OPERA

1
ESRC NCRM Research Methods Festival, Oxford, 1
July 2008 Session 27 Resources for Data
Management and Handling Social Science Data
Social care data exploring issues David Bell,
Alison Bowes and Alison Dawson
2

• This case study will
• Identify some of the substantive issues in
  research on social care using the example of
  unpaid care
• Outline a microsimulation approach to
  considering social care issues
• Briefly discuss general data-related issues in
  microsimulation
• Describe data management operations that may be
  required for microsimulation modelling, in
  particular data fusion
• Illustrate potential complexities with data
  fusion using examples drawn from the Family
  Resources Survey (FRS) and the Scottish Household
  Survey (SHS)

3
Example unpaid care

• The continuing importance of unpaid care
• The balance of care
• Supporting unpaid carers
• The dynamics of unpaid caring

4
Key findings to date on unpaid care

• The impact of recent policy change
• Attitudes to caring
• Provision of care, proximity and education
• Significance of housing wealth

5
Some key issues

• Paying for care
• Impact of various co-payment regimes
• Impact of changing patterns of wealth
• Questions about the best expenditure of resources
• Supply of unpaid care
• Attitudes
• Education
• Migration of parents and children
• Delivery of unpaid care
• What carers do
• How to support carers
• Dynamics of unpaid care need for
  microsimulation
• Population movements
• Educational attainment
• New types of care provision
• Resources for unpaid care

6
Approach

• Using a microsimulation model (OPeRA Older
  Peoples Resource Allocation Model) in the
  context of Scottish demographic change
• drawing on data from various censuses and
  surveys, including FRS, census data, data on
  actual caring histories over time (from Welsh
  survey)

7
Microsimulation

• A microsimulation model is a model which uses
  simulation techniques and which takes micro level
  units as the basic unit of analysis when
  investigating the effects of social and economic
  policies
• Dynamic microsimulation
• These models project samples of the population
  forward in time, enabling, for example the
  examination of future income distributions under
  different scenarios

8
OPERA Key elements
9
OPERA - Potential outputs

• Individual life histories, including record of
  events
• Census on any type of event and calculation of
  cost by local authority over user-determined
  period
• Preparation of budget for local authority based
  on current care requirements and costs profile

10
Data related microsimulation issues

• What variables/behaviours to simulate?
• Over what period?
• What level of geographical disaggregation?
• Can all relevant information be found in a single
  sample survey?
• If not how do we form encompassing datasets?

11
Data management operations for microsimulation
modelling

• Cleaning data
• Checking data
• Recoding data
• Missing data / case selection
• Operationalising variables
• Linking data
• Matching data files together
• Data fusion

12
Data Fusion
Common variable(s)
Values of unobserved variable(s) are imputed for
recipient sample
Before data fusion No values for Var Y
After data fusion Imputed values for Var Y (not
true values)
Fusion proceeds by a number of (mostly
numerically intensive) procedures. The objective
is to define new variables whose properties
differ as little as possible from those of the
(unobserved) underlying data.
13
Potential difficulties with data fusion
Examples using FRS and SHS

• Family Resources Survey (FRS)
• Cross-sectional study, annual sampling frame
• Covers GB and N Ireland
• Disaggregation possible to Government Office
  Regions
• Observation unit households/families
• Interviews all adults in household
• 28,029 households in 2005/06
• 2,016 variables (253 derived)

• Scottish Household Survey (SHS)
• Cross-sectional study, two year sampling frame
• Covers Scotland only
• Disaggregation possible to Local Authority areas
• Observation unit individual,
  households/families
• Interviews householder or spouse plus one
  random adult
• 31, 013 households, 28,261 random adults in
  sample to 2005/06
• 2,456 variables (80 administrative and derived)

14
Example 1 FRS and SHS common variables
Ethnicity
15
Example 2 FRS and SHS common variables
Marital status
FRS Variable label Marital status Value label
information for MS 1 Single, never married 2
Married and living with husband/wife 3 Married
and separated 4 Divorced 5 Widowed
SHS Variable label Marital status of the
HIH Value label information for hih_stat 1
Married 2 Cohabiting (living together) 3
Single / never been married 4 Widowed 5
Divorced 6 Separated
FRS Variable label De facto marital
status Value label information for DVMARDF 1
Married 2 Cohabiting 3 Single 4 Widowed
5 Divorced 6 Separated 7 Same sex couple
16
Summary

• Techniques such as microsimulation can play an
  important part in considering substantive issues
  in research on social care.
• There are many relevant variables when modelling
  social care. We identified only a subset of these
  when considering the example of unpaid care. No
  single sample survey contains all the variables
  we would want to use.
• We need to identify relevant sample survey
  datasets and carry out data management operations
  such as cleaning data, operationalising
  variables and linking datasets prior to
  modelling.
• Metadata provides valuable information that
  helps us to identify and address practical
  problems in the data fusion processes necessary
  to produce better microsimulation models