WYNC's Radiolab on Streaks - and the non-independence of individual events

[1e Corbeau]

Radiolab, what I consider the best show on radio these days, did an hour-long episode on stochasticity a few weeks ago. Jonah Lehrer (Proust was a Neuroscientist) touched on streaks in sports. He made some assertions that people were coins and behaved as such: athletes were going to achieve at their base rate (shooting %, average, etc.) when taken as individual events - i.e. an athlete isn't more likely to make a shot if they're "hot."

 

The follow up 18:46 below deal with reversing this assumption: an analyst talks about simulating streaks, such as DiMaggio's 56-game hitting streak, with historical data/probabilities. The big news to me (not a statistician, but I enjoy math and have a few years of calculus/Diff EQ somewhere in the back of the memory) was that Trent McCotter has "proved" that individual events are not independent of the previous outcome, specifically with respect to baseball. It seems, from cursory googling, that McCotter has won SABR Jack Kavanagh awards the previous 4 years.

 

Link to the MP3 or visit the episode page to listen to embedded flash audio.

[rluzinski]

Thanks for the link. I'll have to check it out.

 

"The Book" provided evidence that players in the midst of a hot streak do in fact hit better than their expected performance, but the difference was quite small. I've read other studies that came to similar results. Assuming a constant probability of success isn't 100% right but it's a damn good estimate.

[logan82]

I never really liked the coin flip analogy. I always preferred to think of baseball players more like 20 sided dice. For a .300 hitter their normal performance would be a 1-6 being a hit. They have up days where maybe a 1-8 gets them a hit or down days where a 1-4 gets them a hit. Still a very crude way to look at things given all the variables that go into hitting.