"Why do we teach Statistical Hypothesis Inference Tests?"
with Paul Hewson, Plymouth University (Engand)
Hosted by: Staci White, The Ohio State University
Whilst the author would prefer all thing Bayesian, the requirements of serviced schools prevent this. Nevertheless, the standard inferential fare of t-test, ANOVA seems clumsy in a modern statistical literacy class, and seems to require many big ideas are taken on trust. Modern software makes randomisation tests trivial, but this requires that students have an understanding of computing. As a first exercise, we conduct actual and thought experiments in class, demonstrating threshold concept that under the null, group allocation should be arbitrary. Given that this works so well, we speculate why randomisation tests aren't routinely used in the remainder of the students early learning.
Having trouble viewing? Try: Download MPEG-4 Podcast (.mp4)
Comments
Emil Friedman:
Randomization tests were in chapter four of Box, Hunter & Hunters 1970's book Statistics for Experiments and their first mention of a hypothesis test (section 2.3) is conceptually similar to a randomization test.
Interestingly, my recollection is that we did not cover those chapters in the short course and semester course I took way back when using that text. At least one of the instructors said that those chapters were well worth reading but I suspect that only a handful of the students actually read them.
Paul Hewson:
Hi, thanks. I'll check out that text. As you say, interesting that it was there.
Laura Schultz:
The current edition of Moore, McCabe, and Craig's "Introduction to the Practice of Statistics" includes both randomization tests and bootstrapping confidence intervals in a supplementary (i.e., online) chapter. Also, the newest edition of the textbook that I use for Biostatistics (Samuels et al.) now introduces the logic of hypothesis testing via a randomization test. I hope that since randomization methods are finally finding their way into mainstream introductory statistics textbook, more instructors will start teaching these extremely useful modern techniques. I enjoyed your poster (and always like to hear from like-minded faculty who dare to teach randomization tests to introductory statistics students).
Paul Hewson:
Thanks, I hadn't picked up the online chapter (and I was using a different book, but another post-GAISE book). The irony of all this is that I reckon I have students who are clearer about what is going on as a result of working with randomisation tests, but the expectation is that they have to do t-tests (whether appropriate or not, whether they understand what's going on or not). You can see where Ziliak and McCloskey (The Cult of Statistical Significance) get their raw material.