Hacker News Reader

Comment by frenchtoast8 | Hacker News Reader

I recently learned Go for the first time and I have played almost 50 games of 9x9 on Online Go Server so far. I’m finding it a lot of fun but it has been very humbling.

I learned chess in 7th or 8th grade and was easily able to get to about 700 Elo on chess.com after barely learning the rules, which is about the 60th percentile on the site. I only play a couple games a year now but can still hold my own against 1200 Elo opponents, which is in the 90th percentile.

I feel like I have put in just as much effort into learning Go. I bought a book and have been doing exercises. But I’m still in the 0.1 percentile on the site! (Yes, that’s not a typo.)

I’m sticking with it because it’s fun and that’s all that matters. But I definitely have a lot to learn.

View on HN

The average FIDE elo is 1550; chess.com's average of 1200 elo is much lower because it has many beginner players, including children. The gap between those is huge--much larger than the numbers may suggest. At chess.com 1200 elo people still have terrible board vision and routinely drop pieces; by the time one reaches FIDE 1550 elo that is no longer true and people are starting to plot out complex tactics requiring accurate visualization several moves ahead.

According to the elo formula, a 700 elo player is expected to "hold their own"--draw or win--against a 1200 elo player 1 in 19 games (so they will lose 18 of 19 games), and against a 1550 elo player 1 in 134 games (so they will lose 133 out of 134 games). A 1200 elo player is expected to "hold their own" against a 1550 elo player about 1 in 18 games (they will lose 17 out of 18 games). However, the chess.com and FIDE elos are from very different pools--1200 elo at chess.com is probably equivalent to about 600-800 elo FIDE.

> The average FIDE elo is 1550; chess.com's average of 1200 elo is much lower because it has many beginner players, including children.

Why would that lower the average elo? Elo is a conserved quantity: for you to gain points, your opponent must lose that many points, and vice versa.

So there are two obvious facts about the average rating:

1. In a closed system, the average rating can never change, not up or down, not by any infinitesimal fraction of a point.

2. In an open system, the average rating can change, but for it to go down would require players with an above-average rating to leave the system.

In the much more common scenario where beginners come in, lose a bunch of games, and then leave, the average rating will go up over time.

> Why would that lower the average elo?

Because the average chess.com player is considerably weaker than the average FIDE player.

> In the much more common scenario where beginners come in, lose a bunch of games, and then leave, the average rating will go up over time.

Weak players come into the chess.com pool faster than they leave.

> Weak players come into the chess.com pool faster than they leave.

By definition, someone entering the pool enters with the average elo (for a closed system). The rate of entry and exit doesn't affect the direction of average (open system) elo. (It does affect the rate of divergence; the faster entry is, the longer it will take for the open average to diverge from the closed average.) What matters to the direction is the elo players have when they leave, not how long they take to leave.

Consider what happens if people enter with 1000 elo at a rate of 10 per day, and leave with 800 elo at a rate of 3 per day:

    +-----+---------+-----------+-------------+
    | day | players | total elo | average elo |
    +-----+---------+-----------+-------------+
    |   0 |       0 |         0 |         0.0 |
    +-----+---------+-----------+-------------+
    |   1 |      10 |     10000 |      1000.0 |
    +-----+---------+-----------+-------------+
    |   2 |      17 |     17600 |      1035.3 |
    +-----+---------+-----------+-------------+
    |   5 |      38 |     40400 |      1063.2 |
    +-----+---------+-----------+-------------+
    |  10 |      73 |     78400 |      1074.0 |
    +-----+---------+-----------+-------------+
    |  20 |     143 |    154400 |      1079.7 |
    +-----+---------+-----------+-------------+

> Because the average chess.com player is considerably weaker than the average FIDE player.

This... is nonsense. That has nothing to do with the average elo.

The situation you describe is mathematically impossible. Please consider that before you repeat the claim.

P.S. The response is hilarious (and rude).

The percentile stuff on chess.com has something funky going on with it. Lichess and Chess.com ratings are comparable (not the same, but at least ballpark comparable) and the median elo on lichess is about 1500. [1]

700 elo on chess.com is completely normal for a new player who knows the rules, but not much more. I have a good sized sample to back this up based on coaching classes of complete beginners and getting them to setup a class specific account.

Chess.com seems to have 1.3 million players with 100 ELO which is their hard cap minimum, so that is literally the same, if not worse, as a random move generator. So there has to be some weird bias there.

[1] - https://lichess.org/stat/rating/distribution/rapid

The skill that transfers from chess to go is reading -- "if I play here, my opponent is likely to play there, and then I will..."

why is this downvoted?

Reply on news.ycombinator.com