Prelude

This series of Fantasy Football analyses hopes to provide some quantitative support and make tangible sense of some of the basic principles in Fantasy Football. These analyses will cover a broad range of concepts using both descriptive and predictive analytical methods, while requiring little to no familiarity with these methodologies for the readers.

The first installment in this series will cover some of the basic drafting strategies and help give the reader an understanding of positional value in Fantasy Football - it is important to remember that Fantasy Football value does not equate real life Football value and vice versa. A good Football player in real life can have little Fantasy value (see Kyle Juszczyk) and an average one can have great Fantasy value (see Mitch Trubisky.) As such, every time I will talk about “value” in this series, I will be referring to Fantasy Football value.

This series will assume a Half PPR scoring system, and the typical roster of 1QB, 2RB, 2WR, 1FLX, 1TE, DST and K. Going forward, we will also filter on certain position for our analyses so as to only focus on relevant Fantasy Players, see below:

  1. Running Backs - We will only look at RB with at least 47 Total Rushing Attempts
  2. Quarter Backs - We will only look at QB with at least 212 Total Passing Attempts
  3. Wide Receiver - We will only look at WR with at least 71 Total Receiving Targets
  4. Tight End - We will only look at TE with at least 52 Total Receiving Targets

Introduction and Positional Intuition

If you’re new to Fantasy Football, the only thing you need to focus on right now is the upcoming draft. This installment will go over and provide some analytical support to help you understand why certain positions are highly valuable, and why some aren’t. At the end of this analysis, the goal is for you to understand three things:

  1. The relative value of RB, WR, QB and TE
  2. Why a stud RB is vital
  3. Why over-drafting a QB is not necessary

To begin, let’s take a look at the value of each position, as told by the average PPG by position last year.

Now, we can clearly observe that QB are the most valuable position on a Point per Game (PPG) basis. So naturally, you would expect them to be more valuable in Fantasy, and hence be drafted very early on. We also see RB emerging as the least valuable position, so you would expect RBs to be drafted late. To verify those claims, let’s make use of the ADP - which is the average draft position of every player, essentially a measure of where each player is being drafted on average right now. Let’s look at which position is being drafted most often in the first five rounds of a draft.

So in the first five rounds, 23 RBs are being drafted…despite being the least valuable position, as shown on the previous plot. Similarly, we saw QB being the most valuable position, and yet only one QB is being drafted in the first five rounds.

In order to really evaluate the positional value of the various positions in Fantasy Football, we will discuss and take a closer look at “Tiers” within and the insights we can gain from them. These tiers will help use understand the disparities between good and average players at a given position, and shed some light as to why the ADP is RB heavy.

Tiers in Fantasy Football

As you being your journey in Fantasy Football, you’ll often hear people discussing tiers, i.e RB1, RB2, QB1, QB2, QB3 and so on. Intuitively these tiers are groupings of players by performance, so a Quarter Back that falls in the QB1 tier will be more valuable than a QB2 Quarter Back.

The definition of a given tier is arbitrary and varies from person to person. Some people might consider to top 12 RB to be RB1, while others only the top 6-8. There are other, non quantifiable decisions that often go into these tiers, such as coaching changes, usage changes, health issues and so on. While the definitions will vary, the concepts does not - the higher the tier a player is in, the more valuable he is.

As mentioned above, going into the new season, there are a lot of factors affecting tiers. For the sake of this analysis, however, we will define our tier purely based on last years performance. So our tiers will not necessarily be an indicator of next year’s tiers. That is to say that someone who we identify as a QB1 with or analyses, will not necessarily be a QB1 going into the next season, but that is besides the point.

We are establishing tiers purely to demonstrate positional value, which will in turn help us prove some of those fundamental concepts in Fantasy Football.

Defining Tiers

We will keep our tiers indicators simple. We will make use of the average point per game (PPG), and the variance of points per game. Note that across this document, I will use the word “Variance” “Deviation” and “Standard Deviation” interchangeably. Conceptually, they all mean the same thing - a player with a very high variance in PPG will have a high Standard Deviation in PPG as well. Mathematically, however, they differ. But we’re not going into the mathematical definition, so for our purposes they mean the same thing i.e. High variance means high deviation, means very little consistency. Similarly, low variance means high consistency.

Simply put, we will look at how good players are on any given game, and how consistent they are throughout the season. We will begin by analyzing the data for the Quarter Back position for the year 2018.

There are a couple things to note about this graph. First you will observe that both axes are on a scale of -2 to +2. That is because we have normalized the data. This allowed a cleaner read and puts the data on a similar scale. A player that has a value of 0 on the horizontal axis, means that player has a very average PPG. Someone to the far right means they have a much higher PPG than the average, and conversely, the far left means much lower than average. Similarly, a player at the 0 line on the vertical axis means that they have average consistency. Someone higher up in the graph, has higher variance that the average QB (less consistent than the average QB), and someone near the bottom has a lower variance than the average QB (more consistent than the average QB)

Essentially, the farther right a player is, the higher his PPG is, and the farther left the lower his PPG is. The higher up a player is, the more variance he has (which makes him less consistent), and the lower down, the less variance he has (which makes him more consistent). The ideal player would be on the bottom right - high PPG and low variance and hence high consistency.

Let’s take a closer look at a few player.

  • Josh Rosen (Bottom, Left) - A perfect example of someone you don’t want. Terrible PPG (far left) and low variance (bottom) which in this case means that he’s bad… and consistently so
  • Ryan Fitzpatrick (Top, Center) - What you would call a roller coaster. He has an average PPG, but an extremely high variance and little consistency (Very high up)0. Which means than a given week he could get you 40 points, and on another week 0, which would average at 20PPG in those two weeks. So not necessarily a player that you want.
  • Philip Rivers (Bottom, Center) - One of my favorite Fantasy QBs. Average PPG (close to the x = 0 line) and less variance and hence more consistent than the average QB (lower down). Essentially, someone who gets the job done, but not a flashy guy. Definitely won’t get you a 40PPG, but very unlikely to get you a sub par week either. In one word, decent, and consistent
  • Patrick Mahomes (Center, Right) - The outlier. Patrick Mahomes had an incredibly season where he was significantly better than the average QB on a PPG game basis (far right) and as consistent as any QB (center.) The next closest player is Matt Ryan, and he’s not really that close either. Mahomes was a literal Fantasy god last year.

Clustering the Scatter Plot

The natural next step to this plot, is identifying “clusters”, or simply put, groups within this plot that are very similar. We could intuitively try to group some of the high performers with average consistency, like Patrick Mahomes and Matt Ryan together, and do the same for above average performers with low consistency (Fitzpatrick, Trubisky, Brees.) We would then have to find every one of these clusters and group players manually. While this is doable and intuitively makes sense, there is another way around it. There are certain algorithms that we can use that can help us identify the optimal number of cluster, and what those clusters are.

In this analysis, we will make use of one of these algorithms, called a K-means. As mentioned, the K-means algorithm will identify how many clusters there are in our data, and build those clusters, grouping players with similar PPG and Standard Deviation. See below for the results.

Now, our algorithm has clustered players together, based on how similarly they perform, both in terms of PPG and variance. Now let’s see which cluster is the higher performing one, based on the players within.

##  Cluster Tier  Tier.PPG   Tier.SD
##        5    1 22.601562  7.843288
##        7    2 19.228000 10.912020
##        1    3 18.961034  7.356099
##        4    4 16.589730  5.808217
##        6    5 14.008571  4.517360
##        2    6 13.678667  6.667779
##        8    7 13.202632  8.948080
##        3    8  8.651429  4.267737

We can see from the above that cluster 5 is our highest performing cluster in terms of PPG - unsurprisingly where Patrick Mahomes is. We will hence identify this cluster as our Tier 1 cluster - or rather, our QB1 tier. The next highest cluster in terms of PPG will be the QB2 Tier, and so on.

Running Back, Wide Receiver and Tight End Clusters and Tiers

Now that we have established how to Tier our QBs, let’s apply the same principles to the other relevant Fantasy position. Below are the cluster plot for Running Backs, Wide Receivers and Tight Ends.

One quick note on these clusters. Looking at the WR cluster, we can clearly see that Cluster 3 is very expansive. That is because we have forced Amari Cooper, Tyreek Hill and Michael Thomas into that cluster. Initially, the algorithm recognize the three of them as a distinct cluster and rightly so because of how high their variance is. But from a Fantasy stand point, it does not make sense to group these three into a distinct Tier, so we will group them with the other WR2 for a more realistic outlook.

Similarly, we have forced a similar cluster for the RB. In this case, we forced Derrick Henry and Nick Chubb into the closest cluster, for the same reason highlighted above.

Let’s now look at the Tier 1 for each of the position and this will shed some light on the very first conflict we introduced in this document.

##  Tier Tier.PPG  Tier.SD
##   QB1 22.60156 7.843288
##   RB1 19.57321 9.162231
##   WR1 18.07167 6.153436
##   TE1 16.53704 8.007311

So, we can now see RB climbing up the ranks in terms of value. We saw initially that the average RB was not very valuable, but we can see how that does not apply to the elite RB. We still observe that the QB1 is the most valuable, so why then are they not drafted more often in the first round? Let’s take a closer look at the other tiers, and comparison across positions to answer that question.

Normal Distributions

For the next part of this analysis, we will make use of normal distributions to interpret the position value in Fantasy Football. If you don’t know what those are, here’s a quick explanation - feel free to skip ahead if you are familiar with the concept.

A normal distribution, shown below, is simply a way to summarize a data set, looking at the average and the variance in that data set. Let us only focus on two aspects of the normal distribution plot: the peak and its breadth

In the above example, we can see two very different distributions.Let’s assume that they represent the Fantasy Points of two players, on a game to game basis. The peak (highest point) represent the average Point per Game. So the Red Player has a lower PPG compared to the Blue player.

The breadth of the plot, is an indication of variance (or consistency.) The narrower the plot, the more consistent the player is, and similarly, the broader, the less consistent. So while Blue Player averages more PPG, they are significantly less consistent that Red Player and hence less reliable. Note that the height inversely correlates to the width - a wide distribution will be shorter than a narrow distribution.

Tiers Value

Quarter Back Tiers Analysis

Now that we have established our Tiers and brushed up on Normal Distributions, let’s see what insights we can gain from the positional tiers. Let’s first take a look at the QB Tiers.

So, what can we observe by comparing those distributions? For starters, we see the QB1 Tier having a much higher PPG than the other 3 tiers. However, QB2 and QB3, have a very similar average to each other, but differ greatly in variance. QB4 has a much lower average and a higher consistency as well.

Let’s quantify those differences in averages and variance. Let’s look at the % difference in PPG and variance for each tier, relative to the QB1 tier.

Essentially, we observe that QB2 average about 18% less than the QB1 PPG, and are less consistent. Again, we see that QB3 are very close to QB2, in that they average roughly 19% less than the QB1 PPG, but are slightly more consistent. We do see a big falloff in PPG when it comes down to QB3, and the higher consistency of QB4 only means that, we are dealing with the Josh Rosen kind of QB - Bad, and reliably so.

This data in itself, however, does not tell us much. Is 18% difference from Tier 1 and Tier 2 large, is it very small? Purely from this data, we don’t know. So we need to put it in perspective. Let’s see how those numbers differ from position to position.

All Positions Tier Analysis

First, let’s look at what the normal distribution for all tiers across all positions look like.

Let’s focus on a given position graph. To recap the QB position that we have discussed, we observe that QB1 has a higher PPG, but that QB2 and QB3 are fairly close to each other.

For the RB position We can see the huge disparity between RB1 and any other RB tiers (looking at purely the average PPG.) RB1 Do have a lot more variance than the other tiers, but the difference in PPG is huge and makes up for it.

At the WR position, we see less of a steep drop off from WR1 to WR2, but we do observe that WR1 are very consistent - so good, and consistently so.

The TE position is a bit more ambiguous. We see that TE1 have a higher PPG, but still fairly close to TE2, with similar consistency. TE3 and TE4 however are much more removed - they offer a much lower PPG and consistently so.

Now let’s quantify and compare those differences.

This chart answers a question we had asked earlier. Is the 18% difference from QB1 to QB2 PPG, or the 19% difference from QB1 to QB3 PPG considered big, or small? Well, looking at the other differences across other positions, we can clearly say that it’s a pretty small difference.

The steepest drop off from Tier 1 to Tier 2 occurs at the Running Back position, where an RB2 on average, is almost 65% less valuable than an RB1. While the drop off from Tier 1 to Tier 2 is similar from WR or TE to QB, the drop off at Tier 3 is a lot more substantial, and even more so at Tier 4.

Another common theme in this grid, is the consistency of the Tier 4. This is essentially saying that a Tier 4 player has a much lower PPG than a Tier 1 player, and does so more consistently. So all in all, we are mostly interested in Tier1-3 players, and Tier 4 players are more so fillers for bye weeks or injuries.

  1. The relative value of RB,WR,QB and TE
  2. Why a stud RB is vital
  3. Why over-drafting a QB is not necessary

From the charts above, we can easily understand point 1 and 2. There is a huge discrepancy between an elite RB and any other RB, so having one on your team is vital. Together with the high PPG that we see RB1 bring in, we can easily establish the RB position as the most valuable Fantasy position, since ultimately, you want to have as many RB1 potential RB on your team/bench, more so than any other positions.

We also saw that WR1 have a higher PPG than TE1. However, the drop off from tiers across WR and TE are very similar.Historically, however, they have not been. Last year was a very strange year for TE, whereby only 2-3 TEs emerged as useful from a Fantasy standpoint. So as of now, the data set is skewed by these guys.

In short, bare with me when I say that WR1 is more valuable than a TE1, and that WR is more valuable than TE in general. I will conduct a specific analysis on TE very soon.

As for why QBs are valued a lot less, we’ve already established a baseline. There’s very little difference, relatively, between a QB1 and a QB3, so you do not need to reach for a QB1 when a QB3 will do a very similar job.

Draftnig Scenarios

To prove our third point, let’s look at some concrete drafting scenarios. First, let’s look what two different Fantasy players may do in two rounds, namely, round 3 and round 7. Player 1 will reach for a QB in the early round, while player two will get a RB, and their later round choices are inverted.

Scenario 1:

  • Player 1: Drafts a QB1 in Round 3 and an RB4 in round 7

  • Player 2: Drafts an RB2 in round 3 and a QB3 in round 7

First, let’s take a quick look at the PPG and Variance for each tiers within QB and RB. We will also calculate the “floor” and “ceiling” of each tiers - how low is low and high is high for a given tier. For those interested, the floor will be mean - (1.2*SD) and the ceiling will be mean + (1.2*SD). This will statistically capture over 75% of the expected outcomes for the tiers, and hence be great proxy for floors and ceilings.

## [1] "Quarter Back Tiers 1-4"
##  Pos.Tier Tier.PPG     Floor  Ceiling
##       QB1 22.60156 13.189617 32.01351
##       QB2 19.22800  6.133576 32.32242
##       QB3 18.96103 10.133716 27.78835
##       QB4 16.58973  9.619869 23.55959
## [1] "Running Back Tiers 1-4"
##  Pos.Tier  Tier.PPG      Floor  Ceiling
##       RB1 19.573214  8.5785367 30.56789
##       RB2 11.881951  1.9433858 21.82052
##       RB3  8.752419 -0.1461172 17.65096
##       RB4  7.640964  1.6176191 13.66431

Using that information, let’s calculate the ranges for the two players:

## [1] "Player 1 and 2 Averages, Floors and Ceilings: "
##  Player.and.Selection  Average    Floor  Ceiling
##         P1: QB1 + RB4 30.24253 14.80724 45.67782
##         P2: RB2 + QB3 30.84299 12.07710 49.60887
## [1] "Differential between Player 2 and Player 1 (P2 - P1): "
##    Average     Floor  Ceiling
##  0.6004593 -2.730135 3.931053

What the differential tells us, is that Player 2 has a very slightly higher PPG - essentially, breaking even. But Player 2 has a higher floor, at the expense of a lower ceiling.

Let’s add a layer to this scenario. In reality, the RB2 that you are drafting in the 3rd round, will be someone who has high RB1 potential - that is, someone who you expects will have a breakout season and jump tier. So let’s assume that the RB2 of player 1 becomes an RB1 and similarly, that the RB4 of Player 2, becomes an RB3.

Scenario 2:

  • Player 1: Drafts a QB1 in Round 3 and an RB4 (who jumps tier to an RB3 through the season) in round 7

  • Player 2: Drafts an RB2 (who jumps tier to an RB1 through the season) in round 3 and a QB3 in round 7

Now let’s evaluate the ranges for both players.

## [1] "Quarter Back Tiers 1-4"
##  Pos.Tier Tier.PPG     Floor  Ceiling
##       QB1 22.60156 13.189617 32.01351
##       QB2 19.22800  6.133576 32.32242
##       QB3 18.96103 10.133716 27.78835
##       QB4 16.58973  9.619869 23.55959
## [1] "Running Back Tiers 1-4"
##  Pos.Tier  Tier.PPG      Floor  Ceiling
##       RB1 19.573214  8.5785367 30.56789
##       RB2 11.881951  1.9433858 21.82052
##       RB3  8.752419 -0.1461172 17.65096
##       RB4  7.640964  1.6176191 13.66431

Using that information, let’s calculate the ranges for the two players:

## [1] "Player 1 and 2 Averages, Floors and Ceilings: "
##  Player.and.Selection  Average    Floor  Ceiling
##         P1: QB1 + RB4 31.35398 13.04350 49.66446
##         P2: RB2 + QB3 38.53425 18.71225 58.35624
## [1] "Differential between Player 2 and Player 1 (P2 - P1): "
##   Average    Floor  Ceiling
##  7.180267 5.668753 8.691781

Now we see the true advantage that player 2 has. In the first scenario, where the RB stay within their tiers in the season, Player 2 breaks even. In the scenario where the RB jumps tier, Player 2 emerges with a huge advantage of +7 PPG in any given week, with a higher floor and ceiling. This is the reason why you don’t draft Patrick Mahomes in the 3rd, but rather an Aaron Jones or Marlon Mack - for the massive upside that they have.

This analysis is not limited to the RB position only. Let’s look at the returns for WR and TE as well.

## [1] "Scenario 1 Differential (No Potential Jump): QB1 + RB4/WR4/TE4 or QB3 + RB2/WR2/TE2"
##      Average   Minimum  Maximum
## RB 0.6004593 -2.730135 3.931053
## WR 1.4120057 -1.471466 4.295477
## TE 0.7451033 -2.217701 3.707907
## [1] "Scenario 2 Differential (With Potential Jump): QB1 + RB3/WR3/TE3 or QB3 + RB1/WR1/TE1"
##     Average  Minimum   Maximum
## RB 7.180267 5.668753 8.6917813
## WR 2.316051 3.673565 0.9585369
## TE 2.366879 1.008282 3.7254765

We can identify very similar trends across positions. In scenario 1, Player 2 breaks even, and in scenario 2, Player 2 has a higher PPG, floor and ceiling - although not as big across WR/TE relative to RB.

Conclusion

We have now evaluated the positional values of QB, RB, WR and TE in Fantasy Football. Despite QB averaging more PPG, we have seen why it is more important to get a RB1, WR1 and TE1 ahead of a QB1, as well as why it is not in your best interest to reach for a QB. Instead, you are better off betting on the upside of other players.

We have also observed the big discrepancies in tiers across RB, WR and TE and why RB emerges as the most valuable position in the league, followed by WR, TE and QB, in that order.

In the next installments, we’ll go over specific drafting scenarios and answer questions such as: 1. should you always go RB/RB? 2. How far do you let a QB1 fall before they become valuable at their current spot? 3. When do you draft TE given the duopoly of good TEs last year?

Stay tuned!