I wrote some new code that pulled all ladder players' ranks by day to be used for better filtering of games.

Here's a quick table showing how players ranked in the top 100 have fared against each other, divided into sets of 10.

From the first table, for example, you can see that players in the top 10 have won 79.3% of games versus players ranked 41-50.

This is all 1v1 ladder games between top 100 players (rank at time of game), about 42,000 games.

Winning Percentage:

	1-10	11-20	21-30	31-40	41-50	51-60	61-70	71-80	81-90	91-100
  1-10	50.0	60.7	69.6	76.3	79.3	78.0	76.5	75.9	81.9	82.2
 11-20	39.3	50.0	58.2	66.2	74.2	73.4	76.0	76.6	78.4	86.1
 21-30	30.4	41.8	50.0	57.8	64.4	69.9	67.0	72.2	73.4	80.4
 31-40	23.7	33.8	42.2	50.0	58.4	64.5	69.0	68.8	72.9	72.1
 41-50	20.7	25.8	35.6	41.6	50.0	58.0	62.7	67.5	71.2	68.7
 51-60	22.0	26.6	30.1	35.5	42.0	50.0	57.8	59.9	64.7	69.7
 61-70	23.5	24.0	33.0	31.0	37.3	42.2	50.0	55.8	60.0	70.7
 71-80	24.1	23.4	27.8	31.2	32.5	40.1	44.2	50.0	59.1	61.8
 81-90	18.1	21.6	26.6	27.1	28.8	35.3	40.0	40.9	50.0	56.5
91-100	17.8	13.9	19.6	27.9	31.3	30.3	29.3	38.2	43.5	50.0

Actual Wins:

	0-10	11-20	21-30	31-40	41-50	51-60	61-70	71-80	81-90	91-100
  1-10	 853	1065	1038	 922	 526	 301	 202	  82	  68	  37
 11-20	 690	 797	 991	 962	 789	 431	 301	 203	 109	  62
 21-30	 454	 713	 697	 845	 844	 675	 410	 306	 218	 119
 31-40	 286	 491	 618	 714	 887	 896	 664	 405	 285	 215
 41-50	 137	 274	 466	 633	 690	 838	 731	 585	 390	 257
 51-60	  85	 156	 290	 493	 608	 717	 758	 625	 525	 364
 61-70	  62	  95	 202	 299	 434	 554	 657	 612	 537	 548
 71-80	  26	  62	 118	 184	 282	 419	 484	 517	 560	 564
 81-90	  15	  30	  79	 106	 158	 287	 358	 387	 523	 558
91-100	   8	  10	  29	  83	 117	 158	 227	 349	 430	 512

Is this based on current rank or the ranks at the time of the game?

@Beren, from the OP:

This is all 1v1 ladder games between top 100 players (rank at time of game), about 42,000 games.

There are fewer than 42,000 1v1 ladder games (41713 last time I checked but it's almost certainly different now), btw, where both of the players have a non-zero (non-expired) score on the ladder, and even fewer games have both players still ranked on the ladder. So it would be impossible to get this many games with players of top 100 rank.

Edited 6/30/2015 12:45:28

Correct, it's at time of game.

Technically, it's their rank from their ladder profile page on the day the game starts.

There are fewer than 42,000 1v1 ladder games (41713 last time I checked but it's almost certainly different now),

I can see:

Games 1 - 50 of 111289

for 1v1 ladder games

Edited 6/30/2015 12:50:20

My data set has 108,825 games 1v1 ladder games.

Of those, 73,452 had both players ranked when the game started.

Of those, 42,433 had both players ranked in the top 100.

(I haven't pulled the last 3 weeks of games yet.)

Edited 6/30/2015 12:53:18

@Krzychu: that count includes games which include players with expired ratings (the vast majority of them, actually). I ran an analysis on this earlier (see my thread titled "What makes a template strategic?") and discarded such games- which left me with fewer than 42000 usable games for all players that currently have a rating (let alone a ranking).

Keep in mind that this doesn't mean smileyleg had less than 42000 usable games total- he appears to have data about ranking/ratings from when the game started (the host message?), while I only looked at data about rankings/ratings that we have right now.

My original idea a while back was to use the host message but this is not easily available, so for I while I didn't think there would be a practical way to get point-in-time rank data.

Then I was looking at the ladder profile pages (for example, mine: https://www.warlight.net/LadderTeam?LadderTeamID=4992 ) and notice that the graphs have a point for rank and rating for each day. I checked the page source and found the data in a super-easy-to-consume array called LadderHistoryData.

So just a little bit more code and I was able to pull all players' complete ranks and ratings history.

Ah. I didn't realize you could get Ladder Team IDs from the Game data API. Out of curiosity, how long did the parts of your analysis take to run?

The Ladder Team IDs are not actually in the game data. However, there are less than 11,000 ladder "teams" so I just pulled all of them, ignored ones that weren't the 1v1 ladder, and matched them up by Player ID.

The initial run of 80,000 games or so took probably 12 hours to pull and parse. I only used 1 thread and connection so as not to burden the servers.

The queries mostly run in a few seconds.

Ah. Haha I should've expected that after you said you stored all the data in an SQL database.

The initial run of 80,000 games or so took probably 12 hours to pull and parse. I only used 1 thread and connection so as not to burden the servers.

Thanks! That's a bit faster than I'd expected for 80k games.

Here's a chart. (Mostly to see how easy it would be to do charts for these.)

Thanks for the chart!

Why are all the equal-rank win percentages exactly 50%? Does this mean these stats are totally redundant?

if the player in position 2 wins with the player on position 7, 1 person won 1 lost so it is always 50 %

yeah, data for same rank-area against eachother is meaningless

Well, it's nice to include, else there would be a weird gap in the middle of the chart...

I see! Thanks.

This is very cool, though I misread it as an analysis of rank and whining percentage. I'd like to see that too...

11-20 ranks have better win rates against some of the lower ranks than 1-10 ranks...

Edited 7/5/2015 18:51:28