Average opponents' rating bug?

Sort:
Elroch

13 months ago, the average opponents' rating was removed from the chess statistics, then returned in response to many complaints. After it was returned several people complained that the calculations were no longer correct. I cannot independently verify these observations, but I suspect this may be related to the gross errors now visible in this statistic for some players.

For example, the average opponent rating for my online chess is hundreds of points higher than what chess.com displays as the average opponent rating. I see a similar huge inaccuracy in the average opponent rating of my current opponent in an ongoing online chess game.

In my entire time on the site I have only played one or two players a few points below the displayed average rating, and the other opponent ratings go up to about 1000 points above the supposed average. The displayed supposed average of online chess opponent ratings is completely unrelated to their actual ratings. Certainly no other data should be included in this statistic, so it is definitely wrong.

I would expect that if such a clear bug was reported to any company that was concerned about having an image of integrity and professionalism, it would be treated seriously and dealt with promptly. What does chess.com manage? Well, several other people reported this bug 13 months ago. There are likely many others who have done so in the intervening year.

How can it be that chess.com does not have access to a single programmer who can implement trivial arithmetic correctly? Or does this really demonstrate the complete contempt for user feedback over the past 13 months that it appears to?

I should make clear that the reason I am posting this is that I posted a bug report in a conscientious and helpful way, but have not received a professional response. The person who responded had clearly not looked at the numbers at all, but made a wild (and completely incorrect) guess as to an explanation. I replied to this response, explaining why the explanation made no sense and that there was a real bug to be dealt with. I have not received a response to this communication. Is this really the sort of incompetent impression chess.com wants to give to users?

Scottrf

I've just checked with my blitz games and they are correct.

Won Lost  
1454 1269  
1267 1250  
1156 1323  
953 1318  
1193 1312  
1240 1170  
1052    
1016    
     
9331 7642 16973
1166.375 1273.667 1212.357

Perhaps you're including unrated games whereas they are not?

Elroch

scottrf, the fact that your average opponent rating for blitz is correct is interesting, but does not affect the fact that my displayed average opponent rating for online chess is inaccurate by around 400 points.

Amusingly you made the same guess as the chess.com admin. However, I have played only a handful of unrated games, and just under a hundred rated games, and this is definitely not the explanation. [Moreover almost every opponent in both categories had a rating well above the supposed average].

One factor that may be relevant is low frequency of play. But while this is  likely to be related to the reason for the statistics sometimes being completely wrong, it is certainly not a way to argue that they might be correct (my recent opponents have been all been hundreds of points higher rated than the erroneous average).

Scottrf

From a quick scan yours does almost certainly appear incorrect.

Martin0

Mine for chess960 must be wrong. Only looking at my statistics makes it look wrong (I should have a higher rating if it would be true). I have played 2 games with 1 opponent near the average rating, but everyone else is clearly below (I don't have the statistics though). When I won against the highest rated player I met he had 1971 and at the other game he had 1991 (same as average rating). The highest rated opponent I have played is the same as the average opponents rating. Notable is that that game was the last chess960 game I have played.

ilikeflags

this is a scandal

Martin0

Mine would be right though if it would be average opponent rating last 90 days (with only 1 game available), but that doesn't seem to be the case for everyone (and if that would be true it would still be wrong to not tell that it is based on the last 90 days).

Some time ago (maybe a half year ago) I noticed several current games stats was wrong. I sended a report and they answered they had "synchronized my games" and then my stat became correct, but still wrong for several other players. I'm not sure what synchronize mean, but I think it should have been right in the first place and no need to synchronize specific games.

Metastable

Interesting. Mine seems roughly correct. I tried to verify it, though. The stats summary shows that my average online opponent is 1531. But when I download my games archive ("Download all games as one PGN"), cull out the 960 games and live games, and take the mean of the opponent's Elo score in the PGN, I get 1499. This would be roughly since last June or so.

nola2172

I wonder if perhaps they are not correctly calculating games against opponents who have had their account banned for cheating (which upon checking, you seem to have had the frequent misfortune of participating in).  As you get in the higher ratings on here, the number of games you play against those eventually banned seems to go up, so if chess.com counted those as being worth zero or some other number that is not the rating displayed (or using their eventual rating after all their games timed out), that could be causing the problem.

ilmago
 
 

Some of the stats refer to just the period of the last three months.

This has been so for the timeout percentage, for example.

Possibly, this is now also the case for some other statistics numbers?

 

Also, I can imagine that  completely recalculating every single statistics number every single time a game is finished - or even every time a move is made - would be taking up quite a lot of server time, so maybe instead, there are now some more efficient algorithms in place instead, which give exact or at least roughly correct numbers for most cases while using much less calculation time?

 
Bubatz

I think statistics does take into account the average opponent rating during the last 90 days only.  

Elroch

My displayed average opponent rating would be incorrect by a few hundred points over any period, including the last 90 days, since about 98% of my opponents have higher ratings than it (most of them several hundred points higher).

ilmago
 
 

Okay, so that looks like something different than the ideas mentioned so far.

Recently, chess.com has been changing the structure of their game databases, putting together games from Online Chess and Live Chess for some purposes.

Maybe some kind of mix-up with live chess statistics could account for the differences observed here?

 
Timotheous

My wild guess is that there are several interconnected calculations involved based on how they have many features of the site generated, and that they are aware of some glitches, but are working on a way to fix one glitch without generating a dozen more. So a single type of calculation, looked at in isolation, may seem easy to fix, but perhaps the actual programming involved is much more complicated.

Martin0

At least the last 90 days would make sence for my chess960 rating, but not your rating Elroch. Maybe there are an error in the calculation if you haven't finnished any game within the last 90 days against anyone with a non-closed account. At least thats the only theory I can come up with that sounds a bit likely.

Elroch

I suspect you have identified a key factor in the offending statistic, Martin0. I am not sure, but it is possible that I have not finished any games recently against players who have not had their accounts closed (I have played very few games recently - only one or two at a time, and one of my most recent opponents got banned for cheating).

It could be that in this situation, the average opponent rating statistic gets replaced by one that is nothing to do with online chess, perhaps based on blitz chess or something. Why they would do this is a complete mystery to me - I suspect it would have to be an unintentional side effect of more general code.

I wonder if the statistic will suddenly change when I finally finish my sole current game that is not against a banned opponent. Smile If this is so, I'm less bothered about this problem.

Bubatz
[COMMENT DELETED]
Elroch

Update on this bug. The average opponent rating did change when I finished the game. However, the day after it changed back again to nonsense.

Surely it won't be long before a chess.com software expert figures out where their software is going wrong and fixes it?

Martin0

It surprised me it could change to be right over a day and then go back again. Seriusly some staff should be able to understand the problem and fix it as soon as possible.

Elroch

Surprised me too. I am looking forward to someone finally sorting this out.