The Thirty-Four Scrolls
Double machine learning for targeted email promotions: why the raw numbers lie, which parts of a subject line actually drive sales, and how better targeting earns more.
Each paper is retold as a flip-through storybook set in the world of Dungeons & Datasets. Every villain stands for one idea in the paper, and the numbers are the paper's own.
Double machine learning for targeted email promotions: why the raw numbers lie, which parts of a subject line actually drive sales, and how better targeting earns more.
Positive versus negative political ads in the 2000 and 2004 presidential elections: negative ads persuade, positive ads bring voters out, and tone alone can swing a close race.
The sequel to The Thirty-Four Scrolls: using what past emails taught you to judge new subject lines written by generative AI, and knowing when to run a real test instead.
Walmart's entry into U.S. grocery: shoppers value nearness as well as price, and rival chains would have built more stores without it, so the real gain is about a fifth of the leading estimate.
Can salespeople's CRM activity logs improve sales forecasts? Pooled, they make forecasts slightly worse, but a minority of reps write logs that genuinely see the future.
How to control for group fixed effects in a causal forest, re-estimating whether a credit rating lets firms borrow more: the forest only sees how groups are ranked, so give it each group's fixed-effect coefficient.