[war]Parsing +10 dex allotment, basic "just 50" gear, currently maximizing for parry.

Quote:

---

Originally Posted by Hachiko

I forget how to calculate statistical significance, but the difference if it is dependent (i.e. simulates a multi-roll table) is less than if it were independent (.34/6.58 < .35/4.35). The fact that it's closer if you use that model AND the fact that the Parry rates work out if you use the other model seems to make me lean towards concluding that they are dependent variables, and, effectively, having a higher block rate would reduce the number of crits you receive. Might be worth testing with a very high block rate shield v. a very low block rate shield.

---

That's not a way to evaluate probabilities.

Quote:

---

I think the crazier thing is that in both sambles (though yours moreso) is that the Shield gave a significant amount of evasion...

---

No, it didn't.

Quote:

---

I don't understand how you can say that the evasion "matches up almost perfectly" when there's a relative difference of 13.8% to 12.5%, that's 1.3% points difference, or roughly a 10% relative difference. The difference if the crit chance is dependent was only .34%, or a 5% relevant difference.

---

Then don't run the numbers. Assuming 13% is the actual evasion, the StDev is 15. sqrt(2001*.13*(1-.13)) = 15. Expected value is 260. 252 is within 1 sigma. 277 is slightly more than 1 sigma. "Relative difference" is meaningless. For block then parry, we're in very slightly under 2 sigma. Acceptable, but moving away. Larger sample sizes are needed to reduce StDev.

Quote:

---

Originally Posted by Coramac

Assuming 13% is the actual evasion, the StDev is 15. sqrt(2001*.13*(1-.13)) = 15. Expected value is 260. 252 is within 1 sigma. 277 is slightly more than 1 sigma. "Relative difference" is meaningless. For block then parry, we're in very slightly under 2 sigma. Acceptable, but moving away. Larger sample sizes are needed to reduce StDev.

---

Seems like both crit and parry work out better in the situation I've described, then.

The standard deviation for crits, if it's a multi roll table is (assuming "6% crit rate") 8.9. Expected crits for the WAR is then 83. 87 is within less than one standard deviation. For Paladin, you would have a pool of 1215, you have a Std Deviation of 8.27, and a expected value of 72, which again, puts the crit rate of the paladin within slightly less than one standard deviation.

For parry, then, we would have to assume the numbers are completely off in terms of % in the APP. So instead we wold divide Parry/(Non-miss hits), i.e. 356/ (2001-252) = 20.3%. Actual parries. Since this is so close I dont think it's necessary to worry about expected value / standard deviation, since we're basically exactly on where the expected value is due to just guessing we are right.

Now check it on the PLD 308/(2001-277-201) = 20.2% actual parries.

The data fit the multi-roll simulation table FAR better than any single roll table. I don't see how you can say block then parry puts it slightly under 2 sigma. If you do block then parry they have almost exactly the same actual parry rate, and essentially the same crit rate.

I agree it's multi-roll now without a doubt. Ninjii's idea of testing on lower level mobs to drastically increase evasion rates was really smart. I agree the check is: miss (dodge) -- Block -- Parry -- Hit. Where Crit goes I'm not as confident yet. My testing while being attack from behind to eliminate parry shows crit rates well below 6%. If he'd posted his raw data, this would probably be easy to determine with sample sizes of ~10000 attacks.

Quote:

---

Originally Posted by Coramac

I agree it's multi-roll now without a doubt. Ninjii's idea of testing on lower level mobs to drastically increase evasion rates was really smart. I agree the check is: miss (dodge) -- Block -- Parry -- Hit. Where Crit goes I'm not as confident yet. My testing while being attack from behind to eliminate parry shows crit rates well below 6%. If he'd posted his raw data, this would probably be easy to determine with sample sizes of ~10000 attacks.

---

I didn't know to save that *&%*. I wasn't testing that! :-P

If I get time later this week I will replicate it again.

I don't really have anything to contribute at this time, but keep up the good work!

Quote:

---

Originally Posted by Ninjiitstu

I didn't know to save that *&%*. I wasn't testing that! :-P

If I get time later this week I will replicate it again.

---

What a waste... I'm guessing based on my parsing that you took 10000 total normal attacks (parser doesn't do well with non-normal attacks); I was getting ~50% evasion on mobs up to 36 (didn't test 37 but accuracy increased on 38, must be a cap around 50%). That's like 6+ hours a parse?

I'll take a look at your dex stuff tomorrow; left my laptop (and Excel) at the office over the weekend.

Quote:

---

Originally Posted by Coramac

What a waste... I'm guessing based on my parsing that you took 10000 total normal attacks (parser doesn't do well with non-normal attacks); I was getting ~50% evasion on mobs up to 36 (didn't test 37 but accuracy increased on 38, must be a cap around 50%). That's like 6+ hours a parse?

I'll take a look at your dex stuff tomorrow; left my laptop (and Excel) at the office over the weekend.

---

I took 5000 normal attacks per test. Like 25k total hits. It took, basically from early morning till when I posted. Something like 14 hours total, though I wasn't sitting there watching the entire time, lol.

Quote:

---

Originally Posted by Coramac

I agree it's multi-roll now without a doubt. Ninjii's idea of testing on lower level mobs to drastically increase evasion rates was really smart. I agree the check is: miss (dodge) -- Block -- Parry -- Hit. Where Crit goes I'm not as confident yet. My testing while being attack from behind to eliminate parry shows crit rates well below 6%. If he'd posted his raw data, this would probably be easy to determine with sample sizes of ~10000 attacks.

---

Crit is more then likely a 1 roll table rule based of hit and then directly weighted by % modifications to crit.

Don't forget, that deviations will always be present, this is the law of a *chance* based system. It just boils down to the user at where they draw a line on what they feel to be acceptable or not at the range of deviation that is present.