Rating inflation because of 1800 option as provisional rating

Hi everyone,

I've taken a relatively long break from chess and have noticed I frequently blunder. Subsequently, I've seen my rating drop from 2000-2100, to 1900ish on another site, yet my chess.com rating seems to be higher than it has ever been before. My opponents here seem to miss obvious tactics, use flawed tactics or simply just drop a piece frequently.

Now I've been wondering why this is the case, but I think I've found the cause of it. I've been paired against a few 1800ish players that played absolutely terrible(for the given rating), so I checked  their account and it turns out their rating was still provisional! Apparently, new players can now choose between different ratings represented by the following descriptions: new, beginner, intermediate, advanced and expert.

Now this system imo is fundamentally flawed to begin with, as beginner intermediate and expert are very relative terms. Besides this, I think that handing out different provisional ratings mean that players with accurate and relatively high ratings that are close to these "intermediate", "advanced" and "expert" provisional ratings get paired against overrated players too often(considering how many  players join this site vs how many people are accurately rated that high), making the rating less meaningful.

Anyone else think this is a problem? Let me know what you guys think.

Doesn't regular Glicko use only one relatively low rating for new players though?

The problem wouldn't necessarily be these new players being inaccurately rated, rather the players they are paired up against. As the amount of people that join this site is pretty large compared to the amount of people actually rated as high as the initial rating of these new players that describe themselves as "intermediate", "advanced" and "expert"

btw, I'm not saying these new players are necessarily arrogant, I think that for many of these new members, "expert" and "advanced" mean something different than to chesscom. 

If chesscom is going to give new members the option of different initial ratings, they should at least give these new members some tactic tests or something to kind of prevent people from just randomly getting a high rating. 

That simply isn't true. There is no difference in elo points earned playing against a provisional player and a regular player. 

Just look at my last rapid game(which is the reason I found out about this system).

I played against an extremely overrated provisional player and I still won 9 points!

I'll try to explain with some statistics of the rapid time control and math. The last 24 hours(when I checked at 6:41 GMT) 10,206 people have joined our humble chess community. There are 5 rating options these players can choose to start with, and I will assume for the sake of keeping this simple that the distribution is about even. 

So that is : 10,206(members) / 5(options) = 2041.2(members that have chosen any given category)

Now I am  #23,213 of the "world" with a rating of 1771.

From #3,375, these players are all over 1999 in rating though, so I will exclude these players as they probably barely play 1800 players. 

So that's #23,213 - #3,375 = 19838(players that will likely get paired against 1800 players).

(Mind you we have a few accuracy issues. For one, I do not know how many of these players are actually new players, but I haven't found any way to see this, so I'll simply ignore this. Second of all, I should not choose my own rating as the starting rating, but I cannot see other people's stats and looking through the world leaderboard will take an eternity. Third, I do not know how many of these provisional players of the last 24 hours, actually play rapid games. So I'd say, 7.5% with a deviation of 5%)

Then 2041.2 / 19,838 * 100 = 10%

This 10% means that of all the players that get paired against each other from 1771 to 1999, 10% of these games will be against provisional players! 

So an actual 1771-1999 player, will get "free elo points" every 10 games!

And the amount of players with a certain rating decreases exponentially so you can count on it that this number will only increase with higher ratings. 

I liked the old system where they just start all at 1200, which is kind of an average rating.

Nice work...but you'll never convince anyone .  I pointed out this problem when they first introduced 1800 as a self reported ratings choice.  You have a flood of players creating 1800 rated accounts and losing points, then bailing on that account and starting a new one, all in the name of trying to luck their way to , say, 2000+...just by getting enough other 1800 patzers in a row.  Every time these 1800 players drop hundreds of points and then bail, they are adding helium to the balloon.

In a few years when the ratings get crazy, they will probably fix the system but grandfather all the much-to-highly-rated accounts that already benefited from the mistake, because they won't want the uproar of dropping ratings across the board, and so, the ratings will remain permanently "off".

@universityofpawns Exactly, the point of these ratings starting at 1200 is to spread out these players over the members with the most common rating of this website, to prevent accurately rated players from encountering these provisional players too often!

@btickler, finally someone who understands! 

Your first argument I already mentioned myself, and a news article is not a reliable method of measuring how these new players are distributed. 

Your second argument is literally "you're wrong because who cares about rating inflation?!?" and then go on to say that the point of these ratings is to get paired with people that are of about the same skill level, yet fail to see how this type of inflation affects this probability of getting paired up against an evenly matched opponent.

"free points every 10 games at that level barely inflate anything over the long run" so the entire pool of players between 1771 and 1999 getting 10 points every 10 games would not affect inflation? I'm fairly sure you're just making that up, or did you actually calculate the effect?

The percentile only represents the ratio of your rating against all other members, not the amount of players actually rated a certain rating, nor is it compared to only active accounts. 

Besides this, new players on this site, are by far, mostly just beginners and people who barely play, so the percentage of players that are better than you will keep getting smaller without deflation or inflation.

Agree with Bobby T.....I've been playing here a long time, had a different i.d. long time ago when I was a tactical player new to the game (pinorforkit).....the ratings seem to go back to an approximation of what your "actual rating" is over time, even if there are some bumps along the way. Just plain old natural statistical variation, nothing fancy. Remember Bokonon says: "All of the true things that I am about to tell you are shameless lies."

I agree with the OP... this "self-selected 1800 provisional rating" is a legitimate cause for concern.

Consider the game he posted. Sure, he only earned 9 points by beating that player... but how many points would he have earned instead if the opponent had a "fair" rating of 1200 or so? Two points?

If players are earning four or five times too many points for a victory (or losing far too little with a draw or loss), then the rating system is sliding out of kilter.

BobbyTalparov wrote:. If you have 20M people in a rating pool, 1% of that is 200k. Your current rating is 1771, just under the starting point for new members who select "expert". If all these new players were selecting expert and inflating the ratings, your percentile would continue to get lower (meaning there would be more players better than you, not less).

This is assuming that the number of members on chesscom is static, which it is not, as a matter of fact, the rating inflation issue I am tackling in this thread is caused by new players joining.

"First, it is more accurate than literally making up numbers by assuming an even distribution when one does not exist. My second point was not an argument, rather pointing out that your worry is asinine. And using your own calculations,(roughly 10 points every 10 games which - assuming your other opponents were even matches, works out to be roughly 1 point extra per win), you see it is not much. If such inflation existed, there would be more players over 1800"

1. A news article without calculations is not more reliable than a pretty intuitive assumption(imo) with calculations based off of actual statistics. Also, you still assume yourself the distribution not to be equal, so kinda hypocritical?

2. If your second point is simply pointing out how it's not important, I don't see why it's used in a discussion in a topic about said issue, or let alone, why you're here in the first place. Also, importance is subjective and relative, you can't just dictate what is and what isn't important. Chess isn't really an important subject in the first place. Does that mean that we can't complain if chesscom has a poorly made website? Or if FIDE would just decide to randomly make their president world champion? Are we not allowed to ask/suggest them ways of improving, because "who cares it's just chess, man"?

3. First of all, you meant to say 1 point per game, not 1 point per win. Second of all, spreading out the points over multiple games doesn't lessen the inflation. If you play 100 games, you will have increased 100 points in rating. I'd say that is quite the inflation. It would mean that if these conditions were a constant, I would now be double the rating of stockfish. And yes this is based off of a hypothetical calculation that I myself have already said to be quite inaccurate, but the value of this is to illustrate the principle and importance of keeping new players to a single original rating.

4. You assume there aren't more highly rated players, but I don't know if this is the case. Have you checked? You can't really assume without any evidence. Now THAT is literally making stuff up...

I chose beginner rating because I am a beginner. Why some individuals would pick an expert rating when they clearly are not an expert is a mystery to me. However in the long run it doesn't make much of a difference.

While that is probably true, that is an assumption

Early on, I did some spot checks on new accounts and the vast majority of them were choosing the lower 3 tiers rating strengths. As others have mentioned, their RD is very high and they will quickly drop to where their strength should be. Established players will get less of a rating gain due to that high RD and their much lower RD. In the end, it will be pretty even and if there is inflation, it will be minimal.

As far as I am aware, the leaderboards (and rankings/percentiles) do not include players that haven't played games in a given time control. So non-playing accounts, for a particular pool will also have no impact on that pool, until they play a game and even then, due to RD, the impact will be very minimal.

A master doesn't want to slog their way up against 1200 players and 1200 players really don't want to be dominated against a player they think is also 1200.

Yes, for someone that is a beginner choosing 1800 is going to get dominated and the true 1800's will get an easier game but the players that do choose more accurately, which are very likely the majority, will get games close to their strength much more quickly than the old method where everyone starts at the same place.

@BobbyTalparov ummm ok

Seems to be pretty reasonable to me, but maybe I'm just crazy?

I mean do you really think it's fair to players if I was let's say... rated 2400 and said my rating was 800 to start out? That's not exactly nice.... Obviously I could say "here's my title" (I wish I had one lol) but the issue is to make sure people who are truly chess.com's version of 1200, are playing people around the same level. This won't stop trolls but it'll save people from quitting chess because someone "cheated" etc. This may have had a side affect in the inflation, but it is only people of lower skills. In daily chess, everyone I meet above 1600 relatively knowns their things. In blitz however, I've been playing 1700s who did not seem like 1700s (won both games in 14 moves). So it's a trade-off. Prevent people from being happy? or maybe feeding egos?