Football

Using MFL10 Data to Empirically Derive Value Based Drafting Baselines

JamaalCharles

A bit of a nerd post here as I continue to work through some of the MFL10 data that I’ve downloaded.

Probably the most useful thing that I’ve found in the data so far is that by aggregating information from 350 leagues, it’s really possible to see broad trends in data that might not have even been recognizable to you if you played in a few leagues.  This is an important point I think and one that was underlined to me once in a Twitter exchange where I was complaining about missing the playoffs in 2012 in a league where I had drafted RGIII and Vincent Jackson. Someone on Twitter responded that I should have drafted Peyton Manning and Demaryius Thomas instead. It took me a few seconds to realize that RGIII and Vincent Jackson had scored essentially the same number of points as Peyton/Demaryius and yet my guys had lower ADPs. The rando Twitter instigator was drawing from a likely sample size of 1 in terms of recording his fantasy football lessons for the year.

In order to avoid drawing the wrong conclusions from small samples sizes I took some time over the weekend to do a few things all aimed at figuring out how Value Based Drafting baselines could be pegged to winning a league. I went through the following steps:

  1. Calculate a percent for each player that is equal to the total times that player appeared on a winning MFL10 team, divided by the number of times that player was drafted. So WINRATE = TOTAL FIRST PLACE FINISHED / TOTAL TIMES DRAFTED
  2. Use WINRATE as the dependent variable in a regression that I’ll eventually use to derive VBD baselines.
  3. The next step was to run a multiple regression where the formula is something like WINRATE ~ POINTS OVER BASELINE + ADP, where Points Over Baseline is a player’s fantasy points less the amount of fantasy points scored by some lower ranked player, and where I’m going to use the regression to figure out what ranking that lower ranked player will have. Note also that ADP as a variable in the regression isn’t significant on its own. It’s only a significant variable if it’s included in the same regression with POB. This may sound complicated but it’s not as I’m sure you can think about how fantasy drafts work and understand that the results you have are going to be based on the performance of your players adjusted for where you drafted them. To put it another way, Jamaal Charles outscored Knowshon Moreno, but because Moreno was drafted much later he appeared on slightly more winning teams than Charles did.  That’s all the regression is attempting to explain.
  4. Iteratively test various baselines for each position by determining which baselines yield the best fit in the regression. So I set the QB baseline at QB1, which makes every QB’s POB equal to their points minus Peyton Manning’s points, then run the regression, see how explanatory it is, and then move on to setting the QB baseline to QB2 and testing that baseline in the same manner. I do that with all of QB, RB, WR and TE.
  5. Ultimately I landed at the following baselines: QB16, RB39, WR40 and TE11 (note the starter format in MFL10 is 1QB/2RB/3WR/1TE/1FLEX).

If you use those baselines as well as a ton of hindsight 20/20, you could re-order this year’s draft board using the baselines and player point production where it will come out looking like this:

Overall PLAYER POS POSRNK FP Base Pts Points Over Base
1 Charles, Jamaal KCC RB RB RB1 378 101.7 276.3
2 McCoy, LeSean PHI RB RB RB2 310.2 101.7 208.5
3 Forte, Matt CHI RB RB RB3 299.6 101.7 197.9
4 Manning, Peyton DEN QB QB QB1 435.1 244.8 190.3
5 Moreno, Knowshon DEN RB RB RB4 279.2 101.7 177.5
6 Johnson, Calvin DET WR WR WR1 303.2 134.7 168.5
7 Gordon, Josh CLE WR WR WR2 299.2 134.7 164.5
8 Graham, Jimmy NOS TE TE TE1 285.4 122.2 163.2
9 Brown, Antonio PIT WR WR WR3 291.3 134.7 156.6
10 Lynch, Marshawn SEA RB RB RB5 258.2 101.7 156.5
11 Green, A.J. CIN WR WR WR4 290.5 134.7 155.8
12 Thomas, Demaryius DEN WR WR WR5 289.7 134.7 155
13 Marshall, Brandon CHI WR WR WR6 286.1 134.7 151.4
14 Murray, DeMarco DAL RB RB RB6 246.4 101.7 144.7
15 Jeffery, Alshon CHI WR WR WR7 274.6 134.7 139.9
16 Johnson, Andre HOU WR WR WR8 268.8 134.7 134.1
17 Bryant, Dez DAL WR WR WR9 268.5 134.7 133.8
18 Decker, Eric DEN WR WR WR10 267.1 134.7 132.4
19 Garcon, Pierre WAS WR WR WR11 265.9 134.7 131.2
20 Peterson, Adrian MIN RB RB RB7 232.7 101.7 131
21 Jackson, DeSean PHI WR WR WR12 263.6 134.7 128.9
22 Lacy, Eddie GBP RB RB RB8 226.4 101.7 124.7
23 Brees, Drew NOS QB QB QB2 367.1 244.8 122.3
24 Bush, Reggie DET RB RB RB9 221.7 101.7 120
25 Johnson, Chris TEN RB RB RB10 217.1 101.7 115.4
26 Jackson, Fred BUF RB RB RB11 215 101.7 113.3
27 Bernard, Giovani CIN RB RB RB12 210.6 101.7 108.9
28 Woodhead, Danny SDC RB RB RB13 210.4 101.7 108.7
29 Bell, Joique DET RB RB RB14 207.1 101.7 105.4
30 Nelson, Jordy GBP WR WR WR13 238.3 134.7 103.6
31 Thomas, Pierre NOS RB RB RB15 203.9 101.7 102.2
32 Jackson, Vincent TBB WR WR WR14 234.9 134.7 100.2
33 Edelman, Julian NEP WR WR WR15 234.2 134.7 99.5
34 Bell, Le’Veon PIT RB RB RB16 200.3 101.7 98.6
35 Mathews, Ryan SDC RB RB RB17 192.9 101.7 91.2
36 Gore, Frank SFO RB RB RB18 189.5 101.7 87.8
37 Gonzalez, Tony ATL TE TE TE2 209.3 122.2 87.1
38 Fitzgerald, Larry ARI WR WR WR16 218.9 134.7 84.2
39 Boldin, Anquan SFO WR WR WR17 215 134.7 80.3
40 Cameron, Jordan CLE TE TE TE3 201.8 122.2 79.6
42 Davis, Vernon SFO TE TE TE4 199.5 122.2 77.3
43 Welker, Wes DEN WR WR WR18 210.8 134.7 76.1
44 Stacy, Zac STL RB RB RB19 175.6 101.7 73.9
45 Jones-Drew, Maurice JAC RB RB RB20 175.1 101.7 73.4
46 Allen, Keenan SDC WR WR WR19 207.7 134.7 73
47 Wright, Kendall TEN WR WR WR20 203.9 134.7 69.2
48 Sproles, Darren NOS RB RB RB21 170.5 101.7 68.8
49 Morris, Alfred WAS RB RB RB22 170 101.7 68.3
50 Luck, Andrew IND QB QB QB3 311.6 244.8 66.8
51 Newton, Cam CAR QB QB QB4 310.3 244.8 65.5
52 Stafford, Matthew DET QB QB QB5 310.2 244.8 65.4
53 Rivers, Philip SDC QB QB QB6 310.1 244.8 65.3
54 Dalton, Andy CIN QB QB QB7 308.5 244.8 63.7
55 Smith, Torrey BAL WR WR WR21 198.1 134.7 63.4
56 Rice, Ray BAL RB RB RB23 164.1 101.7 62.4
57 Hartline, Brian MIA WR WR WR22 195.8 134.7 61.1
59 Cruz, Victor NYG WR WR WR23 194.8 134.7 60.1
60 Hilton, T.Y. IND WR WR WR24 194.4 134.7 59.7
61 Witten, Jason DAL TE TE TE5 180.6 122.2 58.4
62 Romo, Tony DAL QB QB QB8 299 244.8 54.2
63 Olsen, Greg CAR TE TE TE6 174.4 122.2 52.2
64 Colston, Marques NOS WR WR WR25 186.6 134.7 51.9
65 Floyd, Michael ARI WR WR WR26 186.3 134.7 51.6
66 Williams, DeAngelo CAR RB RB RB24 152.8 101.7 51.1
67 Wallace, Mike MIA WR WR WR27 185.8 134.7 51.1
68 Gates, Antonio SDC TE TE TE7 171.1 122.2 48.9
69 Roethlisberger, Ben PIT QB QB QB9 293.1 244.8 48.3
70 Wilson, Russell SEA QB QB QB10 290.9 244.8 46.1
72 Bennett, Martellus CHI TE TE TE8 166.4 122.2 44.2
73 Tate, Ben HOU RB RB RB25 145.1 101.7 43.4
74 Sanders, Emmanuel PIT WR WR WR28 176.9 134.7 42.2
75 Smith, Alex KCC QB QB QB11 286.9 244.8 42.1
77 Brady, Tom NEP QB QB QB12 284.9 244.8 40.1
78 Rodgers, Jacquizz ATL RB RB RB26 141.4 101.7 39.7
80 Mendenhall, Rashard ARI RB RB RB27 138.1 101.7 36.4
83 Spiller, C.J. BUF RB RB RB28 135.9 101.7 34.2
86 Jackson, Steven ATL RB RB RB29 133.9 101.7 32.2
88 Richardson, Trent IND RB RB RB30 132.7 101.7 31
90 Powell, Bilal NYJ RB RB RB31 130.7 101.7 29
91 Jennings, Greg MIN WR WR WR29 163 134.7 28.3
92 Ryan, Matt ATL QB QB QB13 273 244.8 28.2
93 Smith, Steve CAR WR WR WR30 162.5 134.7 27.8
94 Shorts, Cecil JAC WR WR WR31 161.7 134.7 27
96 Tate, Golden SEA WR WR WR32 159.8 134.7 25.1
98 Kaepernick, Colin SFO QB QB QB14 269.8 244.8 25
100 Streater, Rod OAK WR WR WR33 157.5 134.7 22.8
101 Tannehill, Ryan MIA QB QB QB15 267.1 244.8 22.3
102 Bowe, Dwayne KCC WR WR WR34 154.3 134.7 19.6
103 Vereen, Shane NEP RB RB RB32 121 101.7 19.3
104 Ridley, Stevan NEP RB RB RB33 120.1 101.7 18.4
106 Fleener, Coby IND TE TE TE9 137.3 122.2 15.1
107 Miller, Lamar MIA RB RB RB34 115.2 101.7 13.5
109 Jones, James GBP WR WR WR35 146.6 134.7 11.9
110 Green-Ellis, BenJarvus CIN RB RB RB35 113.2 101.7 11.5
112 Cook, Jared STL TE TE TE10 132.1 122.2 9.9
113 LaFell, Brandon CAR WR WR WR36 143.2 134.7 8.5
114 Reece, Marcel OAK RB RB RB36 109.9 101.7 8.2
117 Foster, Arian HOU RB RB RB37 108.5 101.7 6.8
119 Nicks, Hakeem NYG WR WR WR37 138.6 134.7 3.9
120 Randle, Rueben NYG WR WR WR38 138.1 134.7 3.4
123 Ivory, Chris NYJ RB RB RB38 102.4 101.7 0.7
124 Moore, Denarius OAK WR WR WR39 135.2 134.7 0.5
127 Griffin III, Robert WAS QB QB QB16 244.8 244.8 0
128 Helu, Roy WAS RB RB RB39 101.7 101.7 0
129 Gronkowski, Rob NEP TE TE TE11 122.2 122.2 0
130 Hopkins, DeAndre HOU WR WR WR40 134.7 134.7 0

A reasonable question to ask might be what good any of this is after the fact. It doesn’t do you much good to know that this would have been a more correct draft order for August, unless I also include instructions for time travel in this post. But this exercise did accomplish the task of empirically deriving VBD baselines, which have been very difficult to pin down over the years.  It also advanced a way to think about drafting players that is essentially the points they produce, offset by their draft cost, but also tied to league winning rates.  If you want to know whether a difference in methodology between what I’ve done and the traditionally accepted Top 100 method for calculating VBD baselines actually results in differences in value, it does. The Top 100 method would have yielded the baselines of QB12, RB36, WR43, and TE9.

If you want to think about how meaningful it might be to move the baseline player from say QB16, to QB17, or WR40 to WR30, I’ve created the following graph which shows points for each position by position rank. So you can see points for Jamaal Charles and then a decent sized drop (on the RB graph).  Generally I think that each of the baselines resides at a place on the curve where scoring is fairly flat.

Rplot58

Forecasting player points will always be the part of fantasy prognostication that will have the most value. But knowing which baselines to use is at least a part of the puzzle. There are a few more steps that I could go through to complete this exercise, which I hope to do in time. The easiest would be to also include data on point per game production for fantasy players as I’m sure that some part of Win Rates is also tied up in having players that might not play every game, but produce a lot when they do play. But I expect that component to add only a small amount of explanation and even that small amount might be illusory. It might be illusory because it’s difficult to know how many games you can expect a hurt player to play. The explanatory powers of that component are likely to be meager because most of what we’re looking for is already included in season long point production. If an example helps, Zac Stacy was drafted very late in MFL10s when he was drafted, and he produced quite a bit when he did play, and yet he was on about as many winning teams as you would expect just from chance.

Another way to advance the analysis that I’ve done here would be to extend it to other formats. I did this with MFL10 data because it was available, but I would love to be able to do the same thing for standard scoring leagues, tight end premium leagues with a double flex, or really any league where I could get a bunch of data.

Subscribe for a constant stream of league-beating articles available only with a Premium Pass.

By RotoViz Staff | @rotoviz | Archive

No Comment

Leave a reply