Opportunities & Challenges of Applying Data Sciences to Global Ageing Data
Longitudinal Studies (Demographic, health, mental-health, psychosocial, physiological)
-
HRS 1992 -
51-61 year old US residents
Over 22000 respondents
Multiple studies merged together -
SHARE 2004 -
50 years+ EU residents
28296 initial residents
From 11 in 2004 to 29 countries -
ELSA 1998 -
50 years+ English residents
12099 initial respondents
Established by US NIA & UK Gov
Opportunities
-
Source of secondary data
Test multiple hypothesis
Comparative studies -
Change over time
Predict trajectory of variables
Multi-level & Growth models -
Mediation Analysis
Hypothesized causal mechanisms
Causal (X) Mediator (M) Outcome (Y) -
Time to event
Survival (event history) analysis
Discrete & Continuous time models -
Prediction of disease
Modern machine learning methods
Web-based data science tools
Challenges
- Time consuming and expensive data collection
- Complex samplign design rather than simple random sampling
- Weight adjustments to avoid biased estimates & incorrect inference
- Attrition due to increasing probability of morbidity and mortality
- Respondent unreachable, refuses or event doesn't occur
- Modern methods include Maximum likelihood & Multiple imputation
- High probability of measurement errors
- Responses conditioned due to participation in the study
- Large number of variables and relatively small sample sizes
- Time series data