
“In its simplest form: The value of a player is determined not by the number of points he scores, but by how much he outscores his peers at his particular position.”
With these words, Joe Bryant launched into what has become one of the seminal works of the Fantasy Football industry, his introduction to the concept of VBD, a measure of a player's value above his peers. It is perhaps the most cited, most quoted work in the field. It has reached such ubiquity that the term “VBD”, or “value-based drafting”, has entered the general fantasy lexicon; it has appeared in articles on ESPN and earned a prominent place in the glossary and on the player pages at www.pro-football-reference.com, the leading repository of football statistics. It would not be a stretch to suggest that the idea of VBD was one of the bedrock principles that first launched Footballguys fourteen years ago.
Given the place VBD has earned in the field, it would seem unwise for a relative nobody like myself to take a few swings at it. (This lack of wisdom only compounds when one realizes that Joe Bryant is the guy who signs my paychecks.) If one simply wants a hamburger, one doesn’t usually decide to grind up the sacred cow. Alas, fools rush in where angels fear to tread.
To start off with, I should say that the concept of VBD is above reproach. A player’s value is not a reflection of the points he scores, but of the degree to which he outscores his peers. Arguments over what sort of baseline we should use to calculate VBD are different than whether we should use a baseline at all. We should.
My problem with VBD is not that the underlying principles are unsound, but rather that the means by which we calculate it is starting to show its age. In 2001, measuring a player’s season-long point total over the point total of the worst fantasy starter was a huge step forward. In 2015, though, it could use a bit of a facelift.
Changing How We Calculate VBD
My first and most important criticism of VBD, as it is traditionally understood, is that it is calculated using season-long statistics. In 2013, Brent Celek played 16 games and accumulated 502 receiving yards, 6 touchdowns, and one fumble. In standard scoring, that works out to 84.2 fantasy points. Rob Gronkowski played 7 games and finished with 592 receiving yards, 4 touchdowns, and no fumbles. That is 83.2 fantasy points. VBD would therefore suggest that Brent Celek was one point more valuable than Rob Gronkowski in 2013.
This would make sense, if you set your starting lineup in week 1 and left it unchanged for the entire season. In real fantasy leagues, though, you are free to start someone else if one of your primary starters is out injured. If, in the nine weeks Gronkowski was out, his owner started another tight end who scored even two points, then the Rob Gronkowski owner would have outscored the Brent Celek owner at the tight end position.
This is a fundamental flaw of VBD, (the statistic, not the concept), and Rob Gronkowski is the poster child for it. VBD is calculated on a per-season basis, while start decisions are made on a per-game basis.
How do we correct for this flaw? The simplest method is to calculate VBD on a per-game basis, instead. In short, instead of saying “VBD = (Total points) - (Baseline points)”, let’s say “VBD = (PPG - Baseline PPG) * (games played)”.
Here’s how it works in practice: going back to 2013, let’s say that Brent Celek is the “baseline” for tight end production. Celek, as I mentioned, scored 84.2 points in 16 games, or 5.26 points per game. Rob Gronkowski scored 83.2 points in 7 games, which is 11.89 points per game. That means Gronkowski provided 6.63 VBD per game played. Since he played 7 games, his total VBD was 7 * 6.63, or 46.4. Suddenly, instead of a below-replacement player, Gronkowski stands out as one of the stronger performers at his position.
But wait, someone might ask, what about the nine games Gronkowski missed? That’s the beauty of this method: it naturally assumes that when Gronkowski is not starting, his owner receives baseline production from the position. That’s the point of the baseline, to serve as a stand-in for “what you might get when your starter is out”.
This understanding also explains why it’s impossible for a player to have negative VBD. Since the baseline is a stand-in for “minimal startable production”, if a player is producing below the baseline, VBD just assumes you will start someone who gives baseline production, instead. Remember, a player can only gain or lose value for you in games where you actually start him.
Changing How We Think of Baselines
So now we have solved one flaw in VBD, as it is commonly calculated. By calculating VBD per-game instead of per-season, we get a number that better reflects a player’s actual contributions. There’s one other issue with VBD that I’d like to raise, though. Let’s explain it with a thought experiment.
Imagine you play in a 10-team league. The top ten running backs in the NFL average 10.0, 9.9, 9.8, 9.7, 9.6, 9.5, 9.4, 9.3, 9.2, and 5.0 points per game. The top ten wide receivers average 10.0, 5.8, 5.7, 5.6, 5.5, 5.4, 5.3, 5.2, 5.1, and 5.0 points per game.
Using a “worst-starter” baseline, VBD will suggest that the top running back and top wide receiver are equivalently valuable. The top players score 10.0 points per game, the baseline scores 5.0, and so both player provide 5.0 VBD per game.
The problem is that the top players clearly aren’t equally valuable. Outside of the top running back, the other nine starters average 9.04 points per game. Outside of the top wide receiver, the other nine starters average 5.4 points per game. In short, the top running back is giving just under 1 point per game advantage over the average starting running back, while the top receiver is giving a 4.6 point per game advantage over the average starting receiver. The top receiver is much more valuable!
This suggests a second fix in how we calculate VBD. Instead of comparing against a single static baseline, we can compare players against the average of all starters at his position, minus the player himself.
Is this “value over average starter” approach always better than a “value over replacement” approach? Well… no, not always. Let’s go back to our 10-team hypothetical. Imagine the 11th best running back averaged 4.0 points per game, while the 11th best wide receiver averaged 4.5 points per game. Now, if you lost your top running back to injury, it would cost you 6.0 points per week. If you lost your top wideout, it would only cost you 5.5 points!
“Value over average starter” is a great way to measure how much someone is contributing to your wins, but “value above replacement” is a better measure of how irreplaceable someone is. And no, the two are not always the same thing. Also, as a consequence, an “average starter” approach is often better when comparing between positions, while the “replacement” approach produces superior results when comparing within a position.
Another flaw of “value over average starter” is that the baseline of comparison becomes much higher. There were only 42 players who outscored an “average starter” last year. There were 108 players who outproduced a “worst starter” replacement level, though. When trying to determine the 60th-most-valuable player, “value over average starter” is worthless. (Even when determining, say, the 35th most valuable player, it can get a little bit weird; you’ll soon see what I mean.)
For the top of the distribution, value over average starter probably better describes a player’s contributions. As you go further down the rankings, though, value over replacement produces a more accurate measurement.
Applying Our Changes
So I’ve described two ways to improve VBD calculations, but just what do those changes look like? Well, the world needs more football acronyms like the NFL needs more court cases, but if you’ll forgive me for a second… let’s let “VBD” be traditional VBD taken directly from www.pro-football-reference.com. Let’s call my per-game value stat “EVoB” for "Estimated Value over Baseline". Finally, let’s call my final value over average stat “EVoS” for “Estimated Value over Starters”.
I calculated EVoB and EVoS using the same scoring system as Pro Football Reference, and used a similar “worst starter” type baseline. When setting my baselines, I assumed a 12 team league that started 1 QB, 2 RBs, 3 WRs, 1 TE, and 1 flex. Here’s the top 36 players by each stat. You can see for yourself how they impact the rankings.
As you can see, the switch between systems does not change very much, especially at the top. This highlights just how solid the old method of calculating VBD really was. We’re not completely overhauling the system so much as we are tweaking it a bit here and there.
The results of switching from VBD to EVoB are pretty much what we’d expect them to be. Guys who missed time move up the rankings, (Arian Foster rises from 6th to 4th, Odell Beckham from 16th to 7th, etc). Guys who played all 16 games slide down a bit to accommodate these new risers.
The more interesting switch, in my opinion, is the move from EVoB to EVoS. The top players at both QB and TE make a strong move up the charts. This is because of how the baseline works.
At quarterback, there are 11 players who produce above a “worst starter” baseline, (assuming the 12th best QB is the baseline). At running back, there are 29 “above-baseline” players, (assuming teams start 24 running backs every week, plus six more in their flex spot). That works out to a ratio of 2.63 running backs for every quarterback, which is why top running backs are so much more valuable.
On the other hand, when switching to EVoS, it turns out that there were six quarterbacks who produced above an “average starter”. At running back, that total is thirteen players. That works out to a ratio of 2.16 “above-average” running backs for every one “above-average” quarterbacks, a ratio which is much more favorable to starters at one of the “single-starter” positions. Top performers at QB and TE therefore rise in the rankings as a result.
But the bottom of the EVoS list demonstrates how it starts to break down a little bit as you approach the baseline. Martavis Bryant had a surprisingly good year, all things considered. He averaged 10.41 fantasy points per game over his 10 games, just a hair ahead of DeSean Jackson’s 10.24 fantasy points per game over 15 games. But since the “average starter” baseline is so high, that 0.17 point per game advantage for Bryant manages to trump the five extra games played by Jackson.
In essence, EVoS is saying “I’d rather have 10 games of a guy who was just the tiniest hair over average instead of 15 games of a guy who was average”. Which I don't think is a fair reflection of actual value. In short, while I prefer the top half of the EVoS list, I prefer the bottom half of the EVoB list.
I think both methods provide a good solution to the problem of how we calculate value over the baseline, and both have merit. The other challenge becomes how we determine the appropriate baseline in the first place. But that’s a challenge for another article.
Note: In an earlier version of this article, the statistics were called VAR and VOAS. After receiving some feedback that the names made the statistics sounded unrelated and confusing, I opted to change to Estimated Value over Baseline (EVoB) and Estimated Value over Starters (EVoS) in the name of clarity. Thanks to those who have contacted me via email and Twitter to let me know how they felt!