Sports Reference Blog

Features: Expected Points

Posted by Mike on March 5, 2012

Today we're rolling out a new feature -- expected points on boxscores, year pages, and team pages back through 2000. What are expected points, you ask? Inspired by a concept introduced by Bob Carroll, Pete Palmer, & John Thorn in their 1988 book The Hidden Game of Football, our crack staff of statisticians (Sean and Neil) have used a decade's worth of play-by-play data to develop a formula that estimates the expected number of points given a combination of down, distance, and yard line. Let's look at this year's Super Bowl for an example. The first play after the kickoff looks like this:

Down ToGo Location Detail NYG NWE EPB EPA
1 10 NYG 23 Ahmad Bradshaw right tackle for no gain (tackle by Brandon Spikes) 0 0 0.48 -0.07
Provided by Pro-Football-Reference.com: View Original Table
Generated 3/2/2012.

That means that given a first and ten from their own 23, the expected value of the next score in the game is roughly 0.48 points by the Giants. With a run for no gain (resulting in second and ten from their own 23), the play changed that to -0.07 expected points, meaning that the play itself had a value of -0.56 expected points. Looking at the first touchdown of the game, we see:

Down ToGo Location Detail NYG NWE EPB EPA
2 2 NWE 2 Eli Manning pass complete short middle to Victor Cruz for 2 yards, touchdown 8 0 5.72 7
Lawrence Tynes kicks extra point good 9 0 0 0
Provided by Pro-Football-Reference.com: View Original Table
Generated 3/2/2012.

Before the play was run, the expected points for a second and goal play from the 2 yard line is 5.72 points, so scoring the touchdown there netted the Giants 1.28 expected points (note that for touchdowns, 7 points are assumed and a made extra point is not worth any additional points, though a missed extra point subtracts one and a made two-point conversion adds one).

What can we do with these? Well, when you sum all of the differences together (minus an adjustment for the end of half and end of game plays), you should get the scoring margin of the game, which you can see in the table below the box score:

Offense Defense Special Teams
Tm Total Tot Pass Rush TOvr Tot Pass Rush TOvr Tot KO KR P PR FG/XP
New England Patriots -4.00 10.70 9.71 0.99 -1.42 -13.35 -15.90 0.22 0.00 -1.47 -3.11 1.71 0.95 0.14 -1.16
New York Giants 4.00 13.35 15.90 -0.22 0.00 -10.70 -9.71 -0.99 1.42 1.47 -1.71 3.11 -0.14 -0.95 1.16
Provided by Pro-Football-Reference.com: View Original Table
Generated 3/2/2012.

Given this table, we can break down the contributions each team's various squads made to the margin of victory -- the Giants' passing offense was their largest positive contributor, while their rushing offense actually contributed negatively overall. In addition to the box score pages, team totals have been added to team and year pages so you can see what cumulative totals each teams' offensive and defensive squads have.

Keep an eye out for further features using this play-by-play data as the year progresses.

9 Responses to “Features: Expected Points”

  1. Topher Doll Says:

    I love all this data you guys are adding these new metrics are fantastic. One thing I've been wondering, though it would be a massive effort. When the Hidden Game of Football was written in 1988 the NFL was a slightly different game, it would be interesting to see an undated version of the book with modern statistics. Looks like you are already doing great work, and the NFL has only made minor changes since then, though some of their data would be grossly outdated in comparison to the modern NFL.

    As always, great work.

  2. James Says:

    The real question: How do Pro Football Reference Expected Points differ from AdvancedNFLStats Expected points?

    Is this going to be a thing like how we have FanGraphs WAR and Baseball Reference WAR? Are we going to get simultaneously cool and annoying acronyms like fWAR and rWAR? Maybe aEPA and rEPA?

  3. Dave D Says:

    Does this mean more PI sections are coming soon?

  4. Bill Says:

    It's interesting how little value is given to a successful coffin-corner kick in this system. Weatherford had negative punting value on three punts downed inside the 8 and one touchback.

  5. P Man Says:

    I'm guessing this means the expected points score on the very next PLAY, not the ultimate expected value of that DRIVE from that situation.

    For example, 1st and 10 at your 23 has an EPB of 0.48 points. Does this mean that on that 2nd down play ONLY, the average number of points scored (taking into account defensive points scored) is 0.48 points? And if so, it would seem like the majority of outcomes in that situation is zero.

    I can see the value I guess in that if you progress through a drive, you can see the value of each play in that drive. But I'm wondering if knowing what the outcome of that drive from that situation would be more informative. In other words, drives that start at the 23 net 2.4 points. At 2nd and 10, they net 2.1 points. And so on.

    Or maybe it doesn't make a difference?

    But I like baseball references win pct for each at bat. For example, leading off a game with a single moves you from a 50% chance of winning the game to a 54%. Leading off with a double moves it to 57%. Leading 1-0 after one inning gets you to a 60% chance.

    But maybe baseball is easier to categorize that way....?

  6. admin Says:

    It's the expected value of the next score to the team. So when you are on your opponents goal line 1st and 1. Your expected benefit is nearly 7 points because you are very likely to score the next touchdown. From other parts of the field, you may score a FG, or a TD or your opponent may score a FG or TD. This tries to weigh the probabilities of those various outcomes.

    Most punting plays are negatives. The team that punts sees their expected points do down by punting. Teams punt way too much. So Weatherford did as well as he could given the situation, but the situations were chosen poorly, imho. We are actually working on a punting metric comparing each punter to the league 99th percentile result for that field position.

    For example, in the 4th the NYG punted from the Pats 43 on 4th and 10. and the Pats got the ball at the 8. If you look historically, those to situations are about equivalent for the teams as to who will score next.

  7. Ben Says:

    Very cool. Especially enjoyed the play-by-play sheets. Looking forward to further implementations of this concept!

    My suggestions:

    -In the box containing game EPs for teams' passing, rushing, special teams, etc., add EPs for off/def penalties so we can see how much they cost the teams.

    -In the box containing stats for each of a team's games in a season (the one with the final scores, passing yards, rushing yards, turnovers, and now EPs) - include EPs for turnovers and penalties so we can see at a glance how costly they were to the teams who committed them in each game.

    It'd be also great to apply this to the final standing tables so we can see not only how good/bad the teams are but how much random (turnovers) and dumb (penalties) mistakes hold them back.

    Keep up the great work, guys!

  8. Mark Growcott Says:

    These new features are great especially the Play by Play data but what I would like to see and others have been requesting for a couple of years now is Team and Player Streak Finders which can be found at other S-R sites notably at the Basketball site.

    Also can we have the Play by Play data begin with who won the toss to determine whether they elected to Receive or if they deferred to the 2nd half.

    A few other suggestions - It would be great to be able to identify Left Handed QBs and also having Time of Possession data in the Box Score and have this as criteria when using the Team Game Finder.

    Keep up the good work.

    Thanks

  9. mick Says:

    need this for basketball. simpler, maybe less interesting but still needed.