Posted on 04 December 2015 - 09:15 AM
I've been trying to design a formula for a number that accurately represents the inflation of a ladder in all cases. Currently, the only relevant things I've thought to include in this formula are # of players, elo of #1, elo of #20, what rank you achieve if you have 2000 (or any other arbitrary number) elo, elo gap between #1 and #2, the lowest elo, the 20th lowest elo, the gap between the lowest and 20th lowest elo, the elo range of the ladder (highest minus lowest), and the IQR (a range statistic less affected by outliers). Help me?Posted on 04 December 2015 - 01:50 PM
Sorry my brain explodedPosted on 04 December 2015 - 02:33 PM
Dayum… Lemme thinkPosted on 04 December 2015 - 02:45 PM
Hivlik wrote
All elo quartiles @Hivlik
Posted on 04 December 2015 - 04:29 PM
standard deviationis that a thing?
Posted on 04 December 2015 - 06:11 PM
um can u put this in dummy terms for mePosted on 04 December 2015 - 06:14 PM
Savagebaut wrote
is that a thing?
Assuming we're going to get a normal distribution (or something very close to it), you can use it to find the probability of people in certain Elo ranges: above a certain number, below a certain number, in range of two numbers
Posted on 05 December 2015 - 01:59 PM
confusedPosted on 05 December 2015 - 09:45 PM
Not really sure what you are looking for…When I think of Elo inflation I think of ways in which the total amount of Elo in a ladder increases. Normally Elo in a ladder = 1400*(# of players). Now with Elo resets and elo punishments it isn't exactly that, but pretty close.
What I think you are looking for is Elo distribution?
Posted on 06 December 2015 - 02:03 AM
qazzy1122 wrote
When I think of Elo inflation I think of ways in which the total amount of Elo in a ladder increases. Normally Elo in a ladder = 1400*(# of players). Now with Elo resets and elo punishments it isn't exactly that, but pretty close.
What I think you are looking for is Elo distribution?
No, you're exactly right. The elo inflation I'm referring to is the expansion of the elo pool. For example, a lot more players have played Build UHC than Iron Soup. Build's elo pool is much larger and therefore inflated. 2000 in build is much less impressive than 2000 in iron soup. I want to set up a thing at the end of the season where I can align different ladders' results while removing the elo inflation factor, which makes them incomparable.
I could do something where I just multiply all the elos in the less inflated by a certain ratio thing based on the other, more inflated ladder, but I'm not able to work my head around exactly what a good formula would be.
Posted on 06 December 2015 - 02:26 AM
Hm, interesting.Posted on 06 December 2015 - 02:52 AM
1. Can elo distribution be modeled by a standard bell curve? Because few players are outliers in both very low/high elo compared to the majority of players2. What significance does 2000/whatever number you use have on the formula? That part confuses me.
Posted on 06 December 2015 - 03:24 AM
Knite wrote
2. What significance does 2000/whatever number you use have on the formula? That part confuses me.
Posted on 07 December 2015 - 09:23 PM
Knite wrote
By definition, bell curves (AKA normal distributions) have a large percentage of their bulk in the center, and small amounts at the very fringes of the left and right. So very high/low players fit right into what one looks for in such a distribution.
However, I think Elo can't be fit to a normal distribution as nicely as one might hope because of human nature. People with very high Elo continue to play for far longer than those with very low Elo. This majorly extends that "tail" of the bell curve, while those with comparably rare LOW Elo's don't really put in the same time and effort to keep LOSING. The resulting curve would probably look more like the right curve than the left curve:
http://www.cdc.gov/ophss/csels/dsepd/ss1978/lesson2/images/figure2.10.jpg
Posted on 08 December 2015 - 09:41 AM
qazzy1122 wrote
Knite wrote...
By definition, bell curves (AKA normal distributions) have a large percentage of their bulk in the center, and small amounts at the very fringes of the left and right. So very high/low players fit right into what one looks for in such a distribution.
However, I think Elo can't be fit to a normal distribution as nicely as one might hope because of human nature. People with very high Elo continue to play for far longer than those with very low Elo. This majorly extends that "tail" of the bell curve, while those with comparably rare LOW Elo's don't really put in the same time and effort to keep LOSING. The resulting curve would probably look more like the right curve than the left curve:
http://www.cdc.gov/ophss/csels/dsepd/ss1978/lesson2/images/figure2.10.jpg
which way do you think the distribution would be skewed?
Posted on 08 December 2015 - 06:01 PM
qazzy1122 wrote
Agh, can't believe I didn't realize that
This makes calculating these so much more tedious
Posted on 08 December 2015 - 06:46 PM
WatWatInTheButt wrote
Posted on 09 December 2015 - 07:13 AM
Badware wrote
WatWatInTheButt wrote...
what
Posted on 09 December 2015 - 10:42 AM
.