In-Game Win Probability
test


Name:

Comment:


Refresh comments


Reno said...
For the love of God, keep writing these artlices.


Reno said...
For the love of God, keep writing these artlices.


tunesmith said...
This is a little tricky to figure out, but if WPA is consistent up until just before the Fells long pass, then I think we can make the following assumptions. From the 3 yard line, there's a TD possibility of 78% (although I think this assumes enough time on the clock for all four downs), and two-point conversions are what, 47%? and that's to get to a 46% WPA (Prater's kickoff). So, multiplying the three together gives a .168, and we started (before the long pass) with a .10 WPA. So if I'm looking at this right, that works out to a +.07 WPA for the long Fells pass. Then the TD was worth +.05 (to 0.22), and then the two-point conversion was worth 0.24 or so. That's a pretty huge play right there.


tunesmith said...
Ah... bummer. That makes it impossible to find per-play WPA for every play, at least from the xml. Still, it looks like it's counting WPA right for the players in question, like it appears Fells got a lot of positive credit for that catch (unless that 0.36 is just from the TD).


Brian said...
That's a quirk with my system. It is not very good (as in not good at all) at anticipating when a team will go for 2. It anticipates it until a team gets inside the 15, and then the system changes over to a Markov model where it is directly calculating the probability of a TD. It's at this point it loses it's ability to anticipate a 2-pt conversion until it actually occurs. But in the end, the erratic swing in WPA cancels itself out.


tunesmith said...
Why is the WPA behavior as it is in this three-play-sequence? 1:07 left, WPA goes from 11% to 19% on a short completion to Eric Decker. 1:02 left, WPA goes from 19% to 10% on an incomplete pass to Willis. 0:29 left, WPA goes from 10% to 1% on a long completion to Daniel Fells to the three-yard-line??? It seems 1% comes out of nowhere, why would such a positive play be worth -0.09?


James said...
Guess I was wrong!


Brian said...
3.7 AYPA will do it. His TD at the end put him in positive WPA territory, but the sack he took in OT dropped him.


James said...
How does Tebow have negative WPA? I don't think it's possible for him to cancel out the 0.45 WPA he earned on the final TD alone.


Jonathan said...
The CBS analysts are going off because, apparently, the Dolphins went for 2 and missed (up 12 instead of 13) and someone said that this is a "fireable offense." The entire crew is just ripping him. It was awful, even by MSM standards. They said "Miami would have lead by 9 at the end instead of 8," while missing the fact that Denver would have just went for 2 twice. So "outcome bias" should just indicate that you made them go for it instead of going for it yourself. I gleaned all of that in 30 seconds. Unbelievable.


49ersFan said...
I saw MIA with a win probability of 1.00 at one point. Does that mean this game has an infinite comeback factor?


Brian said...
True. The automatic algorithm isn't smart enough (yet) to know when teams are going to go for 2, so it assumes a XP.


Jonathan said...
In real time, WPA is a bit wonky in the middle of end game scenarios...it usually corrects itself.
0:17 in the 4th qtr

Jonathan said...
Woah, we have a candidate for the first (?) .01 WPA comeback of the year.
0:17 in the 4th qtr

James said...
Looks like WPA isn't sure how to handle 2-pt conversions, as the WP plunges as the Broncos score the TD.
0:17 in the 4th qtr





2011102300