Should you stack players in redraft Fantasy Football Leagues?
Stacking. It is a controversial idea. For those who haven't heard the term before, stacking is the idea of having both a quarterback and his receivers (whether WRs or TEs) on your team. It became really popular through Daily Fantasy where hitting on the right stack became a near necessity for finishing in the top 5% of thousands of lineups. As the strategy gained popularity, season-long fantasy players began to wonder, should they be stacking players as well? After all, isn't the goal to score as many points as possible?
Personally, I disagree with this. As I've argued before, the point is not to score the most points, but to finish with the most wins. Since the goal each week is to beat your opponent scoring lots of points is important, but not busting is also important. In season-long fantasy you don't need to score in the top 5% of thousands of leagues, you just need to be better than your opponent. On top of this, the downside of stacking is what do you do when the stack has a bad matchup? In daily fantasy, you just don't choose a stack on a bad matchup, but in season long, your options are far more limited. Choosing against the stack you drafted often means playing inferior players.
One last argument you will hear though is that even in season-long fantasy it is a winner take all game. Most leagues only reward the champion. 2nd place is first loser. So while stacking players may be riskier, it raises your team's ceiling and therefore can lead to more championships (while also increasing the number of last-place finishes). The combination of more points and a riskier strategy is exactly what Mike Leone of Establish The Run argues here. He ends by concluding that drafting a stack is worth it because if one player in the stack hits, normally both players do (stud QB leads to better performance by his WRs and vice versa). I wanted to explore this question a little differently, specifically by looking at the road not traveled.
So today we are going to explore stacking in season-long fantasy football. Should you draft a stack? Does stacking actually create more variance and is that even something you want? and how does a stacked team compare with a non-stacked team?
To determine whether you should draft a stack I pulled 2019 and 2020 ADP data from Fantasy Football Calculator, and for each player, I determined ADP swaps, other players at the same position whose ADP were within 1 standard deviation of the original player. This gives an accurate description of who the debate actually was between (for example, a potential swap for Tyreek Hill last year would be Julio Jones, but not Chris Godwin). The goal here is to compare who players were likely debating between so we can compare how their team without a stack would have performed if they had drafted the stack. The other benefit of using standard deviation is that it means the range of options increases as we get later in the draft (so a 1st or 2nd round swap needs to be within a pick or 2 in terms of ADP while a 5th or 6th round pick maybe 5 picks away in terms of ADP). This more accurately reflects who the drafter was choosing between. I then collected weekly performance data from 2019 and 2020 for ~2300 public ESPN teams. Finally, for each week I looked to see if there were options to create a stack on the team by changing the QB, WR, or TE. For instance, if a team had Matt Ryan starting at QB and Tyreek Hill at WR, I calculated how many points that team would have scored in a given week if they had started Julio Jones instead of Tyreek Hill. (Note I ignored any swap that had a projected score for the week of 0 meaning they were either on a bye or hurt for the week). This then produces a set of non stacked and stacked teams where the only difference is one player swapped for another with a similar ADP.
How does a stacked team compare to a non-stacked team?
Violin plot of the difference between a stacked and unstacked weekly score. Positive values indicate better performance with the stack, negative values indicate better performance with the unstacked team. inner white dot represents median with a box plot around it.
On the whole, the stacked team performed slightly worse than the unstacked team (mean difference stacked team scored 2.5 points less than the unstacked team). Now, this frankly is not enough of a difference to matter in my opinion, especially when you look at the distribution above. However, there are likely some low-value stacks occurring here (think stacking Drew Lock with Jerry Jeudy). The conclusions from Establish The Run are about the players that hit and assume you will ignore any bad stack. We can explore this by limiting our sample to creating a stack where one of the players finished as a top-half starter (so top 6 QB/TE or top 12 WR).
Violin plot of the difference in weekly score between stacked and unstacked teams with only high-value stacks included. The white dot represents the median with a box plot around it as the thick black line.
Again we see basically no difference between the stacked and unstacked team, though with a good amount of variance. So at least in terms of points when accounting for the entire season, drafting a stack does not seem to matter. but I've said many times, points are good but what really matters are wins, so let's look at wins.
Violin plots of difference in the number of wins for stacked and unstacked teams looking at all stacks (first graph) and only high-value stacks (second graph).
Similar to points, on average, there is no real difference between stacked and unstacked (though stacking again performs a little worse on average with 0.3 fewer wins). We also get the same result when we only look at the high-value stacks. So again there is not a significant difference between stacked and unstacked teams, so it probably does not matter. Based on this, I am going to ignore the high-value stacks for the rest of this post, just remember that the results don't really change when you focus on the high-value stacks.
What about the win total odds?
Bar plot of the odds of different regular-season win totals for stacked and unstacked teams
Ok, so a few points. First, the overall shape of this graph is a little random because I am not including every team from every league I pulled from, only teams that had the option for an ADP swap in their starting lineup. That said, the main takeaway I have is stacked teams (blue) consistently recorded fewer wins than their unstacked version (orange). This is in line with what we just saw but is a little more visible here. High-value stacks perform a little better, but the story is the same.
Ok but what about Variance?
Bar plot of the week to week variance in team score for stacked and unstacked teams
And how about that. The stacked lineups actually had less, not more variance. I will be honest and say that this surprises me. On the one hand, there is probably fewer degrees of freedom on the stacked team (with 2 players being on the same team and therefore having their performance connected), but still, at a team level, I expected the stacked team to be more variable on a week to week basis.
Does weekly variance even matter?
Bar plot of the week to week team score variance by final standing, error bars represent 95% confidence intervals
And again we see no real trends. Teams that finish first do not have a higher weekly score variance than those that finish last. So even if stacked teams were more variable, I'm not convinced it is something we even should be aspiring for. This point was actually made last year with simulation by Larry Stenger over at stathouse. Dont worry about variance, just get the best players.
So one limitation we have already discussed, the fact that I am looking at a nonrandom subset of individual teams rather than all possible teams across all of the leagues. This may lead to some weird graphs, but should not change the takeaway point because the comparison between stacked and unstacked rosters is at the individual team level. A second limitation is that the high variance of a stack may be at a season-long level (a good season vs a bad season) rather than on a weekly score basis as I looked at here. This would be more in line with the logic from Establish The Run, with the idea that you would only play high-value stacks. While I think that is fair, these results did not significantly change when we only included high-value stacks. Even so, we care about comparing stacked and unstacked versions of the same team, and consistently, the unstacked team performed the same or better than the stacked version. One final limitation is that maybe the ADP swap is at a different position, with the idea to wait for the stack option in the next round rather than draft the better player in a vacuum. This is certainly possible but is much harder to analyze given the butterfly effect such a change would incur.
At the end of the day, I am not convinced stacking is worth it or frankly matters in season-long fantasy football. You should draft the best player available and whether or not you are creating a stack should not factor into that decision. If you thought Matt Ryan in the 8th was a better pick than Patrick Mahomes in the second, then you should take Matt Ryan regardless if you had already drafted Julio Jones or Tyreek Hill.