tag:blogger.com,1999:blog-6894866515532737257.post103689584576297401..comments2024-04-22T21:33:32.590-07:00Comments on Probably Overthinking It: Bayesian Statistics for UndergradsAllen Downeyhttp://www.blogger.com/profile/01633071333405221858noreply@blogger.comBlogger2125tag:blogger.com,1999:blog-6894866515532737257.post-11331479637563599052016-06-18T08:46:21.927-07:002016-06-18T08:46:21.927-07:00"Bayesian methods don't do the same thing..."Bayesian methods don't do the same things better; they do different things, which are better."<br />...<br />They do different things, indeed. So why insist that one is universally better?<br /><br />One thing I've been mulling over lately: Classical methods were invented for, and by, people in the business of designing and running experiments. "Hypothesis tests" aren't really testing hypotheses about the sample you collected, or the population you sampled, but about *the design of the experiment* you ran. In the simplest case, the question isn't "Is theta 0 or not?" but rather "Is this sample big enough to tell if theta is positive or negative?"<br />This is not always "the wrong question," as you call it. Sometimes yes, but other times it's crucial.<br /><br />Perhaps the main benefit of inventing hypothesis tests was so you could *imagine* doing them as you make power calculations to choose the sample size (and other design details).<br />Engineers find it useful to know their instrument's operating characteristics before they choose which lathe or radio or oscilloscope to buy/use. Same with doing classical power calculations to design a study for science, or A/B testing, or medical trials, or what have you.<br /><br />I don't mean to say the Bayes framework can't do this too (I know nothing about Bayesian experimental design) but just that traditional classical methods have legitimate uses different from, not worse than, traditional Bayesian methods.Jerzy Wieczorekhttps://www.blogger.com/profile/03611849167717252118noreply@blogger.comtag:blogger.com,1999:blog-6894866515532737257.post-86795590804703442212016-06-15T05:32:36.924-07:002016-06-15T05:32:36.924-07:00Thank you, Allen (and Sanjoy)! Your workshop gave...Thank you, Allen (and Sanjoy)! Your workshop gave me a lot to think about. Jessicahttps://www.blogger.com/profile/12370010794505516795noreply@blogger.com