Expected Points Added in Football

Can any of you explain the football statistic, expected points (EP) and expected points added (EPA)? I read an explanation, but I’m still confused. Here’s the section where I get lost

We can measure the values of situations and, by extension, the outcomes of plays by establishing an equivalence in terms of points. To do this we can start by looking back through recent NFL history at the ‘next points scored’ for all plays. For example, if we look at all 1st and 10s from an offense’ own 20-yard line, the team on offense will score next slightly more often than its opponent. If we add up all the ‘next points’ scored for and against the offense’s team, whether on the current drive or subsequent drives, we can estimate the net point advantage an offense can expect for any football situation. For a 1st and 10 at an offense’s own 20, it’s +0.4 net points, and at the opponent’s 20, it’s +4.0 net points. These net point values are called Expected Points (EP), and every down-distance-field position situation has a corresponding EP value.
(Note: I’m going to quote sentences from above and ask questions after them.)
We can measure the values of situations and, by extension, the outcomes of plays by establishing an equivalence in terms of points.
I just wanted to be sure–“points” refers to scoring, not a point or number system (e.g., 1.0)
To do this we can start by looking back through recent NFL history at the ‘next points scored’ for all plays. For example, if we look at all 1st and 10s from an offense’ own 20-yard line, the team on offense will score next slightly more often than its opponent.
If I’m understanding this correctly, they looked at all the 1st and 10 situations on an offense’s on 20 yard line, and counted how often they went on to score on that possession. For example, if there were 100 instances of this situation, and if an offense scored on 50 of these, then I guess they’d say the EP for 1st and 10 on one’s own 20 would equal 50% or .5? But I’m a little confused about the “slightly more than it’s opponent” line.” I assume this means that the team has a better chance because the other team doesn’t have the ball at that moment. Does EP factor in this relative difference? I guess they could do that by looking at how often a team scored when they started a drive defending an opponent on 1st and 10 on the opponent’s 20. For example, suppose on 100 of these 1st and 10 situations, the opponent ended up scoring 10 times, which would mean the opponent’s EP would be 10% or .1? Let me pause here for a moment, and examine whether the process is sound–specifically the notion of looking at all the combinations of downs, distances, and field positions and then creating a probability based on the number of times a team scored over during those situations. This seems to treat each specific instance as equal to all others. That is, it ignores the differences of those situations, e.g., the specific teams and how good they are. The 2018 Chiefs chances of scoring isn’t the same as the 2018 Cardinals. And it would also make a difference if the former faced the 2018 Ravens defense or the 2018 Buccaneers defense. Shouldn’t the EP take these differences into account? And would this change if the sample size was extremely large? (It seems like it wouldn’t.)
If we add up all the ‘next points’ scored for and against the offense’s team, whether on the current drive or subsequent drives, we can estimate the net point advantage an offense can expect for any football situation. For a 1st and 10 at an offense’s own 20, it’s +0.4 net points, and at the opponent’s 20, it’s +4.0 net points. These net point values are called Expected Points (EP)…
OK my definition of EP above seems wrong. EP seems to be taking the probability(?) of one team scoring versus the opposing team and then subtracting the two. If my understanding is correct, then the team with the ball should almost be on the positive side. (Thought: Supposed team has the ball on their own one. I would expect EP to be low, but I’m now wondering if the drive/possession should be the “end point,” if that makes sense. What I mean is that, while the EP for the team that has to ball may still be on the plus side, for that drive, if you go beyond that drive, you could argue that when a team starts at their own 1 yard line the opponent’s EP would ultimately be better. Then again, if the team with the ball gains yards and has a good punt, that would not be the case. Also, if the opponent gets the ball back with good field position, the EP at that point should reflect a high probability of scoring.) Another question: Earlier, I assume that the EP is a probability, but I’m assuming that is now wrong, since an EP can be over 1.0. Is that correct, or am I’m wrong about this? More later.

7 thoughts on “Expected Points Added in Football

  1. I haven’t read this whole thing yet, but does it change your understanding at all if I point out that you seem to be misunderstanding the beginning? Early in your response you say

    If I’m understanding this correctly, they looked at all the 1st and 10 situations on an offense’s on 20 yard line, and counted how often they went on to score on that possession.

    But the beginning of the explanation says

    if we look at all 1st and 10s from an offense’ own 20-yard line, the team on offense will score next slightly more often than its opponent.

    Without context (I’ll read the linked article later), they’re saying the team with the ball is slightly more likely to score NEXT, not slightly more likely to score on this drive. I would expect writers explaining their stats to be clear about “next.” Next as in “on that play” or next as in “during that possession” or next as in “the next points to go on the board.”

    1. I haven’t read this whole thing yet, but does it change your understanding at all if I point out that you seem to be misunderstanding the beginning?

      It definitely could–and I’d want you to tell me if you think I’m misunderstanding something.

      I would expect writers explaining their stats to be clear about “next.” Next as in “on that play” or next as in “during that possession” or next as in “the next points to go on the board.”

      This is a good point. Either I didn’t read the article carefully, or they didn’t specify. I assumed they meant score on the next possession, but maybe they mean score on play….But “next” sounds like they mean during that drive or possession. If they meant scoring which team would be more likely to score on that play then I would think they would say that; using “next” seems odd.

      (By the way, I’m writing questions as they occur while I’m reading the article. That is, I didn’t read the entire article first and then ask questions. For what that’s worth.)

  2. I thought I’d tweet the writer and just ask, but his Twitter account is suspended and the website’s FB hasn’t been updated in five years. So I searched elsewhere for a good explanation. This ESPN article from 2012 says

    The value it puts out is on a scale from about minus-3 to 7, and it basically represents “which team is likely to score next, and how many points?” It represents the likely points not just on the current drive but also on the next drive or any subsequent drive until the score changes or the half ends.

    So that answers my question. I’ll finish reading the other material later. It’s an interesting concept.

    1. I’m confused at how it would extend beyond the current drive…or is the EP something that is changing as the game goes on? And there’s only one EP, not one for each team, right?

    2. Yeah, you got it. There’s an Expected Points for each situation in the game, and it changes for each change in down, yardage, and field position, and it’s a constant, meaning that whatever the EP is for 1st and 10 on your own 20, it’s the same EP every time it’s 1st and 10 on your own 20 for every team in the league.

      And yeah, there’s one EP and it’s for the offense, since we’re measuring offensive effectiveness, if I understand correctly, although you could conceivably use it as a defensive metric as well, based on how well the defense lowers the offensive EP. I guess that would be Expected Points Subtracted.

      I hope this doesn’t throw you off, but this concept reminds me of the way we get the “chance of precipitation” in the weather forecast. When the forecast says there’s a 100% chance of rain tomorrow, the meteorologist isn’t saying it is absolutely certain to rain. He or she is saying that historically, when tomorrow’s expected conditions existed, it has rained every time. So you can see that using historical data to predict outcomes is not foolproof, but it gives you a baseline of what you can expect.

      1. And yeah, there’s one EP and it’s for the offense, since we’re measuring offensive effectiveness, if I understand correctly, although you could conceivably use it as a defensive metric as well, based on how well the defense lowers the offensive EP.

        I’m a little confused by this….I think you’re saying that EP is more about the offense–because, really, EP is a number that represents a relationship–the chances that the team with the ball with the ball will score next versus the chances the team without the ball with score next.

        He or she is saying that historically, when tomorrow’s expected conditions existed, it has rained every time. So you can see that using historical data to predict outcomes is not foolproof, but it gives you a baseline of what you can expect.

        I think I understand. But this raises a problem I mentioned earlier. The historical data in football involves different teams–in terms of personnel, quality, style, etc.–in different combinations. The approach seems to treat all the teams as if they’re basically the same, which doesn’t seem sound. When a great offense faces a the worst defense, is the EP really the same when the worst offense is facing the best defense? I mean, I know the number will be the same–but is this is a sound approach?

        Another question: How did they get the -3 to +7 scale?

  3. On to the concept of Expected Points Added or EPA:

    Suppose the offense has a 1st and 10 at midfield. This situation is worth +2.0 EP. A 5-yard gain would set up a 2nd and 5 from the 45, which corresponds to a +2.1 EP. Therefore, that 5-yard gain in that particular situation represents a +0.1 gain in EP. This gain is called Expected Points Added (EPA). Likewise, a 5-yard loss on 1st down at midfield would create a 2nd and 15 from the offense’s own 45. That situation is worth +1.2 EP, representing a net difference of -0.8 EPA.

    Basically, EPA is the extent to which a play has either improved or hurt the EP. The more yards gained, the greater chances for scoring next; hence the better the EPA.

    Based on the way I’ve seen people use EPA, they’re basically using this as a primary metric to evaluate an offense or a specific offensive play, although I’m not entirely sure about this. Actually, the article lists some specific uses of EP and EPA:

    EP and EPA have a variety of applications. We can use EP to measure and compare the relative value of runs vs. passes in various situations. We can tally up the EPA for individual players and for teams for a more accurate valuation than what traditional stats can tell us. Perhaps the most useful application of EP is in the analysis of fourth down decisions, which suggests teams should be going for it far more often.

    The problem I sense is putting everything on scoring–specifically, which team will score next. I don’t think offense or the game can be reduced to this metric. The other variable(s) involve time and snaps. There is value in a team’s offense’s ability to run a lot of snaps and consume a lot of time (especially when protecting a lead) because this will contribute to reducing the team’s defense’s number of snaps and game time they’ll play. I don’t think any other sport has this exact feature. (It can occur in basketball, but in basketball, those who play on defense are the same people playing on offense.) My sense is that EP and EPA ignores this dimension.

    Here’s an indication that this occurs, and the way they go about ignoring the time and snap dimensions:

    But there’s a problem. We all know that being up by 1 point at the end of a game is just as good as being up by 50. Not all points are equally valuable. Teams well ahead will sacrifice point advantages in exchange for running time off the clock, which in the end helps them win.

    To mitigate that problem, the baseline EP values for each down-distance-field position situation must be created based on real game situations when points are equally valuable and time is not yet a factor. The baseline EP values are therefore based only on game situations when the score was within 10 points and in the first and third quarters. This eliminates situations like ‘trash time,’ and other distortions.

    The parameters of game situations where score is within 10 points and in the first and third quarters seem extremely narrow. How useful is this. If the measurements can’t be applied in the 2nd and 4th quarters that seems to really limit the value of these stats.

    I’m not clear on why 10 points is a paramter. In a 14 point difference during the 1st quarter, why wouldn’t using EP and EPA be appropriate?

Leave a Reply

Your email address will not be published.