Analysing change based on two measures taken under different conditions.

Area of Science:

• Longitudinal data analysis
• Biostatistics
• Psychometrics

Background:

• Analyzing change over time in longitudinal studies is complex when measurement conditions differ between periods.
• Measurement condition differences can confound true change effects with 'conditions effects' (e.g., practice effects), leading to biased estimates.
• Similar to age-period-cohort identification problems, isolating conditions effects requires specific modeling assumptions.

Purpose of the Study:

• To develop a statistical model for estimating true change effects in longitudinal data despite confounding from differing measurement conditions.
• To demonstrate that conditions effects are identifiable under empirically unverifiable assumptions regarding confounding, age-related change, and conditions-effect factors.
• To introduce the Conditions-Effect Adjustment Model (CEAM) as a framework for change analysis.

Main Methods:

• Development of the Conditions-Effect Adjustment Model (CEAM).
• Identification of necessary assumptions for estimating conditions effects: sources of confounding bias, functional form of age-related change, and conditions-effect related factors.
• Utilizing sensitivity analysis to assess robustness to these assumptions.

Main Results:

• The study shows conditions effects are identifiable under specific, though unverifiable, assumptions.
• The CEAM provides a method to estimate change effects when measurement conditions vary across study periods.
• Sensitivity analysis allows for evaluating the impact of different assumptions on the results.

Conclusions:

• Estimating change in longitudinal studies with varying conditions requires explicit modeling assumptions.
• The CEAM offers a structured approach to address conditions effects and assess the robustness of findings.
• The model's utility is demonstrated using cognitive test data from the Whitehall II study.