Leagues
DEL Daily News
DEL Forum
Coach Tools
MyDEL
Search Coaches
Coach Records
Change Requests
Changes
Newbie Help
Help Pages
DEL Time: 18:03
|
DEL uses computer ranking systems to help decide quite a bit in the leagues, such as postseason tournament selections and BCS bowl bids in football. You can find a detailed description of the ratings in the Dolphin Ratings pages, which give real-life sports ratings using similar algorithms. On the rankings page, each college team is given up to six ratings, which are called standard, win-loss, predictive, RPI, pairwise, and poll, and are defined as follows: Standard. This is the best measure of the quality of a team's season, based on mathematical techniques. The computation of these ratings is a two-step process. First, team strength ratings are calculated using game scores and locations. Second, team ratings are calculated using the opponents' strength ratings, locations, and game outcomes (mostly win/loss, with margin of victory considered only weakly). The schedule strength is weighted by a game's significance; mismatches (a great team playing a poor team) are given little weight in this rating. This rating is used to determine the "computer champion" in the champions file. Win-loss. The "win-loss" rating is a similar ranking to the standard, except that win-loss information is used instead of game scores. Predictive. This rating provides the best measure of the team's strength, and is called "predictive" because it is the best at predicting future outcomes. The team's rating, as well as that of its opponents, is determined using game locations, and scores. The lines for the next game day are calculated using this rating. RPI (college only). In place of the real-life RPI is the "improved" RPI, which carries the calculation to infinite depth instead of making statistical assumptions about the strength of schedule. While those assumptions have some merit, they are not always right and the errors can be 20 places or more in the rankings. In addition, home field advantage is accounted for. To give comparable numbers to real-life, the calculations are scaled by 0.25 (basketball, baseball, hockey) and 0.334 (football). The football and hockey calculations are also adjusted for biases in the real-life RPI formulas to give a good approximation of the RPI without its errors. The NCAA basketball selection process adds additional adjustments into their RPI equation, which are added to the college basketball RPI scores in DEL. These adjustments are not published; my best attempt at reproducing those is as follows:
Note that non-dI schools are not in DEL, so that adjustment is never made. Also, the (*) value refers to teams in the bottom half of the standings. With the current 324-team CBEL, this means #163 and lower. A second adjustment is made based on the strength of a team's non-conference schedule. If half or more of a team's non-conference opponents are in the RPI top 50 (win or lose), 0.0024 is added. If half or more are in the bottom half of the league (again, #163 or lower right now), 0.0024 is subtracted. NCAA baseball makes adjustments as well, which are approximated as follows:
Pairwise (college hockey). Pairwise ratings compare each team in the league to all postseason-eligible teams (at least at 0.500 and not under sanctions). A team's pairwise rating equals the fraction of comparisons that it wins, with RPI ratings used to rank teams with equal pairwise ratings. A comparison is determined based on five factors, each of which can give one of the teams points:
A critique of the pairwise ratings is that they provide an excellent indicator of a team's postseason worthiness relative to other top teams, but do not sufficiently take into account schedule strengths. That is, a team should play an easy set of opponents because having a good record is more important than playing good opponents. This is unfortunately also true of hockey polls, which are "worse" in that regard than those of any other sport, so perhaps the pairwise system is a reasonable reflection of the polls. With the changes to the RPI and the removal of the "last 16 game" criterion, this is less pronounced than in the past but is still present to some degree. Pairwise (college baseball/basketball). The pairwise ratings in the other college sports are based on analysis of real-life NCAA tournament selection processes. Input data include RPI, standard ranking, conference RPI, record (coverall and conference), and head-to-head against other teams under consideration. Poll (college only). The poll attempts to mimic real-life voted polls, based on detailed analysis of voting tendencies. The treatment of teams is fundamentally different than the ratings listed above, thus leading to significant systematic differences between polls and computer ratings (seen in DEL as well as in real-life). Offense and Defense. These ratings are given on a separate page, and measure how many points each team tends to score and allow. The overall rating listed here is the predictive rating, and if one wanted to predict future game outcomes (odds of winning, expected scores, and expected total score), one could do so with the data on this page.
|