Comment by bccdee

> I remember reading about how in Chicago someone had noticed that kids who did better had more books at home, so they decided to give poor kids books.

The problem here was not trying to infer causality from population-level data, but rather insufficiently controlling that data for correlated variables. If that study had controlled for the income and education of those kids' parents, it would have been much more able to predict the actual impact of giving kids books.

This visual essay thing doesn't present a particularly detailed data analysis, but I wouldn't be surprised if the original study, being properly academic, did dive into this kind of regression analysis.