Compared to the chart for the full bracket, things stretch out a bit when we restrict it to just the top ⅛th of bracket. Three characters, Min Min, Snake and Palutena, are now all over three times the average usage. A couple of characters on the other side of the spectrum, Ice Climbers (0.4% pick rate, 67% win rate) and Chrom (0.2% pick rate, 33% win rate) have what appear to be outlier win rates because they're picked so rarely.
Interestingly, only four characters who have a pick rate of at least twice average have win rates under 50%: Bowser (2.7% pick rate, 49.6% win rate), Sephiroth (2.5% pick rate, 46% win rate), Sonic (2.4% pick rate, 46% win rate) and Steve (2.4% pick rate, 44% win rate). Of the five characters to have a pick rate of twice average in full brackets, three (60%) had win rates under 50%: Bowser, Joker and Sephiroth. It seems that those kinds of "popularity picks," what I referred to as the "Cool DLC effect" in the previous article, dissipate when the going gets tough deeper in bracket. That's unsurprising to me; the people who are picking characters for the cool factor above all else are typically going to be eliminated by people who are more familiar with their characters, even if they belong to a lower tier.
For ease of picking out characters in the middle where pick rates are so similar, here's a version of the chart with pick rate rank as the x-axis:
Next, let's take a look at the changes in pick rate between full bracket and late in bracket:
The horizontal axis shows pick rates in full bracket; the vertical axis shows pick rates in just the top ⅛th of bracket. Two characters enjoy a pick rate increase of a full percentage point deep in bracket: Min Min and Snake. These are characters that already enjoyed decently high win and pick rates across the full bracket, but are also strong enough to have multiple players consistently make it deep into bracket and win once there. The following characters all enjoy at least a 0.5% bump in pick rate deep in bracket: Palutena, ROB, King DeDeDe, Richter (Simon sees only a 0.25% bump), Ken, Isabelle, Luigi, Ness and Sonic.
Joker is the only character whose pick rate dips by a full percentage point deep in bracket. This is perhaps the greatest proof of Joker's inability to perform up to his offline top tier status in online brackets: even those who are capable of making it deep into brackets with him often find themselves stopped by better online characters once there. The next biggest decrease in pick rate comes from Random, down from 1.0% usage in full bracket to a mere 0.1% in top cut. Cowards. The following characters all also suffer at least a 0.5% drop in pick rate deep in bracket: King K. Rool, Cloud, Little Mac, Inkling, Pokémon Trainer, Piranha Plant, Yoshi, Captain Falcon, Byleth and Lucina.
Finally, let's take a look at how win rates change from full bracket to just deep in bracket:
Because pick rate is critical context whenever we talk about win rates, this chart includes stock icon size as a variable to show pick rates as well. As with the above chart, if the character is above the line, their win rate is higher deeper in bracket, and vice-versa. Most of the characters with massive deviations from their full bracket performance don't see much usage. The following all see win rate bumps of at least 10% deep in bracket: Corrin (best at +18.7%), Mewtwo, Lucario, Falco, Ice Climbers, Byleth and Rosalina. Falco is the only one of these characters with an above-average pick rate, as he becomes one of the winningest characters deep in bracket, assuming he can make it there.