Rapid Rating vs. Average Game Accuracy Correlation

Hello all!

I am interested in the correlation between rapid rating and average game accuracy.

If you meet these criteria:

1. Played at least 25 games in the last 30 days.
2. "analyzed" all of your recent games in the last 30 days so that the 'accuracy' is populated in your game history

please feel free to post your current rapid rating and your average accuracy from the last 30 days (by selecting the '30 days' option in your rapid rating stat page and finding the appropriate statistic) into this thread.

After I get enough peoples' responses, I'll compile the information and show the trends.  It'll be an interesting thing to see.

I’m 1425 rapid and on average score 81% accuracy consistently in the last year:

You are probably already aware of the rough correlation between Elo rating and accuracy, yes?

re: that picture

I have extreme doubts, lol.

It matches my Rapid accuracy very well.

Of the last 100 rapid games you've had 7 analyzed (I didn't count games with 10 or fewer moves).

Your CAPS average from those games is 64, which is not on that chart.

I was just looking at my Rapid 7, 30 and 90 day average accuracies without excluding any matches. Which range from 72% to 74%.

Keep in mind that was when they were analysing OTB classical games with it.  That doesn't correlate well with rapid or blitz online.  For example, Levy Rosman is an IM (~2400 FIDE) and the last 30 days has a CAPS of 81 in his blitz games (putting him in the 1600 range in the graph).  Obviously that is nonsense.

The left columns are titled "Typical OTB" so obviously not onlne.

Still, it's a ridiculous table, but since most people are ridiculous themselves it's fine. It does its job as a marketing gimmick.

I believe that graphic was based on an older version of the algorithm, so don't know how well it translates to the newer version.

Agreed, and when they came up with it several years ago, many of us thought it was nonsense then.  There were countless threads like this one discussing it.

Not a lot of useful information from the above posts, but I was able to personally find many players and plot their rating vs average accuracy.  A numerical rating value is somewhat arbitrary based on the mathematics used in rating calculations.  For example, since rating values are arbitrarily based on the constants used in the logistic function to calculate rating changes each game, Magnus Carlsen could have a rating of 2800 or 1000000 depending on the constants used.  The below chart shows that rating is actually an indirect measurement of average accuracy, with some variability of course.

Pretty cool!

It's interesting that the quadratic term in the best fit line is extremely small, meaning that average accuracy is mostly linearly related to rating.

The y intercept is 53.06, meaning someone with the lowest rating on chess.com rapid has an accuracy of only about 50% (simply put).

The trendline also suggests average accuracy increases about linearly at about 1.81% per 100 rating points.

Easier said than done, but if one would like to increase their rating 100 points, they only have to be play a just a bit more accurately over many games.  There's some variability in the chart above, but that's the general trend.  This information suggests that it's more important to try to play more slowly between moves to increase your accuracy and increase your rating, even if you loose from time-loss more often.

This growth pattern looks a lot more logarithmic than linear to my eyes however? Which is would intuition would tell you as you approach ratings above 1800

I think the linear fit is a good approximation for the rating ranges that most people fall in, however with more data at the higher end the quadratic term would very most likely increase

R squares 0.90 is not indirect. It sounds like an error, but the chart shows rather direct correlation. R² 0.4 would be significant, 0.9 is like comparing the same thing over time.

Ofc if you make less accurate moves, a la worth 2 pawns, probability of losing increases. That's kind of the whole idea of chess.

Because accuracy doesn't mean much outside of high level classical chess. 
In all other cases rating doesn't matter anymore than time control, playstyle, and large factors that happen in some games but not others. Some of these large factors can be long-drawn out endgames, a long series of forced obvious captures that might end up being the correct line. An opposite colored bishop game is the most obvious and classical example. 
Long story short your accuracy tells you more about these external factors and less about your overall strength, which is something that your rating does. 

My rapid rating is 2031 and 30 day average accuracy clocks in at 77.8. 

edit: Nvm. I haven't played 25 rapid games in the past 30 days.