Getting Down To Business

March 30, 2014

All too soon it's that time again - April is the beginning of Ballot Projection Season. I don't try to guess what the committee will do, but I do produce my own version of the "Nitty Gritty" report that summarizes the input to their process.

"The committee may consider comparing individual teams based upon":1

The real Nitty Gritty report is ordered by RPI rank and includes the top 100 or so teams. I include the top 75 plus any outside the top 75 that are "in line for" their conference's automatic bid. I order the teams by the results of each team's pairwise comparison against each of the others.

The input is a "pairwise matrix" suggested by an subscriber based upon the pairwise rankings that do a good job of predicting the at-large selections and seedings for NCAA hockey. (Thanks to Jerry Palm for the link.)

I assign a value to each of those criteria and compare each team to every other team then order the report in decreasing number of pairwise "wins." The actual weights are arbitrary (and meaningless) but head-to-head, record vs common opponents and record vs top 25, 26-50, etc. matter more than any of the others.

I also include a variation of "record vs teams under consideration" - instead of opponents' rank, this is each team's record vs teams with a rating more than one standard deviation above average. This is approximately "record vs the top 50" so gives that part of the "records vs top..." comparison a little more weight.


Data for each team is displayed in a two-line format:
Line 1
#The "rank" according to number of pairwise wins
ConfThe team's conference. If the team is currently at the top of its conference standings, the conference is preceeded by a † and the field links to to a list of the conference teams in the running for at-large bids.
TeamLink to the "gory details" page for the team.
PW%The pairwise "winning percentage" resulting from the comparisons.
PW RecThe #of teams that this team "won vs" and "lost to" in the pairwise comparisons
21 April update
This links to a page that displays the team-pair-specific values used in the comparison
ratingThe team's rank by the rating used in the comparisons. I produce the report for both the RPI and ISR.
CRThe ranking of the team's conference using rating.
C RecThe team's conference record
TRThe team's relative rank within its conference. No attempt is made to deal with conference tiebreakers for this report (overall record is used instead.)
NC ratingThe team's non-conference RPI (or ISR) rank.
NC RecThe team's non-conference record
Rd RecThe team's road record
Last 15The team's record over the last 15 games
RecordOverall record.
1-25Record vs teams in the rating's top 25
26-50Record vs teams ranked 26-50
51-75Record vs teams ranked 51-75
76-100Record vs teams ranked 76-100
101-150Record vs teams ranked 101-150
151+Record vs teams in the bottom half of D1
Rec vs TUCRecord vs teams whose rating is more than one standard deviation better than the average rating.
TUC%For teams who've played more than 10 games against the top teams, this is the winning percentage associated with Rec vs TUC. For teams that haven't, it is the number of wins against such teams divided by 10. (This prevents a 1-0 record from trumping a 9-1, for instance.)
Line 2Relevant Results Lists the opponents and team's record for the games used in the TUC comparisons.

There isn't a practical way to display the records vs common opponents for each of the thousands of team-pairs, but that data is presented (for all teams) in a separate report.

Yes there is. For top-75 teams with an overall winning percentage the PW Rec field now links to a page that displays the values used to compare the team to each other team in that category. The "Pairwise Results" report displays for each "team1":

H2H# of wins against team2 "N/A" means the teams have not played.
v COTeam's record vs common opponents: w-l-t (#) where # is the number of common opponents.
Scoreteam's score-team2's score. The background is green for a team1 "win" and red otherwise.
Team2Other (team2) team. The list is ordered by the other team's pairwise win count vs all teams considered.
ratingteam2's rating rank
H2H# of team2 wins over team, or "N/A" if the teams haven't played.
v COTeam2's record vs common opponents (same format as above)
v TUCThe heading contains team's record vs opponents whose rating value is better than one standard deviation above avergae, the row contain's team2's record vs such opponents. Note that this may include team vs team2 results.
bw-wlThe way I compare teams records vs the top 25, 26-50, 51-75, etc. is to just assign a descending weight to top 25, 50, 75, etc. wins, and ascending weight to losses to top 25, 50, etc. teams. This calculation gives zero weight to wins over teams in the bottom half of the field, and any loss to a bottom-half team gets a maximum penalty.

The header contain's team's value and the detail contain's team2's value.

 1Emphasis mine. There's no way to tell in advance which criteria a given committee will consider when selecting at-large teams.
 2I have never seen a definition for "Road RPI" so I have taken a best-guess for my implementation. It uses the base formula, since a home/road adjustment in this case would be redundant.

In memory of

Paul Kislanko