The Fantasy Football Mythbusters: Making the Most of Matchups
One of my favorite statistics to look at when making weekly start/sit decisions is the positional ‘Opponent Rank’; that is, how the defense stacks up against my player’s position. I love to play the ‘matchup’ game: targeting a QB, for instance, going against a weak secondary, or pivoting away from a running back facing a stout front seven (or, of course, starting T.Y. Hilton against the Houston Texans!). To me, it’s always felt like an overlooked metric, one that can make the difference between a win and a loss.
However, do these matchups really matter? Is it fair to assume that great players will stumble against solid defenses and that less desirable options facing weak defenses are a better bet?
In the most recent entry in the Fantasy Mythbusters series, we explored how much (if at all) NFL game script dictates fantasy output. This time, we’ll be focused entirely on matchups. Data is from 2005 and comes from nflfastR; all fantasy points are in Half PPR scoring.
For each game, at each position, I calculate an offense rank and a defensive rank. This approach is simple: how many points that NFL team has scored (or allowed) at that position compared to other teams in the past five games. This is great way to think especially when you bet on your teams on sites like 해외축구.
So, for example, if Dalvin Cook goes off Weeks 1-5 and, in this span, the Vikings have the most total fantasy points at the RB position compared to all other NFL teams, they would have an offensive RB rank of 1. Say they are playing the Packers in Week 6, with a defensive rank of 10: this means that, in Weeks 1-5, the Packers gave up the 10th least number of points to the running back position.
There are certainly flaws with this approach; for example, a team might go through an easy stretch of games and thus have an overrated offense or defense ranking (here’s looking at you 2019 New England Patriots). It also doesn’t account for injuries to key players between weeks that might turn a vaunted defense into a pedestrian one. Still, in the long run, this should mostly balance out. I use a 5-game window instead of the full ‘season to date’ to try to get the ‘form’ that the team is currently in.
Without further ado, let’s look at some positional charts.
We’ll start with signal-callers, partly because their charts are the most intuitive to read. The first chart, a heat map, gives the average PPG for QBs (number in the box) based on how strong the defense is (x-axis) and how strong the offense is (y-axis). The ranks are rounded to the nearest five (i.e., teams ranked 1-5 are all just lumped into the Top-5) to make this chart easier to look at. So, for example, when a top-5 QB offense goes against a Bottom-5 QB defense, the QB scores on average 19.3 points (top right box). Aside from some broad trends, which we explore below, we can use this to answer fun questions: would you rather have a great (top-5) QB offense against a great (top-5) QB defense, or a terrible (Bottom-5) QB offense against a terrible (Bottom-5) QB defense? According to this map, the answer is to go with the great QB offense (14.5 points vs. 12 points).
A simpler look breaks out the average scoring for both offensive and defensive ranks in separate charts (fantasy points on the y-axis, ranks on the x-axis):
- All told, it’s clear that both offense and defense matter for QBs. In our heat map, we see that boxes get redder as we move up (stronger offense) and to the right (weaker defense). We see the same effect in our dot plots: points go down as offense rank gets worse, and up as defensive rank gets worse.
- Another way to consider this is to build a regression model that predicts QB PPG based on the rankings. The output: each one-spot improvement in QB offensive rank brings an extra 0.22 expected PPG for QBs, and each one-spot improvement in QB defensive rank deducts 0.07 expected PPG (both are highly statistically significant). This is interesting since, while both matter, offensive rank matters more. About three times more, in fact, which makes sense: when predicting how an offense will do, it’s more important to know how that offense is performing, not how the defense it is facing has performed against other offenses. We can also see this effect in the chart: box redness drops off quicker when we move from up to down (good QB offense to bad) than right to left (bad QB defense to good) and the relationship to fantasy points is much stronger for offensive rank than defensive in the dotplots.
- There is a sort of ‘superstar’ effect here: the #1 overall ranked QB offense is notably far away from the rest of the field, and the top-3 or so QB defenses are significantly stingier than the other defenses. This makes sense since football is not a linear game!
Here are the RB charts, with the discussion below. These point totals represent the total points scored by all RBs on a roster, not just the RB1 of a team.
- The first major difference – at least to me – in this chart is how much more heat (red boxes) is clustered in the top row (Top 5 RB offenses) compared to QBs. This goes to show that top running backs are very valuable, and there is a reason drafts usually start RB-heavy!
- We see a similar effect here: stronger offense and weaker defense both mean more points, and vice versa. To quantify what’s going on, we can use the same regression model as before. For each single-spot improvement in RB offense ranking, expected points increases by .22 points, which is identical to QBs. However, for each single-spot improvement in RB defense ranking, expected points decreases by .13 points, which is twice the effect of defense on quarterbacks. This is a difference of nearly 4.2 PPG to the RB position between the #1 ranked RB defense and the #32 ranked RB defense; certainly not huge, but definitely a solid difference.
- Although defenses make a bigger difference for running backs, note that Top-5 RB teams are the most resilient to defensive matchups. To see this, examine the difference between a Top-5 RB offense going against a Top and Bottom-5 RB defense. The difference is just 3.6 points: 26.6 points against a Bottom-5 defense and 23 points against a Top-5 defense. For worse running backs – basically any team below a Top-5 RB offense – this difference is larger, with a drop-off of about 5-7 points when going from the worst to best defenses. These numbers suggest top backs are largely ‘matchup-proof: on average, they still perform quite well even against tough defenses.
Time for the wideout charts, which look pretty strange. Again, points here represent the total points scored by all WRs on a roster, not just the WR1 of a team.
- What is immediately odd about the heat map is that the top left box – Top 5 WR offense, Top 5 WR defense – is the reddest, with nearly 41 WR PPG. This is likely due to a small sample size; it turns out great WR offenses don’t play great WR defenses as much as some of the other combinations, probably in part because of a divisional effect. Still, this segues into the main message of this chart: offense appears to matter much, much more than defense (box color is more dependent on moving up and down than side to side). We see a similar result in our dot plots: the relationship between offensive rank and points scored is strong, but not so much with defensive rank.
- Our regression analysis confirms this hunch. While the defensive output is similar to QBs (-.09 points expected for every improvement in defensive rank, which is relatively less because WRs score more), the offensive output is extreme: for every improvement in offensive rank, WRs are expected to put up .41 more points (team total). This is partly a function of WRs just scoring more in general (same yardage and touchdowns as QBs, but more points for it) and just emphasizes that offense is far more important (the size of the pie!) for wideouts than defensive matchups.
- We see another offensive ‘superstar effect’ for wide receivers: the top-3 offenses are far beyond the rest of the field in the dot plots.
Finally, let’s turn to the tight end charts.
- What’s crazy – and, perhaps, maddening – about Tight Ends is that defense doesn’t seem to matter! The redness of the boxes in the heat map doesn’t depend on moving left to right, and the defensive dot plot doesn’t seem to have any sort of relationship either up or down. Indeed, in the regression analysis, there is no expected point change for an improvement in defensive rankings; what’s more, that coefficient isn’t even significant! This could be because TE is a strange position, but it could also be because it is a sparse one: there are not many elite tight end options in the league, so a team could bolster its TE defensive ranking just by not playing the Kansas City Chiefs or Las Vegas Raiders for a few weeks.
- Offense, as one might expect, is a different story. Per the regression, for each improvement in offensive ranking, TEs are expected to score .23 more points a game (which is a lot at this low-scoring position). What’s more, there is a very strong superstar effect here, as the top two ranked TE offenses are very far removed from the pack. This is intuitive: we’ve seen how grabbing an elite TE like Travis Kelce or Darren Waller can make a huge positional difference for your team.
PLAUSIBLE, and closer to confirmed than busted. While previous offense performance is – naturally – more predictive of future offensive performance than the previous defensive performance of the opposing team, we still see expected points starting to decrease as defenses get stronger, especially for running backs and quarterbacks (less so for wideouts). TEs, interestingly, are exempt from this trend: it doesn’t seem to matter the strength of the defense they are facing (although, as discussed above, this could be a quirk of this weird position).
However, what surprised me – and what prevents this from being fully confirmed – is that difference isn’t that large. For RBs, the position at which matchups have the largest effect, each improvement in defensive ranking means just .13 fewer expected points. The distance, as discussed, between the absolute best and worst ranked defenses is just over four expected fantasy points in aggregate for all RBs on an offense (although, as we saw, for non-Top 5 ‘matchup proof’ RB offenses, the difference is greater: about 5-7 points). Still, while the improvement might not be very big, it’s significant (definitely there statistically) and, anyways, fantasy is about finding every edge you possibly can.
Did I miss anything? Have a myth you want to explore more? Message me on Twitter.
Bonus: in writing this article, I uncovered some fun stats about individual teams. One is the Buffalo Bills: they’ve had the 3rd best QB defense (in terms of fantasy points allowed to the position) since 2005, and yet the 31st best (second-worst) run defense in that span! This could of course be a function of weather in Buffalo, but either way, it’s certainly been easy to game plan for the Bills…