1
Social surveys and new forms of data: the case of the UKHLS
Stephen P. Jenkins
Institute for Social and Economic Research University of Essex Email: stephenj@essex.ac.uk
2
Will new forms of data replace the social survey?
No! • ‘New’ and ‘old’ data sources have different strengths
– use all the sources, depending on research goal(s)
• Intrinsic strengths of social surveys
– representativeness of ‘populations’, and random probability sampling – governance; – mechanisms for non-disclosure, confidentiality, release, access
• Social surveys are innovating to incorporate ‘new forms of data
– the UKHLS as an example
3
United Kingdom Household Longitudinal Study
http://www.iser.essex.ac.uk/ukhls/
• New longitudinal study based on a household panel design, i.e.
– sample based on all residents (adults and children) at addresses selected at wave one, following them at each wave, including movers and collecting data about new household members
• Similar to household panels around the world, including e.g. BHPS (which it will absorb) SOEP, HILDA, PSID, SoFIE • Target sample size of 40,000 households
– largest household panel study in the world …
4
UKHLS key features
Representativeness and coverage; New types of data
• • • • •
• •
Large sample size Household focus Full age range sample Innovative data collection methods Multi-purpose multi-topic design to meet a wide range of disciplinary and interdisciplinary research needs Ethnic minority research Biomedical research
5
(1) large sample size
40,000 households gives an opportunity to explore issues where other longitudinal surveys are too small • Small subgroups, e.g. teenage parents or disabled people • Smaller areas: analysis at regional and sub-regional levels, allowing examination of the effects of geographical variation • Smaller time periods: large sample size allows highresolution analysis of events in time
– e.g. focus on single-year age cohorts
6
(2) household focus
• Data collected on all members of sampled households • Important for research on e.g.
– consumption and income, where within-household sharing of resources is important – demographic change, where the household itself is often the object of study
• Observing multiple generations and siblings allows examination of long-term transmission processes • Opportunities to explore linkages outside the ‘household’
7
(3) full age range
• full age range at any point in time
– complements age-focused studies sampling elderly people (e.g. ELSA) or young people (e.g. birth cohort studies)
• Provide a unique look at behaviours and transitions in mid-life –
– e.g. for issues of pensions and long-term care, associated with old age, policy setting is influenced by earlier behaviour
• Large sample size all cohorts can be analysed at a common point in time
8
(4) innovative data collection
Continuous development benefiting from:
– – – – experience from other longitudinal surveys introduction of new types of data and data collection methods additional modes of interviewing (e.g. CATI, Web, …) external record linkage, subject to relevant consents
• E.g. Registration statistics, Hospital Episode Statistics, Education (PLASC; HESA; and institutional); earnings, benefits and NIC records E.g. Census-based info, climate, land use, …
This will potentially involve, for example :
– –
Geo-coded linked data
•
collection of qualitative and visual data
•
An Innovation Panel to allow experimentation and methodological development
9
(5) broad interdisciplinary topic coverage
• Multi-purpose survey supporting a very wide range of research agenda • cannot have the focus of more specialist surveys • Strength arises from bringing together information on many life course domains • Interdisciplinary
– aiming both to meet needs of traditional panel-use disciplines (economics, social policy and sociology) and support more interdisciplinary work within the social sciences (e.g. geography and economics); within the biomedical sciences (e.g. psychology and genetics); and between the two
10
(6) ethnic minority research
• Ethnicity strand includes:
– Boost sample for five key groups (Indian, Pakistani, Bangladeshi, Caribbean, Black African) – Ethnic minority group members selected for main sample – Questions focused on ethnicity issues
• Recognises the increasing prominence of research into ethnic differences for understanding the make-up of British society and issues of diversity and commonality
11
(7) biomedical research
•
•
•
Collection of a wide range of biomarkers and health indicators to be supported Opportunity to assess exposure and antecedent factors of health status, understanding disease mechanisms (e.g. gene-environment interaction, gene-to-function links), household and socioeconomic effects and analysis of outcomes using direct assessments or data linkage Opens up prospects for advances at the interface between social science and biomedical research
12
All data source types are imperfect
As with the ‘old’ forms of data, so too with the ‘new’ – none is perfect!
• New forms designed for purposes other than research
– – – – ‘100% coverage’ of what (sub-)population? What variables (depvar and covariates)? Inconvenient forms and size for research end-user? ‘Ownership’ and release/access issues
• All data source types have problems of
– missing data (item non-response and unit non-response) – measurement error (admin data aren’t perfect)
13
Summary: the UKHLS case
• demonstrates the potential for ‘old types’ of data to evolve in an innovative fashion • incorporating several good features of ‘new types’ of data, together with • traditional strengths of random probability social survey • NB ‘description/data-mining’ is modelling (of a sort), and vice versa
– Overoptimistic to believe that useful generalizations will arise more naturally from ‘new types’ of data than the old (pace Savage & Burrows)
14
Thank you
United Kingdom Household Longitudinal Study
http://www.iser.essex.ac.uk/ukhls/
15
UKHLS study design
• Randomly sample UK addresses: select as sample members all members of private households found • At each wave, all sample members above a threshold age eligible for interview (threshold to be decided) • Individuals followed as they move and form new households • Other individuals who form households with sample members after wave 1 eligible for interview • Following rules mean that the UKHLS will remain representative of the UK population as it changes, subject to weighting and except for new immigrants to the UK
16
UKHLS sample composition
1. A new equal probability main panel achieved sample of 28,000–29,000 households
– Fieldwork to commence in January 2009
2. A new boost ethnic minority sample, to provide 1,000 adult individuals in each of the five main ethnic minority groups 3. The BHPS sample of approximately 8,400 households
– Data collection as part of the UKHLS will start with wave 2 in October 2009
4. An Innovation Panel of 1500 households to enable methodological research
– Fieldwork to commence in January 2008
17
UKHLS design
• Some aspects remain to be finalised and depend partly on co-funding • Expected features include:
– 12 month intervals between interviews – Continuous fieldwork (implications for reference periods for retrospective questions) [Possible 24 month field period, with second wave overlapping with first] – Face-to-face interview at wave 1; mixed mode at wave 2 – Wave 1 individual interview not more than 40 minutes, wave 2 depends on budget, unlikely to exceed 40 minutes and may be shorter – Some data collection directly from children aged less than 16 (unclear when to be implemented)
18
The UKHLS questionnaire
• Length constraints are likely to be particularly acute, given broad scope of UKHLS and wide range of demands
– move away from BHPS structure where most people are eligible to be asked all questions and most questions repeated each wave
• More use of questions asked regularly, but not every wave • More use of questions asked only after key events or at particular ages • More use of sub-samples, perhaps random sub-samples, where full sample unnecessary, or demographic subsamples