@WoodyVU66 alluded to the the FAMU-ISU upset last night, and ESPN was making much of the supposed magnitude of this upset in their crawl. But to me the numbers don't really add up.
SHORT QUESTION: Does anyone know where they got this "biggest upset since '08 season" thing? Betting lines?
It's really hard to judge whether a probability is "correct" for a one-off event. With lots of data, though, you can calibrate. If you assess 1000 events at 1% each, you'd expect around 10 of them to happen. If 50 of them actual happen, you're almost certainly too pessimistic. If only one does, it's less clear but probable that you're too optimistic.
(Simple arithmetic ahead....)
How many games each year involve a team of something like a 1.0% to 1.3% win probability? Iowa State was no great shakes but FAMU is pretty bad. Conservatively maybe half the top 50 plays a game that mismatched or worse. (Most of the top 50 is better than ISU.) So, 25 games a season for 12 seasons. 300 games.
We'd expect to see four bigger upsets out of those 300 games. Instead we supposedly got zero.
Couple of reasons why it's not as miscalibrated as I thought:
SHORT QUESTION: Does anyone know where they got this "biggest upset since '08 season" thing? Betting lines?
It's really hard to judge whether a probability is "correct" for a one-off event. With lots of data, though, you can calibrate. If you assess 1000 events at 1% each, you'd expect around 10 of them to happen. If 50 of them actual happen, you're almost certainly too pessimistic. If only one does, it's less clear but probable that you're too optimistic.
(Simple arithmetic ahead....)
How many games each year involve a team of something like a 1.0% to 1.3% win probability? Iowa State was no great shakes but FAMU is pretty bad. Conservatively maybe half the top 50 plays a game that mismatched or worse. (Most of the top 50 is better than ISU.) So, 25 games a season for 12 seasons. 300 games.
We'd expect to see four bigger upsets out of those 300 games. Instead we supposedly got zero.
Couple of reasons why it's not as miscalibrated as I thought:
- Small sample. Four expected upsets isn't that much.
- I figured there'd be like 1000 such games, not 300. Maybe it's somewhere in between.
- Structurally the nature of college basketball may be that 1.3% is about as low as you can go. Maybe there just aren't that many 1.2s and 1.1s as candidates for an upset. So maybe even 300 candidates is too high.
