≡ Menu

Guest Post: Bryan Frye on Adjusted Drive Yards

Friend of the program Bryan Frye is back for another guest post. As regular readers know, Bryan operates his own fantastic site, http://www.thegridfe.com. You can view all of Bryan’s guest posts here, and follow him on twitter @LaverneusDingle.


For some time, I have wanted to create a new metric that used elements from Total Adjusted Yards (TAY) in order to quantify a team’s production on each drive. Past work from both Chase and Brian Burke has given us insight into the value of touchdowns, interceptions, fumbles, and first downs, translated into yards. This work has been fundamental in the development of stats like Adjusted Net Yards per Attempt, Adjusted Rushing YardsAdjusted Catch Yards, and TAY.

Those metrics have given us valuable insight regarding statistical measurement of individual player performance. I’ve also used TAY to measure the output of offenses and defenses.

However, I wanted to attach generic values to every way a drive can end.1 This is not a rigorous study, and it is meant to be a starting point for future research rather than a conclusive formula to govern the way anyone interprets on-field action.

With that in mind, I’ll briefly cover the generic yardage values for various drive outcomes.

First Down: Based on a 2014 study by Brian Burke, and accepted as the value in both Adjusted Rushing and Adjusted Receiving statistics, a first down is worth 9 yards. First downs are not yet included in ADY, but their value is an important part of other variables.

Touchdown: Based on work from Chase Stuart in a 2008 study, and subsequently accepted by Pro Football Reference as the value of a touchdown in the ANY/A metric, a touchdown is worth 20 yards.

Interception: Based on work from John Carroll and Pete Palmer in the influential book, The Hidden Game of Football, and still used as the generic yardage value of a pick, an interception is worth -45 yards. The primary objection to this number is that it is based on numbers from the 1980s and is somewhat dated. Logically, the fact that most punts net about 35-40 yards, and a first down is worth about 9 yards, means a roughly 45 yard penalty makes sense. However, given the increased likelihood of an opponent drive ending in a score in 2015 is higher than it was in 1988, one could argue that the penalty is not severe enough.2 For the time being, I am sticking with the status quo.

Fumble: Based on Chase’s 2014 study, the penalty for a fumble should be roughly five yards more severe than the penalty for an interception. It is not perfectly clear that the value for an interception should still be -45, but because we are using that as the accepted value, a lost fumble is worth -50 yards. When using fumble values in ARY and TAY, I use -25 yards, given a roughly 50/50 chance at recovery. However, because a lost fumble is the only type of fumble that will end a drive (and still be marked as a fumble in the play by play data), it makes sense to use the full value for team drives.

Field Goal: Here, I did some simple calculations before adding a subjective change. Research by Brian Burke suggests that kicking the ball off to an opponent is worth about -0.7 points, meaning you must account for that when crediting scoring plays. A touchdown and field goal are, thus, really worth about 6.3 and 2.3 points. Since we’re using 20 yards as the generic value for touchdowns, the simplest way to give value for field goals is to use third grade math: 20 yards / 6.3 points is 3.2 y/p; 3.2 * 2.3 is 7.3 (numbers off due to rounding). After arriving at 7.3, I made the subjective decision to add 1.7 yards to make a field goal at least equal to a first down, or 9 yards.

Punt: The opportunity cost of a punt is a first down. Because we’ve already established that a first down is worth nine generic yards, the value I’m using for a punt is -9 yards.

Downs: This one is pretty simple. A turnover on downs means a team gave the ball to the opponent with the opportunity cost being a punt. Because punts net around 35 yards, the penalty for a turnover on downs is roughly -35 yards.

Missed Field Goal:  A missed field goal is worth the points from a field goal plus the difference in opponent field position after the average missed field goal and the average kickoff. I didn’t have time to run all the numbers through years of play by play, so I chose to adopt an intuitive solution for this initial draft of the formula. The average team drive after each kickoff starts at the own 22 yard line. The average missed field goal attempt (that isn’t blocked) occurs around the 29 yard line. Add in the standard seven yards for the snap and hold, and we’re looking at an average opponent starting field position at the 36 yard line (a 14 yard difference in field position). The opportunity cost is the value of a field goal, so 9 yards. When we add the 14 and 9 yard costs to the value of a turnover on downs, we get a 58 yard penalty. Subjectively, I think this is too harsh, so I lowered the value to -50 yards.

Blocked Field Goal: Given that blocked field goals are another type of missed field goal, -50 yards will be our starting point. Most blocks happen at the line of scrimmage and are not returned by the defense. However, the blocks that are returned tend to go pretty far and often see defenders run unencumbered into the endzone. Given an average return of about 8 yards and the threat of a score,-60 yards seems like a reasonable penalty.

Blocked Punt:  A blocked punt means three things. First, the punting team has forfeited the average 35 net yards of a normal punt. Second, they effectively suffered a fumble about twelve yards behind the line of scrimmage. Third, blocked punts that don’t travel past the line of scrimmage are often returned for uncontested scores. These are not common occurrences, so I didn’t put in the level of work for this initial project that I hope to put in later. For now, I’ll use a penalty of -60 yards for blocked punts.

Safety: Remember that a touchdown and a field goal aren’t really worth exactly 7 and 3 points, given that the result of both is that the scoring team gives the ball back to the opponent with the chance to reciprocate. From an expected points perspective, they are really worth about 6.3 and 2.3 points. Conversely, a safety is actually worth more than the points awarded, because the scoring team also gets the privilege of a brand new possession. When we account for the fact that a safety earns a team two points and gives them the ball, on average, sixty yards from the endzone, a safety may be worth up to four expected points. This is where I got a bit lazy. To find the value for a safety, as expressed in generic yards, I divided the 6.3 EP for a TD into the 20 yard bonus, coming to 3.2. Then I multiplied by four (the EP for a safety) and got 12.7. I decided to round to 13 yards for simplicity’s sake.

After coming up with some initial values for all these events, we can add the values to the actual yards gained on each drive to come up with a metric I will call Adjusted Drive Yards (ADY).

Adjusted Drive Yards

The table below contains Adjusted Drive Yards data for each team’s offense, defense, and overall performance. Read it thus: The Arizona Cardinals had 170 drives and gained 6059 ADY, a rate of 35.6 per drive. Their defense saw 180 drives and allowed 3319 ADY, a rate of 18.4 per drive. Overall, they gained 2740 more ADY than they allowed, and their per drive differential was 17.2, tops in the NFL.

RkTmDriveOADYOADY/DODriveDADYDADY/DDADY+-ADY/D+-
1ARI170605935.6180331918.4274017.2
2CAR185533628.8194287214.8246414
3SEA161521632.4167338720.3182912.1
4NE179572132184379820.6192311.3
5CIN171491128.7178351219.713999
6PIT178519929.2180384621.413537.8
7KC172463226.9179346919.411637.6
8DEN19140042119228731511316
9NYJ190447623.6186337018.111065.4
10HOU193406921.1190331117.47583.7
11GNB180450925.1180407922.74302.4
12WAS171457126.7177444525.11261.6
13MIN165442126.8168424025.21811.6
14BUF177450425.4176449125.513-0.1
15ATL164428026.1164431726.3-37-0.2
16NYG180523929.1180561131.2-372-2.1
17OAK185411522.2185463525.1-520-2.8
18TB169434025.7170485128.5-511-2.9
19CHI175435024.9167465827.9-308-3
20DET174440425.3173494328.6-539-3.3
21NOR172540631.4175612635-720-3.6
22SDG174431524.8172492228.6-607-3.8
23IND185366719.8187476925.5-1102-5.7
24BAL185381620.6180473626.3-920-5.7
25JAX189413921.9182506327.8-924-5.9
26MIA181378720.9179482026.9-1033-6
27DAL165359221.8164471028.7-1118-6.9
28PHI194392120.2191532927.9-1408-7.7
29STL187282315.1187433023.2-1507-8.1
30TEN189319616.9173460526.6-1409-9.7
31CLE174341819.6170516730.4-1749-10.8
32SF176336719.1171519930.4-1832-11.3

Using the values from the first draft of the ADY formula puts NFC West rivals Arizona and Seattle at the top of the offensive ratings. The Patriots rate highly as well, as do the Saints – a team carried to a 7-9 record with perhaps the worst defense in modern history.

The only real surprise here may be the Giants ranking so high and the Panthers ranking so low (at least, low for the top scoring team in the league). One reason the Panthers rate relatively low is that their defense and special teams often gave them short fields and eliminated the need for long drives.

The Giants place near the top takes a bit more explaining, although it isn’t really complicated. On a per-drive basis, they were better than average at avoiding blocked punts, turnovers on downs, fumbling, missing field goals, allowing blocked kicks, and ceding safeties. They were better than average at making field goals and scoring touchdowns. They only area where they were worse than average was interception rate, where their 7.78% was slightly inferior to the league average 7.64%. Put all those things together, and you have a shockingly high rated offense.

Defensively, the Super Bowl participants rated one and two, well ahead of the third rated Texans. Nine of the top ten defenses participated in the postseason, while the non-playoff Jets once again suffered the unfortunate fate of playing in the same division as Tom Brady and Bill Belichick.

Similar to the Saints’ stellar offense being fettered by their awful defense, the Rams’ sturdy defense was unable to overcome ineptitude on the other side of the ball. Jeff Fisher hasn’t produced a worthwhile offense since the days of Steve McNair, and 2016 doesn’t look to be much different.

Sorting by differential shows that the top three rated teams in the league played in the NFC. So, naturally, the eighth rated AFC Champion Broncos won the title. This stat comes pretty close to the famous eye test, with all twelve playoff teams ranking in the top thirteen. Once again, the bad luck Jets are the highest rated team left out in the cold.

I don’t want to send too much time pointing out every little pattern or oddity I see. I’d rather hear outside opinions, so leave your thoughts in the comments.

  1. With the exception of kneel down drives to end halves or games, as those don’t demonstrate an offense’s (or defense’s) ability to actually play the game. []
  2. Indeed, Brian Burke has argued that a modern interception is worth about -60 yards. However, Burke did state in a personal communication that this is only applicable to interceptions thrown on first down and that the generic value should be close to 50 []
  • Adam

    Great work, Bryan. Definitely passes the eye test. More thoughts coming later…

    • What I really want to know is…when are we getting the true talent metric? Come on, dude!

      • Adam

        I suffer from tweaking disease 😉

        • I have the same disease, but I like to make the tweaks publicly. I like to see the progressions worked out. That’s why my QB series was a ridiculous 14 parts long.

          • Adam

            You know what, you’re right. Seriously. I’m going to clean up the current iteration I’m working on, and get it out there for public review.

  • Josh Sanford

    This is great work. I think you are owed congratulations for a formulaic expression of offensive competence that “passes the eye test.” Because this clearly does. Do you want armchair quarterbacking? If Yes, read on. If No, then here goes**: the easiest change to make to your formula is also the least scientific one, and the one that will have the least impact on the results: turnover on downs. The offense “usually” only goes for it on 4th down when they are between their opponent’s 45 and 33 yard lines. Let’s say an average of the 37 yard line. You’ve placed the difference at failing to get a first down and punting at 26 yards, which expresses a likely outcome of pinning the other offense at their 11, I think. But the overwhelming majority of those short punts end up as touchbacks. I would put turnover on downs at -19 yards instead of -35. (**Warning: Contains no science. Also, probably just a disguise for expressing my preference to see teams go-for-it on 4th down.)

    • Thanks for the feedback. This is the reason I wanted input.

  • Ken Adams

    On a team or drive level, this seems to be a bit unnecessary. You can use actual net yards (start yard line of team’s drive compared to start yard line of opponent’s ensuing drive) or even better just use expected points.

    An additional point: You said 1st downs are “not yet included” but they definitely should not be included at all. There is no difference on a team level between an 80-yard drive on one play and an 80-yard, 8-first-down drive, so on the team/drive level you don’t want to give an extra bonus for the 1st down.

    The reason you give the bonus on the player or individual play level is because relative to other plays, the 1st down is more valuable. But technically non-1st down plays should be given negative value if we give 1st downs a bonus. On the team/drive level, these will even out. Giving just the bonus (and not the penalty) will incorrectly inflate longer drives with more 1st downs.

    • Thanks, Ken. Omitting first downs makes sense from a descriptive standpoint, but not necessarily from a predictive one. An 80 yard drive with 8 first downs gives the same result as one 80 yard play. Of course it does. No one disputes that, to my knowledge. However, the former is likely to be more predictive of future performance.

      Regarding the first point, you may be right or wrong. I don’t know. This is why I solicited feedback, to discuss these things.

      • Ken Adams

        If you want to make it a predictive stat then the metric should be developed that way. Important to define what you want from a metric. People often get tripped up by creating the metric first and then end up with something in the middle between descriptive and predictive which makes the metric much less useful. FO’s DVOA suffers from this, as an example.

    • Also, and more importantly, the primary goal here is to look at the numbers themselves and determine appropriate bonuses or penalties for certain outcomes (e.g., what is the generic value of a turnover on downs or a blocked punt).

      • Ken Adams

        Again, to my reply below, those are descriptive numbers. I would urge you to decide what question you want the metric to answer first, that will lead you to the best answer as to how to derive these numbers.

        • The question is “what is the generic value of x?” As we’ve come to generally accept 45 yards for interceptions, 9 yards for first downs, or 20 for touchdowns, what are the proper values to assign other types of plays that don’t fall into those categories? I apologize if I didn’t make that clear with my previous reply.

          • Adam

            So the actual AY/D metric is not as important as discerning the value of its inputs?

            • Well if it’s inputs are wrong, the metric’s not worth much.

              • Adam

                True, but I don’t see this as a right / wrong dichotomy as much as predictive / descriptive or macro / micro. The inputs could vary wildly depending on what you’re trying to measure, especially with rare events like blocked punts.

                • I think, for the time being, the right focus is on descriptive stats. Predictive stats are rarely close to as predictive as we want them to be anyway. So a blocked punt counts as being as bad as it is, not as being a predictive event (because not every team is the 2010 Chargers). I feel like this is sort of a job for Brian Burke, but he’s busy working for the four letter company.

                  • Tom

                    I just wouldn’t include blocked punts and field goals at all unless we’re assessing the entire team, which might be what you’re doing and I’ve missed it. I certainly would not have those events any where near an assessment of an offense or defense.

                    • I think it would be appropriate to include them but in a separate category (so off, def, ST). Same goes, like you said, for counting an offense defense with a missed field goal. Sure, the offense failed to score and should be penalized, but they didn’t miss the field goal. Similarly, the defense did a good job holding the offense to a field goal attempt, and they have little to do with the field goal going in or not (I say little because holding the offense from gaining additional yardage does increase the likelihood of a miss).

                      When we look at these with further scrutiny, we see just how interrelated these events are. We see the fluid nature of field position, which can have a big impact on scoring (see last year’s Panthers and their “great” offense).

                    • Tom

                      Yep, agreed. Just to be clear, my viewpoint is that the offense “failed” (to some degree) when the decision was made to kick a FG. So whatever happens after is of no consequence. All I/we need to do is figure out the EP gain or loss for FG attempt.

                      I agree with you 100% that these things are all interrelated…even in light of what I just said, the ability of the offense to get the kicker closer to the posts has a direct impact on whether he makes or misses.

                    • I am assessing the entire team in my personal spreadsheets, but they don’t fully resemble this. My tables are often way more cumbersome than the ones Chase posts, and I am trying to stop overwhelming people with 33 column tables. So I conflated ST stats into O/D and just went with it.

                    • Tom

                      Ok, got it. In that sense, if we’re being descriptive, then yeah, throw in the blocked punts and FG’s. Hate to say it, but probably return TD’s should be included as well? They are similarly fluky?

                  • Adam

                    I agree that predictive stats often fail to adequately predict, so it’s understandable to focus on the descriptive side. There are also various levels of descriptive stats; going back to the blocked punt example, do you use a generic value based on the average of all blocked punts, or calculate the value based on the return length from each individual occurrence? There’s a blurry line between drive stats and expected points, and frankly I’m not always sure where one ends and the other begins.

                    • That is kind of like asking if we count a generic 45 yard penalty for an interception or if we want to use the value of the touchdown that followed the interception. I think there are arguments for both, depending on what your goal is. If we want to measure “true quality” or whatever dumb name we want to give it, then we probably want to omit the fluky stuff. But if we want to just say “this is what happened,” then we should keep it in.

                    • Tom

                      Agreed. I always use the James Harrison example for this because it’s so ridiculous. A loss of 4 EP (standard turnover) or 12 (fluky play where a linebacker rumbles down the field for 99 yards)?

                    • That’s one of two plays Kurt Warner had that you could find stats to say he wasn’t nearly the postseason QB he’s cracked up to be. He threw the deciding pick six in two Super Bowls. The Harrison play was fluky, but the Law one is going to be a touchdown most of the time.

        • Tom

          As someone who’s spent a stupid amount of time on this, both questions (the point of the metric and the value of events) go hand in hand and are worked out as you go. As I’ve noted above, depending on what we’re trying to do, the value we place on a field goal attempt may change, etc.

    • Tom

      Ken – I agree that we can just look at net yards, but then we’re missing bonuses and rewards for TD’s, interceptions, etc. These things matter. As far as EP goes, yes, that would be great, but the data isn’t widely available, and frankly, I have a harder time wrapping my mind around when it comes to this level – drives and plays (for games, yes, I like it).

      • Ken Adams

        The TD (and to a lesser extent, FG) bonus is needed in some form. But the turnovers would take care of themselves since we aren’t using offensive yards, but net yards between start of possession and start of ensuing possession. For example, drive starts at the -30, gain 20 yards and turn the ball over at the 50, that’s a 20-yard drive. Another drive starts at -30, gain 20 yards and punt to +20, that’s a 50-yard drive.

        Still need a way to combine points and yards, but again, that’s why Expected Points is perfect and really what we’re after for a descriptive stat as it captures everything we want and weights it accordingly on the same scale.

        • Tom

          I see…usually drive stats don’t include the yards “gained” on the punt, you’re saying to include them. That’s more in line with EP, which really is a field position/possession stat.

          The problem is data. Before 1994, or whenever they started keeping those drive stats, we don’t know the start and stop of each drive. That’s why we have these values, we don’t know exactly how much field position was lost on a turnover or a punt, etc. Saying a turnover is -45 and a punt is -9, gets us in the ballpark.

          That being said, I’m a huge fan of EP, but it falls apart on some things like pick-sixes: the Cardinals lost about 12 EPA on James Harrison’s 99-yard pick-six in the SB. That’s descriptive, not prescriptive…guess we’re getting back to what you said regarding our intent with these stats.

    • Tom

      Ken, Bryan – apart from the appropriateness of including 1st downs in this study, I agree with Ken that if 1st downs are included, failed 3rd downs should also be included (from my EP studies, 2nd downs can be ignored).

  • Tom

    Bryan, this is awesome. I’ve been messing around with this for a few years now, and it’s great to finally see a post devoted to it!

    OK, my main question is, what are we trying to show/evaluate? How good an offense (or defense) is? If so, then I have these thoughts:

    1. If we’re trying to evaluate offenses and defenses, by themselves, in my mind everything stops once the decision has been made to kick a field goal or punt. I see these as, roughly, “failures” by the offense (and victories by the defense), and what happens afterwards, in a general sense, does not involve that unit (it’s special teams). The offense has nothing to do with a blocked punt, or blocked field goal, and in a less direct way, has nothing to do with a made or missed field goal. But touchdowns, fumbles, interceptions, turnover on downs, and safeties are all drive-ending states directly attributed to the offense.

    2. As far as first downs go, I’m with Ken that perhaps 1st downs aren’t needed here. I realize how important they are, and you’re probably right as far as their predictive value, but I think we just want to answer the question: what was the result when the offense had the ball? How it happened may not matter as much, for this stat anyway.

    My quick thoughts on field goals and punts (my apologies guys, but I really have done some work on this, just could never find the the time to post):

    Field Goals: These are changes of possession in which the team with the ball may or may not get points. I treat all field goal attempts as neutral events for the offense. We know that depending on the game situation, the offense’s #1 goal is to get into field goal range, but overall, the offense’s main goal is to score a touchdown.

    Punts: I think 9 yards works pretty well; I came about it a different way: the team with the ball gains about 37 yards of field position (+37), but loses possession (-45), which comes to -8.

    Great work Bryan, I’ve got some EP data gathered (from Burke’s site a few years back) which could be of use in this study, let me know if you need it!

    • Thanks for the input, Tom. I think I’m coming around to your and Ken’s opinion on first downs for this metric.

    • Adam

      When you say field goals are neutral events for the offense, does that mean you ignore them entirely, or credit the offense for the average points per FG attempt at the given distance?

      • Tom

        Adam – should have been more clear. I mean that when I went through the numbers (gotta find that stuff now) it appeared that the average FG attempt generated 0 EP, or something close to it. So I don’t ignore them, but the team doesn’t gain or lose points/yards.

        Now, I’m talking about the average of all FG attempts, totally ignoring specific instances. I did this so that I could use older data (1970’s).

        I agree that the offense, in theory, should probably get credit for a missed field goal that should have been made…of course, this gets messy, etc.

        • Adam

          How do you figure the average FG attempt generates 0 EP? A made FG is worth 2.3, and accounting for misses takes us down to roughly 1.8. Unless you’re somehow calculating the opportunity cost of not going for it on 4th down and potentially scoring a TD, I don’t see how this adds up.

          • Tom

            OK, Adam, I went back and looked at how the hell I came up with that and here’s what I’ve got. This is long, feel free to ignore.

            A while back I read this post, where the author presents a linear scoring model, based on PFR’s standard -45 INT/+20 TD formula:

            https://codeandfootball.wordpress.com/2013/07/03/adjusted-yards-per-attempt-101-thinking-in-terms-of-a-simple-scoring-model/

            For this model, a team’s own goal line has a value of -2.00 points (corresponding to a safety), the opponent’s goal line has a value of 5.5 points, a yard is equal to 0.075 points, and a turnover is equal to 3.5 points.

            So, now that we have the scoring model, we can just start plugging stuff in to get our yardage values. For starters, we can “prove” that a TD is worth 20 yards: the difference between the goal line at 5.5 points and 7 (a TD), is 1.5 points, 1.5/0.075 = 20 yards.

            Punt: a punt is a loss of possession in which the team with the ball gains about 37 yards.
            -3.5 + (37*0.075) = -0.73 points, which translates to a loss of 9.7 yards. This fits what Bryan has for a punt.

            Successful FG attempt: also a loss of possession, but they also gain 3 points, and may or may not lose field position yardage. Let’s start with a field goal from the 30-yard line (a 48-yard field goal). After the ensuing kickoff, the opponent has the ball at the 22-yard line, so the FG-kicking team gains 8 yards.
            -3.5 + (8*0.075) + 3 = 0.1 points, translating to 1.3 yards.

            If the FG is kicked from the 22-yard line (30-yard kick), the kicking team actually loses yards, about 7.

            I think at this point, I must have stopped and thought, “Well, if a 48-yard made FG only gives you a yard, and there’s a chance you might miss, maybe it’s just a wash”.

            I realize the process here is somewhat goofy and probably flawed – to begin with, there are a lot of problems with even the idea of a linear scoring model – but I wanted to answer your question (and remind myself where I got the idea that a FG attempt is a neutral event).

            Since I mentioned EP, in my original comments – which was probably a mistake, since I wasn’t dealing with EP, but the linear scoring model – I should go through that as well (if anyone is still reading this, I’m amazed). I don’t have the time to go through it all at the moment, but I’m definitely wrong here, a field goal attempt using EP numbers is generally a POSITIVE event for the offense, as you noted, but not as much as you might think.

            Made field goal from opponent 30-yard line:
            4th Down at opponent 30: 1.06 EP
            FG is worth 2.58 EP (using the 22-yard line as average start after kick)
            Net EPA is 1.52, which equals around 24 yards.

            Missed field goal attempt from 30-yard line:
            Opponent having the ball at his own 38-yard line, 1st-10, 1.31 EP
            Loss of 2.37 EPA (1.31+1.06), or 37 yards.

            Since field goals from that distance (48 yards) are made about 76% of the time (2015), we’re left with a net gain of 9.4 yards on a field goal attempt, whether it’s made or missed.

            There’s other things to consider here, but roughly that’s what we’re looking at.

            Thanks.

            • Adam

              Thanks for the answer, Tom, and yes I read all of it! The -3.5 point loss of possession penalty in the linear model doesn’t make sense for scoring plays; I wonder if it’s only supposed to be applied to INTs / fumbles? Otherwise, a TD would only be worth about 3.5 points (7 – 3.5 + kickoff return adjustment), and that doesn’t seem right.

              The EP model makes more sense, but you’re right, +9.4 yards is lower than I’d expect. I’m also skeptical that a 1st down on your own 38 is worth more EP than a 4th down from the opponent’s 30, but I could very well be mistaken.

              • Tom

                Adam – wow, thanks for reading that whole thing…I didn’t think anyone would look at it, was mostly doing it to set myself straight on what I was thinking.

                Regarding the 3.5-point TD: I believe you’re making the comment that it doesn’t seem right in reference to the fact that we normally consider a TD to be worth 7 points (PFR), 6.8 points (Romer I think) or 6.28 (Burke, when the LOS after kickoff was the 27-yard line). What I’m talking about is what is gained on the actual play. A TD score from the 1-yard line doesn’t “gain” 7 points per the linear model (or the EP model for that matter), it gains 7 (or whatever) minus the value at the opponent’s 1-yard line. So yes, the actual scoring of the TD is far less than 7. This is where the 20-yard TD bonus comes from in the linear model: you have the yards gained on the play, then you have the TD itself, which is 7 minus 5.5 – the value, in our model, at the opponent’s goal line.

                OK, how do we explain this using 7 points for the TD and 3.5 for the turnover? Check it out:

                Let’s say you’re on the opponent’s goal line, and then score a TD. You gain 7 points (93.3 yards) for the TD, but you lose 3.5 points (46.7) for the “turnover”. This equals 46.6 yards…this is a lot higher than our standard 20 yards. But remember, you were on the 0-yard line, so after the kick off, you’re losing 27 yards of field position (27 used to be the standard LOS after kickoff). So, now we have:

                93.3 – 46.7 – 27 = 19.6, roughly 20 yards.

                We can do the same thing for a touchdown play from the opposing team’s 40 yard line: you gain 7 (93.3), lose 3.5 (46.7), and gain 13 (net) yards of field position (you gained 40 on the play, getting you to the 0-yard line, then lost 27 after the kick off).

                93.3 – 46.7 + 13 = 60 yards.

                I know this is as goofy as it gets, but again, just trying to address your question and show where all this stuff comes from. Basically the idea is this: every drive ending state (apart from end of half/game) consists of a turnover, a field position change, and possibly points gained or lost.

                Regarding the expected points on your own 38, etc. I checked that again, and the data I have from Burke does say that 1-10 from the 30 is 1.31 EP and a 4th-down from the opponent’s 30 is worth 1.06. But, your skepticism is justified, as EP data I’ve got from Keith Goldner’s Markov Drive Analysis spreadsheet, shows a different story (he has three different models, I’m using the Regressive one, don’t ask me what that means):

                1-10, your own 38: 1.49 EP
                4-1, opp 30: 1.77 EP
                4-10, opp 30: 1.17 EP

                The numbers I downloaded from Burke (this was many years ago mind you) didn’t have different EP states for varying “to-go” situations; a 4th down was a 4th down…this may be why there’s a difference. Certainly a 4th-and-1 has a higher EP than a 4th-and -10. Burke’s calculator is gone, so we can’t really check that.

                In any event, more ramblings on stuff that really doesn’t matter that much, I guess. Hope you found it of interest, though Adam!

                • Adam

                  I read every comment top to bottom. Skimming is for sissies! The 3.5 “turnover” penalty makes perfect sense now that I understand the context, and glad to hear I’m not the only one questioning those EP figures. It’s a damn shame Brian Burke sold out to ESPN, because as far as I know his work has disappeared from the public eye. I don’t know that we’ll ever settle on one version of EP or WP, and given the yearly variations in the NFL environment, it seems all but impossible to nail it down anyway.

  • Tom

    Some other thoughts on your numbers from a different approach:

    Turnover on downs: 35 yards seems about right, but going along with what Josh says below, the number might be brought down a bit. Some EP data: the average for all 4th downs between a team’s 15-yard line and the opponent’s 15-yard line is 0.06, the average first down EP between the 15’s is 1.87. For the team with the ball, that would be -1.87, so the difference is 1.93 EP. Per the first down article by Burke noted above, a yard is equal to 0.064 points, which would give us 30 yards.

    • I’d love to have a look at the EP data if you’re willing to share.