Speed Study Series Presents: Track Bias (by Shagfarms)

Editor’s Note:   Article written by Shagfarms.   Please note the hyperlink below which leads to track-by-track breakdown for horse styles.

Everyone in the sim has witnessed a fluke performance in a sim horse. Your brand new blueblood 2 year old runs a 25 speed figure and is not injured, and then follows it up with nothing under 70. Your bargain bred 2 year old runs an 85 first time out, and then never breaks 60 again. One of your usual claimers, who can’t break a 90 speed figure, all of a sudden puts together a 105 in a losing effort, then regresses back to his usual self in the next race. Those of you who have been around long enough have seen your share of another phenomenon – one of your horses seems to have a preference for a certain track. Your horse is 3 for 3 at Massachusetts, but 0 for 21 at every other track. While statistically possible, it is highly unlikely that it could happen as much as it does. So the question is: Why? This is, as much as some of us would not like to admit it, a computer simulation. Could it be so in depth as to take into account the atmosphere of the track when running a race, like the real world does? Then again, does the real world? Maybe horses like a particular track because of the surface. Then again, maybe it’s just coincidence. There is one thing for certain about tracks in the real world, however. Some of them do have a track bias. If there’s one thing I’ve learned from watching prep races for the Kentucky Derby, it’s to discard the Bluegrass Stakes from being a legitimate prep race. Speed holds up at Keeneland to an extent that can’t be seen at most other tracks in the US. Some tracks play fair; others don’t. The question then becomes; does the sim mimic these trends? Are there tracks that favor closers, and tracks that favor speed demons? A massive study into the history of sim races is where the Speed Study Series headed for this article.

Data only goes back to week 704 for races in the sim. Prior to that, while the data may exist, it is hard to find. But with 4,838 races run since week 704 (up through week 2895) at 60 of our 88 tracks, it is not a major loss that we are missing the first year of races. The race-running in the sim underwent some major changes in the beginning, and it may actually be beneficial that we are missing this data. So data was taken for all 88 tracks, either from week 704, or the week the track was introduced to the sim. For this study, the data was broken down into two groups.

The first group was the overall number of races at the track. This data is regardless of surface, distance, age, class, gender, or any other factor. It is merely the overall totals from the track. This data was then broken down into five categories of lengths back. The first group was “front-runners”, which was any winning horse who was no more than ½ length behind the leader at any time in the race. The next group was “stalkers”, which was any winning horse who was no more than 2.5 lengths behind the leader at any time in the race. The third group was “mid-pack”, which is any winning horse who was no more than 5.5 lengths behind the leader at any point in the race. The fourth group was “closers”, meaning any winning horse that was no more than 8.5 lengths behind the leader at any point in the race. The final group was “deep closers”, which is any winning horse than was more than 8.5 lengths behind the leader at any point during the race. The breakdown was done for every race at every track on every Saturday raceday, and then a total for the track was compiled.

The second group was broken down into distance and surface. 6 groups were formed: Dirt Sprint, Dirt Classic, Dirt Long, Turf Sprint, Turf Classic, and Turf Long. The designations are: Sprint (<8f), Classic (8f-11.5f), and Long (>11.5f). Each distance was then broken down into lengths behind the leader. It was given 15 categories. The first category was any winning horse who was no more than 1 length behind the leader at any time. The second category was any winning horse that was between 1 and 2 lengths behind the leader at its farthest point back in the race. The categories continue like this, with the final group being any horse who was more than 14 lengths behind the leader at any point during the race. The lengths back were calculated for every race at every track for every Saturday raceday, and then assigned to the appropriate surface and distance. Totals were then collected for every lengths back grouping within every surface-distance group. The groups were not broken down into age, gender, class, track condition, field size, or any other categorization.

For those of you who are completely lost, I don’t blame you. What I just listed was for the statistical minds out there who care about the process taken to conduct the study. What I will be using for the actual analysis is the totals from both parts of the study.

In June of 2004, Mike introduced two new tracks into the sim. One, Athens, was due to the location of the Olympics. The other, Argentina, was a personal choice, just like many of the past track selections. The addition of two new tracks was necessary. The sim hadn’t seen new tracks since week 1544, when the Eastern Kentucky through Jamaica tracks were added. It was now week 2202, and the horse pool was growing ever larger. Race fields were drawing 25 horses per race, and many horses would not be able to draw in for 3 or 4 weeks in a row. It was clear that we needed more races. Mike could have added another race to each track. However, the number of races was bumped up to 16 merely 9 months before. Overflooding of races is just as bad, if not worse, than too few, as more races means smaller fields, meaning the likelihood of being forced into creating new horses for the horse pool would go up. So, after looking at the problem, Mike decided that 2 new tracks would fit the need of the sim. These two additions also brought a new change into the sim; the EAAA vs. America divisions.

The Eastern Kentucky, South Africa, Puerto Rico, and Jamaica tracks were added in week 1544, roughly 9 months after the previous contingency of tracks. This was done in the middle of 2002. These four tracks seem like they have been through enough races – they have seen 3038 races run since their inception through week 2895 – but statistics would show otherwise. Of all the tracks, Eastern Kentucky is, by far, the least/most biased of all the tracks. Horses within ½ length of the leader at their farthest point back win at a 30.38% clip, by far the lowest of any track in the sim. Meanwhile, horses coming from more than 8.5 lengths back have won at a 7.60% rate, significantly higher than any other track. Does this show a closing bias? Perhaps. But I am scared to confirm this, as most other tracks have had many more races than Eastern Kentucky. The fact that Eastern Kentucky is so far below the sim average for front runners, and so far above the sim average for closers, makes me think not that the closers have the advantage, but rather that the track has not yet had enough races, and the percentages are on their way to moving closer to the sim average.

Since I cannot tell whether these tracks are actually showing signs of track bias, I will not include them in my Top 5 rankings. However, the data has been collected, and it will be available to use. The data from these 6 tracks are also included in the overall averages of the sim.

From week 704 to week 2895, using only Saturday races, 398,343* total races were run in the sim. Of those, 143,232 (35.96%) were won by “Front-Runners”. 90,916 (22.82%) were won by “Stalkers”, and 93,786 (23.54%) were won by “Mid-Pack”. The “Closers” took home 48,896 (12.27%) and 21,513 (5.40%) were won by “Deep Closers” (For definitions of the groups, see above). 42 of the 88 tracks are above average for Front-Runners, 47 are above-average for Stalkers and Closers (not necessarily the same 47), 46 are above-average for Mid-Pack, and 39 are above-average for Deep Closers. Furthermore, 47 of the 88 tracks show some form of conceivable bias. For example, California’s track is above the sim average for number of Mid-Pack, Closers, and Deep Closers winning races (23.96%, 14.70%, 6.43%), and below average for front-runners and stalkers winning (33.51%, 21.41%). This shows that California appears to have a slight closer’s bias, favoring those coming from behind rather than those staying on the lead. On the other hand, Portugal has an above-average total of Front-Runners and Stalkers winning races (39.13%, 23.04%), and it is below-average for Mid-Pack, Closers, and Deep Closers (22.97%, 10.93%, 3.94%). This shows a high favoritism for those within .5 of the leader, and a clear bias for those within 2.5 of the leader, while sporting the second worst percentage of the 88 tracks for horses who drop to more than 8.5 behind (better than only Montana, who sports a 3.56% win rate for Deep Closers). What does this mean for you? If you have a deep closer running in Portugal, maybe you should change your mind about either the jockey instructions, or the race. So with that, let’s take a look at the best tracks for your horse to race, with our Top 5 Best/Worst for each of the 5 types of horses. Listed is the Track Name, Track number, and percentage of wins from the specific type of horse (Note that tracks 83-88 are omitted from these lists).

1. South Korea (Track 81) – 39.74%

2. North Korea (Track 80) – 39.18%

3. Ostend (Track 36) – 39.17%

4. Portugal (Track 64) – 39.13%

5. Taiwan (Track 68) – 39.10%

1. Kentucky (Track 3) – 32.41%

2. Florida (Track 2) – 33.17%

3. Upper New York (Track 78) – 33.50%

4. California (Track 1) – 33.51%

5. Indiana (Track 20) – 33.61%

1. Missouri (Track 24) – 24.66%

2. Malaysia (Track 56) – 24.56%

3. Montana (Track 72) – 24.48%

4. Mississippi (Track 48) – 24.43%

5. Utah (Track 27) – 24.39%

1. Ascot (Track 33) – 20.15%

2. Ostend (Track 36) – 21.06%

3. South Carolina (Track 70) – 21.18%

4. Spain (Track 57) – 21.37%

5. California (Track 1) – 21.41%

5. Vermont (Track 37) – 21.41%

1. Oklahoma (Track 31) – 25.18%

2. West Virginia (Track 13) – 25.07%

3. Tennessee (Track 74) – 25.03%

4. Upper New York (Track 78) – 24.87%

5. Louisiana (Track 5) – 24.76%

1. Sweden (Track 52) – 21.77%

2. Italy (Track 63) – 21.87%

3. Epsom (Track 34) – 21.97%

4. Newmarket (Track 49) – 22.18%

5. Malaysia (Track 56) – 22.39%

1. California (Track 1) – 14.70%

2. Vermont (Track 37) – 14.39%

3. Kentucky (Track 3) – 13.81%

4. Colorado (Track 71) – 13.72%

5. Illinois (Track 7) – 13.41%

1. Montana (Track 72) – 10.42%

2. New Mexico (Track 14) – 10.69%

3. Idaho (Track 30) – 10.73%

4. Taiwan (Track 68) – 10.79%

5. British Columbia (Track 69) – 10.83%

1. Japan (Track 50) – 7.15%

2. Vermont (Track 37) – 7.07%

3. New York (Track 4) – 7.01%

4. Florida (Track 2) – 6.76%

5. Sweden (Track 52) – 6.59%

1. Montana (Track 72) – 3.56%

2. Portugal (Track 64) – 3.94%

3. North Korea (Track 80) – 4.08%

4. Malaysia (Track 56) – 4.18%

5. Mexico (Track 41) – 4.28%

There are tracks that don’t show signs of track bias, or show conflicting signs. For example, Ostend is the third best track for front-runners (39.17%), but the third worst for Stalkers (21.06%). In other words, once your horse falls to .6 lengths back, it’s chances fell from about as high as they can be to about as low as the can be. However, 39 of the 88 tracks have no horse classification winning percentages that fall with one standard deviation (.2%) of the mean. In other words, 39 of the 88 tracks are statistically different from average for all horse styles. It would seem that these tracks would have a specific tendency that horses running over the course could exploit. Since I found the standard deviation, it then made sense to calculate the 95% Confidence Intervals for each type of horse. Two tracks show absolutely no sign of track bias when compared to the sim average. They are Arizona (Track 15) and Longchamp (Track 35). The 5 types of horses all fall into the 95% Confidence Interval for the average at these tracks, meaning that they are statistically equal to average.

Click here for a table that breaks down the percentages for all tracks.

So what does this all mean? Is there a track bias? Well, looking at the data, it certainly looks like it. Some tracks clearly have above-average win percentages for front-runners, whereas others have above-average win percentages for closers. But there may be something else at work. It’s possible that turf courses play a more speed-favoring role, whereas dirt courses benefit closers. This may be sim wide. The next step is to look at the breakdown of victories for each surface, and for the three types of distances. This analysis will be posted next week, in Part 2 of the Speed Study Series Presents: Track Bias.

Since you’ve gotten this far, it’s obvious that you care about track bias in the sim. As it turns out, there is a near-infinite amount of ways to collect the data. You could look at the bias within different classes (claimers vs. Graded Stakes). You could do off-track vs. firm/fast. It can be broken down into different distances. There is an endless combination of variables to use; I simply don’t have the time. So here is my gift to you, for reading the Speed Study Series thus far: I am releasing the code for the bias program. It is written in Visual Basic 6, so if you don’t know VB, you may still want it (and you can try to convert it), but I can’t promise that it will work “out-of-the-box” with any other language (in fact, I’m fairly confident that it won’t). The code is not the prettiest code in the world, and the way it is written now, requires a large chunk of RAM to properly run (this can be fixed, however). The code is currently set up to handle just about anything you can ask of it; it currently imports class, age, gender, field size, track condition, race condition, owner, color of the horse, and odds (as well as many other things). Furthermore, it also imports post position. So if someone out there is looking to do a study on post position, this program is equally usable. I will offer my support if people need, as it is possible that I am the only one who understands my code (my commenting is not my greatest asset). Please e-mail me if you would like this code.

And that’s not all…Tune in next week for another gift.

Bryan (shagfarms) can be reached at TripleCrown@suscom.net, as can the Track Bias code. He can also be reached through the BTB Message Board, the Main Message Board, the Chat Room, or by Private Message.