Germany will lose like 10 or 2o points. Canda has four wins which each gave them about 30 points. Wow, i guess they could reach rank 4 or so.
Really? Germany will lose points despite winning the whole thing? I know the final trophy has no importance whatsoever for FIFA rankings, but, despite the loss to Canada at group stage, the subsequent wins in knock-out rounds shouldn't have made up for that? Isn't there an additional factor for big international tournament as Olympics? After all, Germany are the only team who won all of their knock-out stage matches in regular time, wihout penalty shootout. I can't believe they lost points just vecause of one draw and one loss at group stage vs fellow-top-10-teams (Australia top 5 and Canada top 10).
They were the higher ranked team in the 3 knock-out thus didn't win many points. Here is the break-down: win v ghana (1487) : 0.2 points (friendly) ---- win v zimbabwe (1218) : -1.2 points (yeah, losing points... rating too low to gain points) draw v australia (2011) : -8.1 points loss v canada (1938) : -34.5 points win v china (1914) : 5.4 points win v canada (1938) : 11.1 points win v sweden (2002) : 11 points
Yes. The match importance factor for the Olympics is the same as the WWC: http://resources.fifa.com/mm/document/fifafacts/r&a-wwr/52/00/99/fs-590_06e_wwr-new.pdf I'm not the rankings expert though, so I don't know what that'll mean for the rankings.
Sob, I know it's just maths, but seems so sad to me... I suppose you factored in all the bonus tied to the Olympic tournament. There's no bonus for final matches or such, is there?
Germany's wins all were very close too, 1 goal, that's doesn't help the ratings as the goal difference counts. There is no bonus for the Olympic-final or so. All matches at the Olympics, group stage or final, are worth the same. And they are worth 4 times the points of a friendly match (double if both nations are in the top10).
Big move: Canada Ignoring the most recent friendlies leading up to the Olympics, Canada's rating could jump over 100 points, putting it around 2040. I expect them to be the 4th (and possibly 3rd) ranked team in the world when the new ratings come out! *** More importantly (tee-hee!), the ratings validated themselves in the small sample of the Olympics games. (Who cares who won, right?! tee-hee!) The median rating difference between the teams in all the matches was 155 points, which corresponds approximately to a (.709) expected win percentage or result for the higher rated team. * The actual win percentage (based on simple win/loss/tie; games which go to PKs count as ties) was (.712) The average game-chart result (based on FIFA's table using goals scored and not simply W/L/T) was a bit lower at (.679) * the mean average difference was (.748) Note - I used a Massey-derived rating for the two lowest ranked teams South Africa and Zimbabwe. This affects the mean average rating difference but not the median average.
posted this elsewhere ; this is where i should have typed it the first time : My rough calculation of rankings for September shows that pretty much everyone at the tournament lost a similar number of points, so there's little change in the top 20 apart from Canada up 6 places. The only other major positive team movement will be South Africa, on the back of the draw against the host nation. 1st USA -21 2nd Germany -15 3rd France -19 4th Canada +87 - up 6 places from 10th 5th England 0 6th Australia -18 7th Japan -5 8th Sweden -23 9th Brazil -29 10th Korea DPR 0 11th Norway 0 12th Netherlands 0 13th China PR -19 14th Spain 0 15th Switzerland 0 16th Iceland 0 17th Italy 0 18th New Zealand -3 19th Korea Republic 0 20th Denmark 0
Estimated performance ratings for the Olympics using the latest (pre-Olympics) FIFA ratings except for South Africa and Zimbabwe, where I used the Massey Ratings to calculate an equivalent FIFA rating. THe 2nd column is the team's performance rating in the Olympics. The 3rd column is the FIFA rating (or Massey-derived rating) nation perf. Rating FIFA rating (*except where noted) Canada 2168 1938 USA 2061 2168 Germany 2040 2115 France 1967 2064 Sweden 1964 2002 Australia 1916 2011 China 1908 1914 Brazil 1887 1982 New Zealand 1863 1848 Colombia 1812 1748 South Africa 1809 1693* Zimbabwe 1501 1508* FIFA's estimated homefield advantage is 100 rating points (which would apply to Brazil for these Olympics, of course). The official FIFA rating for South Africa is 1442 and for Zimbabwe 1208.
Hey @kolabear , I know you probably described it in a previous post, but would you be willing to start a thread on the Massey ratings and how you're using them? I just checked out the Massey website today, and I like their multi-faceted system, (would be fun to track on its own, hence the separate thread,) though I'm not sure how you're translating that to a FIFA-comparable rating.
Good question and, no, I haven't described it earlier. The way I'm doing it is using the "Matchup" feature of the website, where you can plug in two different teams and the website responds with its expected results, which of course can be converted into the Elo scale. Massey Ratings expected win percentages (at least at the time, which was sometime in July, before the Olympics) for teams like the US, Germany, France and a handful of others I checked were similar to what you get using the FIFA rankings, so if you take the expected win percentages of South Africa or Zimbabwe versus these teams you can get an Elo/FIFA-like rating number for them. Massey Ratings has, no doubt, been updating its ratings during the Olympics so the numbers won't work out the same as when I came up with 1693 for South Africa and 1508 for Zimbabwe. (Very very roughly, looking at the current Massey Ratings and assuming Germany, Sweden, China, and New Zealand haven't changed their ratings drastically, I'm getting a rough converted rating of about 1750 now for South Africa, up from 1693 and well over the last FIFA rating of 1440+) (and even more roughly I'm coming up with around 1550-1615 for Zimbabwe, up from 1508 and well over the last FIFA rating of 1208)
One example of getting a data point: According to Massey in the "Matchup" mode, on a neutral field, China is given a 59% chance of winning, South Africa 17%, leaving a 24% chance of a draw. Counting the draws as 1/2 point, that makes the expected win percentage .59 + .12 (which is half of 24%) = .71 which corresponds to approximately a 155 rating point differential. Assuming for now that China's rating of 1914 didn't change dramatically would give us one rough estimate of South Africa's rating as 1914 - 155 = 1759
You might want to consider the draws issue, which cpthomas is working on. How does that correlate to his data that 24% of the matches he tracks are upsets?
I'm not sure what to make of it at the moment. Rather than an overall percentage of upsets, since the Elo ratings imply higher percentages of upsets between teams with close ratings and lower percentages of upsets with further apart ratings, I've put more effort into seeing whether that correlation generally holds true when there's a major tournament like the World Cup or the Olympics. The results on the whole seem satisfactory to me.
I agree with this. The ~25% of games as upsets that I've cited refers to games in which the higher rated team does not win. Elo systems assign likelihoods of the higher rated team winning based on the rating difference between the opponents. My work says that for Elo systems, looking at all games regardless of the rating difference, the higher rated team does not win ~ 25% of the time. (I've actually done this calculation for a bunch of Elo systems.) But, that means for larger rating differences the higher rated team wins a higher % of the time; and for smaller rating differences, the higher rated team wins a lower % of the time. I will say, however, in reference to an earlier post, that a 24% likelihood of a draw, whatever the rating difference, doesn't match the data. For very closely rated teams, the draw likelihood is in the vicinity of 17%, and this is the highest draw likelihood I've seen. But, that doesn't mean that Massey's using 24% is "wrong." His system is predicting results of future games. His predicted 24% draws is what he's come up with as the best % to use. This is a higher % than actual results show, but his % may be the best % to predict, as that % may produce a better correlation with actual results than any other %. If that was too cryptic, I apologize. Perhaps the best thing to say is that it's very hard to predict ties.
I remember trying to figure out how Albyn Jones figured ties. I couldn't tie it to either rating differential or to a fixed percent, but I do remember that 17% worked an awful lot of times.
24% is an unusually high percentage of draws to expect, I agree, and I think it's unusual even within the Massey system. Since I accidentally chose an unusual example of a matchup, I should point out that the Massey system (as I understand it, which isn't much), measures separately different components of a team's strength, such as offense and defense, in making matchup predictions instead of taking one single number -- a team's Elo rating for instance -- to determine the probable outcomes. (Also, it calculates different homefield advantages for each individual team rather than an overall one to be applied equally to all matchups) In theory, as an extreme, I suppose it allows for intransitive predictions -- that is, in theory, it could predict A > B and B > C, but C > A; whereas this couldn't happen in an Elo rating. So for a team like South Africa, the system could predict a higher number of draws if that team seems to play a higher number of draws -- and from what we've seen of South Africa that may make sense, as they're not especially dangerous on offense but very organized on defense, not allowing a lot of goals. Speaking of Albyn Jones, I was thinking of bringing him up in the new Elo Ranking thread which someone started. I still would like to see someone do an Elo-ish rating using recursive techniques (hope I'm using the term accurately). Some/many of you know what I mean by it, but my brain's a little too tired and scattered to try to give a clear, simple description and I wish I could remember where I've tried to before. But the African teams again bring up its relevance. If, as seems the case, their ratings are too low, it is probably useful to have a method which takes their performance rating (especially South Africa's) in this event and loop it back to their earlier games within their confederation, reflecting the fact that this tournament seems to show they are a stronger opponent than the official rating indicates. I believe Albyn Jones' system involved just such recursive techniques or algorithms.
Is 24% of draws really that high? I'm sure it depends on what competition you're looking at, but for all the seasons of MLS, WPS, NWSL, and EPL I've looked at, the draws are usually in the 20-30% range...
Maybe 24% is reasonable, for games with no overtime. Looking at the NWSL right now, for this year, they're at 21.6% ties. One of the reasons South Africa may be showing a lower than expected rating is that (if) they've had a major transition in their program. Perhaps for a while they didn't take women's soccer seriously, but they've had a major change and have substantially upgraded their program. This would be like a new team coming into the system, except starting with South Africa's "not serious team" rating. It then would take a large number of games -- around 30 -- for them to reach their correct rating level. A way to see their likely true rating, in that scenario, would be to figure out where the change occurred. Then, after the change, take the post-change game results and add them into the system again enough times to get up to 30 post-change games. In an Elo-based system, there always will be a lag time if a program experiences a relatively sudden improvement or decline. The fewer the games the team plays, the longer the lag time.
That's a factor, of course. And of course another factor is the one that you especially have identified, particularly in the college game, which is when there are relatively isolated pools of teams which rarely play outside their group. Nothing of course can completely solve the problem (other than the teams playing a more diverse schedule) but a recursive method could/should help. So is 30 the "k-factor" or "k-coefficient" that FIFA is using? I should know this, shouldn't I?! (But then the multipliers based on the match's importance essentially reduces this k-factor for the games with higher multipliers like the World Cup or Olympics?)
In FIFA's documents, K is the weighted imporance you get when you multiply the match importance factor by the given constant - i.e. 15, 30, 45, or 60 depending on what match you play. I don't know if there's a technical term for the time lag it takes a team's rating to stabilize. It's not always 30 games, as it's dependent on all the constants you use to define your Elo system. More important matches will cause your rating to change faster, but won't necessarily decrease the amount of time it takes for your rating to stabilize unless you're starting rating is far from what your "true rating" is and thus have a lot of ground to make up.
So, the new rating is out. One thing bothering me is FIFA including teams back in the rating as best movers, that's silly i think. Further down Uganda lost 110 points to last rating, they only played 1 game (0:4 loss v Kenya). Kenya though (just this 1 game too) has the same points as before. Probably some obscure adjustment?
Well, I would agree it's a silly way of marking movers, but where are you seeing them talked about as such? On the official press release, the "biggest move by ranks" is Canada, since they popped up six spots. If you're looking at the raking pages itself (i.e. the little green triangles), I'm not sure how else you would display it... A little blue circle or something? Definitely odd... Since they just played a single game, neither team should have changed by more than 15pts. I'm guessing it's an oversight for both in some way.
Even further down we have a definite bug. Guam climbs 53 spots by their incredibly increase from 1287 to 1287 point, while Egypt falls 1 spot by doing the exact same thing.
Rating aside, Guam was "inactive" last time like Uganda was, so having their ranking jump makes sense. It's just that the displayed ratings are clearly messed up. Also: it looks like it's only the "new" ratings that are messed up, in all four cases, since the "old" ratings do in fact match the previous release. Egypt didn't play a game between this ranking and last, so it makes sense their points are the same, but just as Uganda and Kenya's point changes make no sense, Guam should be above 1287 now because they beat Macau 5-0 and should have earned a few points from that.