It's been some time since we looked at our database of OHL players to develop a rating system for draft-eligible players. It was two months ago when we last addressed it, so a refresher is in order. I compiled a database of the OHL forwards who were first eligible for the NHL Entry Draft (“age-17” players) between 1989 and 1998, and collected their junior stats, some biographical data, and the results of their post-age-17 hockey careers at all levels. These post-age-17 stats were used to derive each player's Career score, which is an approximation of the player's value as a hockey player after that age. Only the first 10 years after the player's age-17 season are counted, since after that time a player would be an unrestricted free agent at any rate, and free agent seasons are irrelevant for draft valuation.
(Note that the calculation of the Career score has been modified since I last used it, giving more credit to a player's minor-league time. This flattens out the value curve and avoids situations like having fourth-line NHLers being ranked higher than good minor-leaguers who never got a shot due to their size.)
I've spent a few hours playing around with the numbers, using points-per-game as a starting point and adding modifiers based on a variety of factors. Points-per-game has a strong correlation with Career scores, and the various other adjustments are used to refine the estimate.
I won't go into detail about the formula itself at this point, because it's quite preliminary and it's based on me fooling around with the numbers; I haven't applied any rigorous methods at this time. However, this effort did reveal some of the factors that seem to be important in evaluating a draft-eligible player's numbers, and the degree to which they are important. Remember that we're only looking at forwards at this point; the same factors will not necessarily apply to defensemen (or goaltenders).
Age: This is a very important factor in evaluating players, especially the “early birthday” players. In minor and junior hockey, eligibility for each higher level is typically determined based solely on a player's year of birth. For the NHL draft, however, the cutoff date is September 15. This results in a group of players who are playing above their age class, in some sense. A player born in October 1990, for instance, would have spent his minor hockey career playing mostly against players born earlier in that year. The best of those players were drafted in 2008, while the October-born players get another year of junior before being allowed to play in the NHL. This makes their draft year stats seem more impressive than they really are, since they effectively have another year of development than those players born in the early part of the next year, against whom they are competing for draft spots.
This effect is quite pronounced. For most players born between September 15 and December 31, a significant adjustment is required to avoid overestimating their value. The effect of other age differences, such as being born in August as opposed to May, is much smaller but likely still present.
Goals per point, assists per point: Many players who are overrated based solely on their points-per-game rate have a very high assist-per-point rate. Brett Seguin scored 108 points in 1990, but was undrafted and had a middling career. Of his 108 points, 80 were assists, meaning he recorded nearly three assists for every goal he scored. This implies he was more reliant on his teammates to produce his point totals that a player with a more “balanced” goal-assist record.
Players who have a particularly high goals-per-point rate are the converse, and an adjustment is appropriate there as well. However, this effect is not as large as having too many assists per point.
Penalty minutes: Penalty minutes are a small but significant factor. Players with high penalty-minutes-per-game rates need their expected Career scores adjusted slightly downward, and players with low penalty-minutes-per-game rates should be adjusted upward. It's a very small adjustment, but the effect is there.
Team offense: Players who play with high-scoring teammates can have inflated scoring totals, and the formula compensates for this. Again, this is a relatively small adjustment, probably because the players we're looking at are the ones who help make their teams high-scoring teams for the most part, not tag-along players.
One thing I haven't worked in yet is the player's performance in his age-16 season. This will be considered as the method is refined.
To develop the estimate I focused on the top 100 players in Career score, since these are the players that we are really interested in. Beyond them we get into the marginal players, who can be useful as depth players in their primes, but are more or less interchangeable. Of course, when refining the method we will have to consider these lesser players, to make sure we don't suffer from false positives, identifying marginal players as a good one.
Here are the results for the top 30 players in Career score. The “Formula Rank” is the player's rank among the top 100 players in Career score.
Player Career Score Formula Score Career Rank Formula Rank
Thornton, Joe 766 595 1.0 3.0
Arnott, Jason 599 366 2.0 29.0
Lindros, Eric 581 682 3.0 1.0
Gratton, Chris 579 539 4.0 7.0
Nolan, Owen 568 509 5.0 10.0
O'Neill, Jeff 562 545 6.0 6.0
Allison, Jason 542 530 7.0 8.0
Ricci, Mike 526 408 8.0 22.0
Primeau, Keith 523 452 9.5 15.0
Bertuzzi, Todd 523 303 9.5 53.5
Legwand, David 493 622 11.0 2.0
Stillman, Cory 471 355 12.0 33.5
Cheechoo, Jonathan 469 419 13.0 19.0
Savard, Marc 468 568 14.0 4.0
Dionne, Gilbert 442 89 15.0 98.0
Murray, Glen 437 242 16.0 76.0
Peca, Mike 428 282 17.0 62.0
Bulis, Jan 398 469 18.5 14.0
Fisher, Mike 398 313 18.5 48.5
Williams, Jason 396 172 20.0 94.0
Cleary, Daniel 387 365 21.0 30.0
Bell, Mark 374 336 22.0 41.0
Moreau, Ethan 360 303 23.0 53.5
Kostopoulos, Tom 358 209 24.0 89.0
Dawe, Jason 357 305 25.0 51.5
Avery, Sean 353 215 26.5 87.0
Ling, David 353 148 26.5 96.0
Park, Richard 345 395 28.5 23.0
Smyth, Brad 345 129 28.5 97.0
Wren, Bob 344 393 30.0 24.0
We can see that the formula does a pretty decent job of estimating Career score, in most cases. By itself, that's not too impressive, but we need to realize that scouts often miss by a large margin when drafting a player. We can make a list of first-round duds like so, and note that in most cases the formula does a much better job at pegging the players' future values:
Player Career Score Formula Score Drafted (overall)
Thornton, Scott 213 294 #3 in 1989
Warriner, Todd 277 447 #4 in 1992
Kilger, Chad 290 346 #4 in 1995
Tkaczuk, Daniel 233 547 #6 in 1997
Fata, Rico 269 332 #6 in 1998
Devereaux, Boyd 272 322 #6 in 1996
Convery, Brandon 230 368 #8 in 1992
Heerema, Jeff 295 360 #11 in 1998
Ward, Jason 297 337 #11 in 1997
Pearson, Rob 271 284 #12 in 1989
Primeau, Wayne 261 331 #17 in 1994
Rice, Steven 259 315 #20 in 1989
Simon, Chris 227 313 #25 in 1990
Donovan, Shean 253 262 #28 in 1993
Craig, Mike 295 355 #28 in 1989
Of course, the formula is not remotely perfect. It has its share of big misses as well, a couple of which are already included in the table above. Players like Daniel Tkaczuk and Todd Warriner fooled both the scouts and the stats. Other players are more accurately forecasted by the scouts:
Player Career Score Formula Score Drafted (overall)
Roche, David 247 517 #62
Corpse, Keli 235 490 #44
Milley, Norm 298 446 #47
Papineau, Justin 232 417 #46
Cirone, Jason 218 382 #46
Miller, Colin 216 380 Undrafted
Cloutier, Sylvain 243 360 #70
These are players with very good points-per-game figures, and no other apparent statistical factors that would indicate they should be drafted lower than their scoring numbers would suggest. We may be able to limit these misses as the method is refined. These are preliminary results, and should improve over time.
Of course, it wouldn't be an Up And Coming column without discussing height and its effect on evaluating hockey players. Based on the work I've done so far, there does not seem to be any reason to incorporate an adjustment for a player's height. For the formula as it currently stands, among the 100 top Career players, height does not appear to be a factor. About 50% of short players are overestimated by the formula, while 50% are underestimated. The same applies to tall players, and medium-height players. I expected to need a small adjustment due to height, since there is a pervasive bias against short players in the NHL, and this can affect their Career scores. However, that hasn't been required, at least not yet.