Our usual ECAC predictor, Pred, has Cornell in deep trouble.
Place Team Pred 1 Yale 34 2 Clarkson 30 3t RPI 26 3t Vermont 26 5 Harvard 25 6 St. Lawrence 24 7 Brown 22 8 Princeton 19 9 Dartmouth 16 10 Cornell 15 11 Colgate 14 12 Union 12
Here is how we have compared historically to the ECAC Coaches Poll. Below, two scores are compared: the sum of the absolute value of the difference between each team's actual and predicted finish, and a comparable sum of the squares of the difference between each team's actual and predicted finish. In each measure, a lower score indicates a more successful predictor.
Year Sum of ABS:TBRW-Coaches Sum of SQ: TBRW-Coaches 1996 19-28 49- 89.5 1997 38-38 156-150 1998 35-35 198-192This year we also debut a new predictor, RPred, which takes into account the statistical effect regression to the mean. In some cases over the past few seasons, Pred has failed to predict a big move in the standings for one team or another. In many cases, this has been a failure to predict a traditionally strong team returning to form after a few lean years. In other cases, it has been a traditional also-ran which has fallen dramatically, after one or two seasons in the sun, which has foiled our predictor. RPred factors in the historical strength of each ECAC team over the entire history of Div. I ECAC play (since 1963). Effectively, the Pred stat is pulled or pushed half the distance back towards the lifetime winning percentage of the program (there are some normaliziation effects which we need not go into here). The results of RPred for 1999 are as follows:
Place Team RPred 1 Clarkson 31 2 Harvard 27 3 Yale 26 4 RPI 25 5 Vermont 24 6 St. Lawrence 23 7t Brown 22 7t Cornell 22 9 Colgate 18 10 Princeton 17 11 Dartmouth 16 12 Union 13The statistical breakdown of this stat will also be included in the "Never Apologize..." discussion below.
The TBRW? predictions are based on each team's previous year's performance, modified by off-season roster changes.
Each team starts with its previous regular season standing RS.
The number of post-season upset wins or losses PS is then added.
Finally, added to this score is a function of the following modifiers:
The coefficient multiplier for the sum of the modifiers helps balance the effects of roster changes and also generates Pred, the final score which signifies the number of points a team should finish with.5pct Conference winning percentage in 5 seasons previous to past season.
Finally, comes the newest cast member of our statistical stage, RPred. RPred is easily derived from Pred when you have the career records of each team in conference play, EPct.
The average of EPct is slightly different from .500, owing to changing in the length of the schedule and conference membership over its history. Therefore, we also compute MEPct, a modification of EPct that pulls the average EPct back to .500. MEPts is just MEPct * 44. In other words, the number of points you would get with the MEPct over a 22 game slate.
Finally, RPred is then derived from simply averaging the old stat, Pred, and the historical measure, MEPts. RPred = (Pred + MEPts)/2.