What Is the Effectiveness of Ranked Placement Games? A Study
- This topic has 5 replies, 1 voice, and was last updated 1 year, 11 months ago by
.
-
Posts
-
For new members of the community looking to get into online play, the common advice is to play through their 10 placement games to get a more accurate ranking and have fairer and more balanced games.
However, there have been cases where this process didn’t quite work as well as it should.
To explore this further, data from 205 players who had just completed their 1v1 RM placement games (10 games, last played less than 10 days ago) was pulled from aoe2.net.
Results showed that the average win rate for the first 10 games was 36%, with 2-4 wins being the most frequent outcome.
The rating distribution was relatively tight, with players with the same number of placement wins being within 100 points of each other.
Of the 205 players in the set, 76 didn’t play any more 1v1 RM games after their placement.
Of those that did keep playing, most didn’t play much.
For most players their current rating was within 100 points of their placement rating, but players with high placement win rates tended to climb higher than others.
When filtering to the 74 players with 15 or more games, most players did have a post-placement win rate (with the 10 initial games removed) between 40 and 60%, but 12 players (16%) only won a third of their games at most and 13 (18%) won at least two thirds.
This data suggests that the placement system was effective for most players, but tended to underestimate the ratings of the higher skilled players, and players with the lowest placement win rates generally continued to lose more games than they won after their placement.
interesting that you switched the axes for almost every chart compared to how I personally would interpret that data 11
I think placement games are basically trying to square the circle.
If you want them to place the players on the further ends of the spectrum more accurately, they’d be more sensitive and have higher variance and unbalanced games for the players in the middle, and vice versa.
It’s the same with IQ tests.
So basically we’re just trying to find a reasonable middle ground, which will never be perfect for everybody.
chess sites have that option where you can start at a certain higher elo.
But I dont really see AoE2’s need for something like that.
The people who wanna smurf their way up to a certain elo, they will start at 1k either way.
And the few lost souls that come back to the game and have like >1k5 elo skills but just now hit the DE ladder, even they do appreciate some placement games imo, because you’re quite rusty when you’ve been inactive.
​
I enjoyed the read!
I think if you win all 10 of your placement games you’ll end up close to 1500 elo.
So it works reasonably well for intermediate players but I bet for advanced players they have to grind a bit to get to their true elo.
Definitely an inconvenience, but also the lesser evil that works for 90 percent of players.
How I think it should work.
800.
Every consecutive loss is -100.
Every consecutive win is +100.
Once you break the cycle, it moves by 25.
So, 800, 700, 600, 500, 525, 550, 525 END OF PLACEMENT
800, 900, 1000, 1100, 1200, 1175, 1150, 1125, 1150 END OF PLACEMENT
800, 700, 725, 700 END OF PLACEMENT.
Basically the second time you hit any Elo, placement is done, whether it’s wwwwwlw or lllllwwwl or whatever.
Your second change in streak ends it.
Most newbies will only have to lose 3 games before a win (800, 700, 600, 500, 525, 500?) and most average players can find their Elo within just a couple matches (800, 900, 875, 900 as an example)
LLLWWL for low elo and WWWWLW for higher elo.
It doesn’t place you until you stop winning (increments of 25 below 800) or stop losing (increments of 25 above 800)
Don’t care if it makes smurf life easier one bit.
Great analysis, thank you.
​
>Of the 205 players in my set, 76 didn’t play any more 1v1 RM games after their placement.
However, these were fairly evenly distributed across the skill levels, with similar proportions of high, medium and low win-rate players not playing any more ranked 1v1s.
By eye-test, this seems super wrong.
You seem to have roughly the same amount of active players on the left and the right of 1000 elo but you have **tons** of players below 20 games on the left side and only a few on the right side.
On gamepad only I destroyed my first 7 placement opponents as if they were easiest difficulty AI.
The 3 after those destroyed me and one of them was rank 1 gamepad only.
I also understand the ladder for gamepad only is still fresh and will take some time to settle.
- You must be logged in to reply to this topic.
