Table 1.

Summary of data generating rules with dependencies determined from the causal DAG and appropriate functional form for the modelling based on exploration of individual-level data in HRS

Variable	Variable Type	Dependency	Functional Form
Age, Gender, Race	Categorical	Exogenous	Sample proportions in categories defined by gender, race, and 5-year age groups; assumes a uniform distribution within 5 years of age
CSES (HRS only)	Continuous, approximately normally distributed	CSES | Age, Race	Linear regression model
Education	Categorical	Education | Age, Race, Gender, CSES (HRS only)	Multinomial regression model
Diet (ARIC only)	Continuous, approximately normally distributed	Diet | Race, Gender, Education, Age	Linear regression model
HbA1c	Continuous, not normally distributed	HbA1c | Age, Race, Gender, Education, CSES (HRS only), Diet (ARIC only)	Gamma regression model
Rate of Change in Memory	Continuous, repeated measures	Rate of Change in Memory | Age, Race, Gender, Education, CSES (HRS only), Diet (ARIC only), HbA1c	Linear mixed effects model

Abbreviations: ARIC, Atherosclerosis Risk in Communities Study; CSES, childhood SES; DAG, directed acyclic graph; HbA1C, glycated hemoglobin; HRS, Health and Retirement Study