PTR time. Blizzard has also published a longer blog post explaining the upcoming changes in MMR calculation:
I found this to be an interesting read; essentially it acknowledges the so-called “ELO Hell” argument, its smurf corollary and even the “not rewarding individual skill” complaint. As far as HotS is concerned, the ELO Hell debate went a bit like this:
On one side, you had players who maintained the system worked fine and that “ELO Hell” did not exist. To make their case, they would point towards high skill players taking smurf accounts to high levels in a relatively brief time. This, they argued, was the proof that no ELO Hell ever existed, because if the player were more skilled than either their opponents and/or team mates, then that skill would ultimately result in their ascending.
On the other side, there was pointing out that “ELO Hell” was never about the impossibility of ascending, but rather the time it takes for a player account that has a certain amount of matches under their belt.
This retort was relevant, because a player new to MOBAs will initially perform rather poorly, and possibly improve over time; the smurf argument only went as far as to prove that a skilled player, about whom the system knew nothing about, could rise quickly.
What ELO Hell was about is how the system reacted slowly to changes in player’s skill, so that it would take many matches to change its mind:
This section sounds, to me, exactly like Blizzard is answering directly to the sheer tonnage of posts described above.
Another point which I brought up previously in this thread and find important, concerns another key facet of player skill. MMR modeled it as if it was fixed across the hero and battlefield range - what you have instead, is that any single player will be better with certain heroes, and more effective on certain battlefields, and even those two aspects are subject to interplay.
Currently, HotS uses a simple model based on hero level, i.e. it assumes that a player with a higher hero level would be better than another with a lower level. This is contentious as, if you go back to the new-to-MOBA player example mentioned above, they could very well accrue a high hero level during the process of their introduction to the game, without that meaning they are comparatively and meaningfully better with that hero; also, bot games and stimpacks could effectively create a disconnection between hero level and player level.
There’s other interesting points, among which, this resulting in an increased ability to detect “griefing” (feeding, basically), so I recommend everybody to read that post.
It remains to be seen whether how this, rather significant change, at least on paper, will affect the actual rate of ascent. A defeat will still result in MMR being reduced, what Blizzard seems to be focused on doing is penalizing the player who has played ‘well’, less harshly than before. Their learning system appears to adjust MMR changes for wins, as well, but this aspect has, predictably, less screen time than adjustments for defeats.
At the end of the day, the rationale behind the changes seem sound, and a step in the right direction, but the proof is in the pudding, and it remains to be seen how well this will work in practice.
It certainly isn’t going to eliminate all complaints, that’s just impossible; however, if it succeeded in closing the gap between smurfs and existing accounts, it could go a long way.