Given the Flyers are projected to score 226 goals this season, that would put them as a bubble team. However given now that Travis Konecny made the team and will likely add somewhere between 10-20 goals, the Flyers realistically should make the playoffs for the 2nd straight year. To me the most crucial player to the Flyers this year is Jakub Voracek. If Voracek can bounce back from a tough season last year and score around 19 goals, then the Flyers should be in the playoffs for the 2nd straight year.
How the goal projections are done
In order to accurately project the number of goals a player will score you need to take a look at 4 major variables in his past 3 seasons. Individual shots for per 60, shot %, time on ice, and his age. Obviously a player’s most recent season (in this case 2015-2016) is a better predictor of his future stats than that player’s season 3 years ago (2013-2014). Nonetheless, a player’s season 3 years ago (2013-2014) still matters in projecting a player’s goal total for this upcoming season. This process is overall very similar to the Marcel projection model used in the MLB.
The coefficients or the weight each season holds in each one of those 3 respective statistics for Forwards (TOI, Individuals shot for per 60, and shot %) was found by looking at past seasons in the aggregate and seeing how important each past season was in determining the upcoming season. For example, what was the weight of all NHL forwards 2009-2010 ish60 in predicting their 2011-2012 ish60. Below is a chart of all coefficients used in this model for forwards. All coefficients were tested for significance using the T-test and all p-values were p<.05, therefore proving its significance.
To no surprise, the more recent the season, the more weight it holds in determining the future stat for forwards.
So for example, Claude Giroux in 2013-2014 had 7.04 shots per 60 minutes (ish60), in 2014-2015 he had 9.39 shots per 60 minutes, and in 2015-2016 he had 8.05 shots per 60 minutes.
Therefore to determine his projected shots per 60 minutes in 2016-2017 you would do ((2.2*7.04)+(2.2*9.39)+(5.8*8.05))/(10)
This gets you a projection of 8.122 shots per 60 minutes for Claude Giroux on 5v5 for the 2016-2017 season. However, there’s still more to account for than this in his projection.
Players will sometimes have an off shift or have a bad bounce against them or vice versa. Therefore, over the course of an 82 game season, all players no matter their skill level will at some point play at the average level of NHL players. They will regress to the mean in certain statistics. For ish60, players on average regress 16% to the NHL mean, which is 7.2 ish60.
To adjust for this, we will use a weighted sample of 923 minutes of 5v5 ice time. So therefore we will do .16*923 (.16 for the 16% regression to mean) to get 148 minutes. So therefore to adjust Giroux’s ish60 to regress to the mean we will do (775*8.122)+(148*7.2)=7.97. Therefore, a more accurate ish60 projection for Claude Giroux in 2016-2017 is 7.97.
Lastly, you have to adjust for a players age. After using the above method and testing it on previous seasons I found the average difference between the projection for ish60 and the actual ish60 for all players. I then plotted this difference on the Y axis and the age of all the players tested on the x axis and found the linear regression line amongst all the data points. For ish60 the equation of the linear regression line was -0.0494x+1.2501 where X is a players age. So therefore to adjust for age, you plug in Giroux’s age 28 into that equation and get -.1331 which is what you will subtract from Giroux’s regressed ish6o projection which is 7.97. Therefore, you do 7.97-.1331 to get 7.84, which is Giroux’s final ish60 projection for this year which accounts for age, skill, and luck.
You then repeat this process for time on ice and shot percentage. After finding those projections you then know how many shots a player will shoot and the percentage of those shots that will be a goal which is that player’s goal projection for 5v5. Then all I did was average the amount of goals scored on every possible strength (5v4, 3v3, 4v3, etc) over the past 3 seasons and add that average amount of goals to the players 5v5 goals projections to get their final goal projection.
The same process was done with defensemen except the coefficients for their ish60, TOI, and shot % were all different.
As for the players who haven’t played the last 3 seasons in the NHL such as Gostisbehere and Provorov, a simple multivariable regression model was used for them. For example, by running a regression of goals per game in a defensemen’s 2nd year on his rookie seasons iCF, zone starts, CA, goals per game, assists per game, primary points per game, I was able to plug in Gostisbehere’s stats for those metrics in the regression equation to get his goal projection. Although this isn’t the most accurate way to do so, there weren’t many better models out there to project a defensemen’s goal total in his 2nd year.