Median rating not 1500, why? • page 1/2 • General Chess Discussion • lichess.org

Couldn't that just mean that a few are very good and most have lost their twenty games and are at -200? The median could still be 1500.
Graphically, you don't have to look were half of the people are, but were half of the points are.
If you look at the standings in premier league, place 10 means you are better than 50% of teams, but not that your score is on the median.
I, for my part, have always trouble imagining how quitters (site-leavers) influence the overall rating average.

Couldn't that just mean that a few are very good and most have lost their twenty games and are at -200? The median could still be 1500. Graphically, you don't have to look were half of the people are, but were half of the points are. If you look at the standings in premier league, place 10 means you are better than 50% of teams, but not that your score is on the median. I, for my part, have always trouble imagining how quitters (site-leavers) influence the overall rating average.

petri999

#3

Very few player play ultra bullet and list if from active players. so I would assume there is a bian on players who try the game vs player who enter the pool . So the stronger player dont like it hence the actie player pool in weaker. It is bit odd but must relateds on small amount of players

#4

sheckley666

#5

@Haymarket There is no reason for the median to be at 1500. The average is probably close to 1500, but the median can deviate because the distribution is obviously not symmetric, e.g. there is a hard lower limit of 600.
You would have the same effect in your example, if one of the 100 GM's is Magnus Carlsen, and the rest are around 2500 Elo. Then, maybe, Magnus would have 3000 after a while, and all the others could be slightly below 1500.

@Haymarket There is no reason for the median to be at 1500. The *average* is probably close to 1500, but the median can deviate because the distribution is obviously not symmetric, e.g. there is a hard lower limit of 600. You would have the same effect in your example, if one of the 100 GM's is Magnus Carlsen, and the rest are around 2500 Elo. Then, maybe, Magnus would have 3000 after a while, and all the others could be slightly below 1500.

petri999

#6

@Haymarket yes that is how it works if there is no bias in icoming and outgoing players. rating is zero sum game ( not quite but almost) lets pool is polutates so that there is 1000 weak players and 1000 player 1000 points stronger. rating will settle so that weak players get to 1000 and and strong to 2000 and bot median and avererga to 1500 as expected. Now if there is 500 strong players leaving it will not affect anyones rating as they are still relatively just as strong. just that average rating will drop to about 1300 and if rating are normally distributed (as they almost are) the median should be 1300. Obviously in this two peaked distribution median would be just 1000.

so if number players leaving is vastly stroger than staying and flow incomers is small median can move. Actually median is 1500 only because high inflow. well median would stay stable in there is no correlation to strength on players leaving the pool even without inflow

@Haymarket yes that is how it works if there is no bias in icoming and outgoing players. rating is zero sum game ( not quite but almost) lets pool is polutates so that there is 1000 weak players and 1000 player 1000 points stronger. rating will settle so that weak players get to 1000 and and strong to 2000 and bot median and avererga to 1500 as expected. Now if there is 500 strong players leaving it will not affect anyones rating as they are still relatively just as strong. just that average rating will drop to about 1300 and if rating are normally distributed (as they almost are) the median should be 1300. Obviously in this two peaked distribution median would be just 1000. so if number players leaving is vastly stroger than staying and flow incomers is small median can move. Actually median is 1500 only because high inflow. well median would stay stable in there is no correlation to strength on players leaving the pool even without inflow

dboing

#7

Hard to make a model that controls the median
They are usually positing statistical moments like avg or sd etc
But I may be wrong

Hard to make a model that controls the median They are usually positing statistical moments like avg or sd etc But I may be wrong

ungewichtet edited

#8

@Haymarket

Ultrabullet: The median is at ~1300. At 1500, you are better than 70%.

Why do you think the median should be at 1500? The average of all scores should be at 1500.

Doesn't, where the median lies, all depend on the variety of quality of players, to open results?

While I have to admit, in bullet, blitz and rapid the graphs look like a gaussian bell, when in ultra, it does not.
Chess 960 has even less users per week and has no regular graph, either. The median is 1612 in 960 :)

@Haymarket Ultrabullet: The median is at ~1300. At 1500, you are better than 70%. Why do you think the median should be at 1500? The average of all scores should be at 1500. Doesn't, where the median lies, all depend on the variety of quality of players, to open results? While I have to admit, in bullet, blitz and rapid the graphs look like a gaussian bell, when in ultra, it does not. Chess 960 has even less users per week and has no regular graph, either. The median is 1612 in 960 :)

dboing

#9

Not forgetting the bumps

h2b2

edited

#10

the only think I know about statistics is I don't know anything about it and it's often counter intuitive to me.

It looks like in most other time controls, the median is ~1500.

13,653 UltraBullet players this week. I wonder how many total UB games this week, and if total games / unique players in UB is similar to blitz. I'm thinking there's a small percentage of UB players that play most games.

I wonder what the percentage of provisional plays UB has? I suspect it's high. People play a few games, lose, don't play any more.

the only think I know about statistics is I don't know anything about it and it's often counter intuitive to me. It looks like in most other time controls, the median is ~1500. 13,653 UltraBullet players this week. I wonder how many total UB games this week, and if total games / unique players in UB is similar to blitz. I'm thinking there's a small percentage of UB players that play most games. I wonder what the percentage of provisional plays UB has? I suspect it's high. People play a few games, lose, don't play any more.