About Ken Ashby's AccuRatings

Return to homepage

***** Who I am a retired math teacher and engineer living in Dallas, Texas. I spend much of my time running and assisting with races. (I once held a national age record of 175 miles in 48 hours.) I have a dedication to Libertarian politics. I also happen to be a sports fan. I am one of more than a hundred ratings enthusiasts. We are listed on comparison web sites done by David Wilson, Kenneth Massey and Todd Beck. When/Where Growing up in rural Kentucky, I began imitating the Frank Litkenhous ratings while in high school in 1967. It was a hobby to entertain myself, family and friends. I discontinued it for lack of time after 1980. Motivated by the new college football Bowl Championship Series, I resumed ratings in 1998, and began posting them to the internet in 2004. What My initial interest was basketball. Football was much more challenging, due to shorter schedules and less conference interplay. Football games also are less predictable because of fewer possessions and scores. I saw no need to rate MLB or the NBA, whose teams interplay enough for win-loss records to be meaningful. Since 1998 I have rated NFL and NCAA-FBS football teams. I gave up college basketball for a few years, but brought it back in 2014. How I began in an age before personal computers and electronic calculators. I had the ability to do quick mental calculation, but multiple iterations took too much effort, and so my ratings were not as good back then. In 1998 I started using a computer spreadsheet to reiterate. I manually enter data while reviewing box scores and game summaries. I use aggregate data for basketball. I also use database analysis to improve the formula. How Often I update the ratings weekly throughout the season. Why There are two distinct purposes for sports team ratings. The first is to forecast game outcomes, and they are called "predictive." The second is to rank teams on accomplishment, and they are termed "retrodictive." Some ratings systems specialize in one or the other, and some combine the two purposes. I recognized by 2004 that a combination is not great at either, and so I now produce separate figures for predicting and ranking. Predictive Ratings I publish PS (point spread) and OU (over-under) values for each team. The PS values of opposing teams are subtracted to predict the scoring margin. The OU values are added to predict total points. Factors that go into a predictive rating include: (1) win-loss record, (2) strength of schedule, (3) points scored and allowed, (4) game stats, (5) past seasons' ratings, and (6) player experience. I do NOT take into account injuries and suspensions, which are too hard to keep track of. The success of various systems at prediction can be compared according to straight-up wins, wins versus the Vegas Line, and average error. Todd Beck does this at his invaluable web site, ThePredictionTracker.com. Retrodictive Ratings Retrodiction is tougher. It is possible to optimize retrodictive fit using criteria like ranking violations or mean square error, but that does not produce what most observers agree is a fair ranking. Each expert has his own opinion as to the proper weighting of relevant factors. My own retrodictive ratings are based mainly on winning percentage and strength of schedule, with scoring margin considered only to the degree that the game is close. Teams are rewarded less for narrow wins and penalized less for narrow losses, but lopsided margins don't matter. Each team's retrodictive figure is expressed as a percentage between zero and one. It can be thought of as the projected winning percent if the team had instead played a schedule of average teams. Weekly Rankings Before the season, I rank teams by predictive (PS) ratings. After the regular season, I rank them by retrodictive ratings. During the season, the ranking figure is a "predictive mix" of the two in proportion to the amount of the schedule completed, and thus predicts the final ranking. Since 1995, Kenneth Massey has maintained a weekly college "composite" ranking on his site, MasseyRatings.com. In addition to ranking the teams, he ranks each system by how how closely it correlates with the consensus. The Massey site links to Dave Wobus' correlation pages, which show what systems best predicted the latest consensus at each of the earlier weeks. *****