Seed times and results: how do we do it?

As we head into the new year, there is plenty to reflect on. There have been many changes this year, both in our group, and in Quebec athletics. Some of those changes have been linked, in that as race director for the McGill meets, it’s my job to make sure all the athletes in the province (and from elsewhere) have the opportunity to perform at their best when we host them.

There are many aspects of putting on a meet to which I had not given much thought prior to having to organize one. On the other hand, there are some issues that have always come up, whether I was participating as an athlete or a coach. One of these issues, to which I’ve given a fair bit of thought, is the issue of seed times.

Prior to the meet, a coach emailed the federation (another issue: what is the FQA’s role in all this meant to be? The answer is for another time…), and they forwarded the question to me. He thought the seed times in the women’s 1000m were laughable. He proceded to analyse, thoughtfully, some of the entrants, and he made some good points. Many of the athletes had not run the times they (or their coaches) purported they would run on the 15th. This made me think, what is the point of a seed time?

A seed time is meant to organize the heats so that athletes of like ability get to compete together, and have the best chance to run well. It doesn’t do an athlete any good to be in a heat with people who are much stronger than she, nor does it do good, usually, to set an athlete up to win by a large margin (for a younger or new athlete, this might not be a bad idea, but that’s a different story).

The problem arises when we don’t have an accurate prediction of what the performance will be. If a seed time is too old, the athlete may be much faster or slower. If one coach enters a “real” seed time of 3:00 for example, while another estimates a performance of 3:01, putting the two in the same heat may not produce the desired effect. The 3:00 seed time from the summer may represent an old level of fitness. That athlete can run 2:55 now. The 3:01 may be merely aspirational, and the athlete is more likely to run 3:05. So now instead of a 1sec difference (good for competition), we have a 10sec difference (two runners running completely unrelated races).

The solution is simple: all coaches should use the same method. Good luck getting coaches to agree on anything, and good luck getting coaches to tell the truth. This is not a knock on coaches. What they are doing is playing the same game everyone else is, to try to get the athlete they coach into the section they think is best for the athlete.

What we are hoping is that with the new registration system from TrackieReg, past performances will be automatically linked to athletes, so that coaches don’t have a choice in entering a time. This is how it works for national championship meets. Actually, at nationals, each entry is checked. In order to save the meet director (or a minion) from having to check everything, automating the system would be the ideal solution.

That said, for meets at the beginning of the season, it is difficult to rely on performances from months ago for an accurate seed time. A seed from the summer is probably no longer relevant. A coach’s estimation is probably the best we can do at that moment, and it might be more fair. But then, athletes who have competed already will have their seed times be more realistic, while others may be able to rely on the “optimistic” predictions of their coach.

The early season options are:

1) use 2012 seed times that probably don’t reflect the athlete’s current abilities. I don’t think this is fair to athletes who may have worked very hard in the off-season and want to benefit from their recently gained fitness. The same is true of an athlete who was injured, or just out of shape: why should they benefit (or in fact, be harmed) from an out-dated seed time?

2) use translated times from other, more recently contested events. This doesn’t always work, but it is not a bad solution. I suggest rather than Mercier, however. I find McMillan to be far more accurate. (Daniel Mercier gets lionized because he is a local guy who made some big advances in sports science, but most of his ideas that people cite are out of date. It could be that his more recent work is relevant, but when people send me training plans that they are doing that are from him, I am not surprised they are not running well.)

3) Seed all athletes randomly. I can’t see how this is a good idea in distance events. It can work for short sprints where there will be rounds or a final.

4) Seed athletes without a performance as NT. I don’t think this works because it disadvantages athletes without a performance, and likely puts them in a section where they don’t belong. That also hurts the athletes in that section.

5) Allow coaches to estimate the performances of athletes for the first couple meets of the season. This might be fine for indoor season. The stakes are very low. All racing here (except at the CIS level) should be purely preparatory. If coaches want to be unrealistic, that is their business, and their athletes will suffer. I think a system of “public shaming” (so to speak) where an analysis of seed times vs real times are laid out would eventually force coaches to be real about it. I’ve done that below with the women’s 1000m. Surprise, surprise, coaches were unrealistic.

When it comes to outdoor season, we should require real seed times, and using indoor times is not unreasonable, as they are part of the same training cycle. Hopefully by the spring we’ll have the database ready to go, and this is what will happen.

So, we spoke of the women’s 1000m. Here is an analysis of the seed times and the final results:

indiv by diff

What we notice is that out of 39 people, only 7 ran faster than their seed time. Another 3 were within 3 seconds, and another 10 were within 6 (the average difference between seed and real time was 5 seconds). So that means about half the entrants had seed times that were, to put it one way, bullshit.

We can also look at the clubs individually to see which clubs were the best at estimating athletes’ performances:

Club avgs

A couple clubs were not bad at it, but most clubs were pretty bad. Vaudreuil-Dorion was probably the worst offender as they were an average of almost 10sec off, and they had a solid 5 athletes, more than any other club. Lanaudiere had 4 athletes, and averaged faster than their predicted seed times. So clearly some coaches are fudging it more than others.

There are other ways we can look at these numbers, such as difference in placing (since, if everyone lies the same amount, then there won’t be much difference in place, and that’s what counts). The results here are actually not as bad. The big deviations come from clubs who only have one entry, so there’s not really enough info to report. With the clubs that had a few, Phénix managed to outperform the field by an average of 6 places (with 2 athletes). Vaudreuil was again one of the worst, with each athlete losing almost 3 places from their original seed. Depending on where those places are, that could be the difference between heats, which means these mistakes might be preventing athletes from getting in the best heat for them, and that includes the athletes from the club who is doing it wrong.

The full results with original seed times are here: McGill Results December 15, 2012. You can go through all the events and see where you club stands (you have to do the math yourself!).

I think this is a good exercise, as we seldom go back over results, other than to look at how fast we ran. It might be good for coaches to do an analysis on their seed times, and question whether or not they are doing the best job they can for their athletes in that area. Some, clearly, are not.