This week's show
Dave Berri stops by to talk some common issues in regression analysis and expand on issues with evaluating predictive and descriptive stats.
A bit of background. Last week I wrote an article, linked below. We recorded a podcast discussing it, and a lot of the audio was eaten due to technical difficulties. Don't worry, we're working on salvaging what we can. Luckily, Dave Berri was willing to help us out, so you get a bonus episode! Because of the work behind editing stuff, I'll be releasing the bonus episode first and original episode second? It worked for Star Wars! Enjoy.
- Special Guest: Dave Berri (@wagesofwins)
- Andres Alvarez (@nerdnumbers)
- Produced by Brian Foster (@boxscorebrian)
You can watch us live at twitch.tv/nerdnumbers; we usually go Tuesday or Wednesday at 9:30 pm EST. Follow us on Twitter and we'll keep you posted!
Please subscribe to Channel NerdNumbers on YouTube and like us on iTunes!
Profesor Dave Berri stopped by to give a "guest lecture" / interview on regression analysis, using the Nylon Calculus article on Dredge we discussed last week.
As a note, Dave Berri is a tenured professor of economics at Southern Utah University. He has two published books "Wages of Wins" and "Stumbling on Wins". As of this recording he has over 60 published papers in academic publications on sports economics. He regularly teaches regression analysis. In short, he's an expert in this field. As I mention on the show, if you want to disagree with Dave Berri using boxscore statistics and regression analysis as a technique for analyzing sports (obviously we think this is fine) then that's ok. However, if you are going to talk using regression analysis, well Dave Berri knows a thing or two!
Dave Berri discusses many textbook issues with bad regression analysis. It's also worth noting Dave has written a textbook on the subject, which should be out in a year or two.
Full write up of the Wins Produced formula here.
We discuss some of the issues behind the Adjusted Plus-Minus Statistic. Including the issue that it does a poor job explaining what it's supposed to.
We also discuss Regularized Plus-Minus and how in the initial Sloan paper it still doesn't do a good job explaining things by its own volition!
We discuss some issues with misusing the standard error term in regression analysis.
Dave also discusses some issues with multicollinearity and how using larger samples will mask the problem, without actually solving it.
In a problem we will get back to in future posts, Dave brings up the issue of misspecifying a model. Tune in!
It makes sense that this podcast ends with us discussing the merits of Wisconsin and Utah as places to live. A note, in Michigan, if you live on the upper peninsula you are called a "Yooper" and a "troll" if you live in the lower peninsula.
Dave explains why Utah is a lot like Colorado without the traffic. On that note, let's end this show with a poll:
Where would you rather live?— Andrés Alvarez (@NerdNumbers) August 15, 2016