EC5200 Research Methods Lecture 3 Analysis of Longitudinal Datasets

1
EC5200 Research MethodsLecture 3 Analysis of
Longitudinal Datasets

• Prof Peter Dolton
• Room H309
• Office hours Wed 1200-1300, Thurs 1500-1600
• ? peter.dolton_at_rhul.ac.uk ? 01784 443378

Slides and other exercises and handouts available
at http//personal.rhul.ac.uk/UQTE/004/EC5200
2
References

• On data
• ESRC Data Archive
• Tracking people A Guide to Longitudinal Social
  Sources, Govt Statistical Service
• On LS
• Chp 8 in C Hakim (2000), Research Design 2nd
  ed, Routledge
• Regression
• C. Dougherty (2002) Introduction to
  Econometrics Oxford UP
• Panel Longitudinal Data
• Baltagi, B (2003) Econometric Analysis of Panel
  Data, 2nd Ed

3
Data Types

• Cross Section different units at the same point
  in time.
• Time Series same unit over time
• Longitudinal same obs at different points in
  time at least 2.
• May involve admin data merging
• Different time intervals for different purposes.
• May involve tests of the subjects
• May involve data collection from
  teachers/schools/ other organisations.
• Main feature of (prospective) Longitudinal data
  is that allows us to track change and development.

4
LS Types

• Panel Study Is an LS which attempts to be
  nationally representative BHPS, PSID in USA
• Cohort Study particular experience or
  characteristic in common then follow up NCDS
  born in 1958, YCS school leavers.
• Intergenerational Studies PSID, NCDS and NLS
  now have studied the kids of cohort members.
• Retrospective Studies Collect data once asking
  questions about past dates and events problem
  of post-hoc rationalisation bias.
• Synthetic Cohort Pseudo Panel Data
• Cross National Comparative Studies ECHP 12
  countries from 1994 60,000 households 130,000
  persons.

5
Problems with LS

• Big problem with LS is that of initial
  non-response and attrition
• There are different types of non-response
• Voluntary - refuse
• Involuntary death, imprisonment
• Non-contact move away etc
• Solutions
• Weighting
• Structural Methods
• Refreshment Sampling.

6
Problems with LS

• Often researchers apply cross section methods to
  LS
• Seams problem different states at overlap
  points in LS
• LS are often interdisciplinary tribal warfare
  over questions inevitable .
• Expense continuity over funding over long run
  is a huge problem.
• Size of data usually at least 5,000 obs over
  time
• Moral / ethical issues confidentiality
• Merging with admin data eg Crime data!?
• DNA problems

7
Design of LS

• Tracing issue how do you keep in touch to
  minimize drop-out Birthday cards etc for NCDS.
• Optimal gaps between surveys bigger gaps
  mean more drop out and bigger recall issues but
  more often means more expense.
• Data documentation continuity is important.
  this can be a big problem over the extreme long
  run e.g. NCDS
• Type of data collection CAPI, CATI options
• Difficult decisions on Quality /Cost
  trade-off
• Telephone v Face to Face.
• Merging / Sampling from admin.

8
Analysis of LS

• Analysis of LS is complex.
• Permits Duration/Event History modelling
• Multiple state and transition methods.
• Fixed Effects/ Random Effects nets out for
  unobserved heterogeneity.
• LS data is not experimental but it does sometimes
  allow for the assessment of policy change using
  Differences-in-differences those who were and
  were not affected by the change.
• Often having more than one cohort is very useful
  e.g. graduate cohorts 60, 70, 80, 85, 90, 95.
  or YCS or NCDS v BCS

9
British Birth Cohort Studies

• Previous UK/GB National Studies
• 1946 MRC National Survey of Health
  Development
• 1958 National Child Development Study
• 1970 1970 British Birth Cohort Study
• MCS the first national birth cohort study for
  30 years 2000-1
• 1958, 1970 MCS based at IOE

British Birth Cohort Studies
10
Why cohort studies are important

• Continuing birth cohort studies are expensive
  compared to cross-sectional surveys, but they are
  important because
• They tell the linked stories of the lives of the
  members
• They record how long someone occupies a given
  state
• They link events across the life course across
  domains
• They enable investigation of the cause effects

11
Periods ages of data collection in the 3 studies
12
NCDS Follow-ups
13
Figure 3c
Wages scale (pounds per week)

Scores scale
Reading test scale
14
BCS70 Follow-ups
15
Useful Websites for Further Information

• http//www.cls.ioe.ac.uk/Cohort/MCS/mcsmain.htm
• Millennium Cohort Study (MCS) Main page
• Includes Rationale, Surveys, MCS Launch, Data
  Archive, Documentation, Contact details
•  
• http//www.cls.ioe.ac.uk/Cohort/MCS/Publications/m
  ainpubs.htm
• Annual report, Conference Papers, Journal
  Articles
•  
• http//ioewebserver.ioe.ac.uk/ioe/cms/get.asp?cid
  6186
• Bedford Group (BG) on the Institute of Education
  (IoE) Website
•  
• http//ioewebserver.ioe.ac.uk/ioe/cms/get.asp?cid
  61946194_08239
• MCS on BG/IoE Website including the MCS Users
  Guide to Initial Findings (Descriptive Report)