Transcript Conference presentations - Institute for Social and Economic

UKHLS Consultation Conference
Thursday 10th January 2008
University of Essex
The UKHLS team
http://www.iser.essex.ac.uk/ukhls/
Introduction
John Hobcraft
Introduction
• Welcome to the conference
– Rationale and timing
• The UKHLS and social science research
• The structure of the day
– Morning: information and wave 1 update
– Afternoon: looking ahead and getting your views
Overview and update
Nick Buck
Background
• UKHLS is a longitudinal study based on a
household panel design
• Basic design similar to that of British
Household Panel Survey which it will replace
• Target sample size of 40,000 households
• Ethnicity strand – boost sample of 5 groups
plus questions focussed on ethnicity related
issues
• Biomedical strand
• Innovative data collection, data linkage etc.
Where we are now
• Spring 2007: PI team starts work, consultation launched
• June/July: initial meetings of topic groups and ethnicity
strand consultation
• September: Topic groups reported
• September onwards: development of Innovation Panel
questionnaire (NB different from wave 1 questionnaire)
• October: first meeting of Scientific Advisory
Committee, proposals for topic content circulated
• December: measures for wave 1 identified
• January 2008: design of pilot begins; wave 1 innovation
panel starts
Active consultation process: so far more than 30 meetings,
more than 200 written comments.
Key questionnaire constraints
• The following is now expected:
– 12 month intervals between interviews
– Continuous fieldwork over 24 month field period, with
second wave overlapping with first
– Face-to-face interview at wave 1; mixed mode at wave 2,
20%+ face to face only
– Individual interview not more than 30 minutes face to face
interview administered, plus self completion and consents to
link data
– Household roster, plus 10 minutes household questionnaire
– Some data collection by self completion from children aged
10-15 from wave 1
• Questionnaire time for first two waves is short
Potential areas of coverage: Topic
consultation groups
1.
Standard of living measures (income, consumption, material
deprivation, expenditure, financial well-being)
2. Family, social networks and interactions, local contexts, social
support, technology and social contacts
3. Attitudes and behaviours related to environmental issues (energy,
transport, air quality, global warming etc.)
4. Illicit and risky behaviour (crime, drug use, anti-social behaviour etc).
5. Lifestyle, social, political, religious and other participation, identity
and related practices, dimensions of life satisfaction/happiness
6. Psychological attributes, cognitive abilities and behaviour
7. Preferences, beliefs, attitudes and expectations
8. Health outcomes and health related behaviour
9. Education, human capital and work
10. Initial conditions, life history
Key measures for scientific research
Also useful to classify measures in terms of their
place in longitudinal models which researchers
develop
–
–
–
–
–
–
Outcomes
Preferences
Personal endowments and constraints
The wider social and spatial environment
Behaviours
Other variables (e.g. instrumental variables)
Outcomes
• Measures such as money income and consumption
expenditure important for summarising growth in
socio-economic well-being and changes in inequality
and poverty.
• Need to be complemented by other measures, e.g.:
–
–
–
–
‘subjective’ measures of domain satisfaction and happiness,
non-financial measures of deprivation and hardship,
health (mental and physical), and
educational attainment in most general sense.
• These are important contributions to individuals’
‘functionings’ (Sen)
Behaviours
• Panel research has been centrally concerned with the
analysis of behaviour over time, and there is a strong
case for extending the focus. Key areas include:
– Work – market and non-market (including caring) – and pay
– Health and lifestyle related behaviours e.g. smoking, exercise
and diet, medications, pregnancy planning
– Consumptions more generally, including their social and
environmental impacts
– Geographic mobility and (im)migration
– Social, cultural, and political participation
– Criminal, illicit, and anti-social behaviours
– ICT usage, other media usage
Preferences
• Outcomes reflect the interplay of preferences,
opportunities and constraints
• ‘Preferences’ include not only measures of intentions
and stated preferences, also:
– attitudes to risk and uncertainty,
– perceptions, knowledge and awareness
• In a longitudinal context, focus on ex ante intentions,
expectations, plans and aspirations, to see how they
shape future behaviour, and to compare them ex post
with outcomes.
• also a range of underlying psychological and personality
predispositions (e.g. sense of control).
• social identities (e.g. related to ethnicity, religion,
nationality, class, sex, age), and the behavioural norms
associated with these identities.
Personal endowments and constraints
• A wide range of measures summarising the ability of
individuals to realise desired outcomes.
• Person-level measures include physical and mental
health including resilience, cognitive functioning,
genetic endowments and biomarkers.
• The heading also refers to measures of a person’s
human, social and cultural capitals, and of their social
class and family socio-economic background.
• also measures of the ability to even out resources over
time as needs fluctuate, e.g. measures of access to
credit, help from friends, etc.
• Other measures of constraints on participation and
functioning in contemporary society, e.g. access to
transport, or particular forms of media and ICT.
The wider social and spatial
environment
• What individuals can do also depends on the
environment beyond the household in which they live.
• Data about ‘significant others’ outside the household,
and interactions with them will be an important focus.
• Life chances may depend on resources from social
networks outside the household; people maintain links
with former household members after they have left.
• The characteristics of the local neighbourhood are
arguably of substantial importance in shaping
individuals’ lives, including:
– quality of facilities (including housing, schooling, social
services),
– other environmental differences, ranging from air quality (for
health) to prices of goods (for consumption).
Example model structure using different
measure types
Contexts, networks
and enviroments
Preferences,
values, identities
Outcomes
Endowments &
constraints
Behaviours
Example model structure using different
measure types
Contexts, networks
and enviroments
Preferences,
values, identities
Behaviours
Endowments &
constraints
Behaviours
How do we fit everything we want into
UKHLS?
• Research opportunities are enormous / time
constraints are very severe
– Need to be selective in what we include
– Need to focus frequency of inclusion – measures
may have to be collected intermittently
– Maybe do not ask everyone all questions?
– Note some data not collected by questionnaire, e.g.
data linkage: can this substitute for questionnaire
space?
• 10 principles for selecting measures in early
waves…
Principles for selecting measures in UKHLS
1. Longitudinal survey: prioritise measures best used
longitudinally, rather than just at a single point in time,
or repeated cross-section.
2. Household survey: prioritise measures that benefit
from understanding of the household context and
measures from other family members.
3. Do not just duplicate other surveys. Prioritise new
measures not covered elsewhere or where UKHLS
design leads to benefits from replication.
4. Prioritise topic areas that address important and
emerging long term scientific research agendas.
Principles (2)
5.
6.
7.
Have patience! UKHLS represents every age cohort,
and 1st wave is not a baseline survey. Loss from
delaying introduction of measures is not a failure to
collect data at a particular age for the whole sample.
Successful establishment for the long term with low
attrition is priority now. Minimise respondent burden
and avoid measures which may damage response.
Derives from success of BHPS, and benefits from
incorporating BHPS sample. But not a replication, so
BHPS questions not carried unless they address a
topic of continuing importance and no superior
alternative.
Principles (3)
8.
Multi-purpose survey providing a balance of
coverage meeting wide range of needs; must not
focus large share of questionnaire on a few
measures.
9. Resource for UK social science: prioritises social
science research agendas, including policy
applications and agendas crossing traditional
disciplinary boundaries e.g. related aspects of
biomedical research.
10. Priority for topics which most benefit from coexistence on the same survey as other included
topics. In particularly it is important to ensure that
the design maximises the possibilities for crossdisciplinary research.
Overview of sampling and
other design issues
Peter Lynn
Sample components and sizes
Component
Interviewed
Hhds (w1)
1,560
Interviewed
Hhds (w2)
1,300
BHPS Sample
8,350
8,120
New General
Population Sample
Ethnic Minority
Boost Sample
26,780
22,200
4,210
3,300
Total UKHLS
40,900
34,920
Innovation Panel
Structure of Ethnic Minority Boost Sample
Interviewed adults
(w1)
1,240
Interviewed hhds
(w1)
600
Pakistani
1,000
440
Bangladeshi
1,000
430
Caribbean
1,370
1,030
African
1,200
820
Chinese
360
220
Other
1,070
700
Total
7,240
4,240
Indian
Fieldwork Schedule
2008
O
W1
W2
2009
J
A
J
O
2010
J
A
J
O
2011
J
A
J
Incorporating the BHPS Sample
• All BHPS sample households will be included in
UKHLS
• Including Scottish & Welsh boosts and NIHPS
• To be administered standard UKHLS
instruments, starting at wave 2
• Temporal allocation not yet finalised
Sample Design
• New sample will be clustered in a sample of
postcode sectors
• Equal probability sample of addresses in UK
• All persons resident at those addresses are
sample members
• Subsequent to wave 1 interview:
– all sample members are followed
– all children born to female sample members become sample
members
– other members of households of sample members will be
interviewed
• Household associates may also be interviewed
Data Collection Modes
• Wave 1
– face to face interviewing
– Self-completion for 10-15 year olds
– Telephone as last resort for refusal conversion
• Wave 2
– mixed modes:
– telephone where possible;
– face to face elsewhere
• Mixed mode approaches being tested on
Innovation Panel
• Web under consideration for future
Wave 1 topic content
overview
Nick Buck
Components of the wave 1
questionnaire
• Annual repeating measures
• Initial conditions and life history, once only
• Rotating and intermittent measures first
introduced at wave 1
• Young persons questionnaire for sample
members aged 10-15
The Topic Content paper presents settled plans,
but some uncertainties remain (question timings
and detailed question development)
Estimated timings for questionnaire
components
Includes individual questionnaire and selfcompletion
Average for
Wave 1
future waves
Annual repeating
17.8
17.8
questions
Initial conditions
12.4
0.0
Rotating questions
9.7
22.1
See table on page 7 of Topic Content paper for estimated
distribution of the timings by measure type and subject area
How proposals are presented
• Appendix B of Topic Content paper lists
potential measures, grouped by theme and in
the same order as in the Initial proposals paper
(25 October)
• For each measure the table indicates:
1.
2.
3.
4.
Whether proposed for inclusion at wave 1
The proposed frequency of inclusion
When it is likely to be first asked
How we classify it as a measure type
Last three of these are still very provisional
Annual repeating measures
introduced at wave 1
Stephen Jenkins
Principles: frequency of data collection
Optimal frequency for any particular measure
depends on:
(i) the frequency of significant changes in that measure,
and also in associated events that might explain them;
and
(ii) the quality of the information about the measure
collected from of any specific survey instrument in
relation to its cost
• Possibilities include: sub-annual, annual,
biannual, less frequent
Principles (ctd.)
Annual data collection appropriate when:
1.
The dynamics of change per se (e.g. duration in states, factors
explaining transitions from one state to another) are
themselves interesting
and
2. the phenomena themselves are subject to substantive change
from year to year at the individual level, at least for significant
fractions of the population
Annual data collection is less appropriate where the
interest is in long term impacts of earlier conditions,
or where the time to impact is not of the highest
priority
From principles to practice …
• Multi-measure surveys like household panel
studies use a mixture of data collection intervals
• Mixture represents a compromise between:
– optimal collection needs for specific measures, and
– reductions in cost derived by clumping together collection
in interviews (‘waves’) of regular periodicity
• Much analysis uses circumstances at the time of
the interview to derive measures of change or
frequency.
• Survey instruments also include retrospective
histories covering the period between interview
for relatively high frequency measures
From principles to practice …(ctd.)
• Existing research from around the world, including
Britain (BHPS, LFS, various administrative data),
suggests that higher frequency transitions and change
refer to topics such as:
–
–
–
–
labour market participation, hours and earnings
receipt of various kinds of social security benefits
household consumption and income more generally
the onset of disability, and
– other topics for particular groups, e.g.:
• developmental progress among children
• biological and associated changes during puberty
• health service use among elderly people
UKHLS Annual Repeating content
The measures proposed reflect
• existing research
which was reflected in, and supported by,
• contributions to the consultation
The proposed annual content is …
UKHLS Annual Repeating content
• Basic demographic
characteristics and changes,
fertility, partnering,
• Health status (e.g. SF12),
disability,
• Labour market activity and
employment status, job search
• Current job characteristics, basic
employment conditions, hours
of paid work, second jobs
• Childcare, other caring within
and outside household
• Income and earnings
• Life satisfaction
• Political affiliation – basic
measures
• Transport and communication
access
• Education aspirations and
expectations
• Consumption expenditure
• Housing characteristics – basic
• Housing expenditure
• Household facilities, car
ownership
UKHLS Annual Repeating content
• NB. Some annual repeating content will be
introduced at Wave 2
– particularly relevant where Wave 1 establishes
circumstances at the start of the panel, and this is
updated at later waves
• Main topic areas are:
– Activity history over previous year
– Training and skill acquisition, qualifications obtained
– Migration attitudes and behaviour
Initial conditions and life histories
Heather Laurie
Initial conditions and life histories
•
•
UKHLS will provide longitudinal data from the point
at which sample members are selected at wave 1
Need data about people’s earlier life to fully exploit
panel data in analysis
• Initial Conditions
– factual background measures e.g. place of birth, details of
parental background, qualifications
• Life History data
– record all changes in a particular domain over the whole
life-course to date e.g. cohabitation, marital and fertility
history; an employment history; migration history; and
many others
Outcome of consultation
• Best to carry these items at wave 1 if possible
• Items collected once in the life of the survey
• For respondents, is most natural place to collect this
type of data
• If collected at later waves will disrupt the rotating
sequence of other modules
• Provides some longitudinal data immediately for
analysis
• Allows more time for design and development of new
modules/questions for wave 2 and beyond
Critical areas
• Initial conditions:
– Place of birth, national origins, family/parental background,
education and qualifications
• Life Histories
– International migration history, partnership history, fertility
and childbirth history, employment status history, key
previous job
• Non-trivial time constraints for collection of these data
• Timings from wave 1 Innovation Panel for some areas
• May not be able to carry all areas at wave 1
The youth questionnaire
Heather Laurie
Youth Questionnaire
• 10 minute self-completion for 10 – 15 year olds
• Consultation on Wave 1 content still open
• Main design issues:
– Does age range imply two versions of the
questionnaire? E.g. for 10 – 12s and for 13 – 15s?
– Relationship between the content of the youth and
adult questionnaires i.e. comparable measures in each?
– Transition from youth to adult questionnaire to
maximise longitudinal analysis E.g. carry some youth
questions for 16 – 19 year olds in the adult questionnaire?
Areas of coverage
• Need predictors of later outcomes as well as measures
of current views or circumstances
• Establish which questions asked annually, which rotate
in and frequency
• Potentially wide range of areas:
–
–
–
–
–
–
–
Relationships with family and friends
Social networks and illicit/risky behaviour
Experience of education and aspirations
Use of leisure time, health, diet and obesity
Future aspirations for job, family, independence
Social and political attitudes and values
Experience of harassment due to race or religion
Rotating and Intermittent
measures
Lucinda Platt
Rationale
• Use of rotating modules and intermittent
measures for a substantial proportion of
questionnaire time increases the range and
number of questions that can be asked across
the survey (note also subsamples discussion).
• A more infrequent cycle may, anyway, be more
suitable for some measures.
• Some rotating modules will be included at wave
1, but far more in subsequent waves, allowing
time for question development and further (and
ongoing) consultation
Issues
• Trade-off between frequency and depth – more
detailed modules may be feasible less frequently,
more frequent measures may need to be sparing.
• Given less than annual occurrence, some
questions are more suitable for higher frequency
rotations (e.g. behaviours and outcomes), others
can sustain longer periods between (e.g.
relatively stable endowments and preferences).
Suggested frequency of broad topic
areas
• Biennial
• 3-5 yearly
Fuel consumption
Mental health and well-being
Tobacco, alcohol, drug use
Physical activity, fitness, nutrition
Financial/ material well-being
Pensions and savings behaviour
Commuting behaviour
Work aspirations, preferences and expectations
Domestic work
Voluntary work
Family networks outside household
Travel behaviour
ICT usage
Leisure participation
Attitudes and behaviour related to the environment
Housing wealth
Ethnicity and national identity
Fertility intentions
Chronic health conditions
Sleep
Obesity and body mass
Wealth, credit and debt
Employment conditions
Within household organisation
Social relationships within the family
Religion
Social and friendship networks
Social support
Political engagement
Social engagement, social capital
Local neighbourhood
Quality of life measures
Discrimination and racism
Cultural consumption
• 5-10 yearly
Psychological attributes / stable values or preferences
Cognitive ability
Issues (continued)
• Some modules are complementary and are best asked
in the same wave – co-ordination as well as frequency
then becomes an issue.
• Some questions might be most salient for particular
sub-populations – e.g. particular age groups; or may be
suitable for higher frequency at particular ages
• Some questions can be related to events that may occur
in people’s lives and can be asked just at those points.
• Some questions or topic areas still need development –
timing of introduction is in part dependent on that.
• ‘Extra five minutes’ in the ethnic minority boost could
allow additional topics or greater frequency for
especially salient ones (or a bit of both)
What you can tell us…
• What is the right frequency for particular modules and
measures? (in the earlier slide and Appendix B – have
we got it about right?) Why?
• What topics need to be asked at the same time? Why?
• What questions could be targeted at specific age
groups? Why?
• What event triggered questions should we be asking
(other than birth weight following a birth and questions
following a move)? Why?
• What are the best measures for some of these topics?
• At what wave should the measure be introduced? Why?
• Some areas clearly need question development – can
you contribute to that?
Random sub-samples
Nick Buck
Rationale for use of sub-samples
• For many, but not all, purposes 40,000 households is
larger than needed
• Could exploit this by creating random sub-samples
which contain both questions asked of everyone and
subsets of questions asked only in the sub-sample
• This increases the effective length of the questionnaire,
and potentially allows inclusion of more modules or
permits modules to be included with greater frequency
• Contrast this with focussed studies on small nonrandom sub-groups with particularly relevant
characteristics: potential for attrition bias?
Key issues for sub-samples
• Ensuring that right combinations of questions are on
same sub-sample – major design challenge
• Combine ‘light’ measurement of topics for full sample,
with greater depth or higher frequency for sub-sample?
• Issues for ethnic minority boost sample, and separate
analyses of e.g. Scotland, Wales and Northern Ireland
• Context effects?
• NB: we are not proposing to introduce subsampling at wave 1
Different approaches to sub-samples
1. Completely distinct question groups: requires groups
of questions where can assume that there will little
demand for analysis combining data from more than
one group. Implies thematically coherent groups.
2. Overlapping question groups in different sub-samples.
Ensures that every pair of questions is asked in
combination for a random sub-set of respondents.
The time gains are less, but it might cover a higher
proportion of questions.
3. Randomise the allocation of questions at the
individual respondent level. Statistical benefits but
complex for analysis.
Potential design for random sub-samples in
UKHLS (approach 2)
Sample 1
Sample 2
Sample 3
Sample 4
Sample 5
Sample 6
Questions for respondents in all samples, including annual measures
and rotating and intermittent measures
Questions for specific sub-samples:
Group A
Group A
Group A
Group B
Group B
Group C
Group B
Group C
Group D
Group C
Group D
Example question groups:
Group A: Environmental attitudes and related behaviour
Group B: Employment conditions, time-use, work-life balance
Group C: Health related, risky and illicit behaviours
Group D: Social and political engagement
Group D
Implications of using this approach
• 50% of whole sample eligible for every question
• Sub-sample eligible for each two-way combination
larger than BHPS original sample (> 5,000 households)
• Sub-samples defined longitudinally: questions from the
same group would be asked at each wave, mainly
intermittent, but might be possible to have more annual
questions in sub-sample
• Current view is that if we used this approach ethnic
boost samples would also be split
• Potential for other overlapping patterns, e.g. 5 question
groups with each 3-way combination per sub-sample
Questions for discussion
• What views do you have of the advantages and
disadvantages of use of sub-samples as
suggested here?
• Are there other approaches we should consider?
• Which questions could be restricted to subsamples?
• Which questions must be asked of the whole
sample?
Where do we go from here?
• Group 1: LTB4 (downstairs) with Alita
• Group 2: LTB4 (downstairs) with Noah
• Group 3: ISER large seminar room (in ISER
building) with Mark
• Group 4: ISER large seminar room (in ISER
building) with Gundi
• Group 5: ISER seminar room foyer (in ISER
building) with Birgitta
• Group 6: ISER seminar room 4.08 (in ISER
building) with Jon
Other forms of data collection
Birgitta Rabe and Nick Buck
(1) Data Linkage
• Purposes of data linkage
–
–
–
–
Supplement data collected in survey
Substitute data collected in survey
Validate data collected in survey
Survey administration
Types of data linkage
• Individual level
• Organisation level
• Area level
Individual level
• Benefits, earnings, taxes, government schemes
(DWP/HMRC records)
• Education and educational attainment (DCSF
and others)
• Health: hospital episodes, births, deaths, cancer
(HES, NHSCR)
Organisation level
• School indicators
• Firms?
Area level
• Linkage based on geo-codes, e.g. LAD, PCT,
LEA, SOA, grid references
• Wide range of geo-coded data available, e.g.
social and economic characteristics,
environmental quality
Issues in data linkage
• Best timing of obtaining consents for individuallevel linkage given burden and sensitivity issues
• User access plan
(2) Biomarkers and health indicators
• The UKHLS intended to be at the frontiers of research
on social, demographic, behavioural and health
sciences.
• The study needs to engage with the rapid advances in
the biological and life sciences
• Aim is to seek funding for collecting an integrated
package of biomarkers and key related indicators
• Intention is collection of biomarker information by
minimally-invasive methods that can be carried out by
survey interviewers with appropriate training
Development of biomedical strand
• ESRC commissioned John Hobcraft to identify key
potential issues and priorities
• Detailed consultation is needed on which measures
have the highest salience.
• This will involve UKHLS Biomedical and Health
Indicator Advisory Committee (chair Michael Rutter),
and Warwick colleagues on the UKHLS team (Dieter
Wolke and Scott Weich), along with ESRC
• Consultation will focus on ethical issues and
acceptability of this data collection for respondents
• Data collection requires additional funding and will
start at wave 3 at the earliest.
(3) Qualitative and other data from
respondents
• UKLHS has established links with qualitative
longitudinal research and needs to facilitate mixed
methods research
• Opportunities in the following areas should be pursued:
– A range of qualitative sources, including unstructured
interviews, biographies, audio and video diaries and other
forms of visual evidence
– Free text from structured questions
– Platform for linked qualitative studies
– Structured diary data
– Experimental data
• Plans in this area still being developed
What next?
Nick Buck
Consultation will be continuing
• We will make available drafts of wave 1 questionnaires
as soon as possible
• We will circulate more detailed plans for future wave
content, including both more specific indication of
measures to be included, and timing of inclusion;
• Also more detailed plans for sub-sampling to test
potential problems
• We will be approach you for advice on the design of
questions in specific areas.
Other issues for the future
• Data release: wave 1 completed in Autumn 2010, but
potentially release dataset based on first year fieldwork,
during 2010
• We will consult on data structures, documentation
requirements, other support users may need
• Some data may be more confidential than others (e.g.
from data linkage). Likely to be a differentiated release
policy e.g.
– most data available from ESDS using normal license (like BHPS),
– some using special license arrangement,
– some requiring analysis in a secure setting (ESRC has plans for these)
Timetable
January 2008 Consultation on wave one content
concluded; consultation on future waves
continues.
January 2008 Wave One of innovation panel starts
Spring 2008
June 2008
Consultation on Biomarker and Health
indicator component starts
Final survey pilot for wave one
June/July
Design work on wave 2 innovation panel
2008
starts
January 2009 Start of wave one main fieldwork
Thank you for coming and
keep in touch!
The UKHLS Team
http://www.iser.essex.ac.uk/ukhls/
Email us at: [email protected]