- games data set in format:

PlayerA1, PlayerA2, PlayerB1, PlayerB2, PointsATeam, PointsBTeam - and starting ELO list in format:

Player, ELO

In principle ELO rating system is based on this simple idea:

- 1.

If two players (or teams) playing game and each has some power, the difference between powers of this 2 players determining how many points (1 point if win, 0 if lose and for example 0.5 if draw) each player has to do. We call this (in %)**norm**form player. For example if Player P1 is one class (200 ELO) better that player P2 than it means P1 from 100 games will win 76 and lose 24. - 2.
**Games data set**(number of games)**has to be enough big**(then from probability of win in 1 game we going to statistics) to be statistically important, then we can calculate from such games data set the relative power between players, we call it**performance**. The first performance based on enough big data set we can declare as first rating (ELO). - 3.

When we have some rating ELO list, and players playing new games, the rating of players is changing depend on his actual power (some players are improving his ability to play the game and going be better).

(Calculated additive 1 step way)**the new rating = old rating + K*(number of points that player win - norm for this player based on his old rating); where K is usually 15**. How going time, rating of players changing and converge to be mirror of real power of players. This self-organization of rating is also very important. - 4.

If number of players in data set is very big (like whole world), we can say that the power of average player is 2000. (This is then saying the absolut value of ELO, because basically rating in general is relative thing, which saying that this player is better that this and that better that another and saying the rating difference between players, but nowhere is determine if best player rating should be 2500 or 3500; The**2000 as "gravity centre" for world rating system**is just a useful convention.) - 5.

Also this rule is important : If some player played on calculated games data set too big quantity of games and we use additive way to determine his new rating, we have to check if his new rating is not out of border made by his performance. New**rating**for such player has to be determined the same way as**for new players without rating : as his performance**. Or better : calculate ELO more often to decrease size of games data set for aditive calculating new ELO. For example: If Player with rating 2000 will play 1000 games with performance of 2020 his new rating cannot be bigger than 2020, but is set on 2020.

If You need calculate chess ELO from bigger set of games, or maintain Your own chess club rating, You can download for free ELOCalculatorDynamic207.zip (79KB) - when You have set of games results played inside some group of players.