FIDE Chess Rating Inflation

Update Rod Edwards (23 May, 2008)

Since I published the original article on FIDE rating inflation here two years ago, I have seen various suggestions as to the cause of the increase in the average rating of the 11th to 50th ranked players. I am not completely satisfied with any of them as yet.

I decided to post these results in the first place, not just because of the general increase in ratings (which many had observed) but because I found the sudden increase from 1986, after many years of stability, to be startling and rather convincing evidence that something was wrong. And the fact that the increase seems to follow a very simple curve suggests that there is some particular simple cause. Some of the suggestions I've seen could only account for a general increase in the top ratings, not the dramatic onset in 1986. This is my trump card when there is no other easy way to dispute a claim. Some of the otherwise more reasonable ideas in this category are:
  • The average rating of the 11th to 50th ranked players is not appropriate because there are simply more players. Comment: It is true that a larger random sample from a given distribution will increase the average of the top few. The flaw in reasoning here is to suppose that the sampling is random. It is almost certainly true that the more skill players have, the more likely they are to get involved in competitive chess and to stay in it, so players of lesser skill will be sampled less thoroughly, and if the overall numbers increase, the increase in numbers at the very top may still be small. Okay, there may be a certain amount of increase in the sampling even of the very top of the distribution over the years. But the fact that the average rating of the 11th to 50th ranked players in the world remained very stable from 1969 to 1985 indicates that the sampling of the top of the distribution was not increasing much over that period even though the number of rated players increased from a few hundred to a few thousand. And then there's my trump card.
  • The top players in the world are moving furthur from the bulk of average players (as a result of computer training or improvements in opening theory or...). Comment: This was suggested by Mark Weeks. It is true that this would result in an increase in the average rating of the top few players. This would be hard to disprove (though I have a yet-to-be-tried idea about looking at average rating changes from one rating period to the next, as a function of age, and looking for differences in these across time, which would avoid using the 'average of the top few' approach and could be applied to players in all rating classes). Anyway, I still have my trump card.
I don't include arguments that 'ratings should be increasing because chess skill is increasing' in the 'reasonable' category, because rating systems based only on results cannot, even in theory, reflect global advances in knowledge that are available to all. (And I'm not convinced that opening theory etc. has increased skill in general by as much as many claim. The study by Guid and Bratko, using computer evaluations of games themselves is very interesting in this regard.)

Among the suggestions for the sudden onset of inflation in 1986 are:
  • A whole lot of Soviet players, who had been isolated from the FIDE-rated pool, were infused into the system starting in 1986. Comment: Initially, I found this a plausible idea but I now doubt it could have had such a large effect. Essentially, the problem is that there could not have been enough players with enough of an over-estimated rating to have brought the entire international pool of FIDE-rated players up by over 100 rating points. The idea behind this suggestion is that if these players came into the FIDE rating pool with an initial estimate way above their 'true' rating (for which I know of no indication or reason), then they would lose points like crazy to other players, thus inflating the rest of the rating pool. (This could be checked, by the way, by looking to see if Soviet players who appear in the rating pool at this time immediately drop drastically in rating.) But whatever over-estimates there were, the excess points would be diluted in what must have been a larger pool.
  • Almost all women players were given 100 additional points in Nov. 1986. Comment: Aside from being a bit late - the inflation appears to start in 1986, whereas the 100-point increase for women only appeared in the Jan. 1987 list - the same argument applies as for the Soviet players. Even if it were the case that the 100-point increase were entirely unjustified (and I am not taking a position on this one way or the other) there would not be enough rating points injected into the rest of the pool to raise the entire pool by 100 points or more.
  • The rating floor for FIDE ratings has been lowered, and therefore widened, increasing the inflationary effect of the floor itself. Comment: The argument here, if I understand it correctly, is as follows. Players are generally slightly over- or under-rated because of normal statistical fluctuations. But players rated near the floor (below which FIDE does not rate players) are more likely to be over-rated than under-rated, because if they would be under-rated they are simply not on the list. This creates a slight inflationary effect, since this excess of slightly over-rated players then loses their excess rating points to other players, and these rating points gradually diffuse throughout the entire pool. Presumably, this has always been the case, but the difference is that with the rating floors lowered, the number of players near the floor is now vastly larger than it was. On the other hand, the entire pool through which the excess points diffuse is larger too. I do not know, without doing some analysis or simulations, how this trade-off would pan out. And I don't know the exact timing of FIDE's lowering of the rating floor. This has happened several times, I understand. If anyone can fill me in on this I'd appreciate it.
Currently, this last idea seems like the most plausible explanation, if only because I don't have a better one. (I did try to contact FIDE but got no response to my email.)

Original article (edo dot chess at yahoo dot ca)