The h2h vs rw% discussion was brought up internally at RGL about a year ago. As part of that, I ran a statistical analysis that looked at whether h2h or rw% better predicted playoff outcomes using the dataset of S1-S6 playoffs (not including RR divs or am or nc). The results indicated rw% was a better predictor of playoff performance across the board, including when taking into account potential biases. Even when we looked at teams that had faced each other in playoffs, rw% was a better predictor of which team would win - which is a situation where you'd expect h2h to have a significant advantage. This lead to us keeping it as the primary tiebreaking metric. I've got a big writeup draft somewhere with all the stats and methodologies that I wanted to publish back then, but we got busy with S7 and it got put on the back burner.
Through doing the analysis, I figured out that the general reason rw% is better than h2h can be qualitatively stated as "it only takes a single thing happening to mess up a h2h result, but it takes a bunch of things happening to mess up a w% result." (which was backed up by the data and the mathematical modeling I did)
Of course, neither method is perfect - the only way to have perfect competitiveness is to run a double RR division with pick-bans for maps (which is why that's the format for Invite). RGL uses RW% as a tiebreaker because it's correct more of the time than the alternatives. There will always be cases where it's not accurate, no matter which method is used.
If anyone's deeply interested in the methodology, etc, feel free to message me on discord - matchmaking and playoffs structures are one of my main focuses at RGL, and I'm always happy to chat about this kind of thing.
Also, random interesting data drop:
As part of the investigation, we looked into the potential bias that some maps would yield more rounds than others. The data shows a small but not insignificant effect:
Data from adv-im S5 regular season
rt = rounds taken
rtow = rounds taken on win
rtol = rounds taken on loss
rtol/rtow = % of the winning team's rounds the loosing team takes = rtol/rtow
rtol/rp = % of all rounds the losing team takes
#m = number of matches
map | rt | rtow | rotl | # matches | rotl/rotw | rotl/rp | avg. score
cp_gullywash_final1 | 409 | 303 | 106 | 66 | 34.98% | 25.92% | 4.59 - 1.61
cp_metalworks | 407 | 292 | 115 | 64 | 39.38% | 28.26% | 4.56 - 1.80
cp_process_f7 | 364 | 290 | 74 | 65 | 25.52% | 20.33% | 4.46 - 1.14
cp_snakewater_final1 | 369 | 287 | 82 | 60 | 28.57% | 22.22% | 4.78 - 1.37
cp_sunshine | 418 | 297 | 121 | 65 | 40.74% | 28.95% | 4.57 - 1.86
cp_villa_b18 | 424 | 306 | 118 | 63 | 38.56% | 27.83% | 4.86 - 1.87
koth_clearcut_b15d | 305 | 248 | 57 | 65 | 22.98% | 18.69% | 3.82 - 0.88
koth_product_rcx | 320 | 235 | 85 | 61 | 36.17% | 26.56% | 3.85 - 1.39