Mmm, interesting, so now tell me : if you previously wrote this: ..this should mean that we are not counting on Sweden alone to surpass them, are we?
oh gosh, my very rough back-of-the-envelope calculations show Germany losing around 55 points in the World Cup (and another 12 for losing the friendly to Zambia) I think they'll stay above the 2000 mark though.
Alright, now that the R16 is over, I want to do BotE calcs for everyone too and post them - and these will be VERY BotE as I'm ignoring any games with teams beating someone significantly lower (in theory, less than 3pts of movement) and ignoring the pre-WC tune-up matches and I'm rounding, but here's what I've got so far... Groups A/C: NZL (1800**) 1-0 NOR (1908) ; exp. 0.35-0.65, act. 0.85-0.15 ; pts. 60*0.5=30 ; NZL (1730), NOR (1878) PHI (1513) 0-2 SUI (1766) ; negl. ESP (2002) 3-0 CRC (1597) ; negl. JPN (1917) 5-0 ZAM (1298) ; negl. PHI (1513) 1-0 NZL (1830**) ; exp. 0.86-0.14, act. 0.15-0.85 ; pts 60*0.71=43 ; NZL (1687), PHI (1566) SUI (1766) 0-0 NOR (1878) ; exp. 0.34-0.66, act. 0.47-0.47 ; pts. 60*[0.13;-0.19]=[8;-11] ; SUI (1774), NOR (1867) JPN (1917) 2-0 CRC (1597) ; negl. ESP (2002) 5-0 ZAM (1298) ; negl. SUI (1774) 0-0 NZL (1787**) ; exp. 0.48-0.52, act. 0.47-0.47 ; pts 60*[0.01,-0.05]=[1,-3] ; SUI (1775), NZL (1684) NOR (1867) 6-0 PHI (1566) ; negl. JPN (1917) 4-0 ESP (2002) ; exp. 0.38-0.62, act.0.97-0.03 ; pts 60*0.59=35 ; JPN (1952), ESP (1967) CRC (1597) 1-3 ZAM (1298) ; exp. 0.85-0.15, act. 0.09-0.89 ; pts 60*0.76=46 ; CRC (1551), ZAM (1344) SUI (1775) 1-5 ESP (1967) ; exp. 0.25-0.75, act. 0.04-0.96 ; pts 60*0.21=13 ; SUI (1762), ESP (1980) JPN (1952) 3-1 NOR (1867) ; exp. 0.62-0.38, act. 0.89-0.09 ; pts 60*0.27=16 ; JPN (1968), NOR (1851) Groups B/D AUS (2020**) 1-0 IRL (1744) ; negl. NGA (1555) 0-0 CAN (1996) ; exp. 0.07-0.93, act. 0.47-0.47 ; pts 60*[0.40,-0.46]=[24,-28] ; NGA (1579), CAN (1968) ENG (2041) 1-0 HAI (1475) ; exp. 0.96-0.04, act. 0.85-0.15 ; pts 60*.011=7 ; ENG (2034), HAI (1482) DEN (1866) 1-0 CHN (1854) ; exp. 0.52-0.48, act. 0.85-0.15, pts 60*0.33=20 ; DEN (1886), CHN (1834) CAN (1968) 2-1 IRL (1744) ; negl. AUS (2020**) 2-3 NGA (1579) ; exp 0.93-0.07, act. 0.17-0.83 ; pts 60*0.76=46 ; AUS (1874), NGA (1625) ENG (2034) 1-0 DEN (1886) ; exp. 0.70-0.30, act. 0.85-0.15 ; pts 60*0.15=9 ; ENG (2043), DEN (1877) CHN (1834) 1-0 HAI (1482) ; negl. CAN (1968) 0-4 AUS (1974**) ; exp. 0.49-0.51, act. 0.03-0.97 ; pts 60*0.46=28 ; CAN (1940), AUS (1902) IRL (1744) 0-0 NGA (1625) ; negl. CHN (1834) 1-6 ENG (2043) ; exp. 0.25-0.75, act. 0.02-0.98 ; pts 60*0.23=14 ; CHN (1820), ENG (2057) HAI (1482) 0-2 DEN (1877) ; negl. AUS (2002**) 2-0 DEN (1877) ; exp. 0.67-0.33, act. 0.92-0.08 ; pts 60*0.25=15 ; AUS (1917), DEN (1862) ENG (2057) 0-0 NGA (1625) ; exp. 0.92-0.08. act. 0.47-0.47 ; pts 60*[-0.45,0.39]=[-27,23] ; ENG (2030), NGA (1648) Groups E/G USA (2090) 3-0 VIE (1649) ; negl. NED (1980) 1-0 POR (1745) ; negl. SWE (2050) 2-1 RSA (1472) ; exp. 0.97-0.03, act. 0.80-0.20 ; pts 60*0.17=10 ; SWE (2040), RSA (1482) ITA (1847) 1-0 ARG (1682) ; exp. 0.72-0.28, act. 0.85-0.15 ; pts 60*0.13=8 ; ITA (1855), ARG (1674) USA (2090) 1-1 NED (1980) ; exp. 0.65-0.35, act. 0.50-0.50 ; pts 60*0.15=9 ; USA (2081), NED (1989) POR (1745) 2-0 VIE (1649) ; negl. ARG (1674) 2-2 RSA (1482) ; exp. 0.75-0.25, act. 0.51-0.51 ; pts 60*[-0.24,0.26]=[-14,16] ; ARG (1660), RSA (1498) SWE (2040) 5-0 ITA (1855) ; exp. 0.74-0.26, act. 0.98-0.02 ; pts 60*0.24=14 ; SWE (2054), ITA (1841) POR (1754) 0-0 USA (2081) ; exp. 0.13-0.87, act. 0.47-0.47 ; pts 60*[0.34,-0.40]=[20,-24] ; POR (1774), USA (2057) USA now only 3pts above SWE, well within margin of error NED (1989) 7-0 VIE (1649) ; negl. ARG (1660) 0-2 SWE (2054) ; negl. !!! this result is literally just worth1pt of movement if I actually calc this; USA still ahead by 2pts, ofc within margin of error RSA (1498) 3-2 ITA (1841) ; exp. 0.12-0.88, act. 0.83-0.17 ; pts 60*0.71=43 ; RSA (1541), ITA (1798) NED (1989) 2-0 RSA (1541) ; negl. basically same story as ARG 0-2 SWE here SWE (2055) 0-0 USA (2057) ; negl. BUT I'll say exp. is 0.5 both, actual is 0.47 both, so both lose 2pts; this match helped neither SWE (2053), USA (2055) Groups F/H FRA (2027) 0-0 JAM (1537) ; exp. 0.94-0.06, act. 0.47-0.47 ; pts 60*[-0.47,0.41]=[-28,25] ; FRA (1999), JAM (1562) BRA (1995) 4-0 PAN (1483) ; negl. GER (2062) 6-0 MAR (1334) ; negl. COL (1703) 2-0 KOR (1840) ; exp. 0.31-0.69, act. 0.92-0.08 ; pts 60*0.62=37 ; COL (1740), KOR (1803) FRA (1999) 2-1 BRA (1995) ; exp. 0.50-0.50, act. 0.80-0.20 ; pts 60*0.30=18 ; FRA (2017), BRA (1977) PAN (1483) 0-1 JAM (1562) ; exp. 0.39-0.61, act. 0.15-0.85 ; pts 60*0.24=14 ; PAN (1469), JAM (1580) KOR (1803) 0-1 MAR (1334) ; exp. 0.94-0.06, act. 0.15-0.85 ; pts 60*0.79=47 ; KOR (1756), MAR (1381) GER (2062) 1-2 COL (1740) ; exp. 0.86-0.14, act. 0.20-0.80 ; pts 60*0.66=40 ; GER (2022), COL (1780) PAN (1469) 3-6 FRA (2017) ; negl. but fun! JAM (1580) 0-0 BRA (1977) ; exp. 0.09-0.91, act. 0.47-0.47 ; pts 60*[0.38,-0.44]=[23,-26] ; JAM (1603), BRA (1951) KOR (1756) 1-1 GER (2022) ; exp. 0.18-0.82, act. 0.50-0.50 ; pts 60*0.32=19 ; KOR (1775), GER (2003) MAR (1381) 1-0 COL (1780) ; exp. 0.09-0.91, act. 0.85-0.15 ; pts 60*0.76=46 ; MAR (1426), COL (1734) FRA (2017) 4-0 MAR (1381) ; negl. COL (1734) 1-0 JAM (1603) ; exp. 0.68-0.32, act. 0.85-0.15 ; pts 60*0.17=10 ; COL (1744), JAM (1593) = = = = = Knowing that my margin of error is probably 5-10pts, the rough current ranking of all teams that started over 1800 (active in italic underline) are: USA (2055) -35 SWE (2053) +3 ENG (2030) -11 FRA (2017) -10 GER (2003) -59 NED (1989) +9 ESP (1980) -22 JPN (1968) +51 BRA (1951) -44 CAN (1940) -56 AUS (1917) -3 DEN (1862) -4 NOR (1851) -57 CHN (1820) -34 ITA (1798) -49 KOR (1775) -65 A surprising amount of top teams have absolutely bled points, (maybe not so surprising with all of the group stage craziness this WC, especially JAM and the CAF teams raking in points,) with only JPN making any significant gains among top teams GER in 5th and poised to fall further as one of NED/ESP will have to win that match, and JPN also climbing fast... USA is either in 1st or 2nd right now, and could fall to 3rd - but considering that one of the "top four" teams in the WC will likely lose twice (i.e. finishing 4th), I don't think the USA falls below 3rd, and might not even fall at all if none of SWE/ENG/FRA make the SFs. CAN also could fall out of the top 10 if AUS makes a deep run.
Also, I gotta say, now that I've run the numbers, I was *way* off on basically all teams that I thought would be gaining points. With the exception of JPN, everyone I said would gain points has more closely just "broken even", within the margin of error. I guess the GDs in most of those wins weren't enough to overcome the large differences in ratings....
Looking back at this, I think GER is actually already down in 6th as their two pre-WC friendlies (a weak 2-1 win over VIE and that shock 2-3 loss to ZAM) probably cost them another 15pts or so. They could very easily be 7th when all is said and done.
JAM (1537 to 1593) - #43 up to ~36 NGA (1555 to 1648) - #40 up to ~32 RSA (1472 to 1541) - #54 up to ~43 MAR (1334 to 1381) - #72 up to ~66 ZAM (1298 to 1344) - #77 up to ~72
Somebody (and that somebody has to include FIFA) has to seriously think how to reset the ratings for the teams from Africa. FIFA is using the ratings (or rankings) for the purpose of seeding at the World Cup and for qualifying playoffs. Generally speaking, it's not a bad way because, generally speaking, the rating system is pretty good. But it's now a problem because it doesn't give African teams a fair chance at a seeded spot. Also, it's becoming a pressing problem because it isn't fair to the non-African teams that play them — they don't get properly rewarded for victories and... they get hammered in the ratings for a loss. *** The way to do it that comes to mind is to begin by basing the ratings like Nigeria and South Africa which have played in World Cups on their performance ratings against teams outside their confederation, if not exclusively then by heavily weighting them beyond the current weight they're given beyond regular friendlies. Then the adjusted ratings of these teams can be used to calculate the performance ratings of other African teams that played them in official World Cup (or Olympic) qualifying matches.
Yeah, IMO the biggest problem with the FIFA women's rankings is how they treated the "original" ranked teams vs anyone who joined the rankings since. The initial ranking from 2003 had an average team rating of around 1500 for the 108 ranked teams, which tells you that 1500 was the default rating for a newly ranked team. Now, in 2023, the average team rating is closer to 1300 with 188 teams. Small nudging with draws aside, the rankings are *supposed* to be zero-sum, so the average rating of teams shouldn't fall that much *unless* teams that entered the ranking since the original weren't given the same default ranking. Crunch the numbers, and the ~80 teams that have entered the rankings since 2003 have, as a group, an average that's around 1000 instead of around 1500. I can kinda understand the logic there, as "new" teams are almost by default going to be noticeably weaker than established teams, but it gives you problems down the road. CAF has gone from 15 ranked teams in 2003 to 45 ranked teams in 2023. Granted, CAF's original 15 had an average well below 1500 anyway, but it's a *huge* historic weight to the CAF teams that 2/3 of the confederation started at a lower level than the ~100 teams in the original 2003 ranking. This, combined with the issues previously discussed with NGA's lone performances against other confeds holding the rest of CAF down, makes CAF severely underrated. *** I wouldn't even say we need to use "performance" ratings necessarily. FIFA has reset the men's ratings a few times, so IMO a more straightforward thing to do would be to pick a cycle - say, the beginning of the 2011 cycle so we have more than 10 years of data - and just reapply the entire ranking process starting from that point *and* giving everyone the same starting default rating.
SiberianThunderT has an excellent point. Beyond that, however, I do not see unfairness towards the African teams in the rating system. Their federations have chosen not to support their women, so they play virtually no matches outside of Africa. This is unfair to the African women, but it is because of choices their federations have made and not because of the rating system. This always is going to present a rating problem, no matter what system is used. Without enough inter-federation competition, no system can properly rate teams from one federation in relation to teams from other federations.
Fairness to the federations — admittedly corrupt or incompetent in many cases — isn't the main point. Fairness to the women, the players, is much more important. Also, the "fairness" aspect is now spilling over, unfairly affecting teams outside CAF, the African confederation — the other teams around the world in Europe, Asia, the Americas, Australia — who play African teams who appear grossly underrated. England will lose about 30 points for the (official) tie against Nigeria. If Nigeria was rated, say 1855 instead of 1555, England would lose 15-20 points.* The other thing, which perhaps I should've stressed over the "fairness" issue, is simply that we want the ratings to be as credible and accurate as possible. Most of us here on BigSoccer's WoSo threads think (rightly) that FIFA's rating system is, on the whole, pretty good; the big sore thumb which can damage its credibility is the underrating of the African teams. I don't see how we can choose to ignore it, to not try to fix it or ameliorate the problem. I don't know the best way to do it. Maybe the "Siberian Solution" will work, rerun the ratings since 2011 but changing the provisional ratings of new teams to 1500 rather than 1300 (if indeed that is what FIFA really did; how FIFA handles new teams is not something I know about). * another example, for the benefit of @blissett (!), is Sweden and South Africa. Sweden will lose (in my purely back-of-the-envelope calculation) about 8 rating points for only defeating South Africa by one goal. If South Africa was rated about 1772 (a more realistic rating) than 1472, the net effect of the game would be about zero.
Time for an update~ ESP (1980) 2-1 NED (1989) ; exp. 0.49-0.51, act. 80-20 ; pts 60*0.31 = 19 ; ESP (1999), NED (1970) JPN (1968) 1-2 SWE (2053) ; exp. 0.38-0.62, act. 20-80, pts 60*0.18 = 11 ; JPN (1957), SWE (2064) AUS (2017**) 0-0 FRA (2017) ; negl. same story as USA-SWE earlier; both teams lose 2pts AUS (1915), FRA (2015) ENG (2030) 2-1 COL (1744) ; negl. calc'd is a +/- 2pt swing in this case ENG (2032), COL (1742) ESP (1999) 2-1 SWE (2064) ; exp. 0.41-0.59, act. 0.80-0.40 ; pts 60*0.19 = 11 ; ESP (2010), SWE (2053) AUS (2015**) 1-3 ENG (2032) ; exp. 0.48-0.52, act. 0.09-0.91 ; pts 60*0.39 = 24 ; AUS (1891), ENG (2056) Looking at the top 16 present again, and knowing that my margin of error is probably 5-10pts, we have, ENG (2056) +15 SWE (2053) +3 USA (2052) -38* FRA (2015) -12 ESP (2010) +8 GER (1988) -74* NED (1970) -10 JPN (1957) +40 BRA (1951) -44 CAN (1940) -56 AUS (1891) -29 DEN (1862) -4 NOR (1851) -57 CHN (1820) -34 ITA (1798) -49 KOR (1775) -65 *Here I've taken the liberty of docking USA and GER points for USA's "weak" win over VIE and GER's shocking warm-up matches. Thankfully for GER, JPN and NED both losing their respective QFs has slowed GER's fall even with the warm-up matches considered. We're also seeing the "downside" to hosting with AUS's relative difficulty in gaining points - they made it to the semifinal and can still win bronze, but they'll drop out of the top 10 (and CAN survives)! Yes to both! It's incredibly close at the top with just 4pts between 1st and 3rd - well within my margin of error. Both ENG and SWE need an outright win in their remaining matches to remain above USA, as even a draw against lower-ranked sides will lose them points and could push them below the USA (though margin of error could save them in those cases). USA don't deserve #1 after a three-draw performance, so it would definitely be weird seeing them still on top. And on the flipside of the currently-close margins, either of them winning their last match will put some distance between them and USA.
The lack of enthusiasm for resetting Africa's ratings is palpable On the other hand, there's great interest in seeing the USA knocked out of first place in the rankings — and one thing that may keep it from happening is the drag on England and Sweden's ratings from playing African teams in the World Cup (!) According to @SiberianThunderT back-of-the-envelope calculations, England lost 27 points in the 0-0 tie (which went to PKs) with Nigeria; Sweden lost 10 points in their 2-1 victory over South Africa. Off the top of my head, if Nigeria had a more realistic rating going into the tournament of 1855 (instead of the last published 1555), England would've saved about 10 of those 27 points; if South Africa's rating going in was 1772 (instead of 1472), Sweden also would've saved about 10 points. That's 10 points for both England and Sweden in their (much-lauded!) quest to topple the USA in the rankings! So be careful what you're not enthusiastic about Preliminary to suggesting a way to reset Africa's ratings, I very roughly calculated the performance ratings of the African teams in the World Cups going back to 2007 and compared them to their official rating before them (last published rating prior to the World Cup). 2007: Nigeria published rating 1705 / performance rating 1910 Ghana published rating 1511 / performance rating 1420 2011: Nigeria published rating 1672 / performance rating 1980 Equatorial Guinea published rating 1396 / performance rating 1630 2015: Ivory Coast published rating 1373 / performance rating 1520 Cameroon published rating 1455 / performance rating 1820 Nigeria published rating 1633 / performance rating 1850 2019: Nigeria published rating 1599 / performance rating 1540 South Africa published rating 1485 / performance rating 1550 Cameroon published rating 1499 / performance rating 1790 2023: Nigeria published rating 1555 / performance rating 1990 South Africa published rating 1472 / performance rating 1820 Morocco published rating 1334 / performance rating 1860 Zambia published rating 1298 / performance rating 1710 *** the performance rating is the rating which would predict exactly the results obtained by a team, using the game results (converted using the FIFA rating table) and the ratings of the opponents (strength-of-schedule)
The other idea I had for "resetting" the rankings, if you didn't want to go back a full three cycles and set everyone to 1500, would be to a scaled resetting on a more recent ranking and go from there. My specific idea is thus: 1. Take the rankings from 2020 when everything came screeching to a halt 2. Within each confederation, rescale the teams so that the top team has a rating of 1800 and the bottom team has 1200. (e.g. USA, GER, AUS, BRA, NGA, and even NZL would all have a rating of 1800, with the likes of MEX, NED, JPN, and RSA would be in the 1760s, etc.; would be sad days for ARG/COL/CHI, but I bet they can make it up) 3. Just run the 3yrs of results since 2020 as normal In theory, it should have been long enough since the pandemic stop so that the top nations would be back to around 2000pts, and it wouldn't disrupt UEFA/AFC/C'CAF much, but it would vastly help the top CAF teams. It would certainly bring the net average back up to 1500, correcting for having so many "late" teams coming in at 1000 to bring the overall average down to the 1300 we currently see. I don't quite have the time right now, but it's a small enough set of matches that it shouldn't be *too* hard to program up, if I ever get the chance...
It would be an interesting project. One issue I can see is insufficient data for some teams and federations. It take somewhere between 25 and 30 games to get a good rating for a team. In addition, there have to be enough crossover games among federations, but I do not know what that number is although I am pretty confident 15 percent is not enough. Further, the crossover games of a federation probably need to be distributed among the other federations and not just with one of them (especially if that one has its crossover games all with the same partner federation). There is a way to take the current ratings, put together with past game locations and results, and test whether the rating system is properly rating teams from each federation in relation to other federations. I have done this for US college Division 1 women’s soccer, both in relation to conferences and regions. The problem for national teams would be not enough games. For the D 1 women, I have a data base of about 45,000 games, which I break down into the most closely rated 20 percent, the next most closely rated 20 percent, and so on. Then, within each quintile, I look at a region’s out-of-region games and how the game results compared to the opponents’ ratings as adjusted based on game locations. Since I am splitting into quintiles and then looking only at out-of-region games, it necessarily requires a lot of data to get reliable results. For a full description of how I do this, go here. Still, it would be an interesting project and might provide some useful information about whether the federations are rated properly in relation to each other.
The crossover is definitely an issue, though the idea between setting all confeds on an 1800-to-1200 scale was that things *seem* to be pretty equal across the board right now. (We would definitely want more crossover to be confident, though, and getting crossover in general is one of the issues that plagues WoSo to date.) Plus, since the starting point here is based on the existing ranking, it's not like each team would be "starting from 0 games" either, so I'm not *too* worried about that aspect either.
I don't know when I'll get a chance to work on it so I'll mention now an idea for resetting the ratings of teams from Africa. It involves going back to somewhere like 2003 to 2007 and recalculating the ratings using asymmetric weightings for matches between African and non-African teams. For instance, we might double the weight given to those matches but only for calculating the ratings of the African team. (Doubling is just a guess; I haven't ruled out the possibility of even tripling the weight) The system assigns World Cup and Olympic games four times (4X) the weight of a typical friendly. Doubling the weight would mean they would have eight times (8X) the weight of a friendly, but only to calculate the African team's rating. For their non-African opponents, the 4X multiplier still applies. (This also would only apply to matches between African and non-African teams but so far I believe that's every World Cup or Olympic match that African teams have ever played) The rare friendly between an African team and a non-African team would be weighted twice (2X) the typical friendly, again only for calculating the African team's rating. @SiberianThunderT proposed that simply recalculating the ratings and changing the provisional rating given teams without a prior rating would help solve the problem. It's certainly worth looking at. As of right now I have no idea how far it would go to properly recalibrating the ratings of African teams compared to the rest of the world. There's of course many here at BigSoccer who think a reset is unnecessary. They're certainly entitled to their opinion but so far the objections I've heard are trivial and unconvincing so I'm going to proceed on the assumption that we need to do something about it. *** Most folks here are familiar enough with the ratings to know how the system weights games differently depending on their importance, but as a reminder World Cup & Olympic games are weighted four times as important (for the ratings) as a typical friendly. The multiplier for World Cup and Olympic games is (4X) World Cup & Olympic qualifiers; and continental championships (like the Euros and the African Cup of Nations) are weighted three times (3X) Continental qualifiers, qualifiers to continental championships (like the Euros and the African Cup of Nations) are weighted two times (2X) Typical friendlies have a weighting of one, their multiplier is (1X) Inter-confederation playoffs for World Cup qualification would also have their multiplier doubled, under this preliminary idea, for African teams (but only for the African team). The multiplier for the African team would be (6X)
Interesting idea. I wonder how many games there are (without quadrupling them or something like that). The risk, if the games are very few, is that one game then can have tremendous weight. In a game like soccer, with all of its upsets, it could skew things too much -- in either direction. But still it would be worth at least looking at what the ratings look like and how they compare to actual game results -- the latter of which would be critical.