About those "win probability" charts ...

[Quark3]

I've grown leery about these "win probability" ratings, they seem to underestimate  comebacks happening. Especially in situations where they say Team X has some outlandish chance of winning, like 98%.

For example, yesterday I attended the Saints vs Redskins game. With about 6 minutes to go in the game, the Redskins scored a TD to go up 31-16. The game was in New Orleans and my wife and I, who live in New Orleans but grew up in the DC area, were basically neutral because we have old loyalty to the Skins but have also adopted the Saints. As soon as the Skins scored that TD, about 17 of those 18 dedicated Saints fans in our row immediately got up and headed for the exits.

I told me wife "they are crazy, this game is far from over. All that needs to happen to send it to OT is (a) Brees zips the Saints down the field for a TD, (b) the Skins go three and out, and © Brees zips the Saints down the field for another TD, and on one of those TDs they get a 2-point conversion. The Saints can still very well win this game 34-31 in OT".

The Saints ended up winning 34-31 in OT, and it unfolded exactly as i said. Importantly, at the time, when the Skins were up 31-16, i told her it would NOT take some kind of miraculous Big Play, a kickoff return, a Pick Six, a Hail Mary at the buzzer, etc. to make it happen. Just ordinary football where twice Brees goes 7 for 9 for 80 yards and takes the Saints down the field for TDs in a couple of minutes each time. Ordinary football, not amazing plays. And that's what happened.

I'm sure the "win probability chart" at that six minute mark had the Skins at 98% chance of winning, but i clearly saw what i thought was a very credible path to a Saints win. So that leads me to believe that these charts lack some fundamental accuracy. 

I suspect these models take maybe too much data into account? What i mean is, 30 years ago, this kind of comeback would have basically never happened. But rules changes, such as the two-point conversion and the liberal passing rules, make them much more likely. If the models used to develop the win % are using data from more than 10 years back, they would likely be skewed such that they underestimate the chance of a comeback of this kind.

No, when I told my wife that scenario, i didn't *expect* the Saints to come back. The odds were clearly against it. But it wasn't 98% against it like the models probably showed. 

 

[Nozizaki]

It's still highly unlikely. It is exceptionally difficult for a team to drive down the field 2 times and get an immediate stop on defense. It requires almost everything to go your way, and that's just to get it to OT, at which point your win percentage is basically 50/50. Driving down the field against any NFL defense is difficult, that's what the win probabilities indicate. If Washington got a sack, forced a fumbled, picked off a pass, stopped that 2 point conversion, picked up one first down, got a decent kick return... the game is basically over.

[childofpudding]

I believe the models are based on past games where those situations - or very close to them - existed. Where you can argue they might be off is that the models account for all of those situations regardless of personnel. Having Brees at QB alters that, but probably not as much as you think. Like the poster above me said, it also depends on getting a 3-and-out from a defense that had already given up 31 points and a lot of yards.

[scar988]

IT was definitely 98% against it. Because in 98% of those 31-16 games with 6 minutes left, the team up 31-16 wins.

[Soko]

The numbers are dumb. 

They change as the time/score changes, so they essentially mean nothing.

[pwny]

2 minutes ago, Yin-Yang said:

They change as the time/score changes,

..... well yes. That doesn’t mean they mean nothing though. It’s a model based on that current situation. As things change, so does the model. 

[Soko]

1 minute ago, iPwn said:

..... well yes. That doesn’t mean they mean nothing though. It’s a model based on that current situation. As things change, so does the model. 

Yeah, that’s worthless to me. 

So in the 3rd quarter of the SB I imagine the percentage would be 99.9% Falcons victory or something (similar to what it was in a 2 score game with the Skins-Saints). But by OT, I’d guess it was closer to 50-50 odds or maybe even weighted in the Patriots’ favor due to momentum. So what was ever the point of seeing 99.9% earlier? 

It doesn’t do anything. Gives no prediction of who’s going to actually win (whomever is winning is more likely to win....#advancedstats). 

[Adrenaline_Flux]

ITT: People don't understand probability or confirmation bias.

[wackywabbit]

I hate that they usually include some win probability going into the game (i.e. they don't start at 50/50 odds). They are a model for how a game went, so just make it about that game. Especially since there isn't a good way to capture strength going into a game given recent injuries, etc; just remove that variable. 

[pwny]

16 minutes ago, Yin-Yang said:

So what was ever the point of seeing 99.9% earlier? 

Because given the previous situation, 99.9% of teams with that lead in that situation would have held that lead. Just because the Falcons were the 1 out of 1000 that didn’t, doesn’t make the model useless.

[Shanedorf]

2 minutes ago, iPwn said:

doesn’t make the model useless.

"All models are wrong; some of them are useful "

Since we can't predict the future, we create models based on past experiences and use them to extrapolate the future....

Its really not a difficult concept....

[Soko]

13 minutes ago, iPwn said:

Because given the previous situation, 99.9% of teams with that lead in that situation would have held that lead. Just because the Falcons were the 1 out of 1000 that didn’t, doesn’t make the model useless.

The “win probabilities” are insanely high all the time. OP said the Saints-Skins game was within the ballpark of 98% in a two score game. Some part of me doubts that the Saints has a 2/100 chance of coming back. 

You can say the home team has x% win probability because they’re the home team and historically home teams win more often. Still a crappy way to predict a game. 

Useless model to predict wins.

[pwny]

33 minutes ago, Yin-Yang said:

Some part of me doubts that the Saints has a 2/100 chance of coming back. 

A 15 point game with 3 minutes left. Yeah, teams don’t complete that kind of comeback often.

[Jlowe22]

The probabilities are probably based on NFL averages, they go up some when you have a team and a QB that excels at throwing the ball, and a sketchy defense.

Still, it's rare.  15 points, not 14.  That means drive the length of the field quickly, get an immediate defensive stop, drive the length of the field quickly, score a two point conversion.  Make a stop in overtime, drive the length of the field for a field goal.

There was also one more stop the Saints needed to make, since Saints scored so quickly Reskins still had enough time to drive for a field goal.  They probably would have gotten the try off, if not for Cousins intentional grounding. 

There are 250 some odd NFL games played every year.  Something with 1/100 chance has a good chance of happening more than once over the course of a season.

Edit:  When discussing the comebacks, you have to narrow it down to games where a comeback was actually needed.

[Sugashane]

15 seconds in it shows us with a 3% chance to win since we had Fox become our coach.

 

Been pretty accurate so far. lol