One of the most important sources -- if not the most important source -- of arbitrage in playoff fantasy football is in accurately projecting how many games each team is going to play. Last postseason, I detailed a system for doing just that.
And looking back at how the 2014 playoffs actually turned out, the system did quite well. One way to test its accuracy is to compare the expected games for each team that my system projected versus the expected games for each team that was implied by the actual Vegas lines in the games they actually played. For instance, going into last year's playoffs, my article projected 2.30 expected games for Seattle based on the following likelihoods for them playing a given number of games:
- Probability of playing exactly 1 game = 18.7%
- Probability of playing exactly 2 games = 32.4%
- Probability of playing exactly 3 games = 48.9%
- Probability of playing exactly 4 games = 0.0% (They had a first-round bye, so 4 games was impossible.)
The math to convert the above probabilities into expected games is (18.7% * 1) + (32.4% * 2) + (48.9% * 3) + (0.0% * 4) = 2.30.
In comparison, the probabilities based on the Vegas lines for games Seattle actually played were as follows:
- Probability of playing exactly 1 game = 19.6%
- Probability of playing exactly 2 games = 21.2%
- Probability of playing exactly 3 games = 59.2%
- Probability of playing exactly 4 games = 0.0%
And the math to convert these probabilities into expected games is (19.6% * 1) + (21.2% * 2) + (59.2% * 3) + (0.0% * 4) = 2.40.
Based on the above, my system was off by 0.10 expected games, which is good enough for me, especially when you remember that, going into the Wild Card round of last year's playoffs, I didn't know the actual line for any of the games Seattle would end up playing. Now, if we repeat this exercise, calculating "expected game error" for all 12 teams, we get the following list:
|Team||proj Exp G||Vegas Exp G||absolute Error|
The bottom row shows that the average error across all of last year's playoff teams was 0.12 expected games, from which we can reasonably conclude that my system did a good job of projecting Vegas lines in future rounds (and therefore, future win probabilities; and therefore, expected games). The two biggest misses were because Pittsburgh lost as a 60-40 favorite to Baltimore in the Wild Card round and Denver lost as a 70-30 favorite to Indianapolis in the Divisional round. However, if you clicked on the link to last year's column, you'll notice that Baltimore was actually one of the teams I identified as having an expected games total higher than what was implied by their (No. 6) seed. The other team I identified as such was Dallas (i.e., a No. 3 seed that ranked ahead of two No. 2 seeds in expected games). They, like Baltimore, held a fourth-quarter lead on the road in the Divisional Round, only to see controversy -- to put it politely -- dash their hopes and prevent my system from producing even better results. Maybe that's what it's actually good at projecting: soul-crushing defeat!
expected playoff games played
Alright, after looking back, it's time to look forward to this year's playoffs. The procedure I used last year to calculate expected games is the same, except for one wisdom of crowds-related component. This season, Fivethirtyeight.com and ESPN Stats & Info have joined Football Outsiders as statistical outlets that publish playoff win probabilities, so I'm now sourcing them as well. Furthermore, for the purposes of calculating Wild Card win probabilities, in addition to those implied by the Vegas lines, I'm also including as a source the probabilities implied by Pro Football Reference's Simple Rating System (SRS) because the combination of Vegas lines and SRS-implied lines ended up giving me great personal satisfaction in Footballguys' Staff Confidence Pool this season. (Just kidding. Again, it's because the more sources, the better.)
So, without further ado, here are the expected game totals for each team in this year's playoffs:
|Team||P(1)||P(2)||P(3)||P(4)||Exp G||Adj Exp G|
In case you didn't read last year's article, here's how to read the table. In the header row, P(1), for example, means "the probability of playing exactly one game," P(2), means "the probability of playing exactly two games," and so on. EXP G is the number of expected games, and ADJ EXP G is the number of expected games if you're playing in a contest where fantasy points accumulated in the Super Bowl counts double. (Again, refer to last year's article if you want to see how I calculate this ADJustment to EXP Games.)
So, take Arizona, for instance. They have a 0.0% probability of playing four games because it's impossible for teams with a first-round bye to do so. Meanwhile, they have a 27.2% chance of playing exactly one game (i.e., losing in the Divisional round), a 38.8% chance of playing exactly two games (i.e., winning their Divisional round game, but losing the NFC Championship game), and a 34.0% chance of playing exactly three games (i.e., winning both their Divisional round game and the NFC Championship game). Applying the math I did earlier in this article, this set of probabilities means the Cardinals can be expected to play 2.07 games. And for contests that double-count Super Bowl points, Arizona's 2.07 expected games is the equivalent of 2.41 games after adjusting for their 34.0% chance of making the Super Bowl [i.e., P(3)].
In a typical postseason (with last year being a prime example), higher seeds are more likely to win any given game because they're at home and they're probably the "truly" better team. As has been discussed extensively over the past few days, however, this week's matchups (and potential matchups in upcoming weeks) are (and will be) rife with conflict between statistics and seedings. Two teams stand out as having being seeded due to their record far lower than what their statistics suggest: Kansas City and Seattle. According to Football Outsiders, the Seahawks were the most efficient team in the league for the fourth consecutive season (per DVOA) and are the second-most likely to win it all this season, but insurmountable early-season hiccups have made them the No. 6 seed in their conference. Similarly, Kansas City is the No. 5 seed in the AFC despite ranking 5th among all teams in both DVOA and Super Bowl win probability. It's not just Football Outsiders that holds these views, by the way. Fivethirtyeight has Seattle No. 1 overall with the 5th-highest Super Bowl win probability, while Kansas City's ranked 4th and 6th, respectively. Per Pro Football Reference, Seattle and Kansas City are the 2nd- and 4th-best teams in the NFL, respectively.
On the flip side, Denver is overseeded in comparison to their metrics: 8th in the league per DVOA, 5th per Fivethiryeight, and tied for 8th among playoff teams per Pro Football Reference. And by virtue of only being able to play three games maximum, their expected games total falls below both Seattle's and Kansas City's. Even if you're playing in a "Super Bowl counts double" format, the Broncos only ascend into a virtual tie with them.
All of the above illustrates what I think most experts would agree is the key to dominating playoff fantasy contests. NFL rules dictate that four playoff teams can only play three games, while eight can play four games. Most people are going to try to load their roster with players from those three-max teams because a) they're usually the best teams, and b) they start the playoffs closer to the Super Bowl. Therefore, the shortest path to playoff fantasy victory, regardless of format, is to attack your opponents from both sides: Correctly identify the four-max teams that will overperform their seed and correctly avoid the three-max teams that will underperform their seed. It's easier said than done, and as much art as science, but hopefully my expected games system will be the science part that helps you get it done.
projected playoff fantasy points
But that's not all!
Like last year, I've used the expected game totals above to calculate expected FFPC points for both standard (EXP PTS) and "Super Bowl counts double" (ADJ EXP PTS) contests. For the vast majority of players, I've simply multiplied their scoring averages over Weeks 13-16 by the expected games total for their team. The exceptions were players for which injuries have affected their Week 13-16 scoring averages and/or their expected playoff games. In their cases, I did several hours of research to produce better approximations of their expected points.
As an example of forming an educated guess about a player affected by injury in Weeks 13-16, take Tyler Eifert. He missed Weeks 13, 15, and 16, so his relevant scoring average would only be based on one game; that's not ideal. At the same time, all games prior to Week 13, and even the game in Week 14 that he did play, were started by Andy Dalton, who is out this week (and perhaps beyond), so calculating an average using games prior to Week 13 wouldn't be a good reflection of how much he scores with A.J. McCarron at quarterback. Therefore, as loathe as I am to use stats from Week 17 for anything besides attempts at humor, both Cincinnati and their opponent, Baltimore, played their starters the entire way, so I included Eifert's points against the Ravens -- along with his points from Weeks 10-12 -- in his four-game average.
As an example of forming an educated guess about a player whose expected games total will be affected by injury (and will produce cascading effects for other players), take Houston's wide receiver situation. Studying Houston's snap counts, I noticed two recent phenomena relevant to their likely usage rates in the playoffs: 1) Jaelen Strong gets Cecil Shorts' snaps when he's out, but plays about 12 snaps when Shorts is healthy; and 2) Keith Mumphery gets Nate Washington's snaps when he's out, but barely sees the field when Washington's healthy. Aside from Strong having more talent than Mumphery, this pattern occurs (presumably) because Strong can play the second slot receiver in a four-wide formation (and give Shorts a breather in three-wide sets from time to time), while Mumphery is "line up outside or bust." Therefore, with Shorts healthy and Washington likely out against Kansas City, I prorated Mumphery's points-per-snap average to the average number of snaps Washington plays when healthy, and gave Strong his usual modicum of 12 snaps when Shorts is healthy. For subsequent weeks, I'm assuming Washington will be fine, and so Mumphery then goes from a Washington-level number of snaps this week to virtually absent.
Can these educated guesses be wrong? Of course. That's why, in the table below, I've put in bold, red font all of the expected points (and adjusted expected points) for players about whom I made an educated guess. I've also put in bold, blue font one player whose Week 13-16 scoring average was totally unsustainable, so I opted to use his Week 1-16 average instead. Move any of these players up or down in your own personal rankings as you wish:
|Player||Pos||Tm||Exp Pts||Adj Exp Pts|
|Ted Ginn Jr||WR||CAR||40.2||48.0|
|Marvin Jones Jr||WR||CIN||21.6||22.7|
|Kansas City Chiefs||D||KAN||27.2||29.2|
|New England Patriots||D||NWE||20.7||24.0|
|Green Bay Packers||D||GNB||14.7||15.0|
More articles from Danny TuccittoSee all
Expected Playoff Games Played
Expected Playoff Games Played
True Fantasy Points: 2019 Quarterback Projections
More articles on: ProjectionsSee all
3 Lessons Learned After Week 14 - Allen
3 Lessons Learned After Week 12 - Allen
How Teddy Bridgewater and Joe Brady Could Be the Perfect Combo - Larkin