As Leafs fans, we are all used to collapses. In three of the last four seasons, the Leafs have been defined by a collapse of some kind. A byproduct of this is the notion that many of the Leafs players are prone to faltering down the stretch of a season. This leads to criticism of the players for failing to maintain production throughout the season. A lot of these criticisms, fairly or unfairly, have landed at the feet of Phil Kessel (it wasn't that long ago that writers called him Mr. October). Inspired by that, I wanted to take a look at whether this criticism had a basis in reality (i.e. does Kessel's performance does actually drop over the course of a season?). The goal of this analysis is to determine if this criticism is warranted, and if it's not, to identify other factors that may contribute to this perception. For this post, I'm going to focus on Kessel's individual offensive contributions, which, as a goal-scorer and playmaker, represent a large chunk of his value. I may do another post in the future on his defensive contributions, but I think that's a more involved question that requires a more context and thinking.

I chose to look at Kessel's entire career as a Leaf career (excepting the abbreviated lockout season, for obvious reasons). I broke up the season into two buckets (I considered doing it monthly, but that drastically reduces sample sizes). I put the first 60% (or thereabouts) of games Kessel played in one bucket (henceforth referred to as 'Bucket 1'), and the last 40% in another bucket ('Bucket 2'). This works for Kessel, as he's been very healthy over the course of his career (always played at least 70 games per season as a Leaf), and as such, this methodology maintains a good balance between isolating what we want to test in the different buckets, and making sure there's a decent sample size in each. I chose a 60/40 split as it generally coincides with around the All-Star/Olympic break, which is right about when the unofficial drive to the end of the season begins.

The Data

The first thing I looked at was Kessel's scoring rates across each of the buckets, summarized below:

Screen_Shot_2015-03-07_at_7.11.39_PM.0.png

What we see here actually seems to indicate that the opposite of the narrative is true. In both the 2009/2010 and 2010/2011 seasons, Kessel was noticeably more prolific in the latter 40% of games (Bucket 2) than in the first 60% (Bucket 1). In his most recent two seasons, he produced at an almost identical rate in both of the buckets. From a point-production perspective, there is no evidence that Kessel has faded towards the end of a season.

Of course, point production is not the most appropriate metric to use in order to gauge effectiveness and level of play, especially since we are still dealing with relatively small sample sizes within buckets. It's possible that Kessel has simply had the percentages go his way later in the season, or at least moreso than he did in the early stages of it. To get a better picture of Kessel’s level of play and activity, an alternate metric is needed. A good option is the Individual Corsi For (ICF) metric, which is essentially the number of shot attempts made by a player. In the relatively small samples we have, it's a better proxy for an overall level of play and activity than point scoring rates since it’s not influenced by percentages. There are additional benefits to using this metric. For one, ICF per game (ICF/G) is (more or less) normally distributed in large samples for individual players, and even in the small-ish samples that I'm using for Kessel, the ICF/G is relatively symmetric. As such, we can appropriately use a two-sided t-test in order to determine if there is a significant difference in Kessel's level of play between the two buckets.

If you're interested in the formulas and the statistical theory I'm using, this Wikipedia page is a good start. To formalize this slightly, the null hypothesis (H0) that I'm testing is that the ICF/G between each of the buckets is identical (i.e. no difference between the two data buckets). The null hypothesis is rejected if there is a statistically significant difference between Kessel's ICF/G in each of the buckets. If there is no significant difference between the two buckets, we accept the null hypothesis. On to the results:

First the raw values of ICF/G between the two buckets:

And now the results from the t-test:

Understanding the statistical theory behind this isn't all that important in terms of interpreting the results. Essentially, the bottom two rows are the important ones. The p-value represents the probability that we see a result more extreme than what we did, under the assumption of the null hypothesis (which is that Kessel has equal ICF/G between buckets 1 and 2). I'm using a 5% significance level (pretty standard in stats land), so essentially, I reject the null hypothesis if the p-value is less than 0.05, and do not reject otherwise.

As it turns out, Kessel's ICF/G did drop rather noticeably in the latter parts of the season during the 2009/2010 and 2010/2011 seasons. In these two cases, we reject the null hypothesis and conclude that there was a significant difference (p<0.05) in his shot-generation ability between the first 60% and the last 40% of his games. As Kessel is primarily an offensive player, this his heavily linked to his overall level of play. Whether it was variance, luck, lack of conditioning, or anything else, it seems safe to say that his underlying offensive play was worse in the stretch run than it was to start the year. However, in the two most recent seasons, the difference between his ICF/G between buckets 1 and 2 was statistically insignificant (p>0.05).

The Verdict

Based on the evidence shown here, there is no solid basis to the narrative that Kessel consistently underperforms (from an offensive perspective) in the later stages of the NHL season, relative to his 'regular' level of play. While his ICF/G was significantly lower in the last 40% of the games he played in both 2009/2010 and 2010/2011, that trend does not hold in the two most recent seasons I studied. Based on his point rate, we may even conclude the opposite, if anything. Kessel may not rise above and beyond his normal level of play as the season winds to an end, but it seems clear that from an offensive perspective, he doesn't falter either.

Note: I forgot to mention this in the initial draft of the article, but the data I used for this is courtesy of NHL.com and war-on-ice.com