tag:blogger.com,1999:blog-3850643599457806204.post5311999009568175983..comments2023-02-01T15:38:24.309-05:00Comments on Engora Data Blog: The null hypothesis testMike Woodwardhttp://www.blogger.com/profile/16922188923191611428noreply@blogger.comBlogger2125tag:blogger.com,1999:blog-3850643599457806204.post-4191772960386812852020-11-09T08:10:33.542-05:002020-11-09T08:10:33.542-05:00Thanks for your comment. I'm going to talk abo...Thanks for your comment. I'm going to talk about Bayesian methods later. To be honest, I'm not a huge fan of null hypothesis testing; it feels overly complex and reliant on magic numbers. It does work, but... To me, it feels like going to a bad restaurant, yes, the meal fills you up, but you can't help but think another restaurant would have been tastier. Mike Woodwardhttps://www.blogger.com/profile/16922188923191611428noreply@blogger.comtag:blogger.com,1999:blog-3850643599457806204.post-40989873579030270892020-11-04T13:29:45.182-05:002020-11-04T13:29:45.182-05:00Hi Mike!
Here is a simple way to grasp the meanin...Hi Mike!<br /><br />Here is a simple way to grasp the meaning of the null hypothesis.<br /><br />Specifying a hypothesis as "the null hypothesis" is a way to specify that it is supported by lots of "prior data" (data from past experiments that are not directly considered in the present test). <br /><br />Example: You have a medication that has been tested on 1 million patients over many years, and it is effective with probability 0.99. You do a small test on 100 people, and the drug seems ineffective 10% of the time. Would you reject the hypothesis that the drug is 99% effective? Of course not. (You should first try to replicate your result and identify biases in your test.)<br /><br />More precisely,<br /><br />if prob(ineffective in your test) > alpha <br />then retain the drug<br />else reject the drug,<br /><br />where alpha = your estimate of probability of rejecting H0 when H0 is true. <br /><br />The most obvious weakness of this method is that your estimate of alpha does not quantify how much prior testing you rely on. Some methods, including Bayesian methods, count the number of prior trials and combine that number with the number of trials in your test. Anonymoushttps://www.blogger.com/profile/18068160194400919196noreply@blogger.com