Let's Get Naked (Statistically Speaking)

Charles Wheeler likes to get naked . . . he's the author of Naked Economics, which I highly recommend, and I also enjoyed Naked Statistics: Stripping the Dread from the Data, which is full of fun facts and lots of number sense (and it will make you think about all the times you are offered either percent of increase or a number, when you really need both to make an assertion) and here a some random moments I enjoyed:

1) texting while driving causes crashes and laws banning texting while driving may also cause crashes because people can't stop texting while driving, but if there's a law against it, then people will hide their phones down by their crotch and take their eyes off the road;

2) people who buy carbon-monoxide monitors and little felt pads for the bottom of their furniture almost never miss credit card payments;

3) the top 100 grossing films only makes sense when it's adjusted for inflation . . . Hollywood likes to tell the story that each new blockbuster movie is so good it has blown away all the older films, but they like to list the gross (nominal) ticket receipts, not the real, adjusted receipts: here is the real list . . . The Exorcist makes the top ten and Jurassic World makes the top 25 so this list isn't any more cultivated than the gross profit list (though it's less homogenous);

4) our data sets are getting more and more predictive . . . people who buy birdseed are far less likely to default on their loans, but if we can identify drug smugglers 80 times out of 100, is it okay to harass those other twenty people over and over? so statistics generally leads to ethical dilemmas . . .

5) the most dangerous job stress seems to be jobs that have "low control" over their work situations . . . which makes me happy, because teaching and coaching feels highly stressful at times, but I always have control over what's happening . . . but this is only true if we trust the regression analysis, which is the most powerful statistical tool in existence, but very difficult to do well;

6) because you can screw up regression analysis in a number of ways: you can use regression to analyze a nonlinear relationship, you can screw up correlation and causation-- buying birdseed does not cause you to have good credit, those two things are simply correlated-- you can complete reverse the causality, you can omit variables, you can have variables that are so highly correlated that you can't extricate them from each other, you can extrapolate beyond the data, and you can have problems with too many variables;

7) Wheeler concludes with a quick overview of some real-world problems that are going to need clear statistical analysis: the future of NFL football, the rise in autism, the difficulty in assessing good teachers and schools, the best tools for fighting global poverty, and personal data privacy . . . if you're looking for a fairly in depth take on statistics, with more formulas and math than a Freakonomics or Malcolm Gladwell book, this is the one for you.