As John Cleese puts it: "If you're very very stupid, how can you possibly realize that you're very very stupid?" That's the Dunning-Kruger effect - the idea that unskilled people tend to overestimate their own skills. But what if it's not real? That's what Blair Fix argues, convincingly. Take two random values, *x* and *y*. Plot them against each other and you get a random graph. But define a new variable, say, *z=x-y* and then plot *x* and *z*, and you get a correlation. Why? Because you're essentially comparing *x* and *x*, and that's what you're doing when you plot an actual test result and the difference between a predicted and actual test result, which is what Dunning and Kruger did. Why is this relevant? We ran a test last year to evaluate data literacy in a population. We found people's predictions were reliable indicators of their test scores. We were looking for similarity, not difference. Autocorrelation? Via Metafilter.

Today: 0 Total: 79 [Share]

] [