Sab(re)metrics: Snepsts Projections for the 2011-2012 Buffalo Sabres

I know I stated that my second installation would be the thought-provoking piece "How Good Is Thomas Vanek," but I received a nice surprise the other night when Hockey Prospectus and ESPN writer Robert Vollman contacted me about his own hockey projection system called Snepsts. Since I want to analyze the accuracy of the VUKOTA projection system throughout the season, I decided to write a quick article on the Snepsts projection system and what projections are yielded for the Buffalo Sabres this season. This way, we can look out how the accurate these two projection methods will be over our 41 and 82 game regular season. Check after the jump for some links, a brief explanation of the Snepsts projection system, and a comparison of this method's projections with VUKOTA.

To project a players seasonal totals, the Snepsts projection method pro-rates the scoring levels (examining G, A, and Pts. as a per game statistic) of a player and then compares these numbers with players throughout history. Throughout the comparison, Snepsts will weight* the most recent years' stats of the comparable player twice as much as the previous season. For example, if Thomas Vanek was used as a comparable player to project a player's 2011-2012 seasonal totals, his 2010-2011 stats would be weighted twice as heavy as his 2009-2010 stats, which would be twice as important as his 2008-2009, and so forth.

Once multiple comparable players are found, all of their weighted seasons are averaged together to provide the projection of a player's next seasons statistics. In full form, Snepsts will bound the projection stats with maximum and minimum averages of the comparable players (see the Hockey Prospectus article here); for this column, I will only list the averages from this method. In comparison to VUKOTA, which matches the players projection to a curve of similar players, the Snepsts projection method is purely historical, with no mapping or curve fitting to other players statistics. To compare the two projection methods, I will list them in a G/GM, A/GM, and Pts./GM format, since that will make it easier to analyze their totals with real statistics once we reach the middle and end of the season, respectively.

The charts below compare the VUKOTA and Snepsts projection method in a per game format. Any columns with dashes mean that those players do not have a Snepsts projection for this season (this could be from their lack of NHL experience, which leads to 0 matches in the Snepsts projection method). Also, since the Snepsts method only projects goals and assists, it will not include a projection for goalie stats for the upcoming season.

FORWARDS

DEFENSEMEN

A few observations:

• Except for Tyler Myers and Andrej Sekera, all other projections from the Snepsts method are more conservative than the VUKOTA projections.
• The amount of matches a player has is not dependent upon how many years they have played the game. For example, Hecht has played six more NHL seasons than Jason Pominville, yet he has more than three times as few matches when comparing him to players of the past.
• Not all players can have projected stats (this could be due to the lack of playing time for these players, which was touched on in a previous paragraph).

I will like to thank Robert Vollman, who contacted me about the Snepsts projection method and gave me information into how he formulates this method. Please tune in next week for my next piece (I promise!): Sab(re)metrics: How Good Is Thomas Vanek?

*To develop the weights for each season, Vollman compared the correlations of subsequent seasons to a player's previous season. From this analysis, he found that a player's most recent season correlated two times more with their previous season than with the season stats they had two years before.