If you haven’t had the opportunity to read Rob Vollman’s article “Points regression for the Norris finalists”, I suggest that you do. It is a great article that gives a little insight as to how forecasters make their projections. He also describes where the variation in forecasters’ projections arise from. On top of this insight, he more than backs up the points that he makes with three near perfect Pts total projections for Burns, Hedman, and Karlsson in 2017/18!
In the article, Rob Vollman says that “In hockey analytics, there's a standard, four-step process to project a player's production. First, calculate a weighted average of the past several seasons. Second, regress that result toward the League average to remove the calculated impact of random variation. Third, account for the player's age, and finally, make manual adjustments for known factors, such as injuries or new linemates”.
The third step of the process, factoring in the aging curve, is something that has been of particular interest to myself as of late. I have read a couple of articles on this and one of the best I have found (still waiting on Rob’s Stat Shot: The Ultimate Guide to Hockey Analytics to get here in the mail… the NWT is far, far away) is from @EvolvingWild at Hockey-Graphs.com. Through their calculations they determined the peak age of an NHL player to be at age 23, with a negligible decline until the age of 25. I have to admit, as a Junior High Math Specialist I get lost right where the authors note that I should “skip to the positions section” because it “gets rather detailed”. I thank them for the advice, however I ignored it and continued on only then to regret it. To try and summarize, they break down what the changes in WAR looks like from various lenses (forwards, defence, etc) with age.
I am mainly interested in how the Pts Total output of players declines with age. Thus, I decided to see if I could take a shot at it and come up with a simple method to calculate this change in Pts output by age.
My goal was to;
- Determine the Average Change in Pts Total Output by Age for Skaters in the NHL,
- Determine how accurate the Average Change in Pts Total Output by Age could be used to predict Pts Total projections for the upcoming season, and
- Compare these projections to the projections of forecasters that I evaluate.
As I go through this, please remember I am approaching it from a "layman's" knowledge and background of stats. I really like mean, median, and mode.
1st – I collected skaters Pts total, and Age data (obtained from https://www.hockey-reference.com/) from 2006 to 2018 (omitting the 2012-13 lockout data).
2nd – I determined the difference in Pts total output for each player from year to year.
3rd – I determined the average of the difference in Pts total output by Age.
Below is what I determine (through the process above), to be the Average Change in Pts Total Output by Age for Skaters in the NHL. I debated how much data to include in the determination and thus included a couple of options to look at: 10 year average, 5 year average, Average since lockout (up until 2016/17).
My calculations show that an average player increases their Pt totals until they are 28 (10 year average), or 29 (5 year average, Average since Lockout). This is in accordance with some of the other findings which see a decline around age 29 and a little bit later than the calculations mentioned by @EvolvingWild above.
After I determined the Average Change in Pts Total Output by Age for Skaters I wondered how it would look for only those skaters who are in the Top 300 in the league in scoring each year. Most forecasters don’t project past 300 players as most pools do not need projections that deep. So, I did the same steps outlined above but trimmed the data used to be only the Pts total for skaters in the top 300 each of the years. This is what I found.
The Average Change in Pts Total Output by Age for the Top 300 Skaters in the NHL
These findings are a little bit more intriguing than the others which included data for all Pts scorers in the league. They show that an average player increases in their Pt totals until they are about 32 or 33 on all three data time frames. This is a lot different from the Average Change in Pts Total by Age when all player data was included. It shows that perhaps elite players (Top 300 scorers) are able to increase production longer than players in the bottom 5/8ths of the league.
To see how accurate projections would be by solely applying the average change in Pts total output by Age; I took the results from the 2016/17 season and then adjusted those totals by (from what I determined above) to project the 2017/18 totals for players. I then compared those projections to what players actually scored, and determine the average of the absolute error in the projections.
Compared to the average errors that forecasters had in their projections for 2017/18, projections based solely on the average change in Pts output by age fall somewhere from mid to end of the pack. There was a significant difference between taking the Average Change in Pts by Age from the whole league versus just the Top 300 scorers each season. Taking the whole leagues average is almost a half a point better per category of analysis.
Below is a look at the Top 25 Pts getters in2016/17 projections for 2017/18 based on my Average Change in Pts by Age calculation.
Below are the inferior Top 25 Pts Scorers in 2016/17 projections for 2017/18 based on my Average Change in Pts (limited to the Top 300 Pts scorers) by Age calculation. This one actually had a better projection for McDavid.
To compare the projections based solely on the average change in Pts total output by age, I plugged them into the algorithm that I use to compare forecasters projections.
I anticipated that the projections based solely on the Average Change in Pts by age would not do very well compared to other forecasters’ projections. Thus, I was surprised to see that they did do better than some forecasters’ projections.
The projections based on the 10 year average Pts Change by Age actually ranked 9th in a comparison to forecasters’ Top 300 and 200 projections.
Those same projections did not do as well when looking at groups of 100.
The projections based on the 10 year average Pts Change by Age for the Top 300 Skaters did much worse, ranking 12th when compared to forecasters’ Top 300 projections.