# Thread: Calculating Ratings: Order of games played

1. ## Calculating Ratings: Order of games played

I have another question concerning the calculating of ratings. I have a Glicko 1 spreadsheet for each player, where I enter results line by line, (which then updates their Rating and RD line by line).

I'll explain the following scenario, which hopefully will explain my dilemma:

I have Player A (John), who, at the start of a new rating period, played against (Fred) and defeated him. John has played 5 rated games so far (on my master file list he would be classifed as 4g, meaning 4 rated games left to play to get an 'official' rating) so I already have an internal rating calculated for him.

Fred is playing his first rated game against John, which he loses, as mentioned above. (Fred is classified as New on my master file list). I know, however, that Fred scores points in the rating period I am currently working on (with some favourable results later on) so I am deducing that I first need to calculate Fred's performance in this rating period (which includes a number of future games) before I can go back to this game and be able to calculate the (positive) effect of his loss against John's rating.

In other words, right now I cannot enter this game into John's rating spreadsheet because I don't yet have a rating for Fred (but I will do later)

I try to work ahead by working out Fred's results for the rating period (his other games) to find out his performance for the rating period but I soon become unstuck because:

Fred ends up playing against Harry later on. The problem is, Harry also ends up playing John previous to the Fred v Harry game. I can't calculate the Fred v Harry game until I calculate the Harry v John game. However, I can't calculate the Harry v John game because I haven't calculated the John v Fred game at the beginning: I am stuck in a loop.

I am guessing there is a policy/rule that is used in situations like this but I'm not aware of what it is.

Is it a case of having to disregard the result of the Fred v John game (at the beginning) to break the cycle or is there another solution?

OTHER NOTE: The critical factor here is that I am updating a players rating (and RD) match by match, which is also contributing to the problem, I realise. Is it incorrect to try applying rating calculations in this matter? Should a players rating only be updated tournament by tournament, rather than match by match?

EXTRA INFO: The tournaments being played are Swiss tournaments, if that helps.

2. If I understand the issue correctly, there are a few ways you can overcome this conundrum. Perhaps the easiest(?) is to estimate the new ratings (say 1500) run the calculation and see what the result is. Then take the final ratings and then input these as the "estimates" and rerun the calculations. Then repeat this step as often as necessary. At each step the output ratings should converge to the approximate inputs from the previous step. When that happens for all players affected then you have your answer.

The downside of this approach is that certain sets of data lead to unbounded problems (that never settle down to an answer). These are generally problems anyway (like players with no draw and either no wins or no losses).

3. Thanks for your suggestion, I'll see how it goes (though it seems like it might be a bit of a headache )

If Bill has had a look at my post I'd also be interested to see if he has any thoughts or any procedures to follow.

4. I performed one cycle of the calculations and I can see how this will eventually cause the ratings to (mostly) correct themselves. Unfortunately, it is too laborious a job to do manually (in my example the flow on effect that resulted because of 1 player's rating changed the ratings of 7 other players) . I'll need to forgo some accuracy, unfortunately, and apply some rule in this situation. I'm not sure what rule to construct though. Does anybody have any ideas about I could deal with it? It would be most appreciated.

5. I think the simplest approach is to not rate the result of any game where both players don't have a published rating (which I understand is what FIDE do) but obviously this means that the result information is lost for rating purposes.

A 'halfway house' alternative is to rate the game for Fred (using John's internal rating) but not for John (since Fred doesn't have an internal rating).

6. This free Excel add-on should do what you want, although I don't know how it handles unrated players.