Right the numbers from aggregate sites, one example and some statistical analysis rather than sample size too small. In fact we are overstating the necesity of sample sizes and I’m highly doubtful that anyone has done an analysis
Data is for october 18th 2021 from hsreplay for standard
Data Face hunter, 55.7% overall win rate, 4.4% popularity and 250,000 recorded games
Quest mage, 50.9% overall win rate, 13.8% popularity and 810,000 recorded games. There are 29,251 recorded games of this precise match up.
With such large numbers we can make a binomial to normal approximation since both np>5 and npq>5. With n=250000, p=0.138 and q=1-p.
The matchup data suggests that for hunter the encounter rate is only 29,251/250,000 = 11.7% Is this statistically significant? Remember we expect 13.8% based on popularity. Why yes it is, in fact this point lies at 30 times the standard deviation lower.
So if we would state that H0: matchup rate is 13.8% and our H1: matchup rate is definitely not 13.8%, so there’s something fishy.
Then we will need to reject the null hypothesis with and do
So it is not a coincidence that the matchup rate is lower. Something else is going on with an uncertainty much much smaller than your typical 1%.
What that else is, is to make sure people have fun in their matchups, I’m actually fine with these adjusted mm algorithms, I’d just like to know how it works exactly.