Interesting that this game was selected where RSL destroyed Toronto. To the casual observer, you wouldn't be wrong to think that although Bradley is a very good player, he isn't anything compared to Beckerman. Another thing that I find interesting is that Der Kaiser was at the game and watched how RSL and Beckerman took Toronto apart and the next national team game, what does he do? Tries out the RSL diamond with Beckerman playing his normal DM position and Bradley playing the AM position against Mexico. Although he wasn't really saying his mind, from that game, I would think that during the World Cup, we will be seeing the USMNT running the RSL diamond with Berkerman as DM, Bradley as AM and Cletus and Donovan as forwards. Agree or disagree?
He wouldn't fit on the plane. http://www.mlssoccer.com/news/artic...bred-players-celebrate-expansion-move-atlanta Wait for it (second 30).
I meant Dempsey, not our Cletus. Nice beard in a manly, lumberjack sort of way. But, you probably knew that anyway.
I thought it was a slip. I hadn't ever heard Dempsey called Cletus. Or is Cletus a general nickname for anyone named Clint? Either way, dude is fat now.
Has anyone else seen these? http://www.rslsoapbox.com/2014/3/19...hed-decade-for-american-soccer-is-hindered-by Who knew Utah had an identifiable soccer history?
Against the Portuguese, German, and/or Ghanaian midfields, it seems like suicide for 90 minutes. As a tactical adjustment, maybe.
Pray tell; what doesn't seem like suicide against those teams? Not disagreeing with you just not sure we have any good options.
Field the best players in the formation that suits them best. We might as well plan on losing in the group stage if we don't.
They wrote a follow-up to this original article today, trying to pin down why RSL are model-busters. It reveals some interesting stats: http://americansocceranalysis.wordpress.com/2014/04/18/looking-for-the-model-busting-formula/
No offense to the writers of those articles, but I wonder about the efficacy of a model like their's when a single variable can induce such a dramatic shift in expected outcomes. IMO they need to go back to the drawing board and completely revamp their thought process in terms of putting a model together at all. It's one thing to have outliers, it's another thing to have a team who completely breaks your model every year and then to go on and call them lucky or that some unexplained power is giving them the means to break your model....only to go out and do a simple analysis that shows that said team is just simply better at shooting the ball. Basically what I'm trying to say is that I put absolutely no weight or thought into what they have written and/or predicted. It's all hogwash at best. Take for instance this comment from the latest article: To me, in my scientific field, a correlation of 0.47 is awful, there is neither a positive, nor negative correlation with what the author is trying to establish. It's dead smack in the middle. But then he goes on to say this: So basically he is trying to explain away an already established stat and offer "bad news" in terms of RSL's expected goals based on a statistic that has a correlation of 0.47? Uhhhh, ok. I don't know, maybe I'm just reading this all wrong (and have a HUGE bias towards RSL, their stats, and their recent success), but to me the analysis and predictions these guys are doing is....grasping at straws that actually aren't even there. Maybe @BalanceUT can chime in, since he is actually a statistics professor and highly trained in it. In reality I think there are just too many variables for any person to establish a formula that predicts how a team is going to do, especially in MLS. Take the Timbers for instance, they brought in a couple of great looking players over the offseason, adding to what was already a good roster. Expectations were through the roof coming off a season where they had the second most points in the league and the most in the West. They look terrible, among the worst teams in the league right now, nobody saw that coming. This is what happens in MLS. Maybe this guys prediction formula is better suited for leagues that do not have extreme amounts of parity.
Just some random news if you haven't heard yet but go to your AAA office and they are giving away RSL scarves no purchase required, I just picked mine up pretty sweet.
So, what do you guys do with all the scarves? I've got a stack of 'em. STH, several random RSL, free ones from AAA, Continental Tires, America First, etc. During the majority of the summer games, one is already too hot. Do you all have "scarf walls" at your house and use them as wallpaper, or what?
I've seen several people on Reddit who display them on their walls. Some have had really cool designs, others loop them over a curtain rod. Personally, I fold them up and stack them on a shelf. I've usually wear my Inaugural Season scarf to games, but the others just stay on the shelf. The latest STH scarf is too thick to wear to games, IMO. Some of these guys have tons of scarves. I'm tempted to try and get one of the Believe scarves, but I don't know if it's worth the money. http://www.reddit.com/r/MLS/comments/1pv4ib/whats_the_best_way_to_hang_scarves/
That's the difference in the luxury of using data from controlled scientific experiments and the horrible reality of having to use observational data. You ought to relax a little more on what you expect from these kinds of models. Unless it's a time series, you're probably going to see poor correlations, low R-squared values and large standard errors -- at least compared to what you get from lab work. To be fair, they've got a lot of work to do, but you've got to start somewhere, and I think these guys have a nice foundation to build on. Nobody's going to have it 100% correct on their first try but I thought this is pretty impressive. It's more than I've ever been bothered to do, and I bet they can actually refine it to something pretty powerful.
UPinSLC, I had the same thought as you on reading the huge improvement in model (lacking any indication of the actual overall level of improvement, no R-squared, U-Squared or Chi-square difference test, only that it improved the accuracy for RSL) when they'd added what seemed to be a variable that they acted was easy to obtain… Or maybe it was a variable that they'd newly obtained, and plugged in and lookie-loo! Something interesting happened to the model and RSL is better improved. Yeah, they look like folks who've learned how powerful a statistical package can be to use without understanding the concept of modeling behavioral outcomes. And, I noted there was no accounting for dependencies in the data (the fact that a given team provides many data points and always in competition with other teams) which would necessitate a hierarchical modeling process (which maybe they did, who knows?). I will say that in this kind of behavioral work, a correlation of .47 is potentially pretty good. But, lacking tremendously in technical detail, I can't really assess it… except to say that the lack of technical detail leaves me not willing to come back to their site… in comparison to FiveThirtyEight.com, which I visit several times a week.
I was always under the impression that Cletus was a nickname for Clint, but the person also had to fit the nickname also. You wouldn't call Clint Eastwood "Cletus", but a redneck from the backwoods of North Carolina/Georgia who makes moonshine, hell yeah! Since Dempsey is from Nacogdoches, a relatively small town by Texas standards, I would think that the nickname Cletus would be appropriate. Anyway, I have a cousin from I-dy-ho whose name is Cletus, but I don't know if that is his given name or a nickname. Either way, he could pass as a backwoods moonshiner. And for those of you who care, the name Cletus is of Greek origin and was supposibly the name of one of Alexander the Great's generals. There was also a Saint Cletus who was elected pope in 76 AD. I guess he is the patron saint of mullets and moonshiners.....
I went to the internet for the answer as to what people do with all their soccer scarves and the answer is selfies. I however will save them up until I have a daughter and then that said daughter needs a prom dress and make her sew them into one awesome dress cause im not paying (calculate in inflation) 300k for a prom dress.
Isn't that why you just rent the dress for 1/4 of the cost? Or buy something from the internet and not pay ridiculous store prices.
I'm sure this is mostly a reflection of the difference in what our professional lives are, but I think their model is great, as long as you take it for what it is: an estimate of how many goals each team should score given several important factors such as location on the field, type of play and state of the game. No statistical predictor of a soccer match is going to have much accuracy because soccer is inherently noisy. But, that doesn't mean their model is useless. I learned a lot from it. For example, with RSL in particular the model helped us estimate that Rimando saves between 6 and 8 goals per season that the typical keeper in MLS wouldn't save. We also learned that RSL scores more goals than expected, and that led them to investigate why. They came up with a hypothesis (that RSL breaks the model because of better shot placement and because they hold a lead more often in games, resulting in less defensive pressure), and then they tested that with additional data. It turns out that those two things can explain most of the discrepancy. And now we have more to consider: why is RSL better at placing shots? Here again we can form a hypothesis and then try to come up with a way to test it. Isn't that basically what the scientific method is supposed to be? The point is that the original model helped lead to the next question, so it's extremely useful even if it's far from perfect. Also, just as a side note, a correlation of 0.47 is pretty strongly positive. A correlation of 0 means no relationship (and is thus dead smack in the middle). In my field (economics), we rarely see a correlation that high unless it is time series data.
Sure it would be nice to have some standard errors around their estimates and account for dependencies in the data, but like Allez said this is a nice start. And as far as building a hierarchical model, well, you have to remember that most of their readers aren't statisticians. You make it too complicated and nearly everyone stops reading. FiveThirtyEight does a great job of just striking the right balance between statistical rigor and readability for what they do. None of their stuff could ever get published in a serious academic journal, but it's sophisticated enough to be mostly credible. It's not easy to strike that balance, I don't think.
Hah! I had written something almost exactly like this (complete with high correlation with time series models), but erased it before I posted, because I didn't want to look like a dork.