Growth Curve Approaches to Longitudinal Data in Gerontology Research Discussion

1
Growth Curve Approaches to Longitudinal Data in
Gerontology ResearchDiscussion

• Karen Bandeen-Roche
• Department of Biostatistics COAH
• Johns Hopkins University
• GSA Annual Meeting
• November 16, 2001

2
Important Issues

• Jones Memory Maintenance
• Hall Natural history of dementia onset
• Lighthouse Impact of sensory loss
• Newsom Widows depression persistence

3
Alternative analytic approaches?

• Marginal model
• Do populations learn at different rates, on
  average?
• Numeric data ? no distinction in fixed effect
• Autoregressive / transition model
• Does initial performance predict learning rate?
• Does initial depression score predict future
  scores?
• Change point, versus spline, versus curve?
• Treatment of time-varying covariates?

4
A Fine PointAre Multilevel and Latent Growth
Equivalent?

• It depends on the software used
• The distinction treatment of predictor
  variables
• Multilevel, M-plus ADL trajectories in
  populations with different sensory status
• LISREL Covariation between ADL trajectories and
  sensory status, and among sensory variables
• Deep-seated statistical debate
• SEM latent growth predictor measurement model

5
One Nudge
MODEL CHECKING
6
Heart of the MatterWhat Has Been Gained ?

• The truth?
• Avoidance of spurious findings?
• Subtlety?
• Nothing?
• Obfuscation?
• Congratulations to todays speakers!