Realtime Win Probabilities

If you haven't checked out the win probability site during a game in a while, you should check it out. The win probability model itself has been improved incrementally over the past few weeks, plus I've added a few additional features.

For each upcoming play, the probability of a first down is displayed. The Expected Points for each field position, down, and distance are also updated live. Additionally, the probabilities of the current drive culminating in a touchdown and a field goal is displayed.

Any questions, suggestions, or other comments are welcome.

  • Spread The Love
  • Digg This Post
  • Tweet This Post
  • Stumble This Post
  • Submit This Post To Delicious
  • Submit This Post To Reddit
  • Submit This Post To Mixx

5 Responses to “Realtime Win Probabilities”

  1. JJB says:

    I have a quick question. Is there any significance to the "area under the curve" (or whatever the appropriate terminology is for a discrete graph) in the win probabilities? EG, In the CAR/NYG game Sunday night, the area under the curve on the CAR side is MUCH greater than the NYG side, up until the final drive in OT. To me this would indicate that CAR was in control for almost the entire game. And the GB/CHI Monday night game is similar. Yet all the sports yaks are talking about how NYG dominated CAR with the running game. I'm wondering if I'm missing something, or if they are ?

  2. oberon says:

    As usual, they are. The area under the curve (can also be expressed as "average win expectancy over time") is a useful measure of how two teams played in a game, understanding that its predictive power is not necessarily strong. Of course, this curve is not gospel -- it's theoretically plausible that a better model would have had NY in control of the game -- but this model is pretty well refined.

  3. Brian Burke says:

    The WP model is actually pretty simple. It just the likelihood that each team would win, given the game situation at the time (score difference, time remaining, field position, down/distance).

    For example, throughout the 4th quarter, the WP for the Packers bounced around 0.80. This means that a team in their situation would end up winning about 80 out of 100 games.

    It got as high as about 0.90 for Green Bay, which means something really dramatic and uncommon would need to happen for Chicago to win. And that's what we saw. A TD drive, a blocked kick, and then a very quick FG drive in OT was required for the Bears to win.

  4. Lance from Texas says:

    Nice site, here's my question: I'm guessing that you're giving all the team's an equal value, so couldn't generate a model that would allow for a team's strengths vs. the other teams weaknesses and the place the probability filter on top of that, ie CHIC ST-unit vs GB ST-unit in that situation,at that spot, at that temp, with the wind...etc.?

  5. Brian Burke says:

    Lance-Thanks. You're right. The WPs are generic. It would be possible to do all those things with enough data (and time to crunch it). Unfortunately, even 8+ seasons of NFL games slices up the data into very small cells. For example, how many games since 2000 do you think were:
    - 7 point lead
    - 6:30 left in the 4th
    - 2nd and 7
    - at the 50 yd line?

    One. So I have to do a lot of chunking, interpolating, and smoothing of the data to get reliable WP estimates, and the basis for the estimates can get pretty thin. To divide it up further according to team strengths or wind or temp would make it way too thin. Special teams would be especially tough. They are so unpredictable.

    That's a great goal though. Maybe one day. I think adjustments based on team offense vs. defense would be the first place to start.

Leave a Reply

Note: Only a member of this blog may post a comment.