I actually think we know that that was counted as one dribble and that the part at 5:11 was counted as a dribble. That’s because it occurred to me that we actually have data on what minute of the game and where these dribbles happened: https://www.whoscored.com/Matches/847439/Live/International-FIFA-World-Cup-2014-Germany-Argentina Go to that link, go over to the “Chalkboard,” then click on “Dribbles” below the pitch map, filter it to just show Messi (using the player selection stuff on both sides of the pitch), and then go to “Outcome” part at the bottom of the page and filter it just to “Successful.” Now we have a map of where and when each successful Messi “dribble” occurred in this match. The video we have been looking at doesn’t have the match time on the video, but I found a full-match video and cross-referenced things, and it looks to me like the successful dribbles are what I thought they were: Which is to say it is the dribbles you listed in your post, except only one dribble in the 9:30 sequence and a dribble for the 5:11 sequence.
I do agree that football algorithms are devoid of context and in any one example can miss a mark, but please rationalize in what ways are football algorithms so wrong and maybe give an example of clearly misevaluated rating. You have a strong stance against football algorithms but i dont remember you ever clearly explaining why other than ridiculing it by default.
I know that last post was not directed to me, but I’ll give some thoughts, since I’ve been thinking about this recently: I think the value of all-in-one ratings depends on how the formula for the rating is created. One can imagine two different ways to create a formula. The first is to just assign arbitrary weighting to various different actions. This can tell us something in general, particularly when there’s huge differences in score, but isn’t all that useful because the weighting itself is arbitrary and unmoored from anything but someone’s personal feeling about how much something is probably worth. The second is to actually build a model that weights things in a way that is designed to be predictive of out-of-sample match results. As in, you run some sort of regression with all the various statistics and variables you have, and ultimately use weightings on those variables that correlate well with the actual results of matches. This is still not perfect by any means, but at least the weightings would be made in a principled way—i.e. the value given would be related to how much value that sort of action seemed to have in a large sample of matches. To me, the second option is quite a bit better than the first option. It’s not clear to me which bucket SofaScore and WhoScored fall into. I generally assume it’s the latter, but I don’t know for sure and their descriptions of their ratings are far too general to tell what they’re doing. Even that second option isn’t perfect though. For instance, it inherently can only assess the value of discrete actions, and a lot of things happen that aren’t discrete actions that can be assessed like that. Also, even within the realm of discrete actions, the usefulness of this sort of thing depends on the detail of the data. For instance, let’s say we run a regression and it tells us that a “dribble” on average has about X value. That by itself is a really rough measure, because some dribbles are a lot more useful than others. So just using that weighting to value all “dribbles” would overrate a game where a player had “dribbles” but they weren’t particularly dangerous, while it would underrate someone who had few dribbles but the ones he had created a ton of danger. So, ideally, the model would incorporate a lot of detail/context about the action—i.e. where on the pitch it occurred, etc. That way, the model would spit out an appropriately higher value for a dangerous dribble than a non-dangerous one. I think these models do some of that—I definitely know they incorporate location data into it, or at least WhoScored specifically says they do—but the more variables are incorporated the more accurate the kind of regressed model I talked about would actually be.
Do scouting reports refer to Whoscored/Sofascore? Do coaches refer to Whoscored/Sofascore? Do technical analysts refer to Whoscored/Sofascore? If the answer to any of these is yes, there may be a shred of credibility. Otherwise, an equation is fundamentally arbitrary, both in terms of the variables picked, the weighting of said variables, and the fact that you cannot derive any sense of the value of the tabulated actions, nor any sense of decision-making, and activity within the wider narrative and context of the game. Like I don't just think it's 50% rubbish. I think it is COMPLETE rubbish. And Carlito knows better than to bring up Zidane or whatever - I don't even have the slightest sense what his ratings (or other players' ratings) are.
That makes sense. That means that an overall margin of error of rating is an accumulation of margins of error for every action. On average, players for any specific action like dribble fall bellow or above expected, average value assigned to action. If a player consistently produces less value through dribbling than an average player (due to his style or whatever), algorithms will favor him. For any specific game, such margin of error becomes more pronounced so it is to be less trusted. I've always thought that football algorithms favor Neymar and disfavor Modrić for example. I am not sure what that tells us about algorithms, but I think it is plausible to study some examples and reverse engineer which actions algorithms tend to misevaluate or not include at all. Decision-making is an obvious example, but it is difficult to codify in what ways and how much impact it has. Actually, I believe they do use algorithms of some kind to access a big pool of players. I wouldn't be suprised if certain teams have developed their own, more sophisticated algorithms for evaluating performance.. There are certainly examples of successful recruiters in football that use data to evaluate players such as Brentford. (EDIT: I realize now that the video doesn't really talk about data in football, but it is from a guy who worked for Brentford and in other videos discusses data) So your point is misleading. Scouters probably do not use sofascore per se, but most likely is a symbiosis of studying advanced football data with common sense and understanding of the game. Exactly, what we are doing here. Btw, I think football algorithms, for all the flaws they have, are much better, and certainly more conssistent, at evaluating players than an average football fan is because humans are so easily swayed by outcome bias and the whole list of other biases. No football algorithm would rate Varane as the 7th best player in 2018. Lol.
It's not misleading. The question is do they use Sofascore and Whoscored. That is the discussion at hand. I never said they don't use data - data you are not privy to and are not part of our convos.
Still only 21 so time is on his side That said the last ones like him (Messi aside who is incomparable)were a lot more productive at his age http://www.espn.com.br/noticia/2967...cours-e-iguala-feito-so-obtido-antes-por-pele https://amp.sportsmole.co.uk/football/man-utd/news/hazard-wins-ligue-1-player-of-the-year_27532.html
In this context, the fallacy being employed seems more like a combination of several fallacies rather than a single one, but let's break it down: Argument from Ignorance: The person is implying that because there's no evidence provided that scouting reports, coaches, or technical analysts refer to Whoscored/Sofascore, these algorithms must lack credibility. They're essentially arguing that if there's no proof that these sources are used, then the algorithms must be worthless. False Dichotomy: By presenting only two options – either these sources are referred to and have credibility or they are not referred to and are completely rubbish – the person is oversimplifying the situation. They fail to consider the possibility that these sources could have some utility or credibility even if they are not universally used by scouting reports, coaches, or technical analysts. Ad Hominem: The person dismisses the credibility of these algorithms by attacking them rather than addressing their actual utility or methodology. By stating that they are "COMPLETE rubbish," they're attacking the credibility of the algorithms rather than engaging with the evidence or arguments supporting them. Appeal to Authority: The person seems to dismiss the value of these algorithms by suggesting that because they don't have information about whether high-profile figures like Zidane use them or not, they must lack credibility. This implies that the credibility of the algorithms is dependent on whether respected figures in football utilize them, which is not necessarily the case. In summary, the fallacies in this argument stem from drawing conclusions based on a lack of evidence, oversimplifying the situation, attacking the credibility of the algorithms rather than addressing their actual utility, and suggesting that their credibility is contingent upon the actions of high-profile figures. ChatGPT doesnt like you.
2005-2009 greatest dribblers Cristiano Ronaldo 2006/07 CL 86 dribbles attempted 46 dribbles completed 53% dribbling accuracy Lionel Messi 2007/08 CL 86 dribbles attempted 43 dribbles completed 50% dribbling accuracy Ronaldinho 2004/05 CL 64 attempted dribbles 41 completed dribbles 64% dribbling accuracy Lionel Messi 2008/09 CL 91 attempted dribbles 37 completed dribbles 40% dribbling accuracy Ronaldinho 2005/06 CL 83 attempted dribbles 36 completed dribbles 43% dribbling accuracy Cristiano Ronaldo 2007/08 CL 88 dribbles attempted 36 completed dribbles 40% dribbling accuracy Thierry Henry 2005/06 CL 46 attempted dribbles 35 completed dribbles 76% dribbling accuracy Ricardo Kaka 2006/07 CL 77 attempted dribbles 18 dribbles completed 23% dribble completion rate Total number of completed dribbles Cristiano Ronaldo 2006/07 CL 47 Lionel Messi 2007/08 CL 43 Ronaldinho 2004/05 CL 41 Lionel Messi 2008/09 CL 37 Ronaldinho 2005/06 CL 36 Cristiano Ronaldo 2007/08 CL 36 Thierry Henry 2005/06 CL 35 Ricardo Kaka 2006/07 CL 18 Dribbles completed Per 90 Ronaldinho 2004/05 CL 6.11 Lionel Messi 2007/08 CL 5.37 Cristiano Ronaldo 2006/07 CL 4.47 Lionel Messi 2008/09 CL 3.59 Thierry Henry 2005/06 CL 3.39 Cristiano Ronaldo 2007/08 CL 3.21 Ronaldinho 2005/06 CL 3.02 Ricardo Kaka 2006/07 CL 1.42 Dribbling Accuracy/Completion rate Thierry Henry 2005/06 CL 76% Ronaldinho gaucho 2004/05 CL 64% Cristiano Ronaldo 2006/07 CL 53% Lionel Messi 2007/08 CL 50% Ronaldinho gaucho 2005/06 CL 43% Cristiano Ronaldo 2007/08 CL 40% Lionel Messi 2008/09 CL 40% Ricardo Kaka 2006/07 CL 23% https://www.sofascore.com/player/kaka/1770#tab:matches https://www.sofascore.com/player/lionel-messi/12994#tab:matches https://www.sofascore.com/player/cristiano-ronaldo/750#tab:matches https://www.sofascore.com/player/ronaldinho/3581#tab:matches https://www.sofascore.com/player/thierry-henry/11#tab:matches
.... World Cups History ... Dribbles completed ... . 1 Lionel Messi .... 26 matches ... 125 dribbles completed ... . .2. Mané Garrincha .. 12 matches ... 109 dribbles completed 3. Diego Armando Maradona ... 21 matches ... 106 dribbles completed ... 4. Jair Ventura Filho Jairzinho ... 16 matches ... 79 dribbles completed .... 5. Edson Arantes Pelé ... 14 matches ...... 78 dribbles completed ... 6. Eden Hazard ... 65 dribbles completed... 7. Mario Alberto Kempes 18 matches .... 61 dribbles completed .. ..!!!