Dex/Parry tiers - definitive testing methodology needed

[Gamemako]

Pretty messy dataset. Can dump linear regression on it based on deviation from minimum value,

f(x) = 0.400x + 24.01
r^2 = 0.357

I'm wondering if it's a linear increase in DPS relative. I noticed that my block strength with the same shield decreased as I leveled through Toto-Rak, which suggests to me that it might be all normalized by the base level value. If so, the 512 parry would suggest a 15.99% base parry rate with base dex and base parry. Just a hypothesis -- should be easy enough to test using a weathered axe and no +dex or +parry gear.

[Coramac]

Pretty messy dataset. Can dump linear regression on it based on deviation from minimum value,

f(x) = 0.400x + 24.01
r^2 = 0.357

I'm wondering if it's a linear increase in DPS relative. I noticed that my block strength with the same shield decreased as I leveled through Toto-Rak, which suggests to me that it might be all normalized by the base level value. If so, the 512 parry would suggest a 15.99% base parry rate with base dex and base parry. Just a hypothesis -- should be easy enough to test using a weathered axe and no +dex or +parry gear.

It's a step function so your linear regression isn't going to look good.

[Gamemako]

It's a step function so your linear regression isn't going to look good.

Is there any evidence there of a step function? I certainly don't see any in that data or in any data previously collected.

//EDIT: Data visualized:



See anything there suggesting a step anywhere?

[Coramac]

Is there any evidence there of a step function? I certainly don't see any in that data or in any data previously collected.

//EDIT: Data visualized:



See anything there suggesting a step anywhere?

Everything in this game that has been analyzed has worked as a step function. While it may not necessarily be true that this follows, you should know better when dealing with probabilities than to use the method you are using.

Yes, I do see evidence of step function here. There appears to be a significant break with the testing around 240ish. The only issue is whether there is a second break point earlier (which I don't think there is anymore, but Niinjitsu's data indicated there likely was). You do understand that the we're going to see a roughly normal distribution (trials aren't held constant) about the expected value right?

Are you telling me what you showed doesn't look like a step function? Because it most certainly does.