Talk:Elo Rating System for football

Number of goals

In the current system the value of all goals exceeding the difference with more then three are the same. This doesn't seem fair to me. To my opinion a more fair formula for the goal difference index G would be (N is the goal difference; N > 0):

G = 2 - 21-N

Table of examples:

Goal Difference	Coefficient of K (G)	Proposal
+1	1	1
+2	1.5	1.5
+3	1.75	1.75
+4	1.875	1.875
+5	2	1.9375
+6	2.125	1.96875
+7	2.25	1.984375

In this formula the value of every extra goal decreases and the maximum multiplication value by extra goals is 2 instead of infinite as with the current formula. Otto 10:43, 6 May 2006 (UTC)

Your point is fair, the system is flawed, but I don't think Elo are concerned with preparing the formula for the eventuality of dealing with an infinite amount of goals ;-). Philc _T^E_C^I 00:51, 30 May 2006 (UTC)

If the goal difference is two, then you are multiplying by zero overall.

No, if you read the article, if the goal difference is two, you mulitply by ${\frac {3}{2}}$. Philc _T^E_C^I 10:18, 9 June 2006 (UTC)

I corrected the formula and added a table. The formula should now work for N>0. Otto 07:47, 10 June 2006 (UTC)

This article is hardly the proper forum to promote such a change. At any rate, it hardly seems fair that a team winning by a large margin would be limited to a double point gain. While such high scores are uncommon, they are certainly possible against lower-ranked teams.--BassoProfundo 17:26, 10 July 2006 (UTC)

Page Move

I think we should move the page to "World Football Elo Ratings" in accoradance with their website anyone opposed or have any points to make? Philc _T^E_C^I 12:38, 3 June 2006 (UTC)

Seems to me a good idea. Otto 15:45, 3 June 2006 (UTC)

how do you use the graph?

Pardon me if I'm a little slow, but how do you utilise the graph to extrapolate the expected result?

Ok well its annoyingly difficult ot describe but if the teams ratings are equal (i.e. diffrence = 0) then you read of from 0.5, and you get 0.5 so both teams get a W_e, and if the differences are 800 (the maximum) then you read off from 0.99, and the other team gets 0.10. all of the other results are spaced evenly between these two boundries (0.5 and 0.99) or below 0.5 for negative difference (i.e. the lower ranked team).

here are some examples.

Sample Winning Expectancies

Diff   Higher  Lower
0	0.500	0.500
10	0.514	0.486
20	0.529	0.471
30	0.543	0.457
40	0.557	0.443
50	0.571	0.429
60	0.585	0.415
70	0.599	0.401
80	0.613	0.387
90	0.627	0.373
100	0.640	0.360
110	0.653	0.347
120	0.666	0.334
130	0.679	0.321
140	0.691	0.309
150	0.703	0.297
160	0.715	0.285
170	0.727	0.273
180	0.738	0.262
190	0.749	0.251
200	0.760	0.240
210	0.770	0.230
220	0.780	0.220
230	0.790	0.210
240	0.799	0.201
250	0.808	0.192
260	0.817	0.183
270	0.826	0.174
280	0.834	0.166
290	0.841	0.159
300	0.849	0.151
325	0.867	0.133
350	0.882	0.118
375	0.896	0.104
400	0.909	0.091
425	0.920	0.080
450	0.930	0.070
475	0.939	0.061
500	0.947	0.053
525	0.954	0.046
550	0.960	0.040
575	0.965	0.035
600	0.969	0.031
625	0.973	0.027
650	0.977	0.023
675	0.980	0.020
700	0.983	0.017
725	0.985	0.015
750	0.987	0.013
775	0.989	0.011
800	0.990	0.010

--Philc _T^E_C^I 18:19, 7 June 2006 (UTC)

But the graph was $y=1-x$ not $y={\frac {1}{10^{-x/400}+1}}$. --Henrygb 21:46, 7 June 2006 (UTC)

Yes but if you look down the table the relation ship between the columns higher and lower is $y=1-x$. Philc _T^E_C^I 10:15, 9 June 2006 (UTC)

800 is not the maximum point difference. For example, compare Brazil with Palau on the ratings page.--BassoProfundo 17:31, 10 July 2006 (UTC)

Error in examples - loser should lose the same as the winner wins.

I might have missed something, but to me there seems to an error in the examples. The points lost by the losing team should always be the same as those gained by the winning team. At least, this is the case for all the matches I've seen at http://www.eloratings.net/. But, the tables in the examples seem to indicate different amounts for the winner and loser.

The error seems to arise because because a G value of 1 is assigned to the loser, when it should be the same as the G value of the winner, i.e. 1.5 for these examples. Rowandavies

Yeh your right, I just misenterpreted the system when I was wroking the tables out, feel free to change it, though I'll get round to it eventually if you don't but I've got alot infront of me right now. Philc _T^E_C^I 15:54, 14 June 2006 (UTC)

Done. Rowandavies 04:35, 15 June 2006 (UTC)

Thanks Philc _T^E_C^I 13:34, 15 June 2006 (UTC)

Which rankings should we use

http://www.eloratings.net/world.html http://www.eloratings.net/world_cup.html

Should we use the world rankings to update currently or the world cup rankings? W123 01:06, 16 June 2006 (UTC).

The first page linked to has the rankings pertaining to this article. BassoProfundo 17:10, 10 July 2006 (UTC)

Elo Ratings Calculator?

Hey,

Can I someone create an Elo Ratings Calculator that can be used to calculate domestic ratings? For example, it would accept a text document input of something like

ManU 3 Chel 1
Arse 1 Bolt 2

..and would output the overall table and the elo ratings? I know that a 3rd party magazine did this for american baseball, but I was wondering if someone was capable of programming something like this (preferably in C/C++) for other people to use...leave a message on my talk page, --Palffy 03:47, 11 July 2006 (UTC)

Why not do the calculations by hand? The formula is not that complicated. Once you get the hang of it, it should take less than a minute to calculate the new ratings after each match.

As the ratings work now, you would have to change the value of K to reflect domestic matches instead of internationals. You would also need a system for ranking teams as they are promoted and relegated.--BassoProfundo 14:27, 11 July 2006 (UTC)

Well, I could certainly work something out for the last two issues (such as taking Div 1 results into account only and only taking league matches into account--ie, no cup matches/friendlies), however, when you're plan on calculating 240+ matches/season (based on a H-A 16 team league), a program would certainly make this easier... --Palffy 21:06, 11 July 2006 (UTC)

There would still be some issues with that system; presumably you would want to keep the records continuously between seasons, but what about teams being promoted from the last season? And what happens to the ratings of teams that are relegated?

As for the program, I'm afraid that my knowledge of computer programming is far to limited to help you out.--BassoProfundo 02:54, 12 July 2006 (UTC)

Well, I think it would be a good first approximation to keep the ratings the same for the teams demoted from the premier league--ie, those teams would not be of interest in comparing teams (and would be kept separate from the competing teams). If the team would get promoted back to the division, it would simply retain the rating that it once had (which would be a low due to their previous poor season in the league---and would be near-equivalent to their strength level at that moment). What makes the ratings great, is that after 30 games or 1 season, the ratings would be properly calibrated for the club in question. As for clubs being promoted for the 1st time, they would start with an arbitrary ranking (comparable to those of clubs finishing near last in the league) similar to what the current elo rating has done (ie, countries start with an approximate rating and then achieve their real rating after 30 matches).

I could certainly try programming it myself, but my code is very inefficient and I would be unable to utilize the data effectively/efficiently that would allow me to calculate the ratings from 1000s of matches. --Palffy 03:36, 12 July 2006 (UTC)

I did something like this a while back. I parsed results from RSSSF premiership overview pages using regular expressions and incorporated them into an overall table using the ELO formulae. It still needs a bit of tweaking on the parsing and UI side of things, but it does the job. And yeah, I just assigned a provisional ratings to new clubs, like you suggest. Iae 14:26, 12 July 2006 (UTC)

You mind sharing this parser with everyone else? I tried to contact the guys who run EloRatings, and neither has responded to my requests..Quite unfortunate actually.. --Palffy 23:42, 13 July 2006 (UTC)

Yeah, of course. I've been meaning to just finish it off for a while anyway and i've got some free time for a while so i'll try and do that over the next day or two. Also, if you give me the format the files you want parsed are in (or, even better, e-mail me an example file, we shouldn't really be discussing this on the talk page anyway) then i can add some rules to allow it to be read. the more consistent the format then the easier it is to parse. Iae 17:52, 14 July 2006 (UTC)

What language are you programming the parser in? It would really be upto you as to how you want to make it initially...if its a language that I already know or I can pick up easily, I should be able to modify it accordingly (ie, I haven't really decided on how I'd want the input file to look like--that is, the file will probably require comments to separate seasons for easier browsing, but otherwise, anytype of format involving the names or a 3 letter code for teams should suffice followed by a score--after thinking about it, I think this would be the best format--its most similar to the data on RSSSF which I will use afterwards with the program..Post your email on your userpage if you'd like, and we can continue this over email..cheers!

ManU 3-1 Chel
Arse 1-2 Bolt

--Palffy 06:57, 15 July 2006 (UTC)

I've programmed it in .net because it's what i know best. I've never actually distributed a program before so i'm not entirely sure what you'd need for it to work, but at a guess you'd need the .net framework 2.0. Also, that format looks like it would work fine, and the parser gives you an option of converting any names (e.g. ManU to Manchester United) anyway so you wouldn't be forced to keep them in the 4 letter format. By the way, one small nuance is that each season has to go in it's own file and can only contain data for one competition. That shouldn't be too much trouble but just letting you know. Oh and you can e-mail me by clicking the 'Email this user' bit in the toolbox to the left of my userpage... i don't like putting it up available to everyone. Iae 12:22, 15 July 2006 (UTC)

I sent an email response as you requested =) --Palffy 20:10, 16 July 2006 (UTC)

FIFA verses ELO

I think it would be a good idea to put a side by side comparison of the top teams in Elo rating next to the top teams in the current FIFA system. Somthing like this

Top 20 FIFA Rankings compared to Elo
FIFA Rank	Team	Elo Rank
1	Template:BRAf	3
2	Template:ITAf	1
3	Template:ARGf	4
4	Template:FRAf	2
5	Template:ENGf	5
6	Template:NEDf	6
7	Template:ESPf	7
8	Template:PORf	9
9	Template:GERf	8
10	Template:CZEf	11
11	Template:NGAf	22
12	Template:CMRf	21
13	Template:SWIf	12
14	Template:URUf	14
15	Template:UKRf	24
16	Template:MEXf	16
17	Template:DENf	10
18	Template:CIVf	25
19	Template:PARf	28
20	Template:SWEf	13

24.237.198.91 05:58, 24 August 2006 (UTC)

I wouldn't be against this, it could even be implemented into the main table. However, I think there should be a specific section detailing the advantages ELO rating claims to have over FIFA (as opposed to having them dotted all round the article) and vice versa, as well as mentioning about the switch in FIFA system and how it changes things. Other views? Iae 10:13, 24 August 2006 (UTC)

Good idea. I've just added a 'FIFA rank' column. A section explaining differences would also be good, and some examples comparing the two. The examples used on the FIFA ranking page could be a starting point - or perhaps there are too many to be clear (I wrote the explanatory sections on the current FIFA system). -- Wantok 11:30, 24 August 2006 (UTC)