Speculation on WoD hit changes

[Poofah]

Having it be a normal distribution really wouldn't make it feel riskier for the chance at higher damage as you'd very rarely get several high hits in a row or several low hits in a row.....
Most hits seem to be either really big or really small not in the middle.

Yes, I don't see anything wrong with this design choice -- actually I kind of like it. However if the damage ranges are too low, then these abilities will be unattractive regardless of their 'gambling' nature. Reg has a good sample size and the average values seem too low so far.*

The distribution is complicated and we don't know the implementation behind it. So we don't know whether the low average damage is intended, or whether it's a bug in how they've coded the abilities. Also, because the variance is so high, it takes a very large sample size to draw any conclusions about the average value. It will be extremely hard to pinpoint the actual average damage with any confidence -- it would be a lot better if they'd tell us the intention for these abilities (avg dmg and distribution), and we could test whether the implementation matches.



* For Steam Vent we expect an average dmg of 22 (assuming that it's determined by a symmetric distribution), but Reg gets 20.63 +/- 0.727, ie there's only a ~5% chance that the actual average damage is 22. For Shadow Slash we expect an average of 21, but Reg gets 20.66 +/- 0.24. Again it's possible but not likely that the average is actually 21.

[Regillixavatar]

H'ok, time for an update since I finished analyzing yesterday's data. Let me start off by saying that there is nothing quite as exhilarating as starting a tamer battle, getting a single attack in, and then forfeiting...ok maybe doing it 300 times in a row >_>

Instead of jumping around different abilities, I wanted to take a deep dive into Shadow Slash (old 90% accuracy) to see how the damage distribution really breaks down. As an initial attempt at variety I tested the ability with 7 different pets (Level 1 Sen'jin Fetish, Level 7 Voodoo Figurine, Level 9 Crawling Claw, Level 14 Eye of the Legion, Level 18 Landro's Lichling, Level 25 Ghostly Skull, and Level 25 Unborn Val'kyr). Below are the results across all pets



I did some additional analysis on the data set and it looks like the computed base damage fell into the lower half of the distribution roughly 48% of the time into the upper half of the distribution 52% of the time (damn close to 50%). I have not done any statistical analysis to see if this significantly deviates from the 50% we would expect (until we hear otherwise) but that will happen later.

Out of curiosity I decided to look at the average values of both the overall data as well as the average values of the two separate halves of our distribution. The average of all computed base damage was 21.05 (damn close to the expected 21!), the average of the lower half was 17.7, and the average of the upper half was 24.13. This means that if we assume you have a 50% chance to get a hit in the lower half the expected damage would be (0.5*17.7) + (0.5*24.13) = 21.08 (damn close to 21 again!)

As before I am willing to share my raw data with anyone interested or I can show you how to help me make my data set even larger!

it would be a lot better if they'd tell us the intention for these abilities (avg dmg and distribution), and we could test whether the implementation matches.

I posted in the Theorycrafting thread on the Beta forums to see if Celestalon is able to help us out (secret hopes for our own sticky thread). If anything comes of that I'll let you know...in the meantime I'll continue to post my findings here.


Oh, and if you made it all the way through this post you are going to see some mind-blowing stuff in my next post! Currently the damage distribution from crits is either bugged or not what you would expect

[Regillixavatar]

While I was testing, I noticed that when I had a critical hit it never seemed to be for a value on the low end of the damage range so I decided to flush this out with some testing. To help myself from going mad I opted to use a Jademist Dancer with the Rain Dance buff to test out what the damage distribution looks like. Below is a breakdown of my analysis

First, the chart of overall data which shows something similar to what we saw with Shadow Shock



Next, the chart of computed base damage for only critical hits



The thing that jumps out right away is that the minimum observed damage was 23.3...way way above the expected min of 15.4

Finally, I wanted to see if there was how the damage distribution with the Rain Dance buff look this is the part that might blow your mind!



Again we see the minimum observed damage was 23.3! I understand that 62 trials isn't the largest data set ever but 0 of 62 results landed in our lower half (where we would expect 50% to be there...). The crit rate in the data set was only around 66% so the issue cannot be entirely attributed to the higher base damage of crits.

To help illustrate what I am seeing in the data I have included the below summary statistics. Notice how the Overall data and the data w/o rain dance look similar? Notice how the Rain Dance data looks just like the Crit data?!?!

Overall Results - 152 trials
Min: 15.4
Max: 28.6
Average: 23.36
Median: 26.1
Standard Dev: 5.05

Level 3 Enemy w/o Rain Dance Results - 38 trials (was easier to split this way)
Min: 15.5
Max: 28.4
Average: 21.19
Median: 18.55
Standard Dev: 5.13

Level 3 Enemy w/ Rain Dance Results - 62 trials
Computed Min: 23.3
Max: 28.6
Average: 26.99
Median: 27.25
Standard Dev: 1.21

Crit Results - 43 trials
Computed Min: 23.3
Max: 28.6
Average: 27.16
Median: 27.4
Standard Dev: 1.18

I will try to do some additional testing on how abilities like Rain Dance, Nimbus, Hawk Eye, and Adrenal Glands affect the damage distribution and post more for everyone!

[Poofah]

Neat. So a crit always uses the 'high' roll. That makes sense: if you crit a 'low' roll, it would be really close to the damage of a non-crit 'high' roll ( 17.7*1.5 = 26.55 ). Which would be pretty unsatisfying.

Definitely the best explanation at this point is that the game flips a coin, and gives you either a high hit or a low hit, and these both have their own small +/- range. But also, there are strict min/max values, so values are piling up on the extremes. But since this is still a symmetric distribution, it should be very straightforward to predict the average damage -- it should be (min+max)/2, which seems spot on at 21, at least for Shadow Slash.

Right now, it appears that for Shadow Slash, a low hit is 85% of the average damage, and a high hit is 115% of the average damage (which is 21).
For Steam Vent, it appears that a low hit is 75% of the average damage, and a high hit is 125% of the average damage (which should be 22).


The Raindance data is crazy. It's almost as if 'accuracy' is still a mechanic, but instead of determining how often you hit, it determines how often you get a 'high' roll. This would help keep accuracy buffs worthwhile (whereas they were going to be fairly useless in a world of all 100% accuracy abilities). Here's how it works out, assuming that +hit% converts into 'chance for a high hit'.

old Steam Vent = 0.95(28*0.8) + 0.05*1.5*(28*0.8) = 22.96
old Steam Vent with Raindance = 0.45(28*1) + 0.55*1.5*(28*1) = 35.7
new Steam Vent = 0.475*(0.75*22) + 0.475*(1.25*22) + 0.05*1.5*(1.25*22) = 22.965
new Steam Vent with Raindance = 0*(0.75*22) + 0.45*(1.25*22) + 0.55*1.5*(1.25*22) = 35.0625

So it seems just as balanced as the old version. It also jives with the observed data that Steam Vent + Raindance can never get a low hit, assuming the game rolls for crit before it rolls for 'high hit' -- you'd expect 55% crits (always high hits anyway) and 45% high hits.

If this is correct, then you'd expect Nimbus to give you 40% low hits and 60% high hits, and Accuracy would give 25% low hits/75% high hits. +Crit buffs are tricky because they often come attached to +hit buffs. For example, I'd expect Caw (+50% crit chance) to result in 55% crits (all high hits), 22.5% low hits, and 22.5% high hits. But Lucky Dance (+50% hit, +25% crit) should result in 30% crits (all high hits), 70% high hits, and 0% low hits -- because after adding 25% crits and 50% high hits, there's no room for anything else.

Very useful stuff, thank you Reg.

[Regillixavatar]

Well it looks like the Theorycrafting thread was the right place to post!! Below is what Celestalon posted describing the new 100% accuracy moves

For battle pet abilities with damage variance, here is an example of how it works:

Let's say your ability inflicts 200 to 440. The average would be 320.
We calculate our roll size as (440 - 200) / 2 = 120.
We then roll 120 three times and take the highest value. Our rolls are 14, 56, and 101, so we take 101. This is to give more noticable results and keep us away from the average.

Now we determine if the roll is added to or subtracted from the average.

Your base chance for a positive result is 50%.
If your hit is above 100%, your chance for a positive result is 50% + (Hit - 100) (EDIT: Fixed this formula). So, you always get a positive result at 150% hit and higher.
Critical strikes are ALWAYS positive.

A positive result would be 320 + 101 = 421
A critical strike would be (320 + 101) * 1.5 = 631
Otherwise the result would be 320 - 101 = 219

Please feel free to send questions and concerns to @TheCrafticus on Twitter.

[Pettea]

Won't the changes be very bad on abilities with a high miss chance? Because previously you could pretty much always dodge them by reducing their hit chance, whereas now they're probably still going to be hit.

Like when the enemy pet uses Crush with a 50% hit chance, if you Blind this pet, it will certainly not be able to Crush you.

There will definitely be ramifications in pet battling.

[Kpb321]

It will definitely change things. Assuming a hit debuff acts the same way as a buff it should make all of these large range abilities always hit for the low damage if you have a 50% hit debuff up. So 200ish every time instead of 400ish but a crit would still hit for 600ish.

[Poofah]

Won't the changes be very bad on abilities with a high miss chance?

The 50% accuracy attacks are still 50%, so there should be no change with those.

Assuming a hit debuff acts the same way as a buff it should make all of these large range abilities always hit for the low damage if you have a 50% hit debuff up

That's possible, but I wouldn't assume this is the case. It was necessary to change how these abilities work with hit buffs because they wouldn't benefit from +hit otherwise, and that would have made +hit buffs largely worthless. But -hit debuffs will be just as effective as they are now, so they might not have bothered to change them. Also, the new mechanic is capped at 50%, so it wouldn't accommodate Blinding Poison.

If anything, maybe they changed Darkness/Sandstorm, since the -10% hit isn't a large effect on damage, yet it adds a large amount of aggravating randomness.

[Kpb321]

Yeah it looks like currently hit debuffs work like normal on these new abilities. I took a guardian cub and with Onyx bite and killed some highland skunks. With stench up it shows 75% chance to hit and in a small sample set I did seem to be hitting the high end of the range about 50% of the time and I did miss a couple times.

[Regillixavatar]

Kbp, thanks for doing some additional testing on accuracy reductions! I am still quite curious about what happens when we mix an Accuracy Buff and an Accuracy Debuff.

I know the beta servers went down for maintenance about an hour ago but if anyone is interested in helping me test this today or this weekend, let me know (will save me sooooo much time over tricking Skunks into doing it). Zystyl#1639 is my btag for those who don't have it already...I want to test this so much I may actually make an alliance toon

[Poofah]

mix an Accuracy Buff and an Accuracy Debuff.

Good question. Currently, these calculations are all done additively, ie if you have a +X% accuracy buff and a -Y% accuracy debuff, then you'd have 100+X-Y % accuracy. However it produces unexpected behavior if accuracy goes below 0% (ie you can land hits in this case, aka the Blinding Poison bug or negative hit bug).

My guess is that this will still be the case, ie the game will determine hit chance additively. From that blue quote, it sounds like the new mechanic only takes over when accuracy goes above 100%: below 100%, I'd expect the ability to start missing instead.

If your hit is above 100%, your chance for a positive result is 50% + (Hit - 100) (EDIT: Fixed this formula). So, you always get a positive result at 150% hit and higher.

So now I'd expect this behavior from a +X% buff and -Y% debuff, when using one of the new high-damage-range abilities:

Calculate 100+X-Y % accuracy. If less than or equal to 100%, then your hit chance is now 100+X-Y %. If greater than 100%, then you will always hit, and your chance for a positive result (ie a 'high hit') is 50% + (100+X-Y-100), ie 50+X-Y %.

[Regillixavatar]

So now I'd expect this behavior from a +X% buff and -Y% debuff, when using one of the new high-damage-range abilities:

Calculate 100+X-Y % accuracy. If less than or equal to 100%, then your hit chance is now 100+X-Y %. If greater than 100%, then you will always hit, and your chance for a positive result (ie a 'high hit') is 50% + (100+X-Y-100), ie 50+X-Y %.

That is how I would expect the numbers to shake out. I want to test it to make sure that sum of the hit modifiers affects the roll chance and not just the positive modifiers. When I get a chance, I am planning on testing what happens when I use rain dance in darkness...we expect that the roll will be boosted by 40% but we won't know until we test it!

[Poofah]

The tricky thing with that is the +50% crit from Raindance: after the 55% crits and 45% high hits, there's no room left to see any effect from the -10% from Darkness. Maybe Accuracy (+25% hit) would be less complicated to test -- it's the biggest +hit buff that has no +crit attached afaik (but only Netherwhelp has this). Nimbus would work too, but the effect is small and would require a decent sample size to confirm/deny. With Raindance I'd expect:

(most likely) 55% crits, 45% high hits

(less likely) 10% misses, 49.5% crits, 40.5% high hits