not with 95% confidence

December 15, 2015 6:26 AM

The Student of Student's t-test
posted by the man of twists and turns (15 comments total)

Statistics is the only math I've taken since around 4th grade that I've ever used outside of a classroom. My stats teacher mentioned in class that the creator of the Student's t-test had worked for Guinness, but this filled in a lot more details. Very interesting article.
posted by ralan at 7:02 AM on December 15, 2015

This is most appropriate given that I have a Biostats final in 3 hours (and yes, our prof did talk about this story in our class as well).
posted by Ufez Jones at 7:10 AM on December 15, 2015

Nice post. The guy running the Brulosophy blog uses the single-sided version of this statistical test when trying different techniques in homebrewing beer.
posted by exogenous at 7:19 AM on December 15, 2015

As an engineer who used design of experiments on a daily basis in my work, I loved this article. Some engineers scoff at statistics and Design of Experiments (DoE), but they were really the most interesting part of my education. Engineers spend 4+ years learning how to use calculus and mechanistic models, but it never tells you if the real widget or process or novel material in front of you will work in the real world, or if it's better than the alternatives. That's what statistics can tell you - it's measuring the real world and finding out what's true. What could be more cool or useful?
posted by permiechickie at 7:21 AM on December 15, 2015 [6 favorites]

Yes! In college we all had to use this book.
posted by exogenous at 7:29 AM on December 15, 2015

How did a brewer of dry stout revolutionize statistics? And why is he so little-known?

He's not really little known, amongst people who know statistics (and not just Stata ,robust monkeys). Its often mentioned in histories of the field and is a favorite tidbit of many TAs. Good article nonetheless.
posted by MisantropicPainforest at 7:31 AM on December 15, 2015 [4 favorites]

The 20th century's technological advances are pretty clear to everyone, but so much mathematics was discovered then too. Statistics, information theory, non-paradoxical set theory, chaos theory, fractals...
posted by Rangi at 7:32 AM on December 15, 2015 [2 favorites]

I love how he wasn't a fan of p-values! Wish I'd known this back when my intro stats prof introduced him and then proceeded to teach statistics as a set of ironclad (yet poorly justified and often contradictory) mathematical rules. Took me years to recover and be able to use statistics competently and confidently in my work.
posted by congen at 7:58 AM on December 15, 2015 [2 favorites]

The idea of a p-value is perfectly sensible, the problem is the arbitrary decision that p ≤ 0.05 is the threshold between significant and insigificant.
posted by Elementary Penguin at 8:03 AM on December 15, 2015 [2 favorites]

My statistics professor in college said that p-value meant 'the publish it! value'.
posted by Doroteo Arango II at 8:22 AM on December 15, 2015 [6 favorites]

Elementary Penguin: the idea is perfectly sensible, but unfortunately not only is the arbitrary significance threshold an issue, but also a lot of people don't understand what the p-value is.
posted by edd at 8:33 AM on December 15, 2015

Some engineers scoff at statistics and Design of Experiments (DoE) ...

This. As I sometimes need to tell our engineers: "Anybody can make 1 unit work; show me how you're gonna make 1,000 units work!"
posted by ZenMasterThis at 8:40 AM on December 15, 2015 [3 favorites]

John D. Cook points out why it makes sense that a brewer, not a vintner, came up with this. (The OP links to this but it's a bit buried.)
posted by madcaptenor at 9:08 AM on December 15, 2015 [1 favorite]

Previously on Deming (mentioned in the OP).

Great post. Makes me want to get a pack of Guinness next round of grocery shopping.
posted by JoeXIII007 at 11:41 AM on December 15, 2015

I wish this article had come out a couple of months ago when I was teaching the t-test! I mentioned offhand that a Guinness employee had discovered the t distribution, but I didn't know enough myself to say much more. Bookmarked for next semester.

unfortunately not only is the arbitrary significance threshold an issue, but also a lot of people don't understand what the p-value is.

More broadly, a lot of people don't understand conditional probabilities, and the p-values of hypothesis tests are conditional probabilities. Krämer and Gigerenzer's paper How to Confuse with Statistics or: The Use and Misuse of Conditional Probabilities is a nice and not too technical read on a number of ways that conditional probabilities can confuse.
posted by egregious theorem at 2:05 PM on December 15, 2015 [4 favorites]