Evidence Strong
Q-points better than Sinclair? New weightlifting scoring method by Marianne Huebner
I have a question about the total maximum. So this is the theoretical number calculated for the heaviest athlete. Yeah, so this is a maximum calculated from the function. So it doesn't orient itself around world records. It's just the scaling. Now, it is actually interesting that you bring this up because the question will be where all the cue points for women are going to be lower than the cue points for men. Well, clearly we've had different functions for the men and the different functions for the women. And men live more than women do in general. So it cannot be directly compared. I'm Ariana. Welcome to Ariana's Strong Show. It's my pleasure to have you. And can you brief introduce yourself? Hi, Alex. I am so glad that you contacted me. Thank you so much for inviting me. My name is Ariana Gipna. And I am a professor of statistics and probability at Michigan State University. I've been competing as a master's weight lifters for about 70 years. And I also called to the students, the student weightlifting club. I became interested in weightlifting research topics where I thought there were gaps in the literature. And that then included performance development in youth, master's weightlifting age and the influence of body weight on performance of weightlifting performance. I'm very excited to have you. And today we'll be speaking about your recent paper, comparison of Olympic style weightlifting performances of elite athletes, scaling models that count for body mass. So there are weight classes in weightlifting. And there is a female division and male division and recently additional division for transgender athletes. So why would we want some models to compare weightlifting results specifically? Yeah, there's been a long history to try to compare weightlifting results across body weights because heavier athletes typically can lift more. And many nations can use these to determine the best male or female lifters and also to award best lifters in a competition or across an age category. Let me show you some picture that shows it really nicely, how it works. These performance results from 2017 to 2020 IWF qualifying competitions. These are world and continental competition results. In green, you see the pictures, the performances for the women and in blue for the men. And I drew a line across, it's the 90th percentile. So at the top, it's not the world records, but it's the elite athletes. And so we can see that from the lowest to the highest, the performance increases. Now it's very unlikely that somebody who weighs 50 kilos, all of a sudden is going to be compared to the 160 kilos. But we'd like to see who, okay, if it's somebody your way, your performance lifts much less than somebody at the high end is not necessarily a worse athlete. Can we compare those athletes? And we can also see that women lift less than men in general. There's some overlap. The thought would be, how can we compare the one lifters at the lowest weight categories with the ones at the highest? So let me show you the next picture. This jumps to illustrate it. So these are medalists, one in those very same competitions, world and continental championships. So these are the gray dots where the actual performances and just fit it with a 90th percentile curve. Now, can we scale these performances? So to turn this gray dot into a blue dot all the way across so that we should lift us at this weight category, can we compare it with lifters at that category? And I'm looking for a way to scale these points so that overall the average will be horizontal. That means there will be a fair comparison. Yeah, so this can be used in any number of ways for example in competitions to find the best lifter overall. Within body weight categories, but on a pro problem because we have that exact performance within body weight categories for the medals. But across we can only do it once it's scaled. There were attempts before to do it. Why other models or what other ways of comparing athletes were maybe less successful? So there are a number of methods that have been used in currently and in the past. The most famous ones would be the Robbie points at the Sinclair points. They are currently in use. Now what they do is they use world records, both systems. World record is the final system. So Sinclair tries to find a curve only for 10 points for women and 10 points for men for those world records. And the Robbie points is a scaling system that says what every world record gets 1000 points and everybody else is proportional to that. That's not a bad idea. I'm going to show you. So these are the actual body weights, how the athletes weight in the males at the bottom, the females at the top. So one of the things that you can see here, there are spikes. And spikes means that the athletes try to be at the upper end of the body weight category because it might make a difference in their performance. You can also see a long tail in this distribution. That's the unlimited body weight class. The red points are the actual body weights at which world records were achieved. So one of the things to point out is you can see, well we have all these work, world records and body weights that correspond to the weight classes. And then there's a big get up. And in the unlimited weight classes, the world records, you set at the upper end. The athletes who are most competitive are the heaviest in this one weight class. That makes it a little difficult for any kind of statistical methods to fit because you have so many athletes here that are not really competitive to get a world record. And the way it has been done in the past is with regression models. So you use these points and try to fit a best curve. Now the question is which points do you use? And many, many attempts have been made using the world records to find that as an attempt and fit a curve through it. The problem is we have many more left us that are not world records. And world records may have been a doping problem and that in the world records changes, it gets reset and so on. So can you really use this as a comparison? Now if I try to use, well if we try to use the Ruby points, these are the medalists, blue for males, green for females, and they are Ruby points. And then I'm trying to find a curve to see is this horizontal? So can I really compare somebody with the lower weight class with the ones with the highest weight class? And I certainly can't for the world records because they're all 1000 points. But then anybody below, that's not so easy. And the main credit, it's a great attempt. So within weight classes, you can compare, but then across weight classes, there will be a problem. So for example, if somebody, let's say, then weighs 81 kilos and lives say 340 kilos. That's about 90% of the world record in the 81 kilo class. And this person gains one kilo. As the person gains one kilo, he will be measured it against the next world record, not the same world record, but the world record in the upper next class. And that's going to drop his Ruby points without any change to the performance. That is one of the problems with the Ruby points. And the other problem is the world records change. So if I think the 73 kilo class in 2018, the world records was 360 kilos. And now it's at 364 kilos. So a nifter who weighs the same and does not change his performance, the Ruby points drop because it's a reference to the world record. So we'll see these wild behaviors, but we're trying to cross the body weight categories. That was one of the motivation to look at. And we come up with a system that doesn't change so rapidly. Now, Robbie is not good for comparing across weight classes before sub-res were used. So how do they shape? Yeah, so Sinclair is a good system, other than it is bound to the world records. There has been a problem with how do we compare things. Yeah, here we have, I created this figure. These are the same athletes and the same performances. These are the medal lists of those world and continental championships. And I used to sync clear points, adjustments. And the question was, can we create a point system or a skilled system so that these points would be comparable across the body weight? And we get better than what was before, but not quite because on this left hand side, the figure A, that's the males. We can see, well, the athletes in the lowest body weight class actually will have simpler points below the horizontal line, then at the little higher weight classes, they will have simpler points above the horizontal line. And then later on, I think the heaviest will be above that line. So the gray shaded area, this is confidence bands. So we're going to be perfect, but does the majority fit across, does the horizontal line is that covered by this confidence bands? And for those simpler points, that's not the case. So for this region, they are all above the line. That would mean, if this is the case, that lift us in this body weight range, maybe favorably have simpler points. So they might get a better chance at winning an overall middle. For women, it looks the worst. Women is actually quite hard to fit. So we also see the same phenomenon, but even stronger. So the lightest weight women would have more favorable results with the simpler points. So that was the motivation for the cue points. Can we come up with a better system? And as I mentioned, many have tried. Don't know whether that's the final answer that we came up with, but it looks better than the simpler points, because the red points are the scaled points. And if I draw a horizontal line, this horizontal line is covered by these gray bands, confidence bands, for the men. For the women, we can see that has a little bit of a problem, but the confidence bed covers the horizontal lines. So it's an improvement over the same clear points. And it's an improvement over the roby points, in the sense that it can compare across the body weights. Could you tell us a little bit more how the cue points are calculated? Yeah, it's actually a simple formula. I thought it was going to be much harder, but it worked out. Here's the formula. What we did is we tried lots of different functions of body weight. So including functions that hadn't been thought of before. One very systematically a class of functions that cover a whole range of shapes. And then the formula is as follows. The cue points are calculated. You take the total from the weightlifting competitions, and you multiply it with some maximum total, and divide it by a function of the body weight, the actual body weight. So I need the total for the weightlifting competition, and I need the weight in weight. This reference, maximum, is different for men and for women. It was calculated as the maximum of these functions at the upper end of the body weight range. So the function can be easily calculated in calculators, or in Excel spreadsheets, in competition software. And some competition software have already done that. I also have a web applications to add it to try your own shoe points. I'd prepare to bring the link, but it's very quick to calculate. The element in the show notes. Ah, yeah, right. I have a question about the maximum, the total maximum. So this is the theoretical number calculated for the heaviest athlete. Yeah, so this is a maximum calculated from the function. So it doesn't orient itself around world records. It's just a scaling. Now it is actually interesting that you bring this up, because the question will be where all the cue points for women are going to be lower than the cue points for men. Well, clearly, we've had different functions for the men, and the different functions for the women. And men live more than women do in general. So it cannot be directly compared. So the first question would be, well, can't we just change this scaling maximum? We start to make it equal? Well, unfortunately, we can't for the world champion level athletes, but then everybody would want to use it for the national, for regionals. And that would not be the same maximum, the same scaling point. So at this point, we're stuck with separate formulas for men and women. So would Robbie be less susceptible to this? Because they go from 1000 points. So then 1000 in male class versus 1000 in female class should be the same. So these could be compared across sexes. Right. So that is the proportion. I am already working on a system that would be robust across the body weights, but also can be compared for a sexes. It turns out it's not that easy, because I've observed this in several occasions. But the curve for the men is steeper for the plate than it is for women. So we see this in the model that here we go. So these coefficients are different between the men and the women. So it is not easy to use the same formula, although as we tried for men and women, there are typically it's either going to be a good fit for the men or a good fit for the women. But I believe that something can be done, maybe not with two points, but with a different scaling system. And I'm working on that process right now. Will it work for all age groups? That is an excellent question. And that is the question that applies to rugby points and to sing clip points and everything else. Yeah. Because we think about just from a statistical perspective, what we generated a formula system that is generated for 20 to 30 year olds, because the peak of the age of peak performances in the mid 20s, and elite lifters, those who compete in the world championships and Olympics, and those who can set world records. So why would any system that's developed for this very special group, the applicable to any other group, any other ages? So when we think the peak age is mid 20s, why should this formula work for 15 year olds or 70 year olds? That is nevertheless done with the same formula. And it's not clear how well it works for the different ages. So any kind of formula that one can come up with needs to be tested carefully at different ages. And fortunately, I have good collaborators who help me with this project. And we're trying it in different levels, namely, what does it work in national championships? And we've done it Iceland and USA and it works, fortunately. So it is comparable, makes sense. I've also tried it in master lifters. That also works by a little surprise, but very happy that that also works, because as long as this body weight curve makes sense for an age group, then it should be applicable. So it will also work for junior lifters, like ages 18 onwards. But I would not recommend any of those systems for youth. The reason is that youth with each year of age increase, they will get better, but they will also get heavier. You know, they get weight and they get better overall. So it's a different phenomenon that needs to be considered for that. I actually have done some work on that and I have a system, but we're using that only in local competitions. So that is across the age. So my answer would be, yeah, you can use it for masters, and for juniors and for zenias, but please don't use it for youth. The formula will not be stable for the lightest, for very lightweight, like, you know, 30 kilos or so. The other question that I brought up in the beginning with the body weight, we have the heavy is the unlimited body weight, passed as such a large spread from 110 kilos to 180 kilos for men. That's the same range that we have for the limited body weight. So that makes it difficult for the development of formulas, and because in a sense, they are outliers. And so older models have actually excluded the unlimited range. So there are a number of formulas around that have not even considered the unlimited range, but Sinclair does and my formula does and other formulas to that also. The women weightlifters performances, you could see they have actually a much larger range, much larger variation in their results. And that's why it's difficult for both viewpoints and for Cinque to come up with a good formula. And and viewpoints have done a little bit better on that, especially for the women. Do we know why? Yeah, I have some hypothesis about that. That is the women weightlifters have, you know, Cinque was originally developed when there weren't that many women in the weightlifting field, but it's been updated every so often. But the women participation has just really increased. So what can be is that actually with the increase of the participation in women, actually the performance have gotten over all the better. So we haven't reached a status for that. That might still be improved in heaps and balance as participation changes, while it's as much more settled with the men in the old era. The world records can still improve. But for women, they have more chances of getting medals at different levels. So that is the higher range. That is one theory driven by the participation. And the other is that the variation, larger variation of women has been seen in many other physiological measurements. So it's not just in the performance of the weightlifting. I have some key points here. Well, a few points are really easy to use and they're open access. That means it's not tied to me or anybody else. A formula can be used by anybody. The formula can be updated by anybody. So we still recommend, especially with the change in participation, that the viewpoints update every so often, also perhaps line with the Olympics cycle. But this could be done by an IWF committee, for example. And who wants to exclude, say, doping cases and which years to include. So it can be done, the algorithm is available on how to derive that formula. That's number one, easy to use. And it's open to use. The second point is it's key points are an improvement over the existing approaches. Existing approaches favor some body weight ranges. Or there can be large differences in changing small changes of body weight. And lastly, it can be used for junior seniors and masters, but not for you. Yeah, that's a big improvement because it makes it easier. You type formula ones into this competition software and you can use it across all the athletes and open competitions. Excluding dangerous athletes, right? That's very handy. One thing I would like to mention is that many systems used only world records or just the top three athletes in each body weight category. Key points used all the points. So all the performances in the world in continental championships were part of this formula. Thus, I believe it could be more generalizable than those who only use world records. All right, so you think that key points are applicable for national performances, regional competitions, club competitions and so on, right? Awesome. Okay, what comes next to things? So one is the mixed team that's been on my mind. And I've already made some progress on that. But I could compare men and women back not yet across the whole range of body weights. The other is what should consider. I would really like a different separate formula for youth, especially for for children. And before puberty, there's no difference between boys and girls of what they can lift. But what might happen is it's the same of what I've already developed. If that's a system of points, both age and body weight at the same time. That sounds complicated. As we said for youth, the older you get, the better you get, but also your body weight changes. And so the system that I've worked on with a statistics colleague from the UK is already published and it's out. So also open access and it's not easy to change. We have age and body weight multipliers for ages eight to 20. So that works well. But still, my main goal is the checking new national competitions to see whether it's really as well across across all the national and regional levels. And the second is to have a scaling model for mixed teams. I have two more questions. One is, what is your favorite color? Okay, Michigan State University, I choose green. And the second one is, if people want to look more into your work or ask you any questions where they should go and try new on internet. Oh, yeah, there are multiple places. One is open science framework, OSF, where I put this, this work. There's the code and there's the results. There's the formula. There's an Excel spreadsheet with the formula. And it also has my contact. And the other is my MSU email, if not MSU.edu. So thank you so much. It was my pleasure to finally meet you through the Zoom. Thank you so much, Alex. So if we're keeping trying, no worries. - Right. [BLANK_AUDIO]