Examples of challenges

• Estimating public opinion from an opt-in internet survey - Challenge was extrapolation to general public from data collected from Xbox users. (Section 17.1)

• Randomized experiment on the effect of an educational television program - Challenge is adjusting for pre-treatment differences between treatment and control. (Section 19.2)

• Estimating the effects of United Nations peacekeeping - Challenges include this being an observational study. A pre-treatment ‘badness’ score was used to correct for pre-treatment differences.

• Estimating the effect of gun laws - The challenge here was the difficulty of inference using regression with a large number of predictors (50 data points with 30 predictors!!). Also systematic differences between states were not captured by the model’s other predictors.

The book notes that the peacekeeping study and the gun-control study are of the same type, but the peacekeeping study was more focused and was able to approximately account for systematic differences which does not seem possible with the gun-control study.