[back to Home Page] [back to Articles Page]

TullyRunners.com -  Article


Speed Ratings - Response to Some Common Questions

by Bill Meylan ... (August 18, 2007)


I get questions about "speed ratings" in e-mails and in person. People are curious about certain aspects of speed ratings or may be interested in making their own speed ratings ... Some people wonder about the concept of speed ratings or where they came from ... This article will try to answer some questions ... It is a bit repetitious with other articles here, but does discuss additional aspects.

Question - "What is a 'Speed Rating' and what is the real concept behind Speed Ratings??"

Response - A number of articles discussing speed ratings are posted on this web-site (see the Articles Page) ... I realize some articles might be confusing, so I'm going to attempt a different explanation here.  The "concept" behind speed ratings depends on your point-of-view ... This question comes primarily from coaches and runners with a "track & field" perspective.

Track is a "time-oriented" sport ... The most important thing to a majority of track runners is their final time in the race ... The time itself is the important concept ... although competition clearly has significance, the time of the race has a meaning of its own and a knowledge of "who was in the race" and what they did may not be necessary to evaluate a performance with respect to track ... The stop-watch is the evaluator.

As an example, consider high school "Runner X" who just ran a 1-Mile track race in 4:08.9 ... the time itself tells you is was a really good performance (you don't need to know much more than the time) ... While it might be nice to know who else was in the race, the final time evaluates the performance all by itself (and it gets used to make local and national leaderboards).

Now consider high school "Runner X" who just ran a cross country race in 15:24 - "Was it a good performance?" ..... IT DEPENDS ... The time of a cross country race (all by itself with no other info) tells you very little ... Several questions come to mind immediately:
 (1) What is the distance ?
 (2) Is it an easy course or a hard course ?
 (3) What was the weather ?
 (4) Which race course ?
 (5) Who was in the race & where did the 15:24 finish overall ?

You need to know more than just the time for a cross country race ... The time is important, but it depends on how the time relates to the various questions above ... Now assume the following for the 15:24 cross country time:

 (1) the 15:24 runner finished 2nd
 (2) the winner was Chad Hall (15:20) ... Steve Murdock was 3rd (15:25) ... Matt Tebo was 5th (15:28) and Matt Centrowitz was 8th (15:34)
 (3) The race course was Balboa Park 5K and the race was Footlocker Finals

Knowing these facts allows the 15:24 time to be evaluated as a really good performance ... There are three major factors that allow a cross country time to evaluated:

 (1) finish in relation to other individual runners of known ability
 (2) finish time on a race course of known speed, especially when the speed is known relative to other race courses
 (3) the quality of the race itself in relation to other races

These are the three factors that speed ratings use to adjust the final times of cross country races ... The concept of speed ratings is to adjust the final times of any specific race (using these principles via statistical-experience methods) so the times of a race can be compared to the times of any other cross country race ... This process is necessary due to the variability of XC distances, course difficulties, weather and other factors that affect the final times significantly.

Question - "What does a 'Speed Rating' number mean?"

Response - A speed number comes from an arbitrary scale ... I decided a 26:00 cross country time would be zero on the scale ... For every three seconds faster than 26:00, you get one speed rating point ... Here is a simplified table:

     Time    Rating (with no adjustment)
    -----   ----------------------------
    26:00        0
    25:30       10
    25:00       20
    24:30       30
    24:00       40
    23:30       50
    23:00       60
    22:30       70
    22:00       80
    21:30       90
    21:00      100
    20:30      110
    20:00      120
    19:30      130
    19:00      140
    18:30      150
    18:00      160
    17:30      170
    17:00      180
    16:30      190
    16:00      200
    15:30      210	

This table assumes NO adjustment to the final times ... The table is based on a 5K race course that is slow compared to most 5K courses - This "standard" race course is the SUNY Utica College as it existed in 1999 (it was used as the NY State Championship course in 2000, as well as the Section 3 Championship course for several years) ... the course was fairly challenging, and it was slightly longer than 5000 meters when measured cutting all the tangents (shortest possible distance). 

Many 5K courses are NOT actually 5K in distance if measured "shortest possible" distance (effectively USATF Certified) ... many high school courses are measured "mid-course" or "following a painted line" and are actually less than 5K if runners cut the tangents ... Measurements on various NY courses used for championship races have shown that 50 to 100+ meters can be saved by cutting the tangents ... This is one reason distance can never be used in a speed rating formula.

Here is the actual equation I use to generate speed ratings:

Speed Rating = (1560 - (actual race time in seconds) - (race correction)) / 3

where 1560 is the number of seconds in 26 minutes ... 26 minutes is used because it corresponds to zero in the chart above ... the race correction (adjustment) is how fast or slow (in seconds) a race is in relation to the SUNY Utica standard course ... the entire expression is divided by 3 because one point equals three seconds.

The race correction (adjustment) is determined by the three major factors noted above... A brief description of how the correction is determined is presented in the article "Early Season Speed Ratings and a Brief Overview of Speed Ratings".

Question - "Where did the term 'Speed Rating' come from?"

Response - The term "Speed Rating" comes from horse racing and it's been around for many decades (in one form or another).  Speed ratings apply to both thoroughbred and harness racing .... The "rating" part can apply to (1) rating the speed of a race track or (2) rating the speed of an individual horse.

The concept of speed rating horses developed for two primary reasons:
 (1) to help professional bettors bet money
 (2) to help racing officials classify horses for races based on ability

Historically in harness racing, "speed rating" has applied to the harness tracks themselves ... The US Trotting Association has speed rated tracks for many years (I know speed ratings for tracks were being published in nightly harness racing programs back in the 1950s and 1960s) ... The current official Speed Ratings are based on comparisons of individual times at one track with the same horses' times at other tracks over "fast" or "good" tracks at a mile distance ... Here are some examples (the median win times (half were faster, half were slower) for 2005 on a "fast" or "good" track):

                                        Speed      2005      2005
  Harness Track                         Rating    Fastest   Median
  ------------------------------------  ------    -------   ------
  The Meadowlands, East Rutherford, NJ  1:56.3    1:48.0    1:52.4
  Cal-Expo  Sacramento, CA              2:00.1    1:49.4    1:57.1
  Pompano Park, Pompano Beach, FL       1:59.0    1:50.3    1:56.3
  Freehold Raceway, Freehold, NJ        2:00.0    1:52.0    1:58.2
  Saratoga Harness, Saratoga, NY        2:01.2    1:53.4    2:00.0
  Yonkers Raceway, Yonkers, NY          2:00.0    1:54.0    1:59.2

The median and fastest times are determined by the quality of horses racing at the track (some tracks have better horses than others).  The Speed Rating is an inherent measure of the race track speed based on actual results ... For example, the speed rating for Yonkers raceway is 2:00.0 and the speed rating for Saratoga Harness is 2:01.2 ... this basically means that the same horse (or horses of the same ability) raced 1.2 seconds slower at Saratoga than Yonkers on an average track.

Looks a lot like a high school cross country course conversion table, doesn't it? ... That's because they are very similar ... Here are some numbers from my current high school XC conversion table for boys (my girls adjustments are slightly different) ... rather than listing finish times, I list the correction (adjustment) used for the speed ratings (negative adjustments mean faster than the standard):

  Cross Country Course            Adjustment    Normal Adjustment
                                 (in seconds)   Range (in seconds)
  -----------------------------  ------------   ------------------
  SUNY Utica NY                       0             standard
  Bowdoin Park NY                    -12          -6  to  -15
  Sunken Meadows NY                  -9           -3  to  -12
  Van Cortlandt Park NY (5K)         -26         -21  to  -33
  Van Cortlandt Park NY (2.5-mile)   -235       -230  to  -239
  Saratoga Park NY (3.04 mile)       -75         -70  to  -80
  Holmdel Park NJ                    -28         -25  to  -33
  Woodward Park CA                   -60         -54  to  -69

The US Trotting Association decided to stop generating their "race track speed ratings" at the end of 2006 for this reason ... "Handicappers long ago found that the speed ratings were a handy starting point when comparing times recorded at one track with those that might be recorded at another ... Today, the majority of bettors see speed ratings published in past performance programs printed by such vendors as TrackMaster and Equibase, and those ratings reflect week-to-week changes in that horse’s efforts, and variables that are the result of the changing condition of the racing surface".

And that brings the discussion to Speed Ratings for individual Horses ... Again, this has been done for many years ... But speed ratings for individual horses came to prominence in 1975 when Andrew Beyer authored the book "Picking Winners - A Horseplayer's Guide" ... for the first time, a complete speed rating methodology with detailed instructions and examples was published and available to everybody ... and it actually worked if you were willing to spend the time required to do it correctly.

The Daily Racing Form (the daily Bible for thoroughbred past performance programs sold at race tracks) eventually thought so highly of the method that they replaced their old speed ratings with the Beyer Speed Figures in 1992 ... the Beyer Speed Figure is now shown for every race performance for every thoroughbred horse listed in the Daily Racing Form.

Interestingly, Andrew Beyer (a short term student at Harvard in the Class of 1965) was actually helped with methodology by a mathematics doctoral student at Harvard ... Beyer became a columnist for the Washington Post and Washington Star (and was syndicated as well) ... he has written several other interesting horse racing books that any "well-rounded" horse handicapper should read.

Now this is important and it goes right to the heart of the speed rating concept: ... Speed ratings for individual horses were devised for one purpose - to allow the speed of a horse to be compared to other horses to help make a bet! ... Faster horses generally beat slower horses ... Other handicapping factors come into play, but speed is the one factor that gets measured in every race (via the final times and fractional times).

The concept goes like this ... how fast any particular horse runs is determined by how fast all the other horses run ... The method is designed to compare horses to each other ... a specific race time for any specific horse is meaningless unless you know how fast the other horses ran.

General Response - I began "playing" with the Beyer speed method shortly after the book was published in 1975 ... At that time, I was more interested in harness racing than thoroughbred racing (and that's still true), and I wanted to see if I could adapt the method to harness racing ... Being an engineer and scientist, I "tinkered" with the method somewhat and became generally amazed at how well it worked in thoroughbred racing when used in conjunction with other handicapping factors ... I was NOT impressed with the large amount of time it took to do it correctly (including all the background work required).

Fast forward to 1998 ... one of the Tully girl cross country runners (an 8th-grader) qualified for the NY State Championship meet being held in Lake Placid at Mt. Van Hoevenberg on part of the Olympic biathlon course ... the local people were disappointed the "hoped-for" two feet of snow never fell ... Discussions after the race suggested the Tully girls might have a shot at winning a State class title the following year ... That's when I first decided to apply a comprehensive speed rating methodology to high school cross country runners (so I could see where the Tully girls stood during the 1999 season ... in 1999, we lost to Bronxville at States by two points ... the ratings predicted that result to within one point).

Lake Placid brings back "painful" memories whenever I hear the town's name ... In 1980, the local Chevrolet dealer in Tully obtained a number of complimentary tickets to Olympic events in addition to buying others, especially hockey ... He called me one night and said me had an extra ticket to the USA vs. Russia hockey game and wanted to know if I could go (he lived just a stone's throw away) ... I laughed and said it would take a miracle on ice for the USA to beat the Russians, so I decided to stay home ... Clearly, I am NO genius (true story).

General Response - The Speed Ratings that come from me (via the TullyRunners.com web-site) are based upon my own adaptation of the Beyer Speed Figure method in conjunction with Course Conversion tables and appropriate statistical & iterative methods.

Question - "Is the Speed Rating Method purely statistical? ... One of your articles says it's part science and part art ... Why is it "part art"?

Response - A simplified answer is this ... When you have sufficient good data, the method can be purely statistical (meaning anybody using the same data and statistical methods will arrive at the same result) ... Unfortunately, the data itself can be the source of a problem requiring "part art" intervention (meaning it becomes necessary to make a "best-guess" or "best-fit" based on experience).

Pure statistical results are only as good as the data used to generate the results ... The old saying "What comes out of a sewer depends upon what you put into it" applies to statistics based on data points.

As an environmental scientist, I see lots of experimental data for chemical and physical properties such as melting point and water solubility ... when the same chemical is measured carefully and correctly in the laboratory, any researcher world-wide will get basically the same answer ... if they measure it again next week, they will get the same answer ... if they measure it again next year, they will still get the same answer ... The value of the data point does not change ... that's what most statistics evaluate (static data).

In contrast, human race performance & horse race performance can (and does) change from day-to-day, week-to-week and year-to-year.  Therefore, performance of any individual commonly varies over time ... Each of the individual speed ratings was derived from data that had associated variability ... Because of this variability, the race correction (adjustment) gets determined by some form of statistical averaging or statistical fit ... BUT the whole key is knowing what to average or what to fit (and human judgment is commonly required).

In its purest form, the Beyer Speed method uses a great deal of human judgment ... Once a race season is in progress, the Beyer method takes the recent speed figures for each individual horse and projects a speed figure for the upcoming race ... after the race, the projections are evaluated with respect to the actual race times and a race-day correction is determined.  The projections from experienced handicappers are usually very close (so it is not a guessing game by any means).

For high school XC, I rarely do "projection" adjustments for races (it takes way too much time for the quantity of XC I follow, and I already don't have a life during XC season) ... For NY State, I have a large database of speed ratings for individuals and I use that database and a computer to automatically derive an "overall speed rating" based (in large part) on recent performances ... The "overall speed rating" gets used in place of a Beyer projected speed rating (because typical projections are usually close to recent performances).

Purely Statistical ... Finding race adjustments for some XC races is purely statistical.  The best example for me is NY State in the latter part of the season ... my database has multiple performances for over a 1000 individuals state-wide ... multiple head-to-head performances ... many cross-over comparisons (different leagues, different sections) ... Any statistician using that amount of common data will arrive at the same basic results as me.

Knowing What to Average or Fit ... This would require a multiple-part article (each part being very long and boring) ... Variable human performance causes a problem ... Variable quality of races causes a problem ... If the data points were truly static data, then basic statistics applied comprehensively (or even randomly) could do a much better job ... BUT that's not the case - the data are variable and it must be dealt with on a race-by-race scenario.

Question - "How accurate are your Speed Ratings for races outside of NY State"?

Response - It depends on how much data are available when I make the ratings ...

Important to Know ... All speed ratings made by me are NOT equal in accuracy ... this applies to both in-state and out-of-state ratings ... Some ratings have a much larger "margin-of-error" than others. In the latter part of the NY cross country season, the margins-of-error are very small and the accuracy is very high because there is so much evaluated data available.

Although I don't show the evaluation in my web-databases, I "rate" my speed ratings for each race on a scale of 1 to 5 ... 1 being "uncertain" and 5 being "highly certain" ... This is a carry-over from horse-playing days where experience proved that betting serious money on "uncertain" ratings is a very bad idea.

At the very beginning of the NY cross country season, I consider all my ratings to be "uncertain" until proven otherwise ... To me, it is an iteration process (a process of making sequential "guesses" each week that converge on a "best answer") ... The first ratings of the season "seed" the iteration process (I hope they are good, but they may be off) ... results from additional races help refine the ratings ... "More Good Data = Better Ratings".

Accuracy for rating out-of-state races depends on:
 (1) evaluated, historical data for that particular race (or course)
 (2) current data for individual runners in the race (meaning I have some current speed ratings for some runners to use for comparison)
 (3) a knowledge of the quality of the race in comparison to evaluated races

Some State Championship Meets are fairly easy to speed rate with decent accuracy ... For example, I have good evaluated historical data for the New Jersey Meet of Champions, the PA State Meet and the CA State Meet ... in recent years, these meets have had the same qualifying procedures and generally same "quality pool of runners overall", so speed rating them is straight-forward ... The accuracy is NOT as good as my latter-season NY ratings, but the accuracy is perfectly acceptable for my purposes ... I do not worry about about ratings that might be a couple of points high or low ... To me, a runner with 160 speed rating in one state is equal to runner with a 163 rating in another state ("flip a coin" to guess who might beat who on a neutral course).

Rating a State Meet (or any out-of-state meet) I know nothing about is different: ... It requires me to make the initial guess of an iteration process ... My initial guess may be bad = bad speed ratings ... Refining the ratings requires background work, and the effort I'm willing to spend depends on my purpose ... These ratings will NOT be highly accurate.