Read on Twitter
Inspired by @quantitudepod I'll start to post periodic threads that bring together cognitive neuroscience and statistics--two of my favorite topics to nerd out about.
First up: How do we think about age as a variable?
First up: How do we think about age as a variable?
Chronological age is an index of time: the number of complete revolutions around the sun from birth. As an index of time, it is exogenous (without cause). Time is often not the cause of an outcome and instead affords an opportunity for a process to unfold. /2
Age can be a proxy for bio-psycho-social processes. Some are time-dependent (e.g. sensitive period), some are time-correlated (e.g. DNA mutations). If the process is not measured, age as a covariate is accounting for omitted variable bias and itself is difficult to interpret. /3
If measured, processes may mediate age effects on cognition. Time is linear but non-linear trajectories are consistent with processes unfolding over time. "Biological age” counts time from conception, and “maturation” allows equivalent comparison between sexes (or species). /4
Age is a moderator. As proxy for cohort, a generation that survived 2 world wars may be different than another. Age as proxy moderates itself. Mechanisms are relevant at certain periods: age is a sliding window across the lifespan in longitudinal study of within-person change. /5
This is not “either/or”. Time affords processes unfolding until death. So, age is a clock counting down, not up. In the 1920’s average mortality age was ~54 y; in 2020, ~78 y. The “meaning” of age 60 y was a phenomenal event in 1920’s and mundane today. /6
If time affords processes to unfold, and intervention modifies that, then the effect of chronological age in a model and its process pathways are also conditional upon time period. Here we have multiple forces shaping non-linear trajectories across the lifespan. /7
Interpreting a variable called “age” begins with model-based thinking. Directly following from that we determine how the variable is incorporated: a covariate to account for omitted variable bias, a mediated effect in a causal framework, and/or a condition of another effect. /8
I think of "age" in all of these ways when studying cognitive aging, biomarkers of dementia risk, and lifestyle interventions. We need to in order to translate cog neuro research towards clinical recommendations across the lifespan. /fin
Graphical Summary
