I think Sakurai needs to let more time pass before making such judgments, which he said was "completely free of bias". I'm also curious as to what these results look like, and if their are statisticians who are gathering this information. There's another factor to consider as well, and that's how one character can deal with multiple characters. Where is the data that says character x is balanced with y and z? Or does Sakurai actually think all characters are close to 50:50?
I still do not understand why Elite Smash win rates would have any correlation with character tiers.
Win rates in an Elo-based matchmaking system will gravitate towards a 50% win rate for all skill levels, for all players, and for all characters, unless you're at a high or low enough extreme as to where the system has difficulty match-making you with opponents of whom are appropriate for your level.
The very fact that all characters stand around 50% win/loss in Elite Smash is just proof of it behaving as it should. This would be the case regardless of relative character strengths.
What would be a more interesting observation is what the average rating is for each character.
One must first assume this statement is true. "For the given dataset, player skill has no correlation with character choice." For a big enough sample size, this may very well be true, though I suspect (with no hard data to justify it, I caution) that higher performing characters actually are disproportionately chosen by higher performing players, which would only further increase the positive correlation between average GSP and character tier placement.
Once you assume this to be true, you can make another statement. "For the given dataset, character strength is the only variable driving that particular character's collective Elo rating."
If this is true (which it should be if we choose to assume player strength is randomly distributed across all characters, and that the sample size is large enough such that any variations in this distribution due to chance are insignificant), then Sakurai is looking at the wrong variable here.
He's looking at average win rate. If you have one particular character that's averaging 50% at the bottom of the Elo bracket, that is not akin to another character averaging 50% near the upper end. If we take note that Elite Smash Peach players have, for example, collectively 150,000 more GSP than Little Mac Elite players, something is astray in the balance of these two. Due to the very nature of this being an Elo-based matchmaking system, both will have a 50%'ish win rate, regardless of balance concerns, because the Little Macs will be at the lower end match-making against other lower tier characters, or other, lesser-skilled players of potentially higher tier characters, whereas the Peach players will likely be, on average, in a higher bracket matchmaking against other Peach players, or other fantastic players of lower tier characters! These numbers he grants are, quite literally, meaningless.
Players only have a non-50% win rate in an Elo system under two conditions:
1. The system is not capable of match-making this player with others of equivalent Elo rating due to lack of participants.
2. The system has yet to find a steady-state position to place this player, or this player is continuously improving against their competition.
This happens regardless of character choice.