LACCTiC Returns

I feel like this post was only made to list your credentials

Then I would have identified myself. Leaving me out of it, I'm curious about 800/1500 guys who never race at longer distances but do well at XC; how are their ratings determined by Bijan?

If you've never raced 5k or 10k then its as if you did not race the "track PRs" race.

Ratings are not based on your 5k or 10k times (unless one of those is very good, then it makes up a small percentage). 5k and 10k times are mostly just used to translate relative scores between races to somewhat meaningful numbers.

Maybe you've addressed this? Im curious, I see for example an athlete who's Conference time is over 50% of their rating when its an easy conference and the team coasted, didn't worry about times etc. But their Cowboy Jamboree race gets 0% and this was an important race that every runner would have been going all out? Do you vary by how important/serious the race was. I know many conference races some may actually put a hard effort in, but others its obvious didn't ? Maybe too complicated to tweak so much?

Every time I tweak this, someone has a different complaint. It’s VERY difficult to measure race importance automatically (there are far to many races for me to go through by hand and score them)

Importance right now is measured by how interegional the group that raced is. Essentially, it scales with the number of people are two degrees of separation away from race participants. A local race where the same people race the same people all the time will have a small number. An inter-regional race will cover most of the country (because for each region, those runners have raced a lot of other runners in their region).

Conference races score low on this metric, so there is an additional bonus for all races that in the last few weeks of the season.

Your score is made up of your best race, your most important race, and a track race or previous season rating ( if it helps). A final tweak is that if two races are very close in importance (ie 7.7 and 7.8), the better of the two is taken for the “most important race”.

As of now, I have no way of reliably detecting whether a conference race was taken seriously or not. “Everyone taking it easy” looks exactly the same as the course running slow.

A lot of incredible work done to create the site. Well done!

Question...it does not look as though altitude times / performances are taken into consideration. Any thought to using the one's from the TFRRS qualifying performance lists for track?

Altitude should be taken into account because everyone will run slower by the same amount. The only issue I see is if many of the participants track PRs are from altitude - but I think that would be a relatively small effect since a lot of altitude runners go to sea level to PR. Also, track is just one "race" of many - so its affect on the score shouldn't be that much.

Adjusting altitude manually would be difficult - I don't have a database on the altitudes of every course.

Hi, thanks for your work on the site!

How should we understand 'in season fitness'? For example, if a runner has in season fitness of 15:45, would that be their estimated average 5k time or something else?

Also, what does the range mean?

Thanks!

Bijan,

Do you think you could make a pre-loaded nationals simulation again for this year? It was pretty cool last year and worked EXTREMELY well as a draft board for a sort of "fantasy" pick-em that we did for the XC champs

Can you show us how the algorithm performed historically?

What do you make of the team rankings given for lacctic scores compared to the extremely different results when you score a hypothetical meet using the runners individual rankings?

Ok, answers to 4 questions:

1) The range is based on the standard deviation from the performance LACCTiC used to get the runners score. The score is an average of the runners best and most important races (lots of tweaking to make this look reasonable - not sure I have a method I like best yet). Think of LACCTiC as a heat sheet, only instead of using season bests you use a few more performances.

2) I've now preloaded the nationals fields and put the links on the front page.

3) To see how the algorithm has done historically, go to any meet page (recent meets are on the front page and on team pages, or click on the links in a runner's page). The results in these meet pages are a bit off because I (stupidly) only load the time of the runners, not the place. Basically when people run the same time, the website sometimes gets the place incorrect and scores the race wrong. BUT, generally, you can see how the algorithm did at predicting the race based on the red and green arrows (the arrows show the change in place relative to what was predicted before the race was loaded). As you can see, the algorithm does very well.

4) The scores change when you incorporate individual scores because you are scoring a cross country field instead of running a relay race. In the league pages teams appear according to "pseudo-rank," which is essentially if the top 5 ran a 5x5k relay. Of course, how a team performs is very dependent on the field they race in (bad 5th man isn't terrible in a dual meet but its bad in a nationals field). Hence, the difference. It literally says this on the website if you hover over the (?) next to the words "pseudo-rank." I'm tempted to add the text in all red bolded caps all over the front page.

One more comment about this. If I have time over the next year (big if..I have a couple hundred pages of thesis to write), I'm thinking about implementing a fantasy drafting system that works on top of lacctic. It would hopefully look something like:

1) Everyone makes an account and joins a "group." For an example, let's say the group has 7 people.

2) The group picks a target race using a league or a custom entry of teams.

3) A drafting order is decided on (probably randomly).

4) In each iteration, each person in the group submits their next 7 picks in order (you need 7 picks because the first 6 picks could be taken by people ahead of you in this iteration).

5) Once everyone has submitted their picks, the website automatically drafts runners in order, picking your top available choice when you are up.

6) The website sends everyone an email to start the next round.

7) Once the race happens and the results are loaded, the group can select the race that represents the results they want and the website will score everyone's teams.

I'm also hoping to create a beta NCAA qualification system that is hopefully more transparent than KOLAS. NCAA D2 actually asked me to make this a few months ago and I dropped the ball.

If anyone has thoughts on either of these I'd love to hear about them either here or on twitter.

You could also do a fantasy draft like an auction draft, but where no runners are eliminated based on someone else’s picks.

1) set runner costs

2) set a budget for each entrant

3) pick a race and open the draft

4) each entrant can pick whomever they want within their budget.

5) score it afterwards

Keep up the great work. That's great

Hi Bijan,

Just wondering why Valby's rating does not include the Championships but all the other runners' ratings do at this point?

Thanks

That is simple. She has 2 cross country performances that were rated higher than her nationals rating. The program uses an athletes top two XC races and potentially a track time if the XC ratings are not as good as the track rating.

Any chance we could get a preloaded D2 nationals simulation? If not, no worries. Appreciate the work you have put into this.

The flaw is once the fast ratings are in they are in. Are we to believe she could have actually run 19:05-10 on the OSU course and that was a real bad day for her? Or maybe the other ratings are just too fast fro reasons I have previoulsy postulated? It seems like the AL ladies also got ratings for Nationals slower than those SEC ratings.