[2006alum]

That said, ESPN's win probabilities are often absurd, very frequently proven wrong, and I don't even know why they bother. It's like the algorithm has never heard the (tired) adage, "basketball is a game of runs." One and a half minutes before Virginia Tech was up 13, it was up 5. So going from up 13 to up 4 in another one and a half minutes is very predictable. From there, it's probably 30/70 to go from up 4 to down 1 in another one and a half minutes.

Except. Of course, a bunch of times that have been totally forgotten the miracle threes stop falling or the team ahead makes its free throws or just as mysteriously as it started, the run ends.  I can’t vouch for the math that ESPN uses; it could easily underestimate tail risk which happens a lot in these sorts of models that assume certain distributions of data that don’t wind up being accurate.  On the other hand, they probably have a database of thousands of games (there are 5800 D1 games a year), so having a team up by a dozen or more points with three minutes left is probably not an unusual situation.

BC

Here's ESPN's predictions in just the past two days:
MBB, TCU v. Kansas: TCU 98.6% chance of winning with 12 seconds left up. Loses by 4 in OT.
WBB, Ore. v. U$C: U$C 95.% chance of winning with 3 minutes left; loses by 5.
WBB, UNC v. Stanford: UNC 94.3% chance of winning with 1.5 minutes left; loses by 6 in OT.

I'm just saying, if I were a betting man, after consistently seeing many swings like this, ESPN's 90%+ assurances would not coax me into betting even a single dollar. Basketball is a game of runs and anything less than 10 points in the last minute cannot be considered "safe," let alone 98% assurance.

[2006alum]

That said, ESPN's win probabilities are often absurd, very frequently proven wrong, and I don't even know why they bother. It's like the algorithm has never heard the (tired) adage, "basketball is a game of runs." One and a half minutes before Virginia Tech was up 13, it was up 5. So going from up 13 to up 4 in another one and a half minutes is very predictable. From there, it's probably 30/70 to go from up 4 to down 1 in another one and a half minutes.

Except. Of course, a bunch of times that have been totally forgotten the miracle threes stop falling or the team ahead makes its free throws or just as mysteriously as it started, the run ends.  I can’t vouch for the math that ESPN uses; it could easily underestimate tail risk which happens a lot in these sorts of models that assume certain distributions of data that don’t wind up being accurate.  On the other hand, they probably have a database of thousands of games (there are 5800 D1 games a year), so having a team up by a dozen or more points with three minutes left is probably not an unusual situation.

BC

Here's ESPN's predictions in just the past two days:
MBB, TCU v. Kansas: TCU 98.6% chance of winning with 12 seconds left up. Loses by 4 in OT.
WBB, Ore. v. U$C: U$C 95.% chance of winning with 3 minutes left; loses by 5.
WBB, UNC v. Stanford: UNC 94.3% chance of winning with 1.5 minutes left; loses by 6 in OT.

I'm just saying, if I were a betting man, after consistently seeing many swings like this, ESPN's 90%+ assurances would not coax me into betting even a single dollar. Basketball is a game of runs and anything less than 10 points in the last minute cannot be considered "safe," let alone 98% assurance.