Whatever your projection system, the starting point is probably last year's stats. Whether you're doing this consciously or not, your system likely starts with last year's numbers -- or maybe a 3-year average or something like that -- and then modifies them up or down -- possibly significantly so -- based on a number of factors.
In some of my other articles, I've looked at why you might want to modify a players projections up or down for statistical reasons. For example, a couple of consequences of this article are that RBs whose value comes through the running game are slightly more reliable than RBs who accumulate more of their fantasy points through the air. Also, RBs whose value is yardage-dependent are a little more likely to hold their value than backs whose value is touchdown-dependent.
For example, consider the following three backs:
G RushYD RushTD RecYD RecTD
-----------------------------------------------------
Running Back A 16 1800 5 0 0
Running Back B 16 900 3 900 2
Running Back C 16 600 10 600 5
All three scored the same number of fantasy points (assuming, as I always
do, that fantasy points are defined by yards/10 + 6*TDs). However, if
I knew nothing else about these three backs, I'd be fairly comfortable
betting that Back A will outscore Back B, who will
outscore Back C the following year. Rushing is in general more stable
than receiving, and yards are more stable than TDs.
In addition, we know that, as a group, players who have good fantasy seasons (as all three of our fictitious backs above did) decline a little bit the following year. Thus it's not unreasonable to expect a slight decline out of all three.
There is a way of rolling all these ideas into one simple formula. It's called multivariate linear regression. If you're a regression pro, great. If not, don't sweat it -- I won't bore you with any details.
Here's the punchline: I fed my computer 465 running back seasons (more precisely, I told one computer program to feed that data to another computer program). The data consisted of how many yards rushing and receiving, and how many TDs rushing and receiving each RB had, and how many fantasy points he scored the next year. The computer huffed and puffed for several microseconds, finally coughing up this formula:
Projected Fantasy points in Year N+1
equals
.10*RushYD_n + 3.18*RushTD_n + .082*RecYD_n + 2.79*RecTD_n + 11.3
where RushYD_n = rushing yards per 16 games in year N;
RecYD_n = receiving yards per 16 games in year N;
RushTD_n = rushing TDs per 16 games in year N;
RushYD_n = receiving TDs per 16 games in year N.
Important technical note: all the input data was translated to a
per-16-game scale, so this formula assumes all players will play 16 games
every year. More on that later.
In a generally agreed-upon mathematical sense, this is the best possible formula for predicting next year's fantasy points from last year's stats. The mathematically-inclined among you can study it and see that it does indeed take into account the factors mentioned above. If you're not into that, just run the formula for our three guinea pig backs (or better yet, just sit back while I do):
Projected Pts the next year
-----------------------------------------------
Running Back A 207
Running Back B 190
Running Back C 166
So, knowing nothing else about them, we would expect them all to decline,
but Back A to take the smallest hit, followed by Backs B and C in that
order. Remember, these guys all scored 210 the year before, but just
based on the shape of their production, we forecast much different levels
of success for them in the following campaign.
Very interesting note: I tried including more than one seasons worth of data, but it did not improve the accuracy of the projections at all. That's not to say it isn't relevant in certain cases, but in general it's as likely to point you in the wrong direction as the right one.
Now, this is of course a tiny piece of the projection puzzle. Another factor we can get something of a numerical handle on is age. I went back and ran this formula on all 465 RBs for which I had data. I found, not surprisingly, that it tended to underestimate young RBs and overestimate old RBs. In particular, I found that it underestimated RBs under 26 by about 10%, it overestimated RBs between 26 and 30 by about 5%, and it overestimated RBs over 30 by about 10%.
Thus, a simple but effective way of improving our formula's accuracy is to simply add 10% to the projections of anyone 26 or under, take 50ff the projections of backs aged 26 to 30, and lop 100ff the projections of backs over 30.
And this, I think, is about as far as we can go with statistics alone. Think of projections obtained with this formula as the best possible projections you can get without knowing anything about football. (OK, that's not precisely true. There are more sophisticated mathematical techniques that could be applied, but I'm trying to keep things relatively simple. Please, no emails about nonlinear regression and neural nets and whatnot).
For a humorous example of how little knowledge of the actual football world this projection system would have, consider that it projects Olandis Gary to score 234 fantasy points this year. Obviously, you know better. The point is that this formula makes an excellent starting point for your projections. Then you can use things that you know about the non-statistical issues affecting particular players and teams to modify it. In Gary's case, you'd want to modify it way, way, way down, but in most instances, just a minor amount of tinkering is in order.
Without further ado, I present this formula's 2000 RB projections. Many important disclaimers follow:
1999 1999 2000
LastName FirstName Age FPT16 PROJ
---------------------------------------------
James Edgerrin 21 316 294
Davis Stephen 25 290 268
Faulk Marshall 26 315 258
Gary Olandis 24 232 234
Levens Dorsey 29 252 208
George Eddie 26 254 208
Dillon Corey 24 197 203
Staley Duce 24 193 199
Smith Emmitt 30 245 193
Taylor Fred 23 188 187
Garner Charlie 27 212 187
Martin Curtis 26 202 185
Biakabutuka Tim 25 184 183
Williams Ricky 22 157 172
Enis Curtis 23 166 170
Watters Ricky 30 202 167
Stewart James 28 208 166
Smith Robert 27 160 153
Dunn Warrick 24 141 148
Wheatley Tyrone 27 179 147
Alstott Mike 26 173 146
Bettis Jerome 27 162 144
Holmes Priest 26 146 134
Linton Jonathan 25 128 130
Allen Terry 31 156 125
Rhett Errict 29 144 125
Collins Cecil 113 123
Kirby Terry 29 152 121
Johnson James 25 111 116
Oxendine Ken 24 99 108
Huntley Richard 27 130 107
Hoard Leroy 31 141 107
Bennett Donnell 27 122 103
Barber Tiki 24 99 102
Kaufman Napoleon 26 108 101
Smith Antowain 27 115 101
Beasley Fred 25 98 98
Abdul-Jabbar Karim 25 80 90
Murrell Adrian 29 89 89
Pittman Michael 24 81 88
Lane Fred 24 74 88
Shehee Rashaan 24 77 87
Phillips Lawrence 24 83 86
Hicks Skip 25 81 85
Hanspard Byron 23 71 84
Hill Greg 27 84 83
Holcombe Robert 24 81 81
Morris Bam 27 84 79
Basnight Michael 22 59 73
Bynum Kenny 25 68 72
Christian Bob 31 95 71
Pritchett Stanley 26 88 71
Faulk Kevin 23 65 71
Irvin Sedrick 21 69 69
Warren Chris 31 75 69
Centers Larry 31 78 62
Richardson Tony 28 59 62
Bates Mario 26 78 60
Stephens Tremayne 23 63 59
Mitchell Brian 31 59 55
Fletcher Terrell 26 52 53
Smith Lamar 29 51 52
Anderson Richie 28 57 50
Loville Derek 31 50 50
Crockett Zack 27 55 45
IMPORTANT NOTES:
Add Terrell Davis, Jamal Anderson, and the rookies to the mix, and I come up with a top 20 that looks about like this: