Reading Ratings

March 2, 2014

Although it's a few weeks before too much will be made of these, I have begun calculating various measures of team success in 2014. This is an introduction to some of the analysis included on this site.

These ratings are published daily along with some meta-rating variations:

nRPI The RPI is the "standard" NCAA measure of team results. "Standard" has to be in quotes because the committee responsible for each sport's tournament is allowed to make whatever changes it wants to the basic formula.

nRPI is the variation adopted by the baseball committee beginning in 2013. It replaced a system of bonuses and penalties for top-team wins and bottom-team losses or playing too many games vs non-D1 opponents with a definition of winning percentage1 that gives less value to home wins and road losses and more to road wins and home losses.

The base RPI formula is simpler to state than it is to explain. It is 25% winning percentage plus 50% oppenents' winning percentage (OWP2) plus 25% Opponents' Opponents winning percentage (OOWP3).

I publish the nRPI with a comparison to the base and the SOS components of the formula (which are the same for both variants) but for any analysis that depends upon opponents' values I only use the base RPI values.

ISRm Boyd Nation's ISR as amended for home field advantage (Tweaking the ISR's) and margin of victory.

I label this ISRm to distinguish it from the ISR without the margin of victory adjustment, to which I may occasionally refer. Unless explicitly stated otherwise textual references to ISR are based upon the ISRm calculation.

ISOV The Iterated Strength of Victory is my modification of the ISR algorithm that uses a variable adjustment for each game based upon the SOV4 for each game compared to the mean SOV for all games.

The ISOV is to the (original) ISR as Jeff Sagarin's "pure points" is to his "ELO Chess" rating. It is not as good a measure of past performance as the ISR or RPI but it is useful in that it provides a mechanism for a SOS-adjusted scoring stats, which in turn can be used to provide a predictive rating.

Standard Reports

For each of the ratings these reports are updated daily.
PA(rating) This is the basic report to analyze game results according to the specific rating.
(null)Rank by PA()
PA() Sum of defeated opponents' aSOS values for each game won ÷ games played
TeamTeam name links to the team's gory details page.
RecThe usual W-L-T
ConfTeam's conference affiliation
RankTeam's ordinal ranking for the rating
ratingTeam's rating value
SOSThe average of opponents' rating values
aSOS Sum over all wins of opponents' rating divided by games played. This accounts for the inevitable rating inflation caused by losing to good teams.
dSOS Average of opponents' aSOS values. Possibly more indicative of schedule strength than SOS for rating.
vs Opp Ranks Records vs Opponents' rating Ranks is an index to the "gory details" for each team sorted by a "better wins compared to worse losses"5 algorithm.

R′Team's rank based upon the better-win—worse-loss criterion.
TeamLink to the team's gory details page for the rating.
Recwon-lost(-tied) record
ConfTeam's conference affiliation
RnkTeam's rank for this rating
vs...record vs teams ranked 1-25, 26-50,…
Interconference Analysis Interconference results weighted by opponents' rating values.
 Conference rank
CaSOSSum of opponents' ranks in wins divided by number of games played
ConfLink to details by opponent conference
SOS(W)Average opponents' rating for wins
W# wins vs teams from other conferences
WPWinning Percentage vs teams from other conferences
L# losses to teams from other conferences
SOS(L)Average opponents' rating for losses
SOSAverage opponents' rating for all games vs teams from other conferences
 SOS rank

The detail section includes a line for each opposing conference that serves as an index to the interconference results in descending WP order.

ConfLink to the conference's results against teams from other conferences
SOSAverage opponents' rating for teams from the other conference
(W)Average opponents' rating in the conferences teams' wins
W# wins by conference teams against teams from the other conference
WPWinning percentage against teams from the other conference
L# losses by conference teams to teams from the other conference
(L)Average opponents' rating in the conference teams' losses
Opp CLink to the game results for Conf vs Opp C
SOSAverage Conf team rating in games vs Opp C (the other conference's SOS)

Other reports

These are also published daily, but are not strictly "ratings."

Normalized Scoring This uses the SOS as defined by opponents' ISOV ratings to adjust runs-scored and -allowed.
 The first column is a rank assigned by comparing the teams' adjusted scoring against that of every other team.
TeamLink to the ISOV-based gory details page for the team
RecTeam's record vs D1 opponents
ConfConference affiliation
RSAverage runs scored
Rank for RS
Rank for NRS
NRSAverage runs scored normalized for SOS
NRAAverage runs allowed normalized for SOS
Rank for NRA
Rank for RA
RAAverage runs allowed
Rank for MOV
Rank for NMOV
Second-order WP This report is described in Too early Ratings. The column headings are:
Team rank using WtdWins
WtdWinsWW weighted by number and length of path − weighted LL
RankTeam rank by SOWP
WW# of teams B for which the shortest A⇒B path is shorter than B⇒A, or if the path lengths are equal there are more A⇒B than B⇒ paths
LL# of teams B for which the shortest A⇒B path is longer than B⇒A, or if the path lengths are equal there are more B⇒B than A⇒ paths
TT# of teams B for which the shortest A⇒B path is the same length as B⇒A and there are equal numbers of such paths in each direction
UU# of teams B for which there are no A⇒B or B⇒A paths
TeamLink to team page listing the shortest paths to each other team for which there are paths in either direction
D1W# of series wins (or ties)
D1L# of series losses
ConfConference affiliation
NT1# of teams B such that length A⇒B is 1
NTn# of teams B such that length A⇒B is n

 1RPIwins = 0.7×HomeWins+1.3×RoadWins+NeutralWins
RPITies = 0.7×HomeTies+1.3×RoadTies+NeutralTies
RPILosses = 0.7×RoadLosses+1.3×HomeLosses+NeutralLosses
RPIWP = RPIwins+(RPIties÷2)

 2OWP is the average of opponents' winning percentage not including games against the team for which OWP is calculated.
 3OOWP is the average of opponnets' OWP values. Note that this does include results of games involving the team for which OOWP is calculated.
SOV = Winning Score − Losing Score

Winning Score + Losing Score
 5The "better wins compared to worse losses" algorithm uses the opponent rank to assign a value to the win (or loss).
sort value = (#teans+1 - opponent rank) for wins minus opponent rank for losses over all games played.
Higher values are better.

In memory of

Paul Kislanko