“Hours earlier, as the orange Steelman tumbled through the brush, Tom had slid down the embankment, crashing violently through the leaves. While Officer Dexter was awaiting backup, Tom trudged 50 feet upstream and took cover underneath a bridge, where he discovered a two-foot-wide hole at the water’s edge. It looked as though a beaver had burrowed deep into the creek bank. Tom crawled in headfirst, thorny plants scratching his face and arms, and squirmed 11 feet to the narrow tunnel’s end. Panting in the dark, he heard sirens, then faint voices and the jingling of a dog’s tags. Tom assumed that was the end. But soon he accepted what seemed a miracle: The cops had given up the search.”
“Over the past 35 years, the tobacco industry has been the best performing stock-market industry in the world, despite declining volumes in the developed world, lawsuits, and ever more stringent regulatory restrictions. There have been two reasons for this: one, the economics of the industry, and the favourable effect regulator behaviour has had on the industry’s supply-side dynamics; and two, the fact that investors have continued to misunderstand smokers and smoking (and activist behaviour), not least because most people in the investment business do not smoke. I do (at least some of the time).”
“Classical frequentist statistics typically measures the difference between groups with a t-test, but t-tests are 100+ years old and statistical methods have advanced a lot since 1908. Nowadays, we can use simulation and/or Bayesian methods to get richer information about the differences between two groups without worrying so much about the assumptions and preconditions for classical t-tests.”
“This model makes a lot of sense, because rarely are we in a situation to a priori decide that the variance of scores in Group A is equal to the variance of scores in Group B. If you use the equal variances t-test, you should be prepared to justify and defend this assumption. (Deciding between models–such as between these two t-tests–is one way in which our prior information enters and influences data analysis. This fact should make you less suspicious about priors in Bayesian analyses.)”