Assess education–income effect credibly
Overview
This question evaluates a data scientist's competencies in causal inference, experimental design, model selection, estimand specification (ATE) and sensitivity analysis for estimating treatment effects from observational data in the Analytics & Experimentation domain.
You collected data on 1,000 Mountain View residents: College (binary; attended any college) and Income (annual). Is regressing Income on College alone appropriate for estimating the effect of college on income? Identify issues of external validity (Mountain View not representative), model choice (binary predictor in linear regression vs two-sample t test), and causal identification (confounding by age, occupation, parental income, ability). Propose a better analysis: specify the estimand (ATE of college on income), include covariates in a regression or matching framework, perform balance checks, and discuss alternative designs (instrumental variables, regression discontinuity, randomized encouragement). Explain how you would report uncertainty (confidence intervals) and sensitivity to unobserved confounding.