As fantasy managers, we’re all pretty familiar with positional rankings. We see rankings on every website and use them in every draft. Increasingly, these rankings are enhanced by being grouped into tiers. Yet how many of us have thought about the conceptual differences between tiers versus rankings? And why should we care?
Rankings and tiers each imply unique structures of player value. Rankings imply that player values can be ordered sequentially and that the difference in value between each player is the same. So, for example, an idealized ordinal ranking of 20 QBs might look like this:
The least valuable (QB20) is worth a dollar, the second least valuable (QB19) is worth two dollars, and so on, all the way to the most valuable (QB1), which is worth $20. Nobody actually values players like this, of course, but it is what untiered rankings imply. And this implication can impact our draft behavior in sneaky ways that we don’t always consciously notice.
Let’s now contrast this with tiers. Tiers imply an ordinal ranking of groups. Placing one player above another only makes sense if it indicates membership in a higher tier. With rankings, we want the highest-ranked player. With tiers, we want a player in the highest-ranked tier grouping. An idealized version might look like this:
Again, nobody actually believes this either. Usually, we combine rankings and tiers, resulting in an idealized hybrid that might look like this:
These contrasting structures have important implications for the different approaches required for snake and auction drafts. In a snake draft, you don’t really need to have a robust sense of tiers. Once you’ve chosen what position you want to fill, you can simply choose the highest-ranked player in that position (though the fear of getting bumped into a lower tier with your next pick might influence your approach).
With an auction draft, in contrast, you don’t really need a robust sense of overall ordinal rankings. Once you have your tiers set, every player in a given tier has the same value to you, which means that any variation in prices between them represents a “market failure” that you can capitalize on.
If these claims seem overly abstract it is because they are really just theories that need to be tested with actual data. One way to test them is to look at the actual value that each player produced over a season. To do this, I’ll use the average fantasy points per game, or AFP. If actual AFP looks like Figure 1, we can ignore tiers altogether. If it looks like Figure 2, we can ignore ordinal rankings altogether. (Spoiler alert: reality is messy and it doesn’t completely look like either.) So the reality here actually matters, and the “shape” of these values can depend on both position and scoring system.
Starting with quarterbacks, let’s take a look at how their 2018 values shook out. I’ll first consider the 33 quarterbacks who were actually drafted in one of my 2-QB auction draft non-PPR 12-player leagues. Figure 4 charts their performance ordered from lowest to highest AFP along the x-axis:
The first thing to note is that the stylized graph that this slice of “reality” looks the most like is the ranking-tier hybrid. Mahomes is in a tier of his own. Tier 2 is just Ryan and Roethlisberger. Then we have a large Tier 3 that includes Winston, Trubisky, Brees, Goff, Newton, Watson, and Luck. And finally (for our purposes at least), an even larger Tier 4 that includes Prescott, Brady, Mayfield, Rivers, Rodgers, Wilson, Cousins, Allen, and Wentz.
If you are in a 1-QB league, then you might be mostly interested in the top 12 quarterbacks (the QB1’s). You could then zoom in on the top 12 performers, like this:
From this perspective I think the tiers seem less meaningful and the ranking looks fairly ordinal — with the exception of the extraordinary outlier that Mahomes was last year. We can see that the risk or cost of waiting for the 12th best amounts to 8.4 points per week. This is useful because it allows us to compare with other positions in order to determine which to target most aggressively.
If, however, you are in a 2-QB league, then you might be more interested in the shape of the top 24 quarterbacks (the QB1s and QB2s), like we see here:
Now we are looking at a range of 11.4 points. Despite arguably three discernible tiers, the pattern is still surprisingly linear. To see this just imagine drawing a straight line through all the non-Mahomes data points.
Turning to running backs, we see a very different spread of values in 2018. Here are the top 24 (RB1s and RB2s) based on standard, non-PPR scoring:
The AFPs of the top 24 QBs had a range of 11.4 points. For RBs, the range is nearly the same, at 12 points.
Compared to the QB AFPs, the RB AFPs cluster more distinctly into tiers. Gurley is in a tier of his own. Tier 2 is Gordon, Kamara, and Barkley. Tier 3 is Conner, McCaffrey, and Elliot. And tier 4 is a large ordinally arranged group consisting of Mixon, Mack, Carson, Lindsey, Johnson, Jones, and Fournette. The gap between tiers 3 and 4 is over two points, which suggests some urgency in securing an RB or two on the right side of the gap. This chimes with the common belief that it is important to get top RBs early.
Looking at wide receivers, here are the top 36 (the WR1s, WR2s, and WR3s) again based on non-PPR scoring:
With the receivers, we see a spread of 7.1 AFPs from lowest to highest. This time there is no Mahomes or Gurley pacing the field at the top. The shape of the WR AFPs is unique. Wide receivers 40-8 all fall along a fairly straightforward upward linear path. Receivers 7-1 continue along an upward linear path but with greater spacing between each player (indicating a steeper rise in AFP), so much so that these seven receivers cover about 20% of the range of AFPs.
One upshot of all of this is that, at least in 2018, rankings made the most sense for valuing QBs and WRs, while RBs clustered into meaningful tiers.
This overall finding of rankings over tiers has an important implication for draft strategy that I didn’t anticipate at the outset of this analysis. Smoothing over subtleties, if we are mostly looking at a steady linear (or even logarithmic) decline in AFP as we fall from the top performers, then we need to be most aggressive with the highest scoring positions.
In these data, for example, the average AFP for the top 5 players by position is 14.3 for WRs, 19 for RBs, and 20.4 for QBs. This suggests that we should fade WRs in favor of aggressive moves toward top RBs and QBs — and this is especially true in leagues with flex positions. This will matter less in leagues that have rigid position formats because higher scoring positions tend to be higher than average across their distributions of scores. In leagues with flexible arrangements, however, players who find ways to stack RBs and QBs in place of WRs will have a natural advantage that sometimes flouts name recognition.
Further, in formats that allow QBs in a flex position (superflex), it should be clear from these figures that even a very bad starting quarterback is much better than most running backs and wide receivers. None of the top 24 quarterbacks have an AFP below 12 points.
Compare this to the receivers: 80% of them (32/40) have an AFP below 12 points. This means a rational player in 2018 should’ve wanted to start Andy Dalton over Michael Thomas all else equal. That is quite floutatious. Now compare this to the backs: 60% of them (18/30) have an AFP at or below 12 points. On average, this meant putting QBs like Dalton over high profile RBs like Aaron Jones, Leonard Fournette, and Dalvin Cook.
These are the sorts of general insights that we can get from a broad overview like this. Understanding how well last year’s positions meshed with the assumptions of tiers and rankings is only helpful in so far as our tiers and rankings for this year are accurately arranged and ordered in the first place. The next step, then, is to make sure we get our tiers and rankings right.