There's an interesting discussion posted at Marginal Revolution on how you go about explaining regression analysis to people not trained in statistics in the particularly challenging context of a late night television show where the attention span for anything not involving naked women is probably 15 seconds or less.
I don't think the writer nailed it, so here's my attempt:
Most real world outcomes are caused by multiple factors. To understand the effect of any one factor we need a way to control for the other factors, that is, we need a technique to let us figure out how much of the change in the result is caused by changes in the factor we are studying.
Or as Bert and Ernie demonstrated in an excellent Sesame Street episode, the fact that there are no alligators in Sesame Street does not mean Ernie's technique of sticking a banana in his ear to keep the alligators away actually works.
One of the most egregious violations of this principle occured during the dotcom boom / late 90's stock market bubble, namely the linking of CEO compensation to stock prices without controlling for the general effect of a rising market. Shame on those boards who lined the pockets of their CEOs with millions of dollars in bonuses and stock options when they should have been compensating them only for that part of the rise in the stock price that was caused by the CEO's performance.
What must it have been like to sign such a contract? Well, for a CEO who understands statistics, it must have been like knowing in advance that you've bought the winning lottery ticket.
No comments:
Post a Comment