What's the Science in Data Science?
summary
Argues that credibility comes from concrete design not buzzwords and shows which methods actually rescue observational work: instrumental variables, matching, regression discontinuity and difference-in-differences with diagnostics for spillovers, attrition and power. Cites a Yahoo experiment with 250 million impressions where quasi methods overstate effects (up to 1,200%) versus a 5% randomized control, and a Courtyard by Marriott conjoint that forecasts market share predicted within 4% and drives real revenue and jobs.
AI-written summary · about
chapters (13)
- 0:00Introduction
- 1:20Credibility crisis
- 5:51Defining data science
- 8:16Data science credibility today
- 9:57Research design gold standard
- 13:44Social sciences validity threats
- 16:11Economics research design
- 25:20Importance of research design
- 27:27Machine learning and causal inference
- 29:12Case studies proving science
- 37:54Data science industry impact
- 40:24Conclusion and key takeaways
- 42:50Q&A
