Collegiate cross country computer rankings and simulations (LACCTiC)

Thanks for posting your model and giving this a go but your results are actually worse than a person could predict what will happen this fall. Just because you built a statistical model doesn’t mean it adds real world value. What am I missing here?

iowakidscanrun wrote:

Thanks for posting your model and giving this a go but your results are actually worse than a person could predict what will happen this fall. Just because you built a statistical model doesn’t mean it adds real world value. What am I missing here?

The model outperformed coaches rankings the past 3 years by a number of metrics. I agree the defaults will be inaccurate since we don't have good information about who is competing right now, but I suspect the accuracy will increase when we get some real information. And if you edit who is in the race, you will get more accurate results.

And I don't think a person would want to go through and try to estimate the ability of 50,000 runners.

You don't have to like the site.

Why bother posting rankings for the JV league? DI is the only one anyone should care about.

Right of Center wrote:

Why bother posting rankings for the JV league? DI is the only one anyone should care about.

I know you're joking, but one of the goals of the project was to give some attention/rankings/analysis to every runner in college cross country. Tons of rankings are given for the top 5-10 teams and runners, but nothing for the middle of the road. I thought it might be a nice thing for the sport.

How often will the ratings update? Is it daily, weekly, etc.? I checked some of the teams from the Penn State meet today and it had not changed

Why are some people missing? Wesley Kiptoo doesn't show up when I try to search him by name or when I go to ISU as a team. He is a true senior this year.

Overall, really nice. I appreciate that you can look at rankings for all of DI, DII, etc. down to the last place team. I enjoyed looking at the NCAA list to see which non-DI teams might be able to mix it up. Is there an all college list that includes NCAA + NJCAA + other?

asdasdasd wrote:

How often will the ratings update? Is it daily, weekly, etc.? I checked some of the teams from the Penn State meet today and it had not changed

Why are some people missing? Wesley Kiptoo doesn't show up when I try to search him by name or when I go to ISU as a team. He is a true senior this year.

Overall, really nice. I appreciate that you can look at rankings for all of DI, DII, etc. down to the last place team. I enjoyed looking at the NCAA list to see which non-DI teams might be able to mix it up. Is there an all college list that includes NCAA + NJCAA + other?

I'm planning on updating this about once per week but I can run the code whenever.

Wesley Kiptoo is an interesting case that I haven't figured out yet. He appears here as "Wesley Banguria":

I looked into it and found that his name in a lot of tfrrs results is Wesley Banguria, but is linked to the same tfrrs profile. Do you know why this is? If this is for sure the same person, I can manually edit his last name. As for his team affiliation, that should update automatically when he competes for ISU.

You should be able to search the leagues "Collegiate," "Men," and "Women"

For men and women combined:

For all men:

for all women:

This is the same person. This is exactly my point. Your model has crappy input data so is producing garbage results which is also the same as the coaches poll, also based on bad data. You’re saying it will get better as the season goes on but how is it better than me simply following the results of teams and meets?

Nothing is going to be perfect, and honestly the model looked pretty solid on 2019 results. It’s not going to be perfect, chill. It’s also a work in progress obviously, the man is asking for input. I personally am in love with it, it’s been too long for this not to exist. Felt pretty accurate for my race last week, the only one that was really bad that I saw was a race that didn’t have the correct distance and was over 50% freshman from a conference that didn’t race last year.

I agree, it provides another way to try to compare teams. I would not have thought Georgetown vs Michigan vs Minnesota might be that close, but if you look at this you can see it. Once can see why some think BYU and NC State are clearly the top women's teams. Like any model, the more data the better. The COVID tweaks are an unfortunate need.

Just got the LACCTiC rating from the race I had yesterday, and I want to say I’m impressed. The course we ran on was well over a minute short and even though I set a massive “PR” the rating picked up on it and gave me a 5k time slower than my theoretical 5k split in the “8k”. The effort corresponded to the time quite well.

Passing the eye test so far 👍

Thanks for the more positive feedback. I'm hoping to be able to improve the model over time based on input from others.

I've updated the site with this weekend's results.

A big challenge you will have with accuracy is figuring out how to discount races where some or all of the runners frankly weren’t racing very hard. For example, I’ve heard that a number of these early season races were run as tempo runs and we’re only raced, if at all, at the very end. Other times the top racers may simply sit tactically on a much slower pace than they are capable of and only race the end for the win.

As just one example, look at the winner’s lacctic scores from the BYU meet on Saturday. The lacctic scores are way below what everyone knows they can do, but you’ve weighted them heavily because they are from this year. Recent doesn’t necessarily mean accurate for your purposes.

One way to somewhat account for this is to toss out low individual lactic scores and only keep the highest marks achieved when the scores would suggest it was indeed an all out effort. Maybe average the top 3, weighing the most recent top marks a little higher.

I just like being able to see how good of an xc race my old HS opponents did

The weighting is not only based on recency. Within a season, later races as well as better races are weighed more. The bottom 1/8th of races actually receive around 1/2 the weight of the top 1/8th. This is all based on historical data - how well various types of races predict end of season performance/the next track season's PR.

The cumulative average of a previous season ends up having about the same predicting power as a single good race in the season. The problem is that in this season we only have one race right now. I'm not fully convinced that with only one race, we should assume any bad races are just tempos. What if the runner showed up out of shape? When I remove any performance below 2 standard deviations of their average, historical prediction accuracy decreases.

I don't disagree with the fact that some of the runners have ratings that are worse than they should be. I'm just trying to explain why the solution isn't as simple as it might seem. I'm definitely open to ideas - I'm just not going to change the formula until I have hard evidence from the data.

Washed-up Grad Student wrote:

Tfrrs is kind of a mess of a website.

For the first 7 years the front page of tffrs was HTML tag soup. Multiple body elements (ues, that bad). The Tufts boys like to think highly of themselves but theyre stale generation x just like milesplit, runnerspace, letsrun and athletic.net owners were/are. Dont trust their computational capacity, implementations and database skills. They are one dimensional thinkers and if there was any serious stake in distance running (i.e. it wasnt a charity) we would be having this discussion elsewhere.

Just fyi Weini Kelati has gone pro, so she should not be included on New Mexico team. Ditto now for Cole Hocker of Oregon.

Krissy Gear is listed twice for Arkansas

Nice work, I'll be keeping an eye one it. One likely miss to point out and one suggestion\thought.

Women's individual D1 has no runners from OK State. Injuries not withstanding, there will probably be 4 in the top 100, 2 or 3 of those in top 25.

I'm sure coding would be a nightmare, but, allow visitors to the site "vote" what position they believe a team\individual should be. Given enough honest voting and weighting, it should help steer the rankings. Kind of like if you have 1000 random people guess the weight of a given object, the average of the guesses is usually quite accurate.

Kansas wrote:

Nice work, I'll be keeping an eye one it. One likely miss to point out and one suggestion\thought.

Women's individual D1 has no runners from OK State. Injuries not withstanding, there will probably be 4 in the top 100, 2 or 3 of those in top 25.

I'm sure coding would be a nightmare, but, allow visitors to the site "vote" what position they believe a team\individual should be. Given enough honest voting and weighting, it should help steer the rankings. Kind of like if you have 1000 random people guess the weight of a given object, the average of the guesses is usually quite accurate.

OSU is there The ratings are all pretty slow - looking at Roe maybe this calls for a tweak of the methodology as she has a number of pretty fast 6K times - and some very slow times when she most likely was not pushing hard at all. With 4 sub 16 min 5K equivalents (including 15:30 for 2nd place in March) she still has an average rating of 16:26.