Two big new additions to the in-game win probability site this week. An anonymous commenter suggested adding the expected points of a current drive, so I've done just that. For now, the expected points are for the standard drive and do not account for lack of time remaining in a half, but that won't be hard to improve down the road. I've also added the probability of a first down in the current series.As Carl Spackler would have put it, we now have total football consciousness. So I got that goin' for me, which is nice.
And if you missed it last Monday, I was able to rework how the page updates. It's now smooth and seamless. No more hourglass icons while the page waits to refreshes. No more flicker when the graphs are redrawn. And no more re-scrolling down to the game you were following.
The model itself was patched up a little. I had neglected to finish smoothing the data for situations where a team was trailing but was in the red zone. Now the output won't be as erratic as it was in those cases.
Eventually, I'd like to add a capability to see descriptions of the plays behind the probability graphs. For example, you could hover over a spike (or any part of the graph) and you'd see a small pop-up/tooltip-style description of the play. I think it would add a sense of what really changes a team's chances of winning. ("Wow, getting that last first down is really what put the game away.")
Lastly, congratulations go to commenter and longtime reader 'Chase' who correctly guessed the Play of the Week for week 8. The Kansas City 91-yard touchdown return of a Brett Favre interception was the play with the biggest swing in WP.
This is a fun project. Spread the word.
Does this model factor team quality, or is it the expected outcome for two equal teams? THanks.
ReplyDeleteThis is general.
ReplyDeleteMajor update installed for the Sun night game. Now 4th downs are accounted for using a Markov-like process.
ReplyDeleteFor now, it's a relatively straight forward adjustment. For 4th downs inside FG range (defined as the 35 based on historical tendencies), the system calculates the WP if it's good and the WP if it's missed. Based on the probability the kick will be good, the associated WPs are weighted proportionately and the total expected WP is calculated.
For outside FG range, the system knows the average net punt distance for punts from every yd line on the field. The WP for the other team having a 1st and 10 at that distance is calculated.
Next, I'll factor in adjustments for 3rd and 2nd downs.
I hope everyone digs the expected points and 1st down probability additions.
Wow, looks like BAL-CLE and DET-CHI were pretty exciting games to watch. Is that DET win probability spike at almost 1, as in 100%?
ReplyDeleteOakland was almost flat-lined for much of the game.
ReplyDeleteMIA-DEN decide not to show up for the first 5 minutes?
Kunga Galunga. Kunga Galunga Lunga!
ReplyDeleteHey Brian,
ReplyDeleteKing Kaufman over at Salon.com got me to thinking about the value of a timeout. Surely down by 1 w/2 minutes to go and 3 timeouts should have a different win probability than with 0 timeouts. You probably have the data to evaluate this, and the generic value of a timeout would be interesting in and of itself, esp. with respect to failed challenges.
Hey Biarn:
ReplyDeleteHow it is you not to ever respond to the comments as made by Borat?