by Larry_Warrilow on Thu May 31, 2007 7:07 am
a course rating system as a major factor in rankings has always been high on the wish list for we stat hounds. i agree with ernie that course ratings should evolve out of actual tour players statistics. a year and a half ago i worked up a limited model of a rating system based on the very best players results: the best winning stud score, all four rounds from the back tees, and one windy round, based on stats over three years. as a pilot project, the results for the six courses i used as a demo were pretty convincing, and i published the first half of the project on the forum on 5 april 2006.
shortly afterward indiebuilt bailed out, so publishing the second half seemed moot even though it had all the relevant math, and the method for arriving at a universal rating conversion factor. on the other hand, while basing difficulty ratings on the best players results is a valid method, the best players haven't in large part played the vast majority of apcds in competition, leaving many courses unrated, especially in the near future.
steve pitts, working on much the same problem, had developed an algorithm that could digest an entire difficulty level's stats not only in one event, but over a whole season on any given course, and i believe that that would also give a solid comparative course difficulty rating. even one weeks stats, given a large enough field, would give a good temporary number, as long as the conditions were not overly extreme. in fact, the larger the stat sample, the more any particular combination of conditions will cancel out.
having looked carefully at these stats in the past, i also believe that the relative difficulty rating of a course is a constant for all the links difficulty levels. the LC AI numbers, while consistant within themselves, still do not represent actual tour players experience on a given course, but would be an arbitrary place to start.
if mark still has records of tour play during previous indiebuilt or microsoft seasons, a ratings database for all LS and some apcd courses already exists.
in my rating scheme a tough course is arbitrarily picked as the baseline rating course and is assigned a rating of 1.00 (firestone fit this description in my example)[if another course is rated tougher later it would get a lower number: eg 0.98]. every other course gets a number in relation to this baseline course based on player stats. the higher the number, the easier the course: eg, firestone was 1.00, and innisbrook rated 1.008. thus, if a player shot a 70 at innisbrook he would multiply 70 x 1.008 for a 70.56 which would be the equivalent score at firestone. banf was rated at 1.1513, so a 70 shot at banf [70 x 1.1513=80.5] would be equivalent to an 80.5 at firestone. thus each 18 hole score can be compared in rating difficulty to any other by a constant multiplier (a relative difficulty conversion factor) that spans all levels. as a short example here are the six courses i chose for my initial exposition:
course/difficulty-rating/(70)score equivalent at firestone
firestone/1.00/70
innisbrook/1.008/70.56
castle pines/1.024/71.68
pelican hill/1.045/73.15
bountiful/1.086/76.02
banf/1.1513/80.5
these numbers were derived from actual winning scores shot by the best tour players over three years at these particular courses at champ. an analysis of just one large full field event at pro by the whole field might well yield very similar relative difficulty numbers. the point is that the ratings are based on actual links player experience. lw
______________