In August and September, everyone in the hockey world likes their team’s chances for the upcoming season. This is the year a player breaks out, or this player will rebound from a down season, or that player will be great for one more season. Entering camp, everybody is an optimist since you never know what’s going to happen.
But that doesn’t mean there isn’t a way to predict what’s going to happen, as baseball’s PECOTA system proves to us. So for the past two seasons, Hockey Prospectus has released a projection system called VUKOTA as their answer to Baseball Prospectus’ PECOTA system. Here at Broad Street Hockey, we’ve looked at these for each of the last two seasons, checking in along the way.
After the 2009-10 season, we concluded that the projections seemed accurate, but the results were outliers after looking at each individual player’s past. So how about this year?
All tables are sortable by column.
| Name | GP | G | Pts | Pts/82 | GP | G | Pts | Pts/82 |
|---|---|---|---|---|---|---|---|---|
| Blair Betts | 62.1 | 6.3 | 13.9 | 18.4 | 75 | 5 | 12 | 13.1 |
| Danny Briere | 65.2 | 22.5 | 53.0 | 66.7 | 77 | 34 | 68 | 72.4 |
| Dan Carcillo | 63.7 | 9.1 | 20.6 | 26.5 | 57 | 4 | 6 | 8.6 |
| Jeff Carter | 77.4 | 35.8 | 69.9 | 74.1 | 80 | 36 | 66 | 67.7 |
| Claude Giroux | 69.8 | 19.6 | 51.3 | 60.3 | 82 | 25 | 76 | 76.0 |
| Scott Hartnell | 71.7 | 20.2 | 48.9 | 55.9 | 82 | 24 | 49 | 49.0 |
| Ville Leino | 53.9 | 10.1 | 25.0 | 38.0 | 81 | 19 | 53 | 53.7 |
| Darroll Powe | 60.4 | 8.7 | 16.7 | 22.7 | 81 | 7 | 17 | 17.2 |
| Mike Richards | 77.3 | 30.1 | 71.6 | 76.0 | 81 | 23 | 66 | 66.8 |
| Jody Shelley | 50.3 | 2.5 | 7.4 | 12.1 | 58 | 2 | 4 | 5.7 |
| James van Riemsdyk | 72.8 | 18.7 | 40.9 | 46.1 | 75 | 21 | 40 | 43.7 |
On the left are the projections for games played (GP), goals (G), points (Pts), and points per 82 games (Pts/82). To the right are the actual results.
Just like last year, we see that the biggest flaw in VUKOTA is their method of predicting the number of games each player will miss. Once again, the Flyers were relatively healthy (nine of the team’s eleven project-able players hit 75 games), so VUKOTA’s projecting poor health was a problem.
But the system did have some pretty accurate predictions, specifically the predictions for van Riemsdyk, Carter, and Briere. It came close with Betts, Hartnell, and Powe, while it missed Giroux, Leino, Richards, Carcillo, and Shelley. Not exactly great results, but excuses can be made – including the small sample sizes of Leino’s career to that point and Richards’ injury – so let’s try and identify the problem.
As we did prior to the season, let’s compare the projections to the players’ point per game rates of the last three years:
| Name | 08-09 | 09-10 | VUKOTA | 10-11 |
|---|---|---|---|---|
| Betts | 0.12 | 0.29 | 0.22 | 0.16 |
| Briere | 0.86 | 0.71 | 0.81 | 0.88 |
| Carcillo | 0.19 | 0.29 | 0.32 | 0.11 |
| Carter | 1.02 | 0.82 | 0.90 | 0.82 |
| Giroux | 0.64 | 0.57 | 0.73 | 0.93 |
| Hartnell | 0.73 | 0.54 | 0.68 | 0.60 |
| Leino | 0.69 | 0.20 | 0.46 | 0.65 |
| Powe | 0.18 | 0.24 | 0.28 | 0.21 |
| Richards | 1.01 | 0.76 | 0.93 | 0.81 |
| Shelley | 0.06 | 0.16 | 0.15 | 0.07 |
| van Riemsdyk | — | 0.45 | 0.56 | 0.53 |
What do we see? Briere had his best year of the last three seasons, which wasn’t predicted. Carter repeated last year’s performance rather than improve as predicted, Giroux went from being the fourth highest point per game player to the highest, Richards didn’t rebound like he was expected, and Leino played like his 13-game debut rather than what got him thrown out of Detroit.
What’s this mean? Well, there are obviously still flaws. But to say predictions are useless ignores how close VUKOTA gets. The misses came on players who are on the fourth line (Carcillo and Shelley), had breakout years (Giroux and Leino), or have no heart played through a wrist injury all season (Richards). Maybe that’s too many excuses to make for a projection system, but it wasn’t bad. Consider:
Team Predictions With Results
Not a single Flyer was expected to breach the point-per-game ratio this year. Correct. Claude Giroux came the closest, with 0.93. HIT
Only Carter and Richards were expected to hit 25 goals, though Briere would hit 28 over a full season. Predicting three 25 goal scorers… you get three 25 goal scorers. But Richards only hit 23 while Giroux hit 25. Hartnell hit 24, one more than predicted. HIT
Only three players (Betts, Laperriere, and Shelley) were expected to see their points per game ratio decrease from last year. Betts and Shelley did decrease, but so did Powe and Carcillo. Predicting fourth line production has proven to be difficult for VUKOTA, and there’s really no good explanation for predicting Powe to score 0.28 points per game. CLOSE
Claude Giroux was expected to have a pretty big breakout year, going to 0.735 points per game, up from 0.573. He did in fact have a pretty big breakout year, but it was even bigger than predicted. MISS
This group – including Zherdev – was expected to score 207.1 goals. Even without Laliberte and Laperriere’s 8.9 goals, the group combined to score 216. CLOSE
As a group, the Flyers could expect their defensemen to chip in 38.7 goals. Here’s a big miss. The Flyers defensemen only contributed 22 goals, with Coburn and Carle accounting for half (9) of the difference in performance. Timonen and Pronger missed the mark by a combined six goals, and there’s the majority of your problem. MISS
GVT saw the Flyers scoring the same amount of goals as last year, while VUKOTA saw the Flyers scoring 254. But normalizing for the league numbers, VUKOTA agrees with GVT, predicting 237 goals. Without the normalization, VUKOTA would have nailed it – actual 256 goals for. But it did normalize, so… MISS
So there you have it. VUKOTA still has a lot of work to do – particularly in predicting games played and on defense – but it absolutely has predictive value. How much is a legitimate debate, but the results are there.
