A lot of discussion of Allen's DQ with a reaction time of 0.099 seconds. Here's a bit of background. On average, athletes' reaction times are fastest for the 100m and high hurdles. There is no difference between the two. Cont...
I am a stats guy (former data scientist, doing a phd using big data and machine learning etc) and I will tell you that you need to compare more years than just Tokyo and this year to find anything interesting.
The equipment is all different from last month. The blocks are made by Seiko, and would be the same ones from Doha. The blocks from Tokyo were made by Omega. The blocks from NCAA are Gill with a Lynx sensor on it, while the blocks from USA's are made by UCS with the same lynx sensor on it.
No idea the standards used for calibration but it's possible there is a difference between each of these setups.
Imagine thinking that they just use equipment that in reality doesn't work and give bogus results. Why does some people allways resort to wild explanations rather than just believe the most likely explanation. Makes me hungry for burritos for some reason.
Here I have plotted the distribution of reaction times for the Women's 100m (first set of results I clicked on) in all Tokyo/Eugene heats/semis/finals (so some athletes are counted multiple times). We see that the Eugene times are more broadly distributed than the Tokyo times, maybe this would indicate some sort of measurement error?
I've also plotted the rank vs the measurements. We see that there is a uniform decrease for the middle ranks of the reaction times. This might indicate some sort of measurement error, but we can't be certain.
I will pool together each gender's 100, 200, and 110h reaction times (I think these should all draw from similar distributions) and see if this trend holds. Rescaling by the mean values is probably a good way to see if we are in fact drawing from different distributions.
If that's the case then someone needs to conduct a comprehensive study across different events, controlling for manufacturer, venue, calibration standard, etc. If there really are substantial differences that is a real problem when the decision to DQ comes down to .001.
There is definitely a difference in the setup that Eugene is using right now.
And why didn't they use this setup for at least one meet previously to make sure they got everything working properly? Using new equipment that you haven't used before for the first time at a world championship is a really dumb idea.
Women's 100h hasn't happened yet, but herein lies all the data plotted together and the men's 110 h distribution + rank plot. Very clear that there is a uniform difference in the reaction time measurements in Eugene vs Tokyo.
This Imgur post has each of the plots I've made so far.
you can find results and reaction times at the past world champs here. gotta go to work, but this is really basic/easy data collection. i looked at the previous 4 WCs and Eugene is a massive outlier, to the point there is very obviously a timing error that officials HAVE TO BE AWARE OF. They are trying to sweep this under the rug. They have known since DAY 1!
This has literally become a conspiracy to commit fraud. Heads have to roll because of this, at the very top.
By the way, don't just look at the average reaction times from past WCs. Also look at the times for people who false started and you'll see an even distribution of times from about -0.20 to 0.099 declared as false starts. Then in Eugene the only false starts are all within a few thousandths of one another: .092, .095, .096, .099. Compare that to -0.050, 0.053, 0.071 for instance. So all the false starts are clustered right on top of one another in Eugene? Odds of that happening are close to 0.
Sure, you can run a T test in Excel, but that's not the point. The point is that real stats guys do everything, even the first cut through the data, in a stats package. So if someone's using Excel, she isn't a real stats guy.
PS Just to be clear: I'm not saying the people doing it in Excel are doing it wrong, or getting the wrong answer. I'm just saying the probability they are a stats guy is vanishingly small.
Do you really think excel can't run a T test...? What super advanced statistical test do you think is so much more legitimate?
Sure, you can run a T test in Excel, but that's not the point. The point is that real stats guys do everything, even the first cut through the data, in a stats package. So if someone's using Excel, she isn't a real stats guy.
PS Just to be clear: I'm not saying the people doing it in Excel are doing it wrong, or getting the wrong answer. I'm just saying the probability they are a stats guy is vanishingly small.
More like I don't want to use my company software for a hobby? If I can do all the relevant things in excel (all of us easily can) then I will.
The equipment is all different from last month. The blocks are made by Seiko, and would be the same ones from Doha. The blocks from Tokyo were made by Omega. The blocks from NCAA are Gill with a Lynx sensor on it, while the blocks from USA's are made by UCS with the same lynx sensor on it.
No idea the standards used for calibration but it's possible there is a difference between each of these setups.
This is great info and makes a big difference I think. We can probably disregard the differences between Doha/London and Tokyo in that case.
So if we only look at the last 4 world champs using the same system
Beijing - 149.6ms
London - 154.9ms
Doha - 154.4ms
Eugene - 133.2ms
The difference between Beijing and London/Doha isn't quite significant from a statistical standpoint. T-test says 5.6% chance that difference is random. That's fine
The big one of course is Eugene. The chance that the Eugene RTs are randomly different (to the average of the other 3) to that degree is 1.002*10^-15.