It is a commonly held belief that age is a significant factor on draft eligible players. The theory is that at 17 and 18 years of age, players see a significant growth in their development and that an older player should have a higher production level than that of a younger player. Therefore, if you are of this belief than junior point totals need to be adjusted to reflect how much offensive production is affected by age. I had used behindthenet.ca when doing my 2014 rankings and had been playing around with using Rhys J’s age formula before I started my own project.
To see if age is a factor, I needed to add it into my regression model. In choosing a date to base the age around, I chose the NHL entry draft age cut-off which is “any player who will be age 18 on or before September 15 in the year in which such Entry Draft is held”. To use Sidney Crosby as an example again, he was born on August 7th, making his age 18.11 on Sep 15.
Now were going to delve into stats class for a little refresher about the null hypothesis. The null hypothesis is the hypothesis that the researcher tries to disprove, reject or nullify. For this sample, my null hypothesis is that a player’s age, in his draft year, does not play a factor in their best point per game season in the NHL. To test this, we will take our test samples used in my previous articles and add draft age as another factor to the regression model. To test if the null hypothesis is accepted or rejected we need to choose a significance level for the test and I have chosen 5% for my test. This means that we will accept the null hypothesis, unless the age factor has a p-score of less than 0.05.
This time we will change it up and start with the d-men. When we add age as a factor these are the results of the test:
P-score Non-Power Play Pts/G: 0.214
P-score Age: 0.807
P-score Power Play Pts/G: 0.394
P-score Age: 0.521
In both cases, we see that the p-score for age does not meet the significance level of the test so we will accept the null hypothesis for d-men; that age does not play a factor when it comes to their best Pts/G season in the NHL. In this case, points in junior also does not meet the threshold of <0.05 and showcase the lack of value in their predictability of future offensive potential. This does not mean that age may not play a role in the development and growth of a young defenseman but that it does not play a significant factor in their offensive production.
If we look at purely the correlation and adjusted r-squared numbers from my last blog against when we add age to the equation is:
|Correlation (w/o age)||0.192||Adjusted r-squared (w/o age)||0.018|
|Correlation (w/ age)||0.194||Adjusted r-squared (w/ age)||0.00009|
|Correlation (w/o age)||0.180||Adjusted r-squared (w/o age)||0.014|
|Correlation (w/ age)||0.201||Adjusted r-squared (w/ age)||0.0028|
While the correlation stays around the same, adding age as a factor to non-power play scoring lowers the adjusted r-squared to the point that the data provides no value (0.009% of NHL scoring is explained by junior points and age). For power play scoring, we see more of the same as the correlation goes up slightly but the more important number, which is the adjusted r-squared, explains just 0.28% of their PP totals in their career NHL season. As we can see, age is clearly not a factor when it comes to future offensive potential when evaluating defenseman and no point adjustments are necessary for age.
Now repeating the same test with the forwards, we start to see some different results emerge:
P-score Non-PP Pts/G: 1.04E-12
P-score Age: 0.752
P-score Non-PP Pts/G: 7.36E-12
P-score Age: 0.0414
Well that is interesting as the p-score for age only rejects the null hypothesis when it comes to power play production. I would have thought it would have affected both or neither. Instead, age is not a factor on even strength production and the p-score comes nowhere close to meeting the significance level. To see how the correlation and adjusted r-squared has changed when adding age as a factor:
|Correlation (w/o age)||0.584||Adjusted r-squared (w/o age)||0.336|
|Correlation (w/ age)||0.584||Adjusted r-squared (w/ age)||0.331|
|Correlation (w/o age)||0.551||Adjusted r-squared (w/o age)||0.298|
|Correlation (w/ age)||0.571||Adjusted r-squared (w/ age)||0.315|
Adding age as a factor slightly weakened the model for non-power play production and the p-score did not meet the significance level for the test so the linear equation used previously will remain the same. However, age did meet the significance test and further evidence shows that the correlation and adjusted r-squared improve when age is added as a factor. The new equation to project power play production is:
1.694 + (.363 * PP Pts/G) – (0.089 * Age @ Sep 15 of first eligible draft year)
So basically, if age only matters when it comes to power play scoring of forwards, than the growth and development of younger players is likely not the main factor driving the results. If it was, than we would have expected to see age have a nearly equal effect on even strength production, as it does power play production. Since age is only a factor on a forward’s power play scoring, it makes me think that the biggest factor age makes in juniors is on the amount of power play ice time. Coaches are more likely to reward veteran players with power play time, while the younger players have to earn that ice time. Late born players (Sep 16-Dec 31 birthdays) have one year less of junior hockey left to play so it makes sense that they would have the bigger ice time since the younger guys always have next year. What is a valuable piece of knowledge going forward is that using a blanket age formula on all CHL players does not work and the growth and development of players from the same draft class seems to be highly overrated. To take 2014 as an example I probably over valued Sam Bennett due to this reason. Now if only the CHL had TOI stats so we could see if my theory holds true and prove that age matters…sort of…for forwards…but only on the power play.