“Cross. . . country. . . time is relative” – Albert Einstein
These rankings are calculated using the following algorithm:
- Gather all of the meets which the teams in the region competed in. For each pair of meets, use the individuals who ran in both meets to evaluate the difference in their speed.
- Use least-squares to get an overall adjustment score for each meet.
- Gather the teams from last year’s regional results and form their top 7 from their most recent results. SOME TEAMS DID NOT RUN THEIR TOP 7 IN THE MOST RECENT MEET. IF THIS IS THE CASE FOR YOUR TEAM PLEASE EMAIL ME AND I WILL FIX IT. (bmazaher at caltech dot edu)
- For each member of a team’s top 7, adjust all of their performances according to the meet’s score.
- Average each runner’s performances over the season. More weight is given to recent performances. More weight is also given to performances whose adjustment is more certain.
- Rank all of the individuals in the meet and score the simulated results.
Interpret the results in the following way:
[Place] [Time] [Time behind leader] [Uncertainty of time] [Name] [School]
The uncertainty of the times is in seconds. That is, someone with a “time behind leader” of 30 and an uncertainty of 10 will likely run between 20 and 40 seconds behind the winner.
DI National Rankings
Lindsey Rudden -> India Johnson (Michigan State)
Adoette Vaughan -> Lauren Gregory (Arkansas)
Paxton Smith -> Alec Hornecker (Colorado)
Added noise to bootstrapping simulations for better probabilities
Removed any meet with the word “alumni” in the title
More 11/21 changes:
Fahy, Kristin -> O’Keeffe, Fiona (Stanford)
Sargent, Sadie -> Orton, Whittni (BYU)
Brown, Reed -> Hocker, Cole (Oregon)
DIII National Rankings
National Race Simulations:
The first link gives my normal format. The second link simulates the meets 5,000 times using the distributions of each runners performances (more weight towards recent ones).
Vaporflys at Regionals
DIII Regional Rankings
Over 50% of runners finished within one standard deviation of their expected place and around 80% finished within two standard deviations. A perfect model would have 68% and 95% respectively, in accordance with the behavior of a normal curve. (See the detailed breakdown here)
Note that the “times” here are meaningless – only the time behind the leader is relevant. They are normalized to be the time that someone would run on the fastest course that anyone ran in that region. These are usually obscure meets with short courses. For the women I normalized times to the median course so they would at least look like 6k times.