When I was in high school I conducted a statistical analysis of female students correlating GPA with the use of hair coloring. Unfortunately for my dating prospects, the study showed a correlation in the negative direction.
I was thinking of that experiment after I read two unrelated but similar articles today, one associating breast augmentation surgery with suicide risk and the other associating diabetes with thigh length. In both cases the researchers were wise enough to point out that there was no evidence that one fact caused the other, but surely some readers are measuring their thighs at this very moment.
It makes me wonder if we're entering an era where the proliferation of all kinds of structured data (via the web, primarily) and powerful tools (Excel) will cause a dramatic increase in generally incorrect notions. Whenever data is analyzed there is always a risk of "beta error" -- finding correlations where they don't exist. As more data becomes available and more tools are available for analysis, the risk (or the certainty) of beta error increases. If we all try to reduce our error rate to 5% then one in twenty things we believe will be definitively wrong.
March 7, 2003 12:45 PM
This is Ari Paparo's blog. It generally concerns itself with Internet product development and marketing. 