Although it's still early in the season the first College Hockey Offensive
Defensive Ratings are now available and will be posted in a separate
message.  We've made a couple of important modifications to CHODR this
year which will be discussed briefly here.  For more details, check
the CHODR webpage at  http://it.stlawu.edu/~chodr
 
First, we have added Nebraska-Omaha, Mankato St, and Niagara to the
list of teams included in CHODR.  More teams will be added as
additional full Division I teams emerge.
 
The more substantial change to CHODR itself is a move to a multiplicative
model with maximum likelihood estimation from the previous least squares
additive model.  While we still interpret the offensive/defensive ratings
as goals for/against rates vs. an average team, the ratings are now
computed to maximize the probability of all past game scores (rather than
minimize the sum of the squared differences between predicted and actual
scores).   For those with a bit of probability background we assume a
Poisson scoring model for figuring the probabilities.
 
The more obvious change is to a multiplicative calculation to predict
scoring rates in a particular game.  Suppose that Team A is playing
Team B on neutral ice
 
  OLD CHODR   Team A's scoring = (A's Off) + (B's Def) - AVG
  NEW CHODR   Team A's Scoring = (A's Off) * (B's Def) / AVG
 
In the "old" CHODR, we adjusted for home ice by adding a fixed amount to
the home team's scoring rate and subtracting the same amount from
the visitor's predicted rate.  Under the new scheme we will multiply or
divide by the home ice factor.
 
Example:  Based on games played through 11/23, the average scoring rate
is 3.09 goals and home ice adjustment is 1.032.  To forecast the score
for SLU (Off=3.04, Def=3.19) at Boston College (Off=3.85, Def=2.72) this
Friday
        BC's predicted rate = (3.85*3.19/3.09)*1.032 = 4.11
       SLU's predicted rate = (3.04*2.72/3.09)/1.032 = 2.60
 
Predictions for all upcoming games can be found at the CHODR website.
 
The other big change:  We now compute the Overall Rating as the RATIO
of Offensive to Defensive ratings.  Thus an "average" team will have
a rating near 1.00, stronger teams will be above 1.00 and weaker ones
will have overall ratings below 1.00.
 
We feel that this method may reflect the benefit of a good defense more
accurately.  For example, suppose that the "predicted score" for Team A
vs. Team B is 2-1, while Team C is predicted to beat Team D by 5-4.
Under the old overall rating system, both team A and C would have the
same one goal advantage. A Poisson model would give Team A a much
better chance of defeating B, while C and D are more evenly matched.
 
Comments?
 
Robin Lock             Tim Danehy
[log in to unmask]    [log in to unmask]
 
HOCKEY-L is for discussion of college ice hockey;  send information to
[log in to unmask], The College Hockey Information List.