My thoughts and data analysis of USAT scores and rankings
Since my foray into triathlon is still relatively new, my historical database is relatively small as it seems quite a bit of the data generated is either destroyed or hidden from view at the end of each year. Even simple race data from quite a few of the local triathlons are refreshed and written over each year. USAT does the same thing. During the year, you can see individual race data, but sometime during the beginning of the year (this year it was in the April or March timeframe), the individual race results are removed from public view and replaced with only the current year’s data. So, unfortunately, I only have this year’s data with which to base the theories below.
For the guys/girls racing in the US, we get scores each race and then at the end of the year, your highest 3 scores are averaged to give each person a score that is then used to rank you within your age group nationally. Theoretically the way this scoring system works, they are able to compare one individual to another even when they have never raced the same races. You can see the challenge with coming up with such a ranking system I’m sure.
For a little background, here’s how each race is scored according to USAT: http://www.usatriathlon.org/rankings/rankings-criteria.aspx
I took the data from each race that I participated in this year as well the races I plan to participate in next year. I took a portion of the top participants (usually those that had scores above 70 for that race or the top 100 participants, whichever was smaller) and graphed their USAT score in that race vs. their overall race time. See the chart below for an example.
According to the way the data is calculated with the par time, I believe a straight line should fit to the data. It doesn’t, but I’ve not been too concerned over that. In each instance, I was able to fit a 2nd degree polynomial trendline to the data with an R2 > 0.99, so it was a very good fit. Each race, of course, had a different value for their trendline, but using the equation for the trendline, I can now predict the scores (for that race on that day) based on a given time or vice versa. Later, I’ll explain why I’ve recorded this and what value it is to me.
When I’ve looked back over my past year’s worth of data, the scores seem to be a bit confusing to me at times. There are times when I’ve raced well and received a mediocre score and other times when I thought I didn’t race well and received a great score. Of course, there will be some variation in the scores naturally, but the ballpark should be the same. However, after doing some digging and looking at quite a few races, I noticed quite a few trends (again, for complete disclosure let me state that this is only based on one year’s worth of data).
For one, it seemed certain races are scored higher than others. For example Triple T, Vegas, Kona, and several of the 70.3 races. I’m sure there are more, but these are some of the races I’ve looked into and noticed this. Theoretically, you’d think that Kona and Vegas, because they are the world championships for the 140.6 and 70.3 distances respectively, that they would have a higher average scores, so that explains that. Well… not quite. Of the population of people that I looked at, nearly all of them had their Vegas (or Kona) score as one of their top 3. Again, maybe everyone peaks for this race, so it should be their best race. Possibly, but the probability of that happening on such a grand scale is not likely.
Same thing for Triple T. Most everyone that I looked up who was involved in (and completed) the Triple T series of races, had 2 of the scores from that series used in their top 3 scores. For example, Scott Iott and Adam Zucco, who are the guys from Training bible podcasts (and also happen to be in my age group) both raced many races this year. However, if you look at their scores, their top 3 happen to be 2 scores from Triple T and their score from Kansas 70.3. They did quite a few other races this year and not always the same races. One would think that perhaps another race would be in the mix for them, but those were their top 3 ranked races. Maybe they peaked for those and didn’t peak well for other races, but it seems implausible that only 1 guy of the top 20 from the first race at Triple T didn’t use any of the scores from that weekend in his top 3. They all had one of their best races of the year all on that weekend or at least the data would lead you to believe that. See the chart below for details. It lists the top 20 finishers from the first race at the 2012 Triple T. Beside their name is either a number which signifies how many of the Triple T scores were in their top 3 for the year or an X which signifies they didn’t complete enough races to be ranked this year (i.e. they only did the Triple T).
Here’s another way I’ve proven to myself that the ranking system seems to be slanted toward particular races. Let’s take Andy Potts as an example. He’s ranked #1 in the 35-39 Age group. His average score is 116.5. His lowest score of the year is 109.2. So, you’d think that (if the scoring system were truly level), if he were to race my local races, he would score at least his lowest score of the year. Let’s pick the Markey Race for Women’s Cancer (I could really pick any of them, but for this example, that’s what I’ll use b/c the graphical data and equation are shown in the first chart above). Here’s a description of the course for anyone interested: http://www.markeyraceforwomenscancer.com/course/
For details on why I’m using the times I’m using, T1 is long, the bike course is 12.9 miles with about 30-35 ft/mile elevation gain, and the run is a 5K with about 30 ft/mile (and part of that on a very uneven grass surface).
In order to score a 109.2 on this course, according to this year’s data say you’d need to finish the race in 42:50. Potts is obviously a stellar swimmer, so we’ll put him at world record pace and give him a 3:30 for the 400 meter swim. The fastest T1 this year was 58 seconds, but for sake of argument, we’ll give him a 45 second T1. If we give a 30mph bike speed (the fastest on this course is only about 24mph which includes 2 pros), that adds up to 25:48 Add in a blazing T2 of 15 seconds (once again significantly faster than anyone has ever done it over the last 3 years) and send him out on the run to do a 15 minute 5k. IF he does all those superhuman feats in one day, he would cross the line at 45:18. He still wouldn’t score close to his worse effort of the year (score-wise). In fact, he only score a 105.5 according to the best fit polynomial equation above.
So, viewing the scores in this manner lead me to believe that certain races are definitely ranked higher than others. However, I’m not sure why. It has something to do with the par scores being artificially low (or high) at certain races, but I’m not sure what causes that because the system is set up to theoretically remove the natural variation in times due to weather or any other abhorrent course condition.
It seems that the ranking system could be manipulated if you picked and chose the “right” races, whatever they may be so that you maximize your score. I plan to use the data I’ve collected this year (with the equations for the trendlines) to try to predict my scores for next year based on my time. We’ll see how it turns out and if the “highly scored” races this year turn out to be the outliers next year as well.
In the end, the USAT score is nothing more than a relative comparison that doesn’t mean a whole lot to the average age group racer. But, I still find it interesting to delve into the data and see how it behaves. Can I play games with the system? Can I score higher next year than last? I hope so, but part of that may be race selection and hopefully the rest will be fitness. Until then, we’ll see how my theories hold up. If anyone has actually read this post and has thought on it, I’d love to hear comments or opinions on this.