UT AUSTIN / RELATIONAL DATA13 tables.
One cohort.Performance / time / confoundingCONCEPT COVER · NOT MEASURED OUTPUT
One cohort.Performance / time / confoundingCONCEPT COVER · NOT MEASURED OUTPUT
Who is improving—not just performing well?
I joined 13 relational tables to define a benchmark cohort, then fitted per-subject regressions of performance against time to rank improvement trajectories.
13 relational tables · per-subject regressions
What I worked on
Building the cohort came before fitting the regressions: the project required joining 13 relational tables and defining which observations belonged in the comparison.
Early-stage improvement rates were systematically steeper. I controlled for that effect when ranking improvement trajectories, rather than treating every raw slope as directly comparable.
Tools and methods R · tidyverse · Cohort definition · Regression · Confounding