AHL vs. NHL shot volume

    Andrew Anstey

    The possession vs. shot quality argument has long been a point of contention among fans, media, and NHL management. However, there is little debate that shot volume is an easy indicator of offensive output for hockey players. It’s a simple, tangible and highly visible stat that has been tracked for years. Looking at the best shot producers over the last three seasons, we get a who’s who of stars known for their offensive prowess.

    Untitled

    The same criteria is often used to highlight up-and-coming offensive talent in development leagues such as the CHL and AHL. Frank Vatrano was one of last year’s AHL standouts for the Providence Bruins, scoring at a blistering pace, with 36 goals in 36 games. Vatrano’s goal scoring was driven partly by a high shooting percentage (19.3%), but mostly by his league-high 5.2 shots per game.

    Shot volume is another major talking point surrounding presumptive 2016 top-3 pick Patrik Laine. Laine lit up the Liiga this year for 2.9 shots per game in the regular season as a 17 year old, the fifth best rate among players with >10 games played. He jumped to 3.4 shots per game during the playoffs, scoring 10 goals in 18 games en route to winning the Jari Kurri trophy as the Liiga playoff MVP.

    The question is, does shot volume in development leagues necessarily translate to the NHL? I investigated this by identifying 97 players who played at least 15 games in both the AHL and NHL in the 2015-2016 season. The 15 game cutoff was chosen (mostly arbitrarily) to reduce the effects of small samples skewing the results. All of the NHL data was obtained from Corsica, and all AHL data was obtained from Prospect-Stats. First, I investigated basic shots per game in both leagues using a simple linear regression. The fit was assumed to cross through the origin (ie; a player that couldn’t manage a single shot in the AHL likely wouldn’t do so in the NHL either).

    new

    The R2 value of 0.36 indicates that approximately 36% of the variance in NHL shots/GP is explained by variance in AHL shots/GP. There is a decent correlation between the two, however it’s not as good as I expected. The reason for this is likely the wide variance in ice time that players get in the AHL compared to the NHL. Most AHL call-ups are top players on their AHL team, while in the NHL they tend to earn a place in the bottom six as a replacement for an injured or traded player. As a result, our data here is likely somewhat skewed because it does not account for ice time.

    I corrected the data set by controlling for time on ice, and looking at shots per 60 minutes. The AHL does not keep track of ice time, however Prospect-Stats luckily tracks an estimated time on ice (eTOI) stat, using a formula developed by Stephen Burtch. Using this eTOI, I got an estimated AHL shots per 60 minutes for each of the players.

    old

    This correction improves the correlation of the data significantly – we now have an R2 of 0.52 from a simple linear regression of the data. What does this mean? If we know how well a player generates shots at the AHL level, we have a pretty good idea of how well that player can do it at the NHL level. This fitted model tells us that a player who could generate 100 shots in 60 minutes of AHL hockey would generate around 82 shots in 60 minutes of NHL hockey. Using our regressed model, let’s take a look at a few prospects on the Marlies who could be expected to make the jump next year.

    I selected the 7 forwards below as candidates to make the Leafs roster for 2016-17, based on public opinion and my own observation of their play this year. Nylander is essentially a lock for the team, but the rest will likely be fighting for their spot in the lineup in training camp.

    Next, I applied my regressed model to the estimated AHL shots per 60 minutes generated by each player this year to give us an expected NHL shots per 60 minutes. To estimate goals, we need to know how AHL shooting percentages translate to the NHL – obviously, NHL goalies provide a much bigger challenge than AHL goalies (except for Ondrej Pavelec). Luckily, a similar study from 2012 by Kent Wilson provided an approximate translation factor of 0.79(an AHL player shooting 10% can be expected to shoot around 7.9% in the NHL). As Kent noted, this factor should be taken with a grain of salt due to the small sample size used to calculate it, and the large variance in shooting percentage between samples. Using this factor, we can determine an approximate NHL shooting percentage for each player from their career AHL shooting percentage, and then get an expected NHL goals per 60 minutes.

    marlies

    Most of these players already earned some time on the Leafs last year, so let’s look at how our expected data from AHL numbers compares to their real NHL production.

    marliesprojected

    Despite a very small sample size for most of these players (only Nylander played more than 20 games), the model did a decent job of predicting NHL shot production. Soshnikov was an exception, generating nearly double the shots expected by the model. There was less correlation in the expected and real NHL goal rate, however this is not surprising considering the small number of games played and the effects of variation in shooting percentage. It’s nearly impossible to predict how much NHL ice time these players will get next year (if any), so we can’t take any real guesses at how many goals they’ll score over a full season. However, we can compare their expected NHL goals per 60 minutes to current players to get some context and see how they stack up to NHL regulars. Using Corsica, I looked at the goals per 60 minutes of all forwards with over 500 minutes in the 2015-2016 (in all situations). In the figure, each Marlies player is located along the curve according to their expected NHL G/60.

    vsleague

    By inspection, we can see that this data set is relatively normally distributed (thanks to central limit theorem magic) and somewhat skewed to the left (thanks to Alex Ovechkin). Using this measure, the Marlies all look pretty good. Nylander fits into the top quartile of players, and looks like he’s ready to play in a top line role. Soshnikov, Brown, Hyman, Leipsic and Leivo all look like they could slot into the middle of the lineup. Kapanen is closer to the wrong end of this spectrum, and looks like he’s not ready to crack the top 9 yet.

    To summarize:

    • William Nylander is pretty damn good, and has a wide margin on the rest of the players on this list. He looks like a guarantee to play a major role on the Leafs next year.
    • Soshnikov, Brown, Leipsic, Leivo and Hyman are fairly close together in terms of goal production, with Soshnikov leading the way. They will all be competitive options to fill out Toronto’s second and third lines next year.
    • Kapanen is a long shot to stay with the Leafs next year, and likely needs some more time in the AHL to improve (not surprising, considering he’s only 19).