Collegiate cross country computer rankings and simulations (LACCTiC)

How long is it before someone who is not racing will be removed from the database? There are a couple of people on my team who have graduated but still have eligibility due to COVID, so they appear on the database.

Craig's Mullet wrote:

How long is it before someone who is not racing will be removed from the database? There are a couple of people on my team who have graduated but still have eligibility due to COVID, so they appear on the database.

I think the end of next weekend is reasonable? A lot of top D1 runners still haven't raced.

Is there a way to see what the TiC rating is for a specific race/ result? Or will it only show the season/ current TiC?

Craig's Mullet wrote:

Is there a way to see what the TiC rating is for a specific race/ result? Or will it only show the season/ current TiC?

I purposefully avoided showing entire races results to honor TFRRS's questionable right to be the only results provider. One thing I can do is create a calculator that converts between races (ie put a time in for one race and get the equivalent for another). Another thing I could do is just have a table of races and the times needed to get a 20 minute or 15 minute TiC. Do either of those sound like what you are interest in?

over-the-hill wrote:

Just wanted to give you a big thumbs up for this. I have done similar analysis, albeit on a much, much simpler and smaller basis for my wife who is a high school coach. I found that normalizing the data really helped the kids gauge their improvement better. Too often kids were getting 'PRs' on courses that were obviously fast or upset for running slow on tough courses.

This is great stuff. If I were to make one suggestion it would be to have the ability to plot data graphically. It's an easier way for an athlete / coach to see trends in the data.

I see that you have now moved the runner from the BigTen women's team to the correct team outside the BigTen that she now runs for as a grad student. I assume you moved her after seeing results for her on her new team. Thank you. But even after that update 4 of the top 9 you listed did not complete this season for the team thus far. I know two of those graduated and no longer attend the university. One is a grad student still there but I am not sure if she is still on the team or is healthy. Two others ran well early in the 2019 cross country season but have not competed much since then I would guess due to injuries. If you look at results this season thus far you should be able to start using this season times rather than prior 5K times and should be able to remove athletes that are no longer on the team or out for the season with injury. Some schools are much quicker than others at updating their roster webpages.

Both Michigan and Michigan State's first meet is not run by the full team but now that one or both have run their 2nd meet you can probably use the results to figure out who is likely to make it onto the top 10. There could be a few exceptions for some who are may be held out for minor injuries but most of the top runners should have run now. You have one listed who has not competed much since the early part of the 2019 XC season. You have one who is now a grad student still at the same school but she has not run this season so I am not sure she is still on the team or is healthy. If runners are not racing within the first few meets then they may not be on the team or they may not be healthy and if they do return they may not be as fast as their prior season best 5K time. I don't follow the teams as closely now so I don't know the details of all the runners and don't want to put names out the boards here. The upcoming races will provide much better data than prior season performances.

Washed-up Grad Student wrote:

Craig's Mullet wrote:

Is there a way to see what the TiC rating is for a specific race/ result? Or will it only show the season/ current TiC?

I purposefully avoided showing entire races results to honor TFRRS's questionable right to be the only results provider. One thing I can do is create a calculator that converts between races (ie put a time in for one race and get the equivalent for another). Another thing I could do is just have a table of races and the times needed to get a 20 minute or 15 minute TiC. Do either of those sound like what you are interest in?

The first thing you mentioned sounds quite interesting.

hmmmmmmmmm wrote:

I don't think everyone was out for a tempo that day. It seems like the calculator is something to re-visit around mid-October once more meets have been ran. Right now there are too many discrepancies as it slowly gathers data to correct itself throughout the year. Neat little idea though!

I think I've found a way to handle this. I now no longer look at performances that I classify as "tempos" when adjusting courses. The times for what you put have adjusted to be a bit slower (and more in line with the potential dragonfly distance). Still not perfect, but I think it's better.

Regarding eligibility of runners: I'm not going to respond to vague complaints about eligibility status if the person making those complaints is unsure themselves. Delete the runners in a simulation if you care that much. I'm not going to monitor by hand the activity of almost 60,000 runners -- I'll implement the best heuristics I can (which will involve removing anyone who has not raced by the end of this weekend).

I would suggest you wait until next weekend as I can right off the top think of 5 top runners who have not raced yet, but likely will October 1.

Figured I'd revive this thread since I'm now feeling somewhat good about the weighting system and the team rosters

Oregon's got a 13:12 guy who's rated at 13:30.

xczvzxcv wrote:

Oregon's got a 13:12 guy who's rated at 13:30.

And yet, last weekend he finished within 1 second of Charles Hicks, who is a 13:30 guy.

A track 5k is the "speed rating" standard I'm using to make things relatable - its not predicting what Cooper would do in a 5k on the track. Cooper is better at track than cross and that is represented in the rating.

If you simulate nationals, Cooper's finishes have a large range because of uncertainty in his ability. Lacctic says he has a 10% chance of winning, which seems about right because any 13:12 guy is going to have a shot at winning. But you cant go ranking him #1 because of a track PR.

Final point: Being a 13:12 guy doesn't mean that if you go run a 5k next weekend, we would expect a 13:12. It means we would expect your best race to be 13:12.

While no rankings are perfect, this is a nice tool to give indications of teams or individuals that have not raced, nor really raced any common opponents. It seems like some runner's ranking are based on very recent results, and some on results spread out over several years, which seems a little inconsistent.

@WUGS: What are the odds of you publishing week-by-week full team national rankings? Many teams outside the top 50 would love to see where they rank week over week and year over year. It's a great way to track progression (or the lack thereof). This may be the best tool we have to replicate an RPI that other sports use.

Thanks!

FastTuohy wrote:

While no rankings are perfect, this is a nice tool to give indications of teams or individuals that have not raced, nor really raced any common opponents. It seems like some runner's ranking are based on very recent results, and some on results spread out over several years, which seems a little inconsistent.

There are two filters that I use to remove "tempo" races for more accurate rankings:

1) An overall filter. This removes performances that are slower than the best performances (or gives them less weight depending on how much of an outlier they are). This filter is only used if a weighted average of how recent the meets are is less than 1 year ago. If all your good races are in 2018, the filter will have placed all the weight there and this average will be 2 or 3 years. In cases like this, the algorithm will not focus on 2021 but instead use a more uniform weighting to represent that you are currently less fit, but could quickly get a lot better.

2) A season-based filter. This removes performances within a season that are slower than the best performances in that season. To avoid one randomly noisy TiC from skewing the results, this filter only activates if there are 2 or more "non-tempo" races.

This is the best system I could come up with that gives reasonable results. In future years I may just never consider performances that are more than 2 years old, but COVID created some unique challenges this year.

Regarding weekly rankings, that is definitely something I'd consider doing. I need to think about what "field" to score the slower teams in, since the results will vary depending on the field. One option would be a "non-qualifiers" ranking - a race with teams # 33-64. You can also already search for leagues (or click on leagues on the top right of a team pages) to get rankings within smaller groups.

Thanks for putting in the work for such a useful site. If you are open to ideas, what do you think about enabling simulations based on other data fields? I realize it would all be dependent in what data you can get, of course. I have no idea what data you can easily get or incorporate by matching from different databases.

Examples:

Year (fr, soph, etc.)

HS graduation year (e.g. how is the class of 2020 doing?

Etc.

There used to be a ranking system that would do this weekly that would run a simulation where each team dueled every team, one-by-one and each team's total win-loss record was used to rank everyone.

That or could we just run all 350+ teams in one giant race and see who wins?

Don’t know how feasible this is, but it would be great if we could see end-of-season LACCTiC for previous years (especially as time goes on). Was looking through 2019 NCAA results and noticed how LACCTiC (2021) vs TiC can be pretty skewed for people who improved or dropped off significantly.

an idea wrote:

Don’t know how feasible this is, but it would be great if we could see end-of-season LACCTiC for previous years (especially as time goes on). Was looking through 2019 NCAA results and noticed how LACCTiC (2021) vs TiC can be pretty skewed for people who improved or dropped off significantly.

Do you mean to freeze LACCTiC ratings at the time of the race? As time progresses, the weights change and obviously new performances come in. In the results from the 2019 season, the "rating" is their current rating, not their rating at the time of the race. For more recent races, I have "frozen" the order in which the ratings would have expected them to finish and use that to give the red and green arrows for scores and places.

TiC in 2019 is the equivalent for a 5k in the 2019 season. The adjustment for 2019 TiC to 2021 TiC is the same for everyone and gives the average improvement. Theres a big jump from 2020 to 2021 potentially because of shoe tech.

Any change you've looked into doing a full national ranking? Would still love to get that data, even if it's not 100% perfect