By Alex Vigderman
Welcome to the new SIS DataHub!
We’ll be offering lots of tips, tricks, and explanations of the different stats on the site, but let me take some time to talk about one stat (or suite of stats) in particular.
If you go to any of the Summary leaderboards on the DataHub, you’ll see that the table sorts by one of Total Points Earned or Total Points Saved by default. Those stats are the offensive and defensive versions of the Total Points system, which we introduced at the start of the 2018 NFL season. Total Points exists going back to 2016, when Matt Ryan took home both the MVP and the Total Points crown.
What does Total Points do?
Total Points takes nearly everything that SIS measures about a play and uses it to evaluate each player on a scale that allows you to compare them more easily.
It’s always useful to be able to understand the different ways in which players can be valuable. Does he break a lot of tackles? Does he get a lot of yards after the catch? Does he make the best out of a poor offensive line? Total Points offers the opportunity to take all of those elements and get a quick picture of how well a player is performing overall.
What does the number mean?
All of Total Points uses the Expected Points Added (EPA) framework. EPA works by taking any given situation and finding the odds that each different scoring possibility comes next. For example, if the next scoring play is a field goal by the current defensive team two drives from now, you count that as a -3. Average those values across all instances of the same situation and you get its Expected Points. Take the change in Expected Points on any given play and you get its EPA.
Roughly, you can think of a 0 EPA play as one that “stays on schedule”, an EPA of 1 or more as a big play for the offense, and an EPA of negative-1 or less as a big play for the defense.
Total Points starts by evaluating each player on that scale, where 0 is average. That’s what we call Points Above Average. Then to both reward players who play full seasons and keep the sum of Total Points around what we’d expect a team to score or allow, we scale the results to the league scoring average (around 22 points per game). So when you see Patrick Mahomes’ 155 leading quarterbacks in 2018, you could take that as a rough estimate that he contributed just under 10 points per game to the Chiefs’ offense on his own.
On the defensive side, it’s a little bit harder to wrap your mind around, because the scaling is exactly the same but points are bad for the defense. Aaron Donald’s 66 Points Saved in 2018 suggests that he was responsible for reducing his opponents’ scoring by 66 points over the season.
How does it work?
We won’t go into complete detail here, but let’s run down the different data elements we consider, how they are evaluated in terms of EPA, and how they get bundled together.
Total Points works on each of the passing game and running game as a whole, so we’ll walk through them that way.
Everything starts up front. We start with identifying who was rushing the passer and who was blocking. Then, we estimate how likely each person was to either blow a block (offense) or force a blown block (defense). On each play, credit is assigned to each player based on how they performed compared to that expectation, and the resulting blown block plus-minus value is multiplied by the average EPA of a blown block.
Players are additionally credited or debited if they were involved for good or for bad in a batted pass, deflection, or pressure, based on the average EPA of those events.
Each pass attempt gets split into four portions: the throw, the catch, yards after catch before contact, and yards after contact. For any throw, we take the value of the route at the intended depth in terms of its completion rate and interception rate and give the quarterback a big chunk of that value. A similar basis is used to estimate how much of the two YAC components we expect, and we give the receiver credit based on the difference in EPA between what he achieved and what was expected. The receiver and quarterback split the value of the throw, air yards, and yards after catch before contact, while the receiver gets the after-contact value on his own. Beyond that, we ensure off-target throws don’t hurt the receiver and drops don’t hurt the quarterback.
The defense at large takes responsibility for the throw itself because many factors contribute to the throw that’s selected, but the primary defender in coverage is responsible for the catch and yards after catch. Any broken or missed tackles are evaluated according to their average EPA impact. If the pass is intercepted, the quarterback and defender are equally debited and credited based on where the ball was caught. The defender then gets extra credit for the change in field position from his return.
All players running routes or defending in coverage have an expected target rate based on the coverage scheme, number of routes being run, route type, and alignment. Each player is assigned a value according to how many targets above expectation they had, scaled according to the EPA value of the potential target.
Pressure, Sacks, and Fumbles
Quarterbacks are given full responsibility for the sacks they incur (less the value of any blown blocks by the offensive line). They are given neither credit nor blame for pressure unrelated to blown blocks, with the idea that their throws are made more difficult but they also had some part in the pressure in the first place.
Sacks or evaded sacks are measured using the EPA of the sack (or potential sack). The sacker(s) get full credit, unless it was deemed a coverage sack, in which case the coverage unit splits the credit.
Pass rushers are given credit for how well they generate pressures relative to the average of players lined up at the same position. Any pressure-related events that might have been debited from the line are given back to the receivers (and quarterback in the case of blown blocks), owing to their having a harder job as a result of the pressure.
All fumbles, recovered or lost, are evaluated similarly. The value of the potential turnover from that spot on the field is multiplied by the odds that possession will be lost based on whether it was in the backfield or not. So players like Lamar Jackson in 2018 are docked a lot of value for his 15 fumbles, even if the Ravens recovered most of them. The person who recovers the fumble gets the inverse of the value that would be lost if the offense recovers or the “rest” of the fumble value if the defense fumbles (i.e. the value of the turnover multiplied by the odds that it is recovered).Pass Plays / Run Plays / Additional Adjustments / How to Use It
Like with passing, the first step is to identify the blockers and box defenders. In addition, we use the intended run direction to identify the key blockers and defenders on the play.
From there, we calculate the play’s expected yards before and after contact based on the number of box defenders, the blocking scheme, the run direction, the spot on the field, etc. The blockers are evaluated based on the play’s performance above that expectation, with most of the credit or blame going to the key blockers identified earlier (unless the runner cut the run back or bounced outside, in which case things are more balanced among blockers).
The same value is distributed among the box defenders, again focusing on the defenders at the intended gap. Blown blocks are evaluated similarly to what’s done in the passing game.
The runner is evaluated against the offensive line’s expected performance calculated above. The rusher is given some credit for yards before contact because elusive runners can generate their own space, but most of his value will come after contact. On any play where a broken tackle was charted, we ignore the yards after contact portion and instead give the back a standard EPA amount based on the average value of a broken tackle. Fumbles are treated like they are on pass plays.
Given each defender’s initial alignment, the heaviness of the box, and the run direction, we estimate the EPA that would be expected if each of the possible defenders made the tackle. That expectation is subtracted from the actual EPA on the play to get the value of the tackle actually made.
Broken or missed tackles are taken independent of where they are on the field, so each one is considered worth the value of an average broken or missed tackle in terms of EPA.
At this point it’s common knowledge that run plays are less valuable on average than pass plays. At a basic level we can see this because the average yards per attempt on passes is much higher than it is on runs. At a more granular level, coaches can make inefficient decisions by electing to, for example, run from heavy personnel on second-and-10.
In order to more accurately evaluate the players on a play as opposed to the coaches or situations, we implemented a Play Selection Adjustment, which applies to each player on each play. We take the expected value of the play given the run/pass decision and some personnel and game state information, compare it to an average play, take the difference, and distribute that value among the players involved. That way, a back being run into a heavy box time and again isn’t punished simply for being on the field in a sub-par situation for him.
This adjustment generally moves a player a handful of points one way or the other depending on how often he was involved in pass or run plays.
As mentioned above, after all of the initial calculations are done, we re-scale everything so that the league total is in line with the league’s scoring average, or just over 22 points per team per game. Because the quarterback represents the most obviously critical position, he’s given 1/3 of this adjustment for the offense, and the rest is split among the other offensive players.
Let’s say that you read all this stuff and already kind of forget what you read at the beginning. Here’s a quick-and-dirty version:
- We take Expected Points Added and give individual value to every player on every scrimmage play, starting in 2016
- You can find it on the SIS DataHub player pages and leaderboards. Here’s the leaderboards for quarterbacks, offensive linemen, defensive linemen, and defensive backs as examples.
- Pass Offense: Quarterbacks and receivers split value for the throw, the catch, after-catch yards, and after-contact yards. Additional considerations for offensive line performance, uncatchable passes, and drops.
- Pass Defense: Defensive backs are measured on how often they are targeted above expectation, and much of the value that the receivers or QB get on a completion is correspondingly taken away from the defender. Pass rushers are credited for forcing blown blocks and disruptions at the point of attack.
- Rush Offense: The offensive line and running back both take responsibility for yards before contact (weighted towards the O-line), while yards after contact beyond what’s expected are totally owned by the back. Broken tackles hold a lot of value.
- Rush Defense: Preventing yards before contact is the name of the game for the defensive line, while linebackers and defensive backs get value from making tackles that limit yardage compared to expectation.
- In general, there’s a lot value to be gained and lost from turnovers (or turnover-worthy plays) and plays in key spots (e.g. just outside field goal range, third down).
Now that you’re familiar with what goes into Total Points, what do you do with it?
The first thing you might do is find players whose traditional stats or reputation don’t line up with their rank in Total Points.
How was Lamar Jackson at nearly negative-20 Points Earned rushing in 2018? You saw the reason for that above (his propensity to fumble).
Why was James Conner in the bottom ten in Rushing Points Earned despite scoring nearly a touchdown per game and averaging 4.5 yards per carry? He had a good offensive line, didn’t break a lot of tackles, and made a lot of his hay on outside runs, which have a higher expected value to begin with.
Total Points gives us the opportunity to more critically engage with the stats players compile and consider the context in which he compiled them. And as SIS continues to add more data points to its operation, our assessment of those things will only get better.