WEBVTT 00:00:00.000 --> 00:00:02.740 So really, when it comes to strategizing, it 00:00:02.740 --> 00:00:05.320 could be an instance of, okay, let's open you 00:00:05.320 --> 00:00:07.620 at this number. We know what you're capable of. 00:00:07.639 --> 00:00:09.460 Let's open you at this. This will give you the 00:00:09.460 --> 00:00:12.580 best opportunity to get 300 or more, which might 00:00:12.580 --> 00:00:15.000 lead to a medal. Naturally, when you take that 00:00:15.000 --> 00:00:17.730 information. to a competition, as you see the 00:00:17.730 --> 00:00:20.289 board start to change, that's when you can make 00:00:20.289 --> 00:00:23.750 a little bit more of a freer choice of what load 00:00:23.750 --> 00:00:27.649 to put on because this information is there to 00:00:27.649 --> 00:00:30.629 just provide some form of strategy moving into 00:00:30.629 --> 00:00:33.170 competition. Once you're in competition, it's 00:00:33.170 --> 00:00:35.869 then down to intuition and how to read the board. 00:00:36.030 --> 00:00:38.350 So I think that's one of the steps that would 00:00:38.350 --> 00:00:40.909 be nice to look at in the future. Before I kind 00:00:40.909 --> 00:00:43.289 of get onto the performance zones, I think it's 00:00:43.289 --> 00:00:45.210 important to understand this started from something. 00:00:45.320 --> 00:00:48.600 quite simple so the paper is quite complex even 00:00:48.600 --> 00:00:51.259 when we developed it i had the help of a data 00:00:51.259 --> 00:00:54.020 scientist who happens to be one of my weightlifters 00:00:54.020 --> 00:00:57.799 you know an incredible help in providing information 00:00:57.799 --> 00:01:02.729 on how to make this sort of bus setup Welcome 00:01:02.729 --> 00:01:05.549 to Evidence Strong Show. It's my pleasure to 00:01:05.549 --> 00:01:07.810 have you the second time. The previous time I 00:01:07.810 --> 00:01:11.010 had you, we talked about your paper on stable 00:01:11.010 --> 00:01:13.629 and variable components in Olympic weightlifting 00:01:13.629 --> 00:01:16.450 technique. And today we'll be talking about predicting 00:01:16.450 --> 00:01:18.709 performance. But for three people who haven't 00:01:18.709 --> 00:01:20.530 watched the previous one, could you briefly introduce 00:01:20.530 --> 00:01:23.549 yourself? Firstly, thank you for having me on 00:01:23.549 --> 00:01:26.590 again. I appreciate it. I feel quite privileged. 00:01:27.010 --> 00:01:30.170 So my name is Sean Chavda. I'm the program lead 00:01:30.170 --> 00:01:32.049 for Strength and Conditioning. conditioning, 00:01:32.069 --> 00:01:34.849 distance education, a master's program at Middlesex 00:01:34.849 --> 00:01:36.969 University. I'm also head of weightlifting coach 00:01:36.969 --> 00:01:40.230 at Middlesex University where I coach Silver 00:01:40.230 --> 00:01:42.849 Chachit for Great Britain and Jessica De Silva 00:01:42.849 --> 00:01:46.370 for Portugal. And alongside my academic endeavors, 00:01:46.609 --> 00:01:50.109 I'm also a senior performance scientist at British 00:01:50.109 --> 00:01:52.510 Weightlifting. Awesome. So these are your qualifications. 00:01:52.849 --> 00:01:56.189 You also do a PhD and this study we'll be talking 00:01:56.189 --> 00:01:58.870 about today is a part of your PhD. This is the 00:01:58.870 --> 00:02:01.010 first part of the PhD. Yes, almost. I forgot. 00:02:01.069 --> 00:02:02.709 to mention that I'm doing that. Awesome. It's 00:02:02.709 --> 00:02:06.569 important. So why would we be interested in predicting 00:02:06.569 --> 00:02:09.590 performance? I mean, ultimately, if we can predict 00:02:09.590 --> 00:02:12.849 anything, it's always quite useful to be able 00:02:12.849 --> 00:02:15.250 to do that, right? And, you know, let's not beat 00:02:15.250 --> 00:02:17.150 around the bush around predictions. If there's 00:02:17.150 --> 00:02:19.110 anything we could predict in life, it'd be lottery 00:02:19.110 --> 00:02:21.509 numbers because it's far more lucrative. But 00:02:21.509 --> 00:02:24.090 predicting weightlifting performance actually 00:02:24.090 --> 00:02:26.830 came about because of a real life performance 00:02:26.830 --> 00:02:29.169 problem, which I encountered in the build up 00:02:29.169 --> 00:02:32.039 to Tokyo. And, you know, we weren't coming through 00:02:32.039 --> 00:02:34.740 this traditional route, having qualified, but 00:02:34.740 --> 00:02:36.879 instead through the International Olympic Committee's 00:02:36.879 --> 00:02:39.780 refugee Olympic team. So not many people knew 00:02:39.780 --> 00:02:42.180 about Sil's abilities, as he didn't have the 00:02:42.180 --> 00:02:44.669 privilege to compete at. international events 00:02:44.669 --> 00:02:47.129 for like five years, but we knew his capabilities. 00:02:47.389 --> 00:02:50.370 But what we didn't know is whether those capabilities 00:02:50.370 --> 00:02:54.129 would be worthy of this unspoken target, which 00:02:54.129 --> 00:02:56.849 we'd kind of set for him, which was finishing 00:02:56.849 --> 00:03:00.189 within top 10. So with this ranking in mind, 00:03:00.250 --> 00:03:02.169 in the buildup, I figured it was really important 00:03:02.169 --> 00:03:05.110 to try and find out what total would be required 00:03:05.110 --> 00:03:08.050 to achieve that top 10 ranking. And that's where 00:03:08.050 --> 00:03:09.650 I ended up questioning whether I could actually 00:03:09.650 --> 00:03:13.409 try and predict this. So this is the why. Now, 00:03:13.409 --> 00:03:17.990 you use something, a concept previously not discussed 00:03:17.990 --> 00:03:21.310 that widely. It's called performance zones. So 00:03:21.310 --> 00:03:24.009 could you explain what it's all about? Yeah. 00:03:24.069 --> 00:03:26.349 So before I kind of get onto the performance 00:03:26.349 --> 00:03:28.689 zones, I think it's important to understand this 00:03:28.689 --> 00:03:31.430 started from something quite simple. So the paper 00:03:31.430 --> 00:03:34.050 is quite complex. Even when we developed it, 00:03:34.129 --> 00:03:37.169 I had the help of a data scientist who happens 00:03:37.169 --> 00:03:39.590 to be one of my weightlifters, you know, an incredible 00:03:39.590 --> 00:03:42.550 help. in providing information on how to make 00:03:42.550 --> 00:03:44.490 this sort of a bus setup. But essentially everything 00:03:44.490 --> 00:03:46.590 was done in Excel at the beginning with scatter 00:03:46.590 --> 00:03:49.669 plots and regression lines. And what I did for 00:03:49.669 --> 00:03:52.469 each of the competitors in the men's 96 group 00:03:52.469 --> 00:03:56.569 is I predicted what they might do at the games. 00:03:56.710 --> 00:03:58.870 And if we fast forward a little bit, then it's 00:03:58.870 --> 00:04:00.669 something I'll go into more depth on later. I 00:04:00.669 --> 00:04:04.710 then predicted what total is likely to be achieved 00:04:04.710 --> 00:04:07.370 between the places of ninth and 10th. And that 00:04:07.370 --> 00:04:09.629 ninth and 10th can be termed a performance zone. 00:04:09.819 --> 00:04:11.960 So to answer your original question, what is 00:04:11.960 --> 00:04:15.060 a performance zone? It's a cluster of ranks. 00:04:15.300 --> 00:04:18.980 So in the paper, I take the rank from one to 00:04:18.980 --> 00:04:21.439 15, but I break them down into a medal zone. 00:04:21.579 --> 00:04:23.639 So that would be a performance zone. Fourth to 00:04:23.639 --> 00:04:27.060 fifth, which was a performance zone two. Essentially, 00:04:27.100 --> 00:04:29.560 that's like the outside shot at achieving a medal. 00:04:29.839 --> 00:04:33.639 Sixth to eighth, which I believe top eight is 00:04:33.639 --> 00:04:36.860 the standard. you need to achieve to qualify 00:04:36.860 --> 00:04:39.939 for the Games, but also you achieve an Olympic 00:04:39.939 --> 00:04:43.350 diploma. as well which i didn't find out until 00:04:43.350 --> 00:04:46.689 the after the games um and then ninth to tenth 00:04:46.689 --> 00:04:49.730 is again your outside shot to achieve a top eight 00:04:49.730 --> 00:04:53.350 and then 11th to 15th which was the maximum number 00:04:53.350 --> 00:04:56.509 of lifters within um the olympic categories so 00:04:56.509 --> 00:04:58.850 that's how i split the performance zones but 00:04:58.850 --> 00:05:01.430 what's important is i consulted with the performance 00:05:01.430 --> 00:05:04.069 manager at british weightlifting to ensure that 00:05:04.069 --> 00:05:07.290 these performance zones were highly contextual 00:05:07.290 --> 00:05:09.889 to what we need them for but it's important to 00:05:09.889 --> 00:05:12.189 note that in other Other sports or other governing 00:05:12.189 --> 00:05:14.230 bodies, they might deem performance zones to 00:05:14.230 --> 00:05:16.629 be slightly different. So, for example, you might 00:05:16.629 --> 00:05:19.009 have first and second as a performance zone, 00:05:19.149 --> 00:05:22.290 third and fifth as a performance zone. And whatever 00:05:22.290 --> 00:05:24.470 you decide to use as a performance zone just 00:05:24.470 --> 00:05:26.290 needs to be justified and worked within your 00:05:26.290 --> 00:05:28.110 own practice. But these are the ones that we 00:05:28.110 --> 00:05:30.810 decided were the most logical and the most useful. 00:05:31.089 --> 00:05:34.250 So anytime where the placing matters for qualification 00:05:34.250 --> 00:05:37.629 for other events or making a team or getting 00:05:37.629 --> 00:05:41.949 a scholarship. or whatever, that would be the 00:05:41.949 --> 00:05:44.069 zone qualification you would or classification 00:05:44.069 --> 00:05:47.149 you would use. Right, yeah. Okay, so you tried 00:05:47.149 --> 00:05:50.930 to predict what is the likely total required 00:05:50.930 --> 00:05:54.050 for each zone. How did you go about it? Yeah, 00:05:54.089 --> 00:05:58.250 so the weight class change that happened prior 00:05:58.250 --> 00:06:02.269 to November 2018, I think it was July 2018, might 00:06:02.269 --> 00:06:03.970 have been the last competition at the old weight 00:06:03.970 --> 00:06:05.970 categories. This was kind of the foundation of 00:06:05.970 --> 00:06:09.560 the paper because we have loads of historic information 00:06:09.560 --> 00:06:12.240 on these old weight categories but we have nothing 00:06:12.240 --> 00:06:15.480 on the new ones so one of the objectives of the 00:06:15.480 --> 00:06:19.560 paper was to see if we could predict what totals 00:06:19.560 --> 00:06:23.399 would be achieved at specific performance zones 00:06:23.399 --> 00:06:27.240 in each weight category or well in each new weight 00:06:27.240 --> 00:06:30.360 category which had been introduced by the IWF 00:06:30.360 --> 00:06:34.420 using the old weight category data so essentially 00:06:34.420 --> 00:06:38.069 the new category data didn't exist and when they 00:06:38.069 --> 00:06:40.529 did exist there's far too small of a sample to 00:06:40.529 --> 00:06:43.310 make a prediction so we had to use historic data 00:06:43.310 --> 00:06:45.670 and then once we developed these predictions 00:06:45.670 --> 00:06:48.250 we could then check if those predictions were 00:06:48.250 --> 00:06:52.500 close or not to what actually happened at Tokyo, 00:06:52.660 --> 00:06:55.079 European Champs, and the World Champs, for example. 00:06:55.420 --> 00:06:57.800 So what competitions did you take into account? 00:06:58.180 --> 00:07:03.839 So just the major three for us in Great Britain, 00:07:03.959 --> 00:07:06.899 which would be the Olympic Games, World Championships, 00:07:07.000 --> 00:07:10.639 and European Championships, so continental. Okay, 00:07:10.660 --> 00:07:14.379 and which years? So this was part of the methods, 00:07:14.379 --> 00:07:18.699 which I think was quite important. So if I go 00:07:18.699 --> 00:07:20.639 through the methods, that might help answer the 00:07:20.639 --> 00:07:23.720 question. what years we used and kind of came 00:07:23.720 --> 00:07:26.980 about it. So the data was scraped from the IWF 00:07:26.980 --> 00:07:29.600 website from 1998 all the way through to 2021. 00:07:30.040 --> 00:07:33.019 I took all the totals of all competitors. Then 00:07:33.019 --> 00:07:35.740 I reduced the data down to the top 15 at each 00:07:35.740 --> 00:07:38.240 competition. So World Challenge, European Challenge 00:07:38.240 --> 00:07:40.459 and Olympic Games. With regards to the years, 00:07:40.720 --> 00:07:43.500 what I did is I ran some effect sizes between 00:07:43.500 --> 00:07:47.740 each year within each competition type. So European 00:07:47.740 --> 00:07:51.899 Championships, 1998, 99, so on. And I compared 00:07:51.899 --> 00:07:54.519 those years. So I just took the average achieved 00:07:54.519 --> 00:07:57.199 and just compared those years to ensure that 00:07:57.199 --> 00:08:00.000 in general, there wasn't a large difference between 00:08:00.000 --> 00:08:04.480 each of these years. And this essentially helps 00:08:04.480 --> 00:08:08.490 us identify it. like an outlier year where performances 00:08:08.490 --> 00:08:11.149 might have just completely dropped or just gone 00:08:11.149 --> 00:08:14.370 up by you know a significant amount so once we're 00:08:14.370 --> 00:08:17.089 happy that there are either small or moderate 00:08:17.089 --> 00:08:20.110 changes between them just decide okay that's 00:08:20.110 --> 00:08:23.290 acceptable let's pull the data together for each 00:08:23.290 --> 00:08:26.290 competition so it helps increase the sample that 00:08:26.290 --> 00:08:28.560 we're using for the prediction okay i have two 00:08:28.560 --> 00:08:31.399 questions about the data so one would be how 00:08:31.399 --> 00:08:34.120 did you decide to transform the previous weight 00:08:34.120 --> 00:08:37.179 classes into the new ones if you did or how did 00:08:37.179 --> 00:08:39.720 you kind of group them together and then the 00:08:39.720 --> 00:08:43.559 other would be some athletes will reappear from 00:08:43.559 --> 00:08:46.480 one competition to the other so some data come 00:08:46.480 --> 00:08:50.200 from the same athlete so how how did you handle 00:08:50.200 --> 00:08:52.440 that yeah excellent questions actually questions 00:08:52.440 --> 00:08:55.360 that came up in the review process when prior 00:08:55.360 --> 00:08:58.159 to this being published so the The first question, 00:08:58.220 --> 00:09:01.240 correct me if I misinterpret the question, was 00:09:01.240 --> 00:09:02.759 what was the first question? No, no, the first 00:09:02.759 --> 00:09:05.200 one was white glasses because you mentioned before. 00:09:05.820 --> 00:09:09.120 July 2018, the weight classes changed. We have 00:09:09.120 --> 00:09:13.419 that data and you included data in this analysis 00:09:13.419 --> 00:09:18.019 from 1998, which means you included also the 00:09:18.019 --> 00:09:21.159 data from the previous weight classes scheme. 00:09:21.340 --> 00:09:26.360 So how did you manage? So it was all the data 00:09:26.360 --> 00:09:30.820 from 1998 to 2018. Every single competition that 00:09:30.820 --> 00:09:32.919 had this, they would have the same weight categories, 00:09:33.100 --> 00:09:35.980 right? They were all put in to, they were all 00:09:35.980 --> 00:09:39.039 aggregated for each performance zone, right? 00:09:39.419 --> 00:09:42.220 Now with that, we split them into rank zones. 00:09:42.320 --> 00:09:44.500 So you've got your aggregate for each rank zone. 00:09:44.679 --> 00:09:47.240 We then ran something called a K -fold cross 00:09:47.240 --> 00:09:50.019 validation, which essentially just evaluates 00:09:50.019 --> 00:09:53.580 the model's ability to predict when given unseen 00:09:53.580 --> 00:09:56.539 data. which is what we're going to be putting 00:09:56.539 --> 00:09:58.659 into it at some point. But also what it does, 00:09:58.679 --> 00:10:00.639 it stops us overfitting this regression line. 00:10:00.779 --> 00:10:03.740 So the regression line we fit in was a polynomial 00:10:03.740 --> 00:10:06.740 regression, basically curves through. Similar 00:10:06.740 --> 00:10:09.600 to weightlifting totals and body weight, it's 00:10:09.600 --> 00:10:11.940 polynomial. It goes up, it'll plateau and it'll 00:10:11.940 --> 00:10:15.639 start to come down. Same as strength in relation 00:10:15.639 --> 00:10:18.919 to body weight, it's curve linear, not linear. 00:10:19.200 --> 00:10:21.200 So anyway, we run this polynomial regression. 00:10:21.440 --> 00:10:23.799 Because we're using the old data, that's all. 00:10:23.909 --> 00:10:27.710 plotted and the curve means at any body weight 00:10:27.710 --> 00:10:31.190 category on the bottom axis, we can just go at 00:10:31.190 --> 00:10:34.090 this point. which is a new weight category, go 00:10:34.090 --> 00:10:36.690 up, find the intercept, and then we can predict 00:10:36.690 --> 00:10:42.090 what the total would be for that new weight category 00:10:42.090 --> 00:10:44.789 in that specific performance zone within that 00:10:44.789 --> 00:10:48.350 specific competition type. So you took the actual 00:10:48.350 --> 00:10:52.789 data of the body weight data of each athlete 00:10:52.789 --> 00:10:55.649 in each weight class, and when you predicted 00:10:55.649 --> 00:10:59.049 the total, then you were able to then go back 00:10:59.049 --> 00:11:03.669 and kind of you had a linear. relationship polynomial 00:11:03.669 --> 00:11:07.250 but kind of not by classes so not by categories 00:11:07.250 --> 00:11:10.370 more continuous yes no it was by category so 00:11:10.370 --> 00:11:12.149 if you look at the paper there's some supplemental 00:11:12.149 --> 00:11:16.350 supplementary information which are graphs and 00:11:16.350 --> 00:11:20.269 in those graphs show um on the bottom axis it 00:11:20.269 --> 00:11:23.610 has each olympic weight category or each new 00:11:23.610 --> 00:11:26.090 weight category i should say and on the vertical 00:11:26.090 --> 00:11:28.990 axis it's got the total and it's a simple it's 00:11:28.990 --> 00:11:31.519 a simple case of just intercepting vertically 00:11:31.519 --> 00:11:34.039 and then horizontally to that regression line 00:11:34.039 --> 00:11:37.440 to then find what the predicted total would be 00:11:37.440 --> 00:11:40.740 for that weight category and the new weight category 00:11:40.740 --> 00:11:44.059 we use instead of just putting in the number 00:11:44.059 --> 00:11:49.419 89 or putting in 102 what we did is we took the 00:11:49.419 --> 00:11:52.779 new weight category information that had been 00:11:52.779 --> 00:11:56.080 competed in which was i think only two european 00:11:56.080 --> 00:11:59.100 champs maybe one or two world championships we 00:11:59.100 --> 00:12:01.440 just took the average weight as opposed to just 00:12:01.440 --> 00:12:03.399 taking that body weight the reason we took the 00:12:03.399 --> 00:12:05.960 average body weight is because on the top end 00:12:05.960 --> 00:12:08.779 of the spectrum where you have no end cap to 00:12:08.779 --> 00:12:11.139 be as heavy as you want we felt we felt like 00:12:11.139 --> 00:12:14.220 that would influence the predictability so to 00:12:14.220 --> 00:12:16.379 keep things fair and logical we just went take 00:12:16.379 --> 00:12:18.539 the average of each new weight category because 00:12:18.539 --> 00:12:21.039 we know that we have that information and let's 00:12:21.039 --> 00:12:24.559 put that in because It represents the mean of 00:12:24.559 --> 00:12:27.320 the group of their body weight. So we're not 00:12:27.320 --> 00:12:30.919 just going to put 89 in if most of them fall 00:12:30.919 --> 00:12:34.779 around 88 .7 or 88 .6, for example. So that's 00:12:34.779 --> 00:12:37.320 how we predict it. So old weight category data 00:12:37.320 --> 00:12:41.159 plotted and new weight category data used as 00:12:41.159 --> 00:12:44.320 the intercept. Okay. I'm looking through them 00:12:44.320 --> 00:12:46.820 because I thought maybe we could just put a graph 00:12:46.820 --> 00:12:50.100 on top of the video. The version I have is before 00:12:50.100 --> 00:12:53.350 publication. publisher shouldn't but it's like 00:12:53.350 --> 00:12:56.570 every zone will have a separate i can share my 00:12:56.570 --> 00:12:58.830 screen if you like yeah i've got so i've got 00:12:58.830 --> 00:13:01.210 a graph here for like the olympic medal zone 00:13:01.210 --> 00:13:05.029 which i'm i'm happy to to show you awesome so 00:13:05.029 --> 00:13:08.700 this is part of the supplementary materials. 00:13:09.000 --> 00:13:13.139 For example, let's say this regression line that 00:13:13.139 --> 00:13:15.419 runs through the middle of these squares were 00:13:15.419 --> 00:13:18.580 developed using the old weight category data. 00:13:18.899 --> 00:13:21.279 We've then got some confidence intervals around 00:13:21.279 --> 00:13:23.899 it, 95 % confidence intervals. So if we were 00:13:23.899 --> 00:13:26.000 to run this on, say, a homogeneous population, 00:13:26.360 --> 00:13:30.179 run a regression, 100 times 95 of those regression 00:13:30.179 --> 00:13:32.679 lines will likely fall within this range. We 00:13:32.679 --> 00:13:34.960 have a predictive interval on the outside. Predictive 00:13:34.960 --> 00:13:37.419 intervals are a little bit wider. but these essentially 00:13:37.419 --> 00:13:42.399 account for any future individual cases which 00:13:42.399 --> 00:13:44.779 might come in. And that's why they're extremely 00:13:44.779 --> 00:13:48.639 wide. So for example, in the future, if one person 00:13:48.639 --> 00:13:51.539 lifts in the 96, they're probably going to be 00:13:51.539 --> 00:13:53.779 between this bound and this bound here, the dash 00:13:53.779 --> 00:13:57.120 line. So we can see for the men's Olympic medal 00:13:57.120 --> 00:14:00.659 zone, so first and third, the prediction is this 00:14:00.659 --> 00:14:03.139 value here. So approximately, let's just say 00:14:03.139 --> 00:14:06.000 400 kilos, if you want the actual value. values 00:14:06.000 --> 00:14:09.639 that are available in the tables above paralyzed 00:14:09.639 --> 00:14:12.360 people with loads of numbers. So let's just take 00:14:12.360 --> 00:14:15.399 it as 400. So that's our prediction. And then 00:14:15.399 --> 00:14:18.279 after the Olympics, I took an average of the 00:14:18.279 --> 00:14:21.500 medal zone and I took a standard deviation and 00:14:21.500 --> 00:14:24.460 I simply plotted it on top. So this shows that 00:14:24.460 --> 00:14:27.279 actually it was an overprediction compared to 00:14:27.279 --> 00:14:30.000 what actually happened. However, the top end 00:14:30.000 --> 00:14:34.610 almost reaches that 95%. confidence range so 00:14:34.610 --> 00:14:37.289 while you might look at this and think or somebody 00:14:37.289 --> 00:14:39.190 may look at this and think well that's an over 00:14:39.190 --> 00:14:42.190 prediction it's it's rubbish in actual fact you'd 00:14:42.190 --> 00:14:43.830 rather have an over prediction than an under 00:14:43.830 --> 00:14:46.090 prediction because it means you're trying to 00:14:46.090 --> 00:14:49.350 gear gear up to lift like that weight which is 00:14:49.350 --> 00:14:51.649 you know far heavier than what was actually achieved 00:14:51.649 --> 00:14:54.470 so it might mean on game day if you're expecting 00:14:54.470 --> 00:14:57.950 this to happen and you lift close to it and this 00:14:57.950 --> 00:15:00.529 actually happens, you end up overachieving and 00:15:00.529 --> 00:15:03.230 potentially getting a medal. Whereas if we look 00:15:03.230 --> 00:15:06.230 at the 73, the prediction's almost spot on in 00:15:06.230 --> 00:15:09.509 terms of the average. So second place was really, 00:15:09.669 --> 00:15:11.850 really close. But if you achieve anything between 00:15:11.850 --> 00:15:14.029 these red lines, you're likely to then have got 00:15:14.029 --> 00:15:16.809 a medal. In the same breath with predictive intervals 00:15:16.809 --> 00:15:19.870 here, if you were to achieve this value, it would 00:15:19.870 --> 00:15:22.429 have put you outside the middle zone. So with 00:15:22.429 --> 00:15:25.029 these confidence intervals that we have, we're 00:15:25.029 --> 00:15:26.850 just trying to account for as much error as possible. 00:15:26.889 --> 00:15:29.850 But being at a competition is extremely dynamic 00:15:29.850 --> 00:15:32.490 and anything can happen, especially at highly 00:15:32.490 --> 00:15:35.629 competitive competitions. But with the ability 00:15:35.629 --> 00:15:38.529 to predict, at least we have some opportunity 00:15:38.529 --> 00:15:42.110 to strategize longitudinally or even just going 00:15:42.110 --> 00:15:44.289 into competition. So I think it certainly has 00:15:44.289 --> 00:15:47.769 a utility in sport. Should we have a look at 00:15:47.769 --> 00:15:52.649 the 9th to 10th place zone or to 11th to 15th? 00:15:52.710 --> 00:15:55.750 How in comparison will it be wider? How it will 00:15:55.750 --> 00:16:00.190 look? yeah so as we expected the lower down the 00:16:00.190 --> 00:16:02.549 ranks you get the wider the interval becomes 00:16:02.549 --> 00:16:05.549 and that's for pretty much most competitions 00:16:05.549 --> 00:16:08.750 more so for olympic than the world champs i believe 00:16:08.750 --> 00:16:12.269 and the reason being is olympics is about equality 00:16:12.269 --> 00:16:15.529 and inclusion and representation so you know 00:16:15.529 --> 00:16:19.710 while you are getting very very good weightlifters 00:16:19.710 --> 00:16:22.590 your top eight are the ones that are really world 00:16:22.590 --> 00:16:24.529 leading because they're the ones that have qualified 00:16:24.529 --> 00:16:26.679 by being the best in the world. The rest then 00:16:26.679 --> 00:16:30.539 get allocated to continents. Now, they might 00:16:30.539 --> 00:16:33.399 be the best in the continent. However, in the 00:16:33.399 --> 00:16:35.840 overall world ranking, they may not be so high, 00:16:35.980 --> 00:16:40.419 which means naturally from maybe ninth down, 00:16:40.659 --> 00:16:42.779 you're going to start to see these confidence 00:16:42.779 --> 00:16:45.720 intervals spread a lot more than you would at 00:16:45.720 --> 00:16:49.320 the higher echelons of performance, which is 00:16:49.320 --> 00:16:52.879 quite evident when you get to 11th to 15th. In 00:16:52.879 --> 00:16:55.220 terms of prediction and And the strength of the 00:16:55.220 --> 00:16:58.139 predictions, would you say that they are a little 00:16:58.139 --> 00:17:01.279 bit less accurate for these places too, the prediction 00:17:01.279 --> 00:17:04.000 model? It depends how you view it, because in 00:17:04.000 --> 00:17:06.460 theory, if you were to say, how good is your 00:17:06.460 --> 00:17:08.839 prediction into how good is your predictability? 00:17:09.160 --> 00:17:11.059 Everything that's happened has fallen within 00:17:11.059 --> 00:17:13.099 the predictive intervals. I see. So actually 00:17:13.099 --> 00:17:15.200 it's a hundred percent. Yes, we can predict. 00:17:15.339 --> 00:17:18.700 The problem is, is the actual value itself. The 00:17:18.700 --> 00:17:21.660 range is far too large that we're predicting. 00:17:21.859 --> 00:17:25.660 So, you know, it's... I suppose it'd be the equivalent 00:17:25.660 --> 00:17:29.440 of saying, think of a number between 1 and 3. 00:17:29.599 --> 00:17:32.500 I'm more likely to get the number correct, right? 00:17:32.839 --> 00:17:35.599 But if I open up, think of a number between 1 00:17:35.599 --> 00:17:38.299 and 100, I've made my chance a lot smaller to 00:17:38.299 --> 00:17:41.579 predict that. So it's similar to there. Whereas 00:17:41.579 --> 00:17:44.160 if I just scoot up to, I mean, the medal zone 00:17:44.160 --> 00:17:47.960 here, you can see the 11th to 15th zone is really, 00:17:48.119 --> 00:17:51.500 really wide. And a lot of the actual, actually 00:17:51.500 --> 00:17:53.670 what happened was an underprediction. in comparison, 00:17:54.029 --> 00:17:56.730 sorry, what actually happened was less than what 00:17:56.730 --> 00:17:59.009 was predicted. Whereas when we look at the world 00:17:59.009 --> 00:18:02.789 champs. first and third, a lot of them are very 00:18:02.789 --> 00:18:05.869 close, or the standard deviations cross the prediction, 00:18:06.069 --> 00:18:08.529 right, with the exception of the 89 and the 102. 00:18:08.869 --> 00:18:11.250 And you can see the prediction interval is smaller, 00:18:11.490 --> 00:18:14.289 confidence interval is smaller, and what was 00:18:14.289 --> 00:18:17.269 predicted as a point estimate and what actually 00:18:17.269 --> 00:18:19.809 happened were very close. Would you say it's 00:18:19.809 --> 00:18:22.670 kind of a ceiling effect with the metal zone 00:18:22.670 --> 00:18:27.269 versus 11s to 15s? Yeah, the margins up top are 00:18:27.269 --> 00:18:30.269 a lot smaller, yeah, for sure. sure one of the 00:18:30.269 --> 00:18:33.549 things we have discussed from a practical and 00:18:33.549 --> 00:18:36.109 an applied perspective is looking at first second 00:18:36.109 --> 00:18:39.170 and third and looking at how they may cross over 00:18:39.170 --> 00:18:42.990 using similar models but as soon as you do that 00:18:42.990 --> 00:18:46.490 your sample just gets a lot smaller anyway so 00:18:46.490 --> 00:18:49.710 the predict the ability to predict and generalize 00:18:49.710 --> 00:18:52.289 might be reduced but if that's what you have 00:18:52.289 --> 00:18:55.470 available at least again you have something that 00:18:55.470 --> 00:18:57.849 enables you to strategize as opposed to going 00:18:57.849 --> 00:19:00.450 well it's not robust statistically so let's not 00:19:00.450 --> 00:19:03.029 use it at all i think that's where practice and 00:19:03.029 --> 00:19:06.369 application and statistics and academia need 00:19:06.369 --> 00:19:09.150 to kind of come together and go right this is 00:19:09.150 --> 00:19:11.089 what we should do and this is how we should be 00:19:11.089 --> 00:19:14.009 as robust as possible these are the confinements 00:19:14.009 --> 00:19:16.450 we have or the barriers that we have what's the 00:19:16.450 --> 00:19:19.970 next best thing that we can do so also as we 00:19:19.970 --> 00:19:22.329 will progress there will be more data because 00:19:22.329 --> 00:19:24.930 there will be more competitions with the new 00:19:24.930 --> 00:19:27.720 weight classes so yes exactly and that That's 00:19:27.720 --> 00:19:30.839 one of the things that we're looking to do for 00:19:30.839 --> 00:19:33.599 the women's weight categories. So this was just 00:19:33.599 --> 00:19:35.660 done for the men. But what we'll do this time 00:19:35.660 --> 00:19:38.819 is because these new weight category totals exist, 00:19:39.160 --> 00:19:41.839 we're going to add them to the old data. So you 00:19:41.839 --> 00:19:45.799 then have an entire spectrum of old data and 00:19:45.799 --> 00:19:49.420 new data in. So you'll have a range of body weights 00:19:49.420 --> 00:19:52.319 and then see if our predictability gets better 00:19:52.319 --> 00:19:54.180 that way, because there still won't be enough 00:19:54.180 --> 00:19:57.299 data of the new weight categories. Because, you 00:19:57.299 --> 00:19:59.720 know, we're confining it down to weight categories 00:19:59.720 --> 00:20:02.500 and we're further confining it down to competition 00:20:02.500 --> 00:20:05.119 type. So every time, and then performance zone. 00:20:05.240 --> 00:20:07.059 So every time you're getting smaller and smaller 00:20:07.059 --> 00:20:10.539 and smaller. So that's the next step that we're 00:20:10.539 --> 00:20:12.839 planning to do over the next few months. That's 00:20:12.839 --> 00:20:16.900 exciting. I forgot to get the answer to the question. 00:20:17.079 --> 00:20:20.299 How did you handle the reappearance of the athletes? 00:20:20.440 --> 00:20:24.539 Did you reduce the data set because it took maybe 00:20:24.539 --> 00:20:27.799 the best one or did you... as it was? How did 00:20:27.799 --> 00:20:29.640 you handle that? Yeah, really good question. 00:20:29.859 --> 00:20:31.579 I feel like you might have been the one that 00:20:31.579 --> 00:20:34.299 reviewed my paper because it began with a question. 00:20:34.900 --> 00:20:37.680 And it's a really valid point, especially as 00:20:37.680 --> 00:20:40.579 it affects the generalizability of the models. 00:20:40.740 --> 00:20:44.579 But what we did is we treat each total as an 00:20:44.579 --> 00:20:47.319 individual case and not each athlete. And we 00:20:47.319 --> 00:20:50.099 did that because performance changes over time. 00:20:50.200 --> 00:20:52.980 So it either go up or down and therefore so will 00:20:52.980 --> 00:20:55.529 the rank that. the total achieves or that person 00:20:55.529 --> 00:20:59.009 achieves and given we're trying to predict totals 00:20:59.009 --> 00:21:01.569 in specific performance zones it's important 00:21:01.569 --> 00:21:03.569 to capture this information if we want to ensure 00:21:03.569 --> 00:21:06.829 our model is representative of what might actually 00:21:06.829 --> 00:21:09.690 happen but also like you mentioned if we use 00:21:09.690 --> 00:21:12.849 if we select the athlete and what competition 00:21:12.849 --> 00:21:15.809 and what total we use to analyze then we run 00:21:15.809 --> 00:21:18.089 into selection bias you know which one did we 00:21:18.089 --> 00:21:20.390 choose and why would we have chosen that And 00:21:20.390 --> 00:21:22.869 as a result of that, you also significantly reduce 00:21:22.869 --> 00:21:24.930 the sample. So not only have you got selection 00:21:24.930 --> 00:21:28.069 bias, you've got sample reduction as well. And 00:21:28.069 --> 00:21:30.970 with that in turn, you then reduce your ability 00:21:30.970 --> 00:21:32.849 to predict, which is our point of the paper. 00:21:33.049 --> 00:21:37.309 So using athlete reappearance was just lesser 00:21:37.309 --> 00:21:40.049 of the two evils, I suppose. And we felt like 00:21:40.049 --> 00:21:43.509 it was justified enough to be able to do that. 00:21:43.710 --> 00:21:46.029 So statistically speaking, we would love each 00:21:46.029 --> 00:21:49.549 data point. So each total be from a separate... 00:21:49.549 --> 00:21:51.849 separate different person each time different 00:21:51.849 --> 00:21:54.609 person that's not how the competitions run though 00:21:54.609 --> 00:21:58.849 so probably leaving the total as it appeared 00:21:58.849 --> 00:22:02.390 in the competition was more close to the reality 00:22:02.390 --> 00:22:06.890 than selecting one or the other yeah and i'm 00:22:06.890 --> 00:22:10.390 certainly no statistician so my depth of understanding 00:22:10.390 --> 00:22:12.730 of statistics wouldn't be as definitely nowhere 00:22:12.730 --> 00:22:14.589 near as good as someone who's a data scientist 00:22:14.589 --> 00:22:17.549 but this is where as i mentioned before like 00:22:17.549 --> 00:22:20.319 the robustness of this but making sure we're 00:22:20.319 --> 00:22:22.599 getting enough information that can be utilised 00:22:22.599 --> 00:22:25.680 within our field is. It's really important to 00:22:25.680 --> 00:22:28.680 kind of come to an intermediary conclusion where 00:22:28.680 --> 00:22:30.980 we're like, okay, it's robust enough. It's not 00:22:30.980 --> 00:22:33.700 statistically, it might not be perfect, but because 00:22:33.700 --> 00:22:38.079 of what the data is and how the data is collected. 00:22:38.400 --> 00:22:40.259 Because you could have somebody that medals at 00:22:40.259 --> 00:22:43.900 Europeans and then comes 10th in a world championships, 00:22:44.039 --> 00:22:46.779 for example. So which one are you now taking? 00:22:46.900 --> 00:22:48.299 Are you taking their best one? Are you taking 00:22:48.299 --> 00:22:50.740 their worst one? I think it pulls out more problems 00:22:50.740 --> 00:22:53.759 than it does. solved to be honest yeah fair enough 00:22:53.759 --> 00:22:57.019 i think it was statistics you always are if everything 00:22:57.019 --> 00:23:00.000 is perfect then the prediction is 100 but it 00:23:00.000 --> 00:23:02.759 is far away from reality because reality is never 00:23:02.759 --> 00:23:05.500 perfect so you have to i guess find this balance 00:23:05.500 --> 00:23:09.299 so you can find that exactly and if if if data 00:23:09.299 --> 00:23:12.180 was always so perfect and so easy to use especially 00:23:12.180 --> 00:23:15.039 for predictions like i said before time and cheap 00:23:15.039 --> 00:23:16.839 you know i wouldn't be predicting weightlifting 00:23:16.839 --> 00:23:20.269 i'd be predicting lottery numbers But it gives 00:23:20.269 --> 00:23:22.589 us a good starting point at the very least. And 00:23:22.589 --> 00:23:24.869 I'm sure somebody far smarter than me will come 00:23:24.869 --> 00:23:27.930 up with a model that's... even better and you 00:23:27.930 --> 00:23:31.450 know is more maybe more malleable so it can change 00:23:31.450 --> 00:23:33.769 as the dynamics of weightlifting changes as well 00:23:33.769 --> 00:23:37.029 who knows ai is developing quickly maybe you 00:23:37.029 --> 00:23:40.269 will know definitely yeah very quickly is there 00:23:40.269 --> 00:23:43.650 anything from the methods and setup of the study 00:23:43.650 --> 00:23:46.109 i didn't ask about but you think it's important 00:23:46.109 --> 00:23:50.289 really no i feel like i'm i i've covered feel 00:23:50.289 --> 00:23:52.450 like i've covered it to be honest okay so are 00:23:52.450 --> 00:23:55.150 we ready for the for the results i would like 00:23:55.150 --> 00:23:59.450 us to go on more you to tell us what were the 00:23:59.450 --> 00:24:02.769 prediction across the competitions and across 00:24:02.769 --> 00:24:07.309 the weight classes okay so that's quite a there's 00:24:07.309 --> 00:24:10.049 going to be a quite a big one to to answer because 00:24:10.049 --> 00:24:14.069 there's so many sub categories of of it because 00:24:14.069 --> 00:24:17.859 you know We're predicting category. We're predicting 00:24:17.859 --> 00:24:21.380 competition type, category. And then under those 00:24:21.380 --> 00:24:23.619 categories, we've got subcategories of performance 00:24:23.619 --> 00:24:26.779 zones. Yes. So if I were to answer it in the 00:24:26.779 --> 00:24:29.279 most simplistic way as an overarching answer, 00:24:29.460 --> 00:24:32.700 yes, the predictive models were pretty good at 00:24:32.700 --> 00:24:34.880 predicting weightlifting performance at different 00:24:34.880 --> 00:24:37.799 performance zones. However, these will vary depending 00:24:37.799 --> 00:24:41.000 on the competition type and depending on the 00:24:41.000 --> 00:24:43.440 performance zone and the weight category that 00:24:43.440 --> 00:24:46.619 you're looking at. What we did see, like we mentioned 00:24:46.619 --> 00:24:49.819 before, is as you come down the performance zones, 00:24:49.980 --> 00:24:53.160 so to the lower ranked lifters, so 11th to 15th, 00:24:53.259 --> 00:24:57.240 the predictability there and the fit of the data 00:24:57.240 --> 00:25:00.019 for the regression line was reduced quite a bit. 00:25:00.279 --> 00:25:04.259 So the R -squared value for the four performance 00:25:04.259 --> 00:25:07.299 zones at the top were all pretty much above 0 00:25:07.299 --> 00:25:11.180 .9. Some were close to 1, like 0 .99, 0 .96, 00:25:11.400 --> 00:25:14.160 0 .97. As soon as you get to zone 4. five, 11th 00:25:14.160 --> 00:25:16.700 to 15th, there's a significant drop, you know, 00:25:16.720 --> 00:25:22.539 0 .6, 0 .89, 0 .77 for Olympics, World and European 00:25:22.539 --> 00:25:26.559 champs. I mean, if we were to just pick one at... 00:25:26.750 --> 00:25:32.789 At random, let's say 67 kilo men's category. 00:25:33.130 --> 00:25:36.970 Performance zone two, for example, had a prediction 00:25:36.970 --> 00:25:40.089 of, so between fourth and fifth, the prediction 00:25:40.089 --> 00:25:44.410 was 321 kilos. What actually happened was 321 00:25:44.410 --> 00:25:47.529 kilos plus or minus one kilo. So that's less 00:25:47.529 --> 00:25:52.789 than 0 .05%, 0 .04 % of the difference. And the 00:25:52.789 --> 00:25:54.809 absolute difference was zero. Keep in mind when 00:25:54.809 --> 00:25:58.000 I present the standard DVR, in this paper, it's 00:25:58.000 --> 00:26:00.859 not to any decimal. It's always rounded up because 00:26:00.859 --> 00:26:03.599 in weightlifting, we lift to a kilo or more. 00:26:03.759 --> 00:26:07.579 So actually, that was pretty much on the nose. 00:26:07.839 --> 00:26:11.240 Yeah, it was a bullseye. Whereas if we, let's 00:26:11.240 --> 00:26:15.839 say, move to the 109 plus for zone two, fourth 00:26:15.839 --> 00:26:18.119 to fifth at the Olympic Games, the prediction 00:26:18.119 --> 00:26:22.319 was 442. What actually happened was 414. So the 00:26:22.319 --> 00:26:25.539 difference is just under 6 .5%, which is... 28 00:26:25.539 --> 00:26:27.880 kilo difference that's quite large so in that 00:26:27.880 --> 00:26:30.519 instance if you've got an athlete that's trying 00:26:30.519 --> 00:26:33.519 to achieve fourth to fifth and you're using that 00:26:33.519 --> 00:26:36.519 as a predictive model it's certainly over achieving 00:26:36.519 --> 00:26:40.259 or over predicting so if they go in lifting like 00:26:40.259 --> 00:26:44.319 420 at the very minimum and they're like oh i'm 00:26:44.319 --> 00:26:47.059 never gonna get fourth or fifth you'd be surprised 00:26:47.059 --> 00:26:51.000 because in this instance 414 was the total which 00:26:51.000 --> 00:26:53.200 kind of brings me on to something else which 00:26:53.200 --> 00:26:55.859 i think would like to to look into in more depth 00:26:55.859 --> 00:26:59.799 is how these performance zones cross over. I 00:26:59.799 --> 00:27:03.880 think it's really important. For example, we 00:27:03.880 --> 00:27:06.940 have our standard deviation. We have our average. 00:27:07.099 --> 00:27:09.599 We have our standard deviation. And let's say 00:27:09.599 --> 00:27:12.960 this is our medal zone. Then we'll have our fourth 00:27:12.960 --> 00:27:15.579 to fifth and so on and so forth. So we'll keep 00:27:15.579 --> 00:27:17.720 that down. But they cross over at some point. 00:27:17.859 --> 00:27:20.440 So it just changes the conversation you have. 00:27:20.910 --> 00:27:24.150 like look if you achieve a very random number 00:27:24.150 --> 00:27:27.390 if you achieve 300 kilos you're going to be above 00:27:27.390 --> 00:27:30.130 the average of fourth to fifth but it also means 00:27:30.130 --> 00:27:33.210 you might meddle so really when it comes to strategizing 00:27:33.210 --> 00:27:36.289 it could be an instance of okay let's open you 00:27:36.289 --> 00:27:38.569 at this number we know what you're capable of 00:27:38.569 --> 00:27:40.410 let's open you at this this will give you the 00:27:40.410 --> 00:27:43.549 best opportunity to get 300 or more which might 00:27:43.549 --> 00:27:45.970 lead to a medal naturally when you take that 00:27:45.970 --> 00:27:48.700 information to a competition, as you see the 00:27:48.700 --> 00:27:51.259 board start to change, that's when you can make 00:27:51.259 --> 00:27:54.720 a little bit more of a freer choice of what load 00:27:54.720 --> 00:27:58.619 to put on because this information is there to 00:27:58.619 --> 00:28:01.579 just provide some form of strategy moving into 00:28:01.579 --> 00:28:04.119 competition. Once you're in competition, it's 00:28:04.119 --> 00:28:06.839 then down to intuition and how to read the board. 00:28:07.000 --> 00:28:09.319 So I think that's one of the steps that would 00:28:09.319 --> 00:28:11.680 be nice to look at in the future. So the zone 00:28:11.680 --> 00:28:15.059 will have the number and then confidence and 00:28:15.059 --> 00:28:18.309 then standard deviation. so it's the plus minus 00:28:18.309 --> 00:28:22.750 however many kilos so that's the number in theory 00:28:22.750 --> 00:28:26.470 you would need to total to be in that zone to 00:28:26.470 --> 00:28:29.930 so to place in that zone but it can be less or 00:28:29.930 --> 00:28:33.190 more depending who is coming to the competition 00:28:33.190 --> 00:28:36.329 and exactly it and it all depends on the day 00:28:36.329 --> 00:28:40.069 and i think That's really important. And to come 00:28:40.069 --> 00:28:43.509 back round to the original reason why this idea 00:28:43.509 --> 00:28:46.670 kind of came up is we wanted to know the realities 00:28:46.670 --> 00:28:50.160 of Finnish. finishing within the top 10. But 00:28:50.160 --> 00:28:52.980 we also know that on the day, anything can happen. 00:28:53.180 --> 00:28:54.900 So while we'll have a strategy of, look, if you're 00:28:54.900 --> 00:28:57.759 going to attempt 200 kilo clean and jerk, we 00:28:57.759 --> 00:28:59.900 need to start you at this. Then we look back 00:28:59.900 --> 00:29:02.640 at that individual's information. Okay, the last 00:29:02.640 --> 00:29:06.460 time you hit 200, you started at this. This is 00:29:06.460 --> 00:29:08.180 how training has been going. And it becomes a 00:29:08.180 --> 00:29:11.180 conversation and it has to be very much a relationship 00:29:11.180 --> 00:29:13.440 -based conversation to have because it's a big 00:29:13.440 --> 00:29:16.099 decision to make. I'm not going to go off, completely 00:29:16.099 --> 00:29:18.980 go off of it. predictive model. But like I said, 00:29:19.000 --> 00:29:21.740 it provides a starting point, provides us an 00:29:21.740 --> 00:29:24.680 ability to strategize. And also from a funding 00:29:24.680 --> 00:29:27.920 perspective, models like this might help identify 00:29:27.920 --> 00:29:30.940 trajectories of individuals and their development. 00:29:31.369 --> 00:29:34.329 And if you have a relatively robust predictive 00:29:34.329 --> 00:29:38.269 model and their trajectory is moving to a performance 00:29:38.269 --> 00:29:42.029 zone, which is aligned to funding, then you can 00:29:42.029 --> 00:29:44.789 say, look, this is what we're predicting is going 00:29:44.789 --> 00:29:47.109 to be the middle zone. This is the trajectory 00:29:47.109 --> 00:29:49.650 of the athlete. In the next four years, it's 00:29:49.650 --> 00:29:52.130 very likely they're able to achieve this. So 00:29:52.130 --> 00:29:54.910 it's got multiple applications. Okay, now in 00:29:54.910 --> 00:29:58.849 terms of how the predictions you made were true 00:29:58.849 --> 00:30:02.569 or accurate or close to. accurate in terms of 00:30:02.569 --> 00:30:05.269 type of the competition was it easier to predict 00:30:05.269 --> 00:30:08.690 the totals for olympics awards or for europeans 00:30:08.690 --> 00:30:11.450 it was all easy to predict it's just how strong 00:30:12.089 --> 00:30:14.390 Well, how good were those predictions? For the 00:30:14.390 --> 00:30:16.670 first four performance zones, pretty decent. 00:30:16.869 --> 00:30:19.930 For the fifth performance zone, 11th to 15th, 00:30:19.950 --> 00:30:23.710 they weren't good. But generally speaking, there 00:30:23.710 --> 00:30:26.450 were more overestimations than underestimations. 00:30:26.490 --> 00:30:29.589 A potential reason for that could be the impact 00:30:29.589 --> 00:30:32.549 of COVID and people now just returning back into 00:30:32.549 --> 00:30:35.329 competition. So I'm very, very much excited to 00:30:35.329 --> 00:30:37.880 look at. what's happened this year and comparing 00:30:37.880 --> 00:30:40.420 it to our prediction model but also as i mentioned 00:30:40.420 --> 00:30:42.660 for the women's one where we developed that adding 00:30:42.660 --> 00:30:45.880 the current existing new category data and seeing 00:30:45.880 --> 00:30:48.279 if that has an even better ability to predict 00:30:48.279 --> 00:30:52.980 so was it better for 11 to 15 for the words because 00:30:52.980 --> 00:30:55.480 i i would assume that words would be the most 00:30:55.480 --> 00:31:00.299 stable in terms of yes yeah i'm in on the same 00:31:00.299 --> 00:31:02.619 opinion of that and i'm looking at it now it 00:31:02.619 --> 00:31:06.650 was the confidence intervals smaller okay but 00:31:06.650 --> 00:31:08.670 you still have wide predictive intervals but 00:31:08.670 --> 00:31:10.809 if we forget the predictive intervals for a second 00:31:10.809 --> 00:31:12.710 they were still narrower than they were for the 00:31:12.710 --> 00:31:15.170 olympic one but you're right in saying that they 00:31:15.170 --> 00:31:18.230 in theory should be more stable for worlds because 00:31:18.230 --> 00:31:21.809 you actually have the top 15 in the world so 00:31:21.809 --> 00:31:24.250 the variation will be far less in comparison 00:31:24.250 --> 00:31:27.470 to like i mentioned the olympics where you're 00:31:27.470 --> 00:31:30.720 not necessarily getting the top 15 in the world 00:31:30.720 --> 00:31:33.539 and i say top 15 is because that's the sample 00:31:33.539 --> 00:31:37.200 that we extracted for analysis i had played around 00:31:37.200 --> 00:31:40.640 with commonwealth championships and i mean that 00:31:40.640 --> 00:31:45.059 was incredibly varied but you expect commonwealth 00:31:45.059 --> 00:31:47.799 championships was extremely varied it's not something 00:31:47.799 --> 00:31:50.759 i i ended up pursuing because after doing it 00:31:50.759 --> 00:31:53.579 a couple of times I realized there's probably 00:31:53.579 --> 00:31:56.720 not a huge amount of value in doing it for this 00:31:56.720 --> 00:31:59.799 because the variations were so large, the confidence 00:31:59.799 --> 00:32:02.000 intervals and predictive intervals. So the only 00:32:02.000 --> 00:32:04.180 other way of maybe looking at that one would 00:32:04.180 --> 00:32:07.460 be just the medal zone, possibly. Certainly for 00:32:07.460 --> 00:32:11.180 England, where we are eyeing up medals at that 00:32:11.180 --> 00:32:14.619 specific competition. So possibly it becomes 00:32:14.619 --> 00:32:18.000 a bit more useful. Yeah. The sample is way smaller, 00:32:18.039 --> 00:32:20.619 even comparing to Europeans, less countries. 00:32:21.200 --> 00:32:23.799 it makes sense but you would hope that it's at 00:32:23.799 --> 00:32:26.359 least if you could only predict the metal zone 00:32:26.359 --> 00:32:29.640 it would still be worth it yeah i think so i 00:32:29.640 --> 00:32:32.160 think it would be and again it's only to open 00:32:32.160 --> 00:32:34.799 up the conversation of if first second and third 00:32:34.799 --> 00:32:37.380 will cross over you know it's like okay if you 00:32:37.380 --> 00:32:41.460 hit a 200 total that's a potential bronze medal 00:32:41.460 --> 00:32:45.200 but it also is the lower bound of the silver 00:32:45.200 --> 00:32:48.420 medal so hit 200 and you should be in focus out 00:32:48.420 --> 00:32:50.630 for a medal so you know it drops conversations 00:32:50.630 --> 00:32:53.410 and i think that's that becomes really important 00:32:53.410 --> 00:32:57.970 what a motivation yeah Yeah, exactly. And I think 00:32:57.970 --> 00:33:01.690 sometimes it also helps, potentially can help 00:33:01.690 --> 00:33:05.710 reaffirm what a coach is thinking in terms of 00:33:05.710 --> 00:33:07.849 what they should be achieving going into a comp 00:33:07.849 --> 00:33:09.809 or what they'd like to achieve going into a comp 00:33:09.809 --> 00:33:12.029 or what they think the medal zone will look like. 00:33:12.150 --> 00:33:14.630 Can we talk a little bit about the weight classes? 00:33:14.829 --> 00:33:17.769 Because weight classes are not every five kilos. 00:33:17.910 --> 00:33:20.930 They are differently organized in weightlifting 00:33:20.930 --> 00:33:24.490 because the population is spread. different like 00:33:24.490 --> 00:33:28.889 curves so were there any differences in prediction 00:33:28.889 --> 00:33:31.609 and strength of the prediction in terms of weight 00:33:31.609 --> 00:33:35.630 classes? Yes. So one of the issues we considered 00:33:35.630 --> 00:33:38.609 that we would have is on the two ends of the 00:33:38.609 --> 00:33:40.710 weight spectrum, the lightest and the heaviest. 00:33:40.910 --> 00:33:42.970 But then we realized the lightest probably doesn't 00:33:42.970 --> 00:33:45.049 matter as much because you can be as light as 00:33:45.049 --> 00:33:47.029 you want, but nobody is as light as they want. 00:33:47.130 --> 00:33:51.049 They're as close to the, let's say, 55 kilo threshold 00:33:51.049 --> 00:33:54.910 as possible. Whereas your 109 plus is you're 00:33:54.910 --> 00:33:57.779 going to have people weighing 115 kilos. and 00:33:57.779 --> 00:34:01.900 170 kilos. There's a huge span. That's why we 00:34:01.900 --> 00:34:05.599 opted to take the average weight of those that 00:34:05.599 --> 00:34:08.559 competed in the heavier top -end weight classes, 00:34:08.780 --> 00:34:10.880 because it's the best representation of that 00:34:10.880 --> 00:34:16.119 group average. So what we did see, I mean, I've 00:34:16.119 --> 00:34:18.440 got the tables in front of me only because I'm 00:34:18.440 --> 00:34:19.980 not going to remember these numbers off the top 00:34:19.980 --> 00:34:23.940 of my head, but the World Championships 109s, 00:34:24.000 --> 00:34:27.539 within that category, the performance zone prediction 00:34:27.539 --> 00:34:30.840 became... slightly worse. When you look at the 00:34:30.840 --> 00:34:33.900 same performance zone between different categories, 00:34:34.099 --> 00:34:37.300 it just varies. So for example, the medal zone 00:34:37.300 --> 00:34:39.800 for the 89 kilos, the difference between the 00:34:39.800 --> 00:34:42.829 prediction and the actual was three points. 40%. 00:34:42.829 --> 00:34:46.969 The prediction for the 109 plus was, I mean, 00:34:47.010 --> 00:34:51.510 just over 0 .05%, 0 .06. So it's pretty much, 00:34:51.530 --> 00:34:53.489 again, on the money. And remember, these are 00:34:53.489 --> 00:34:56.150 averages. So if anything, it tells you silver 00:34:56.150 --> 00:34:58.730 medal, not gold or bronze. So yeah, it does vary. 00:34:58.869 --> 00:35:02.590 There's no pattern I could see between the different 00:35:02.590 --> 00:35:05.230 weight categories at different performance zones. 00:35:05.389 --> 00:35:11.190 But there are less lifters in the 55 from what 00:35:11.190 --> 00:35:13.079 I remember. remember the lighter categories which 00:35:13.079 --> 00:35:15.739 means the data is kind of a bit more bit more 00:35:15.739 --> 00:35:18.639 spread oh could you could you repeat um i didn't 00:35:18.639 --> 00:35:22.380 understand the point in the 55 there wasn't in 00:35:22.380 --> 00:35:25.099 the world champs so what i didn't notice so in 00:35:25.099 --> 00:35:28.159 the world champs the 55s had the prediction to 00:35:28.159 --> 00:35:30.679 the actual the percentages were all above five 00:35:30.679 --> 00:35:32.739 percent difference which is quite big but the 00:35:32.739 --> 00:35:34.860 reason being it's not an olympic weight category 00:35:34.860 --> 00:35:39.199 right so and neither was 89 and neither was was 00:35:39.199 --> 00:35:42.769 102 at the time the data was collected. And then 00:35:42.769 --> 00:35:44.469 because people are changing weight category, 00:35:44.590 --> 00:35:47.289 naturally these values will end up changing. 00:35:47.610 --> 00:35:49.909 And that's why when you look at the paper, there's 00:35:49.909 --> 00:35:52.869 no difference between the predicted and the actual 00:35:52.869 --> 00:35:55.789 for the new weight categories at the Olympics. 00:35:56.090 --> 00:36:00.469 Because the 55, the 89 and the 102 will be contested 00:36:00.469 --> 00:36:02.630 in the up and coming Olympics, but they weren't 00:36:02.630 --> 00:36:04.429 in the last Olympics. I don't know. If you're 00:36:04.429 --> 00:36:07.050 a fan of weightlifting, like your head is spinning 00:36:07.050 --> 00:36:09.809 because every Olympic cycle we have changed. 00:36:09.929 --> 00:36:13.690 changes, how others qualify, what the weight 00:36:13.690 --> 00:36:18.730 classes are and so on. So strap in for the ride. 00:36:18.909 --> 00:36:20.889 No, I think it's just important to highlight 00:36:20.889 --> 00:36:24.309 that the predictive model paper is hopefully 00:36:24.309 --> 00:36:27.550 a starting point that can be developed further. 00:36:27.789 --> 00:36:30.739 It's certainly... a good opportunity to help 00:36:30.739 --> 00:36:33.719 strategize going into major competitions opens 00:36:33.719 --> 00:36:37.280 up the doors to try and apply this to other areas 00:36:37.280 --> 00:36:40.900 other sports but also you know masters competitions 00:36:40.900 --> 00:36:45.260 youth juniors and look at how that changes longitudinally 00:36:45.260 --> 00:36:48.559 as well and yeah i think the more the more data 00:36:48.559 --> 00:36:52.099 that we we end up getting hopefully the predictability 00:36:52.099 --> 00:36:55.199 ends up being a lot more refined and a lot more 00:36:55.199 --> 00:36:57.500 accurate and it's also really important to note 00:36:57.500 --> 00:37:01.260 that those had sanctions or were banned were 00:37:01.260 --> 00:37:05.099 extracted from the data when it was known that 00:37:05.099 --> 00:37:08.159 they had a ban in some cases there might be athletes 00:37:08.159 --> 00:37:11.000 in there who haven't yet been sanctioned or might 00:37:11.000 --> 00:37:14.039 be in the future but taking that one person out 00:37:14.039 --> 00:37:17.019 if it's in a medal zone for example could influence 00:37:17.019 --> 00:37:19.320 predictability but there's no controlling that 00:37:19.320 --> 00:37:22.699 yes so like i like i said earlier hopefully some 00:37:22.699 --> 00:37:23.980 you can develop something it's a little bit more 00:37:23.980 --> 00:37:27.219 malleable and as things change within the field 00:37:27.219 --> 00:37:29.889 the data changes in the process of the prediction 00:37:29.889 --> 00:37:33.550 ends up updating as uh as results update but 00:37:33.550 --> 00:37:36.869 i hope i hope people enjoyed the read and found 00:37:36.869 --> 00:37:39.989 it if not useful then at least interesting i 00:37:39.989 --> 00:37:43.070 can't ask you about your favorite color because 00:37:43.070 --> 00:37:45.829 i did it last time but i will ask you about your 00:37:45.829 --> 00:37:49.090 favorite lift and why oh that's a good question 00:37:49.090 --> 00:37:53.070 my favorite lift to do myself or to yes no you 00:37:54.179 --> 00:37:57.539 um i've got to say back squat because i i don't 00:37:57.539 --> 00:37:59.980 weight lift anymore i think i like back squat 00:37:59.980 --> 00:38:02.820 you know it's a good good exercise for for low 00:38:02.820 --> 00:38:05.119 body strength and you can you know add a bit 00:38:05.119 --> 00:38:06.940 of weight weight to it but it also humbles you 00:38:06.940 --> 00:38:10.360 very quickly as well might be a day where you 00:38:10.360 --> 00:38:13.659 don't feel so good and you'll find out very quickly 00:38:13.659 --> 00:38:16.320 so it's got to be a back squat a boring answer 00:38:16.320 --> 00:38:19.980 i'm afraid but no no in terms of coaching do 00:38:19.980 --> 00:38:21.980 you prefer coaching snatch or clean and jerk 00:38:22.989 --> 00:38:27.489 I suppose. Whatever the problem is in front of 00:38:27.489 --> 00:38:33.449 me, both are fun to coach and they both have 00:38:33.449 --> 00:38:35.610 their similarities, but it always depends on 00:38:35.610 --> 00:38:38.969 the person in front, for sure. So politically 00:38:38.969 --> 00:38:46.170 correct. I know, I know. If I were to choose, 00:38:46.489 --> 00:38:49.750 I'd say snatch, if I were to choose, because 00:38:49.750 --> 00:38:51.610 of the speed of movement and the intricacy. And 00:38:51.610 --> 00:38:55.769 I also think, Once someone understands the snatch, 00:38:56.050 --> 00:38:59.190 the clean tends to be developed a little bit 00:38:59.190 --> 00:39:01.150 easier. And that's, I'm talking about a complete 00:39:01.150 --> 00:39:06.789 novice here. Okay, I'm satisfied. We can move 00:39:06.789 --> 00:39:09.940 on to the next question. Okay, where people can 00:39:09.940 --> 00:39:12.480 find you if they want to connect or learn more 00:39:12.480 --> 00:39:17.739 about what you do? Yeah, so I'm contactable via 00:39:17.739 --> 00:39:22.219 email. So my email is s .chavda, that's spelled 00:39:22.219 --> 00:39:26.440 C -H -A -V -D -A, Violet Delta Alpha, at mdx 00:39:26.440 --> 00:39:31.500 .ac .uk. I'm also available on most social media 00:39:31.500 --> 00:39:34.760 platforms or academic platforms, so LinkedIn, 00:39:35.059 --> 00:39:37.849 ResearchGate. just stick my name in i'm sure 00:39:37.849 --> 00:39:41.510 i'll come up twitter my twitter is shy the number 00:39:41.510 --> 00:39:45.449 two tweet underscore to tweet was it shy underscore 00:39:45.449 --> 00:39:49.070 to tweet yep shy underscore to tweet and instagram 00:39:49.070 --> 00:39:53.510 is coach underscore awesome thank you so much 00:39:53.510 --> 00:39:55.750 for today you're welcome thanks for having me 00:39:55.750 --> 00:39:58.190 on a second time as soon as you publish the next 00:39:58.190 --> 00:40:00.510 one there will be a third one so be prepared 00:40:02.760 --> 00:40:06.519 I'm glad somebody's finding the research interesting. 00:40:06.860 --> 00:40:09.519 Yeah, of course. I live for research.