Never Apologize, Never Explain 2016-17

Never2

The predictions use these metrics, described below:

Metric Definition
RS Prior ECAC regular season
PS Prior ECAC post-season
Imp Second-half improvement
PtsRet Pct returning *
Avg10 Traditional program strength

* metric modified from last year, as described below

The Metrics

RS

RS is each team's points in the prior ECAC regular season:

RS  
Brown 12  
Clarkson 23  
Colgate 14  
Cornell 22  
Dartmouth 22  
Harvard 28  
Princeton 9  
Quinnipiac 37  
RPI 23  
SLU 25  
Union 18  
Yale 31  

PS

PS is the number of upset (lower seed) advances (+1) or eliminations (-1) in the previous season's ECAC tournament.  The consolation game is ignored.

The upsets in the last season's ECAC tournament were:

PS  
Brown 0  
Clarkson 0  
Colgate 0  
Cornell 0  
Dartmouth +1  
Harvard 0  
Princeton 0  
Quinnipiac 0  
RPI 0  
SLU 0  
Union 0  
Yale -1  

 

Imp

Imp is the difference in points gained in the second half (games 12-22) minus the first half (games 1-11) of the prior ECAC regular season, divided by 2 (and rounded down).  A positive Imp indicates the team had a better second half:

Pts 2nd ½ 1st ½ Diff Imp  
Brown 12 6 6 0 0  
Clarkson 23 16 7 9 +4  
Colgate 14 6 8 -2 -1  
Cornell 22 8 14 -6 -3  
Dartmouth 22 10 12 -2 -1  
Harvard 28 15 13 +2 +1  
Princeton 9 2 7 -5 -2  
Quinnipiac 37 17 20 -3 -1  
RPI 23 8 15 -7 -3  
SLU 25 14 11 +3 +1  
Union 18 12 6 +6 +3  
Yale 31 18 13 +5 +2  

 

%Ret (Percentage Returning)

The old figure, WRet, captured how much of last year's players' contribution returns (or has been lost to graduation or early departure).

This year we are using a radical departure from prior predictions.  We are replacing the complex formula that was a function of forward, defense and goaltending with a very simple stasistic: the percentage of returning point scoring.  For one thing, the old metric didn't seem to be doing a great job of calculating the fall-off due to graudating (or early departing) talent.  The goaltending figure tended to overwhelm the other components.  Also, the differentiations both into forward vs defense and top N of each introduced arbitrary cutoffs with relatively weeak theoretical justification.

The following players with eligibility remaining are not returning:

%Ret = Points Returning / Total Points

  Total Pts Lost Return %Ret  
Brown 140 68 208 .67  
Clarkson 268 73 195 .73  
Colgate 253 122 131 .52  
Cornell 213 38 175 .82  
Dartmouth 257 106 151 .59  
Harvard 313 115 198 .63  
Princeton 151 20 131 .87  
Quinnipiac 430 177 253 .59  
RPI 264 74 190 .72  
SLU 293 90 203 .69  
Union 242 32 210 .87  
Yale 224 83 141 .63  

Avg10

Avg10 measures traditional program strength, by taking the team's mean number of Points over the prior 10 seasons.  This will be used to compute the expected Incoming factor.

Brn Clk Cgt Cor Drt Hvd Prn Qpc RPI SLU Uni Yal  
2007 16 30 17 24 27 22 22 24 17 33 15 17  
2008 15 33 21 25 15 27 28 22 15 16 25 22  
2009 10 20 17 29 24 24 28 21 13 26 20 32  
2010 16 11 26 31 17 17 18 22 23 23 28 32  
2011 18 19 11 24 26 15 24 19 24 13 36 35  
2012 14 22 23 30 19 25 16 23 17 21 32 22  
2013 20 19 15 19 22 14 20 37 27 22 24 25  
2014 17 24 29 24 16 16 8 28 21 18 37 24  
2015 13 19 26 22 26 25 6 35 18 29 17 28  
2016 12 23 14 22 22 28 9 37 23 25 18 31  
   
Pts 151 220 199 250 214 213 179 268 198 226 252 268  
Avg10 15.1 22.0 19.9 25.0 21.4 21.3 17.9 26.8 19.8 22.6 25.2 26.8  

Summary:

Avg10  
Brown 15.10  
Clarkson 22.00  
Colgate 19.90  
Cornell 25.00  
Dartmouth 21.40  
Harvard 21.30  
Princeton 17.90  
Quinnipiac 26.80  
RPI 19.80  
SLU 22.60  
Union 25.20  
Yale 26.80  

 

Putting it All Together

From these independent metrics, we now generate several calculated metrics:

Metric Name Definition Calculation
Prior "Prior Season" Total of prior season metrics RS + PS +Imp
Ret "Returning Value" Portion of value returning %Ret * Prior
%Inc "Percentage Incoming" Pct of points to be replaced by incoming 1 - %Ret
Inc "Incoming Value" Portion of current value from incoming %Inc * Avg10
Nieu "New Predictor" Sum of veteran and incoming factors Ret + Inc
Pred "Prediction" Predicted order of finish Rank order of Nieu

 

All but the last of these calculated metrics are just convenient conceptual groupings -- the final metric, Nieu, could be calculated directly from the independent metrics.  But we'll sacrifice brevity for clarity of expression (and check our work) by spelling them out in the table below:

RS PS Imp Prior %Ret Ret Avg10 %Inc Inc Nieu Pred  
Brown 12 0 0 12 .67 8.04 15.10 .33 4.98 13.02 11  
Clarkson 23 0 +4 27 .73 19.71 22.00 .27 5.94 25.65 4  
Colgate 14 0 -1 13 .52 6.76 19.90 .48 9.55 16.31 10  
Cornell 22 0 -3 19 .82 15.58 25.00 .18 4.50 20.08 8  
Dartmouth 22 +1 -1 22 .59 12.98 21.40 .41 8.77 21.75 6  
Harvard 28 0 +1 29 .63 18.27 21.30 .37 7.89 26.16 3  
Princeton 9 0 -2 7 .87 6.09 17.90 .13 2.33 8.42 12  
Quinnipiac 37 0 -1 36 .59 21.24 26.80 .41 10.99 32.23 1  
RPI 23 0 -3 20 .72 14.40 19.80 .28 5.54 19.94 9  
SLU 25 0 +1 26 .69 17.94 22.60 .31 7.01 24.95 5  
Union 18 0 +3 21 .87 18.27 25.20 .13 3.28 21.55 7  
Yale 31 -1 +2 32 .63 20.16 26.80 .37 9.92 30.08 2