What are the chances? A look at Win Probability and Graphs

Advanced NFL Stats is a website I’ve grown to like recently. They do a lot with NFL stats and don’t try to bludgeon you over the head with one of them (I’m looking at you, DVOA.) One of the things I particularly like over there is their win probability graphs. I’ll tell you how the graph works (it’s mostly intuitive), but first I need to explain win probability.

On the face of it, win probability is also intuitive; it’s the chance that a team will win the game at any single point in the game. The complexity comes from how it’s calculated. First, all plays were separated into groups based on all their factors (game score, down and distance, time left, and position on the field.) The win probability for any future plays that match those factors is the previous chance that a certain team won. Jump it for some examples!

Suppose the Steelers are down by three, with 2:00 to go in the first half, and they have a first-and-10 on their own 35 yard line. The chance the Steelers win is 42%. This tool is great, by the way.

The one drawback of this is obvious: there’s no built-in talent factor. All teams start at a 50% chance to win the game, when that’s almost always not the case. Take their game probabilities, for example. That might be something I’ll try to tackle during the offseason, though. Stay tuned.

The graph gives you a visual sense of the win probability throughout the game, and you can hover over it to find out which plays moved win probability the most. I’ve found that those are usually scoring plays that give the scoring team the lead.

For a good example of how these graphs work, let’s look at this one, sickening as it may be. This was a classic back-and-forth close game where no team was ever ahead by more than a touchdown. It also shows the effect of one play late in a game, where Joe Flacco’s pass to T.J. Houshmanzahdeh took the Ravens’ chances of winning from 40% to 92%. Meanwhile, Rashard Mendenhall’s touchdown run earlier in the fourth quarter only took the Steelers’ chances from 63% to 72%, because they were already likely to at least tie.

After each Steeler game, I plan to provide a link to the win probability graph and highlight the key plays from the game. During the offseason, I’ll take a look at the best games of the decade, according to their “comeback factor” and “excitement index”, which I’ll explain at that time.

Since you won’t be engrossed in the Steelers game this week, but you’ll still probably watch football anyway (“I need Jahvid Best to come through for my fantasy team!”), I suggest watching the games on TV, and having the win probability graphs among the sites you look at on your computer (along with your fantasy football matchup, of course.)