in country

Liabilities• Triple-barrelled question

– Traditional ‘race’• Asian, Black, White

– National origin• Pakistani, Indian, etc

– Regional origin• African, Caribbean

• Inconsistent and oversimplified• Non-representative

– ‘Nationality’ and ‘Country of birth’ only covers the British Isles

• Unwieldy– 15 Dummy variables for analysis

Changing Ethnic Definitions• Recommended improvements to the ethnicity

measure are slated for the 2011 UK Census– Two ethnic categories to be added

• Arab (under ‘Other’)• ‘Travelers’ and ‘gypsies’ (under White)• Plus a ‘write in’ option for ‘Other’

• In addition, it also captures– Language

• Respondent’s main language• How well respondent speaks English

• What does this mean for the BCS?

ANCESTRY

PhysicalAppearance

ETHNICIDENTITY

ReflectedAppearance

Group History

What’s Missing: Self-Selected Ethnic Identity

Culture&

Language

Community Attachment &Participation

FamilySocialization

CollectiveMemory

Citizenship & Immigration Status

Religion & its Importance

What’s missing?: Ascribed Ethnic Identity

Ascribed Ethnicity

Skin colour and Physical features

Name

Dress

Accent

Religious Practice

Proficiency in English

Social Customs & Practices

Improving BCS Ethnicity Measures• Census ethnicity measure must be incorporated into the

BCS to ensure comparability– This does not mean this need be the only ethnicity measure

included

• Options:– Include separate indicators for each pertinent aspect

• Language as with the Census• Consistently capture comprehensive measures for Nationality and

Country of Birth• Present the ‘write-in’ results for the ‘other’ ethnicities in the data set

– Use existing Multi-faceted Ethnicity Classification Instruments• e.g. Multi-Ethnic Identity Measure (MEIM)

– Combining methods– Do nothing; continue with current classification

ANALYTICAL SOLUTIONS

Ethnicity and the BCS

Anomalies of BCS data • As captured, only ‘Whites’ can be

simply ‘British’ – ‘White-British’

• It is not possible for ‘Black’ or ‘Asian’ respondents to simply be ‘British’ – Rather, specific ethnicity is always

attached. For example• ‘Black or Black-British–

Caribbean’• ‘Asian or Asian-British—Indian’

• The Data do not capture different ‘White’ ethnicities beyond ‘British’, ‘Irish’ and ‘Other’– ‘English,’ ‘Scottish’ and ‘Welsh’

are the ‘Other’ options– Thus, other ‘White’ ethnicities,

such as ‘Polish’, are not captured

Practical Analytical Solutions

• The BCS ethnicity categorisation scheme permits partitioning at two tiers– General Ethnicity

• A larger group differentiated mainly by traditional ‘race’ groupings

– White, Black, Asian– ‘Other’

– Specific Ethnicity• A more refined comparison of

groups falling within a general ethnicity

– Under ‘Asian’, the sub-groups ‘Indian’, ‘Pakistani’, ‘Bangladeshi’ can be compared

– Partially addresses consistency issue

• This split is precise but potentially non-representative of reality

What about ‘Other’?• The general ‘Other’ category

can be treated as another General Ethnicity

• The ‘Black—Other’ and ‘Asian—Other’ can also be treated as other Specific Ethnicity categories

• HOWEVER: currently no meaningful way to talk about these groups because they – Comprise an unknown

number of different ethnicities– The proportion of these

ethnicities is also unknown

And ‘Mixed’?• Because of relative rarity

‘Mixed’ ethnicity cases are difficult to analyse

• Analytically they can be – Merged into the

specifically-named BME group for analysis

• The supposition being that they will appear to be a member of that group to a third party

– Pooled by year (see below) for independent analysis

• Limited techniques available due to low counts

Comparison Problems• Sample Size for Ethnicity

– Majority of respondents are ‘White British’

• Much smaller numbers of the Asian and Black General Ethnicities

• Makes meaningful statistical comparisons between ‘White’ and other General Ethnicities problematic

– There are even fewer of the specific ethnicities

• These are not evenly distributed

– e.g. There are many more Indians than Chinese

– Especially problematic for the ‘follow-on’ modules and sub-modules

Solution?: Pooling the Data• Because BMEs comprise only a small proportion of the

BCS sample (approximately 10% total across all BME groups), can be combined across years to: – Meaningfully analyse responses by specific ethnic group – Provide sufficient numbers to support ethnically separate

analyses• Despite this, for some categories, there are still too few cases

even when all data since 2001 is used– In such instances, an ethnically pooled model is the only option

• NOTE: when pooling by year, only indicators available for all years can be used– e.g. religion and sexual orientation cannot be controlled in

analyses employing data from years prior to their inclusion

Incorporating Location: Criminal Justice Areas • England and Wales are

currently divided into 43 Criminal Justice Areas– These specify jurisdictional

and administrative boundaries for CJS agencies

• The Police Service• The Courts• The Prison Service• The Probation Service

• This is the smallest geographic subunit universally* available in the BCS– Distribution of BMEs varies

substantially across them• Important for multilevel analyses

Representation: PFA & Ethnicity

To understand the complexity, multiply by ten…

PFA Example One: Respondent Ethnicity Counts London (2001-7)

London

Asian: 5039

Indian: 2335

Pakistani: 737

Black: 5628

Caribbean:

2700

African:2736

PFA

General

Specific

PFA Example Two: Respondent Ethnicity Counts Cumbria (2001-7)

Cumbria

Asian: 32

Indian:6

Pakistani: 2

Black: 13

Caribbean: 4

African: 7

PFA

General

Specific

PFA Sample Size Solution?: Analysis by Ethnic Centres• Low counts in minority

ethnicities make the BCS unfit for examining ethnic differences nationwide

• London and the West Midlands have the highest concentrations– Focus analyses on these areas,

which will :• Maximise the number of ethnic

minority respondents• Better represent actual diversity

levels and composition• Better capture how crime-related

outcomes vary by ethnicity– Use Local Authority rather than PFA as

the smallest geographic subunit (special license only)

Conclusions• Using the BCS to analyse the impact of minority

ethnicity status over criminological outcomes is difficult but not impossible– Currently, definitional and methodological compromises

must be made• Improvements to the BCS could make these compromises

unnecessary

– Limited utility for comparisons by context and geographic location• Wider release of the ‘special license’ data would ease this