Tuesday, May 24, 2011

Elo and Kill Death Ratio

So how do you tell if someone is good just by looking at their dota stats?  Sometimes elo (for those who don't know what elo is) is the best identifier and other times its the kill death ratio (KDR). Here are some of my thoughts on how to interpret some of the different stats you might see.

Every new name created in pub leagues starts at 1000 elo. If you win games your elo goes up and vice versa if you lose. So a high elo is generally good. It usually means the player have more wins than losses and means they help win games more than they help lose games. But when you see a high elo, you also have to look at their games played. A 1200 elo with 500 games played most likely means the player is stuck at that rating. 1200 with only 30 or 40 games played means they shot up from 1000 pretty fast and are probably better than what a 1200 elo might indicate. This follows a pretty standard rule of statistics: the bigger the sample size (games played) the more accurate the data is going to be (elo rating). So when you see a player with plenty of games played, the average of their stats should be a pretty accurate estimation of how good they are.

The big problems in trying to pinpoint someones skill level by stats is when they don't have many games played. This is where someones kill/death/assist stats come in. In dota, if you get a kill for every time you die, you'll have a 1.0 kdr. Usually anything below a 1.0 kdr isn't good, but that doesn't account for someones assists. If someone played a support correctly and had a high amount of assists to deaths, their kdr would be low, but they still might be damn good.

Heres an example of a random game in dotacash:
As you can see all of the players except two have enough games played for a good estimate of their skill level. But how would you rate the player living-on-video just by his stats? Well you can see he has won both of his 2 games and has a 9.5 kdr. Also being up 32 elo in 2 games means he won ~16 each game which indicates the teams he was on and played against had around the same average elo. So this player living-on-video helped win 2 games and had a great kdr, I would guess hes a much better player than troll-dad, who only has 1011 elo with 56 games played.

Getting a good estimate and your teams abilities before you play with them is always a good thing to do. A low level player might do things like auto attack, push the lane, fail at harassing, take unnecessary creep damage and etc. If you can expect how your team, as well as the enemy, might play, you will reduce the chance of being surprised by good or bad play from either side.

Next, I will cover a few aspects of laning. And if you have any questions about this post or something you want me to talk about next time, let me know.


  1. everything is a numbers game these days haha

  2. This is good to know. It'll help me know who the really good players are.

  3. Thanks man, I hate people who are too friggin good, this might help.

  4. Every time I read this it makes me want to listen to Basshunter's Dota ha ha.