March 29, 2010
From Daigle To Datsyuk
Sample Size

by Corey Pronman

As hockey statisticians, we would all like to be able to develop perfect projection systems; something that can give us with a very high level of accuracy what we can expect of Player X the following year or even multiple years down the road. GVT and VUKOTA have helped us here at Puck Prospectus in reaching that elusive goal, but we can advance much further as hockey stats continue to advance. What has always been a big debate in the hockey world, however, has been trying to project NHL Draft prospects. NHL executives will tell you of numerous, numerous occasions they’ve been approached with a “system” that could tell them who they should select in the upcoming Entry Draft using some sort of statistical model.

While I would be the last one to sit here and discourage the use of statistics, if anyone tries to create a selection model and use it as the sole way of trying to elevate one prospect over the other, there methodology would be very flawed. While there are many issues that can affect a player’s statistics, be it league, ice time, linemates and more, the biggest issue in my opinion is the sample size of the statistics used.

The sample size of the statistics used to analyze Entry Draft prospects before their selection is unfortunately very limited. I don’t care what anyone ever tells you, in a CHL season where a player probably won’t even play 75 games, you cannot, absolutely cannot, say you have a decent sample of statistics to have full confidence in any sort of predictability value. How many times will you see a NHL player, who plays 80+ games, see his statistics fluctuate from year to year? The fluctuations may not be extreme, but it’s enough to make a difference when trying to project. As an example, let’s take Jarome Iginla’s rate statistics post-lockout:

Year 		Goals/Game		Points/Game
05-06		.43			 .82
06-07		.56			1.34
07-08		.61			1.20
08-09		.43			1.09
09-10		.43			 .92

Year 		Goals/Game		Points/Game
05-06		.68			1.72

Of course he did that playing alongside the amazing 19 year old Alexander Radulov which definitely affected his stats, but factors that affect pre-Draft prospect statistics is a column for a future date. From a production standpoint at the least Esposito looked like a sure bet to go at the top of the 2007 Draft, but this is how he did in his Draft year, and how the rest of his QMJHL career played out:

Year 		Goals/Game		Points/Game
06-07		.45			1.32
07-08		.54			1.23
08-09		.69			1.20

While one may look at those numbers and think that they aren't bad, you have to remember that there should be a natural progression in production between his age 16 season ('05-'06) to his final age 19 season ('08-'09). What we can see is the drop-off from his first year, which was a big reason for his falling Draft stock, on top of other things such as intangible issues, but the point is that we couldn’t use the 57 games he played in '05-'06 as a predictor for Angelo’s future performance, and if we had we probably would have made a big mistake.

Now taking this logic and applying it to the top two prospects in this year’s Draft class in Taylor Hall and Tyler Seguin, here are the two player’s OHL statistics:

Taylor Hall

Year (age)		Goals/Game		Points/Game
07-08 (16)		.71			1.33
08-09 (17)		.60			1.43
09-10 (18)		.70			1.86

Tyler Seguin

Year (age)		Goals/Game		Points/Game
08-09 (16)		.34			1.09
09-10 (17)		.76			1..68

Now from looking at both of their statistics, Tyler Seguin has a small edge in Goals/Game in 09-'10 and Hall the same for Points/Game; however Seguin has a big edge over Hall in their age 17 seasons. The issue here with the statistics when using the sample size argument is that we have three seasons of Hall dominance, and only one of Seguin. The sample size is significant as Hall played 183 games from '07-'08 to the present compared to Seguin’s brilliant 63 games this year. So as a team selecting at the top of the Draft, banking on Seguin continuing his success is much riskier than doing the same for Hall. However, as outlined by comparing their age 17 seasons, if Seguin’s development curve continues to rise, he may end up the better prospect. Again this is just using statistical theory at the moment, and not taking into account their actual skills and abilities.

The major issue here is that outside of the elite prospects who usually have two to three good to great seasons of statistics to look at, many of the Entry Draft prospects usually have only one major season, and if they have a late birth date maybe two. So when trying to make a projection model, the sample size becomes an issue. While statistics can still be used to predict the future, you have to take these kinds of things into account and analyze the player’s talents by watching them in order to get a clearer picture of who a player is and where you should have him on your draft board.

Follow Corey on Twitter at @coreypronman.

Corey Pronman is a contributor to Puck Prospectus and runs the statistical hockey site The Hock Project. You can contact him at CPronman@fau.edu.

Corey Pronman is an author of Hockey Prospectus. You can contact Corey by clicking here or click here to see Corey's other articles.