Writing
Long-form notes about methods, models, and scientific reasoning.
-
Evidence-based dashboarding
A dashboard can be based on data without being evidence-based; dashboard design should begin with intended decisions, testable requirements, and evidence that the interface works.
-
Tree-based machine learning
Tree-based methods can answer similar-sounding questions, but prediction, interpretation, and causal estimation each require different targets, assumptions, and validation.
-
Choosing thresholds for predictive models
Choosing a risk-score threshold requires balancing sensitivity, positive predictive value, alert burden, and the consequences of missed events.
-
Defining neighborhood geographies for clinical research
Census tracts, locally defined neighborhoods, and ZIP Code Tabulation Areas describe different places; choosing among them is a scientific decision, not a data-management shortcut.