Friday, November 30, 2012

Predicting Regression in the 2013 NLL Season

In sport, athletes can break past predictions either performing at astronomical levels or competing below their projected skill level.When given a proper sample size however, one can pick out statistical anomalies as just that, a spike on the chart. As the laws of regression tell us, most things will end up falling or rising closer to the mean in the long run. This is true in lacrosse as shown by pal of the blog Dan Shirley, who is totally serious and prefers to just be called Dan.

The Math: 
He took the data for players with 100+ shots from the years 2012, 2011, and 2010. Next, He ran a regression with a constant of 0 for 2010 and 2011 eS% as the independent variables and 2012 as the dependent. This basically means that it weights the 2010 and 2011 stats to find the best weights to get 2012 results.

The formula would look like this: 2012_eS% = 2011_eS% * w1 + 2010_eS% * w2 + c where year_eS% is the effective shooting percentage for that year and w1/w2 are the weights from the regression and c is the linear constant (0 in our case). So, with simplified numbers plugged in, we get 2012_eS% = 2011_eS% * 0.71 + 2010_eS% * 0.31.


As an aside, the p-values (for alpha = 0.05)  for the weights are 2011: 0.0288 and 2010: 0.3117. So it's technically not a super strong fit, but who cares. I ran a regression for just one year predicting the next (ie 2011 -> 2012 and 2010 -> 2011) and came up with statistically significant coefficients but that's more regression to the mean instead of prediction. The main problem with that is it really hurts players like JT who shoot REALLY well and it treats that as an outlier and penalizes them the following year.


Anyways...I then took players that had 100+ shots in 2011 and 2012 and calculated their 2013 eS%. I could do more, but players with over 100 shots are the ones that are getting the majority of floor time and are the real difference makers on offense.


Here's the results with the columns being: Player name, 2012 eS%, 2011 eS%, Expected 2013 eS%, and change from 2012 (sorted from largest increase (+) to largest decrease (-))



Player
2012
2011
E(x)
Delta 2012
Kevin Buchanan
0.113
0.236
0.154
0.040
Aaron Wilson
0.165
0.271
0.202
0.036
Chad Culp
0.245
0.323
0.274
0.029
Scott Ranger
0.263
0.339
0.292
0.029
Stephan Leblanc
0.246
0.315
0.272
0.026
Shawn Williams
0.209
0.270
0.232
0.023
Cody Jamieson
0.245
0.296
0.266
0.021
Dan Dawson
0.263
0.312
0.284
0.021
Lewis Ratcliff
0.242
0.288
0.261
0.019
Callum Crawford
0.243
0.284
0.261
0.018
Rhys Duch
0.282
0.315
0.298
0.016
Jeff Shattler
0.286
0.316
0.301
0.015
Mike Accursi
0.265
0.292
0.279
0.014
Ryan Benesch
0.334
0.348
0.345
0.011
Ryan Ward
0.276
0.287
0.285
0.009
Curtis Dickson
0.318
0.323
0.326
0.008
Scott Evans
0.238
0.247
0.245
0.008
Corey Small
0.206
0.217
0.214
0.007
Drew Westervelt
0.307
0.307
0.313
0.006
Dane Dobbie
0.254
0.247
0.257
0.003
Luke Wiles
0.321
0.307
0.323
0.002
Athan Iannucci
0.263
0.247
0.263
0.000
Garrett Billings
0.247
0.226
0.246
-0.002
Shawn Evans
0.279
0.254
0.277
-0.002
Colin Doyle
0.284
0.244
0.277
-0.007
Brendan Mundorf
0.236
0.177
0.222
-0.014
John Grant
0.290
0.211
0.271
-0.019
Josh Sanderson
0.290
0.206
0.270
-0.021
John Tavares
0.408
0.292
0.380
-0.028
Mark Steenhuis
0.382
0.204
0.334
-0.048
 

 You’ll see players like Kevin Buchanan, Aaron Wilson, Chad Culp, Scott Ranger, and Stephan Leblanc score 4-6 more goals than last year and on the other end John Grant, Josh Sanderson, John Tavares, and Mark Steenhuis score 3-7 fewer goals (based on if they take 150 shots.)

No comments:

Post a Comment