I am not good in statistic but my common sense tell me a sample of 200 is very small to test something with only a few % different.
Conclusion drawn fm test with such small sample is nothing better than a wild guess.
I hope more ppl will conduct similar tests and one day we can see the trend(or no trend) for dex/acc clearly when we combined all the data.
His sample was 200 per 10 increase giveing a sample of 800 from 10-50 that is plenty. And i'm not using this as proof nor is he, what he was looking for was a pattern, as you can see from the data no pattern exists,
Also SE said in the "ask the Dev's" thing today that DEX affects ACC but on a tiered level, so maybe we need to try DEX tests from like 80 and up, I have no clue but if i was asking the devs I'd show them this data and say dex is broken.