Statistical Significance, Again
Science, Ziliak and McCloskey say, needs to learn both whether and how much. But the putatively common practice of ending the empirical analysis by checking whether a calculated test statistic does or doesn’t reject the null hypothesis according to a 5 percent or some other a-sized crtical region is basically to stop at the question of existence without addressing the scientific issue of magnitude
That is Saul Hymans, reviewing The Cult of Statistical Significance. His review appears in the June 2009 issue of the Journal of Economic Literature, which just came in yesterday’s mail.
Jul 7 2009 at 8:09pm
McCloskey has always struck me as overly strident, yet I continue to read the most daft statistical analysis from even Ivy League business schools.
It seems likely to me that the problem is not so much a matter of confusion over the word “significant” as the inability of the vast majority of humans to understand the three door Monte Hall problem. When strikingly intelligent mathematicians can work it out but still not understand it, what hope is there for the rest of us.
Jul 8 2009 at 9:07am
Thanks for the posts on statistical significance. It’s odd that some people ignore the fact that if you have enough data you can get statistical significance on just about anything. And why choose 95%? Most medical research works on 99%. Some like 90%? The choice depends on how much you want to prove your point.
And what if the p-value is .059? Technically it’s not significant, but it’s darn close!
As in other things, size matters!
Comments are closed.