Thayer, I think it is interesting that your test with FA showed a 6.0% crit reduction. The BC value was 1.2%, and with full hit cap in WLK we are showing 4.8%, which is four times greater. Your test showed results which are five times greater at ~20% crit. My tests showed a 4.8% crit reduction whether hit capped from gear or from using FA, but I was testing at only 10.0% crit.

The only variable between my test and yours is the crit rate, since yours was 21.72% -- we had the same spec of 0/3/0 and the same hit rate of 5% from gear.

These two links seem to go to the same image. Did you already do a 10K shot test at ~20% crit without FA?

My thought is that we need to run a hit capped 0/0/0 test test at ~20% crit, and also tests at ~30%, and see what happens, as you propose in your initial post.

I am in the process of a 0/0/0 test at 8% hit from gear and 20% paperdoll crit, so if you have already run one, I would be interested to see the results! I'll edit in my results here when the testing is complete, but it may take me another day or two to accumulate enough shots.

Whether the test without FA at 20% shows a 4.8% or a 6.0% crit reduction, it will be interesting information. I don't think I have the gear yet to pull off an untalented 30% crit rate with either 8% or 5% hit. If you can, I hope you will go for it and post the results.

Yes, I do have 400 Skinning.

If this has some effect, it would only be .54 percent. You can see the progress of my test in one of the above posts, where I am showing an example of RNR during the 10k test.

I am doing a test in 1000 shot blocks. I will post each block and the cumulative averages. The test is at 27.86 crit, 263 hit from gear.


1: 20.1 crit return --- 7.76 crit reduction --- 7.76 crit reduction average
2: 22.0 crit return --- 5.86 crit reduction --- 6.81 crit reduction average
3: 20.8 crit return --- 7.06 crit reduction --- 6.89 crit reduction average
4: 21.8 crit return --- 6.06 crit reduction --- 6.69 crit reduction average
5: 20.6 crit return --- 7.26 crit reduction --- 6.80 crit reduction average --- 62 shots blocked (6.2 percent)

6: 22.7 crit return --- 5.16 crit reduction --- 6.53 crit reduction average --- 48 shots blocked (4.8 percent)
7: 22.9 crit return --- 4.96 crit reduction --- 6.30 crit reduction average --- 52 shots blocked (5.2 percent)
8: 22.7 crit return --- 5.16 crit reduction --- 6.16 crit reduction average --- 55 shots blocked (5.5 percent)
9: 19.4 crit return --- 8.46 crit reduction --- 6.42 crit reduction average --- 50 shots blocked (5.0 percent)

EDIT: Removed redundant information in order to make room for crit bonus differences. This will be a constantly updated post.

Any further updates on this information have been move to a post further in the thread, in order to make the data easier to parse.


Note: I pay 2g 17s to 3g 43s a repair at 1000 shots, so its costing me 20 to 40 copper or so copper a shot... I guess I am mailing each shot to my target!

Have we considered the possibility of diminishing returns being inadvertently added to Critical Strike rating?

This would suggest that if 10% is below the DR threshold, you may have varying values between critical strike ratings. With the addition of it to defensive characteristics I don't see it being an impossibility.

This speculation would require a very large sample size, but it seems possible. I can get you some small sample data tomorrow if your current tests are consistent with your first.

I'm currently auto-shotting with a 0/2/0 spec (I normally only put 2 points into FA due to my gear set) and after 1000 shots @ 23.26% crit I have a -4.66% crit rate. I'm planning on letting this run till my gun breaks or I run out of ammo (~3000 shots) and resume it tomorrow when I wake up.

Honestly not seeing the numbers you guys are getting, even on small sample sizes.

edit:

Final Number (because I'm an idiot and forgot to reset my Recount so I have to use a segment when I went to reload.)

Total: 2318
Hits 1913
Crits: 405
Char Sheet Crit: 23.26%
Expected Crit: 18.46%
Actual: 17.47%

Difference: -0.99%

Sample size probably too small, I'll let it run all afternoon tomorrow, but I think theres something here...

If I average my first 6000 shot test with my second parsed tests, at 6000 I get a 6.195 average, which is lending itself to negate the factor of FA effecting crit returns. There is now reason to believe that a higher crit rating may result in lower crit returns. This may be a case of DR on crit, it may be associated with blocked damage shots, or it may be a third factor we are not aware of at this moment.

A 6.165 average - 4.80 standard crit reduction is a additional 1.365 crit reduction. Using my previous formulation on blocked damage shots, I assumed a 1.39 crit reduction, and this would seem to fall in line with a -0.025 percent RNR swing.

Of course all this is subject to change, as all things are. Remember I am operating under the assumption of 5 percent blocked damage shots.

Using the assumed equation we would expect a blocked damage shot crit reduction of 1.16 Percent. This leaves a RNR swing of +0.17 percent. This larger swing is acceptable at the moment because of the size of the sample available.

Using these numbers for Auto shot only, we get a total of 5663 shots fired with 1195 crits, for a crit percentage of 21.1, and a crit reduction of 5.36, resulting in a 0.56 additional crit reduction. Using the assumed formulation, we would presume a 1.05 percent additional crit reduction. This gives us the largest RNR swing of -0.49 percent, over ~6000 shots.

Note: this sample was preformed in conjunction with using Steady Shot, and with all talent points spent.



We would need a much larger test pool at 00 / 00 / 00, or any variant up to 00 / 03 / 00, while hit capped, to solidity any theory stated in this thread.

Last edit for the night, time to sleep.

Are you removing all armor not specifically needed to get the stats you're testing, and firing the [Dalaran Rifle] or something similar?

I have completed my test of 10,000 shots using 0/0/0 spec, 8% hit from gear and ~20% crit (closest I could manage with my gear). I am afraid the results muddy the water a bit.

I am seeing a 5.39% crit loss, which is more or less 0.6% more than the expected 4.8% loss, but not as much as Thayer was getting using FA for part of his hit. In fact, it is half as much.

Test done on Darnassus Heroic Dummy:

Heroic Test Dummy -- 10,000 Autoshots
spec 0/0/0
268 (8.17%) hit rating from gear

20.17% Paper Doll crit
14.78% Actual crit
5.39% Crit depression
Screenshot
Screenshot (Including info on blocked damage)

In one of my posts in the old thread, I noted that most of the test results posted in the hunter forums for various reasons showed ~5.3% crit loss on the heroic dummy. I had speculated that this was a built-in higher rate of crit loss for ranged compared to the melee rate of 4.8%. This did not seem to be borne out by further tests I reported in that post and later in the thread, which showed 4.8% crit loss for ranged too.

I have also noted that tests with more or less full specs seemed to incur the 5.3% rate while stripped down specs got 4.8%. However, it is also true that the tests giving the 5.3% rate were with paperdoll crit of 20-22%, while the tests with stripped down specs were also done with stripped down gear and low paperdoll crit rates. It seems to be the crit rate rather than the talent composition that is affecting the results.

While the blocked shots could explain the reduced crit rate found at 20% paperdoll crit, I really don't understand why blocked shots would not also affect a 10% paperdoll crit rate in the same way. I am leaning towards diminished returns as the explanation, as dssurge suggested, except it may not be unintentional. The final test results on the higher crit rate that Thayer is working on will be enlightening, and Focused Aim isn't completely off the hook yet -- the test with FA showed 6.0% crit loss vs 5.4% loss fully hit capped from gear. Unless 10,000 shots is not enough to eliminate the RNG when looking at such a small difference...

I haven't been following the melee crit loss discussion in the rogue forum to see if they have been noticing the same effect.

Well the main difference now is you had on 425 shots blocked, or 4.25 percent while I had 4.96 percent, thus creating a range for blocked damage shots.

My first 5000 shots where I am recording block damage values, is showing 5.34 percent blocked shots, and a higher crit reduction rate then you are showing.

I am going to get 5000 more shots with block values recorded, and then I will spec 5 points into crit, and with the piece of gear I got recently, I can test at 33.52 crit, 263 hit from gear.

This is my raiding block for the week for the next 3 nights, so I probably wont get anymore data until this weekend.

Where are you seeing how many blocked shots I had? I see the amount of damage that was blocked, but not the number of shots.

Do you think the difference in block/crit loss between my test and yours is merely RNG because of the small differences we are looking for, or the result of using FA?

Also, I should clarify that when I said "diminishing returns," that probably wasn't the best choice of wording. I was not referring to pvp style DR against a single target, but meant that the higher the crit rate, the higher the crit loss. During my test, the crit percent fluctuated, first down, then back up (but not as much), then down again (but not as much), so I don't think merely fighting the same target for a long time causes the crit loss to increase. I suppose block could be the mechanic for this diminished effect -- the more crit you have, the higher the block rate used against you?

If we are going to try tests using Lethal Shots, I can reach 30% crit I think with 5% hit from gear and Focused Aim. I'll try to fit it in over the course of the week since I will not be raiding again til the weekend.

From observing my combat log, and changing AP and such, i found that the amount blocked by the dummy is always 76. I took the blocked damage you had, divided by 76, and it always comes out to an integer.

SO 32300 / 76 = 425, and 425 out of 10000 shots is 4.25 percent.





I think we may have a more complex pattern occuring. What i thin is happening is there is some kind of "crit compression" occurring at higher levels of crit, and FA some how interacts to make this more apparent at lower crit levels. Right now i am more concerned at testing with Fully capped gear to find a baseline for the compression, if there is indeed one. I think the blocked damage shots are somehow involved, but we need more data, since s you stated, we are dealing with a small margin of error here, and a large sample size is needed to eliminate RNR as a factor.

So indeed, the hypothesis at this time is that the higher you crit rate becomes, the more crit you lose, due to some of the blocked damage shots being shots that were supposed to crit, and thus the block negated the crit. If you have A higher crit rate, then more of those shots will be blocked. The assumptive value i am operating at now is 5 percent of shots will be blocked, so if 5 percent of your crits become blocked and thus negated as crits, then you will see it become more evident at higher levels of crit.

Diminished returns is something I have though always applied to crit, and I always saw passing evidence that at a certain point, stacking crit no longer had as much of an effect on DPS returns. This may be due to what we may be finding here, or it may be due to some other factors, such as a "cap" on effective crit. In any case, what we are finding here may be a case of "diminished returns" on stacking crit, when it comes to attacking boss level mobs.



This will help as well, as it will expand our range table for FA tests, and we can see if there is a higher "crit compression" at this level with FA as well.

If you look at my more recent tests, I am doing them in 1000 shot blocks, then doing the average myself, so I can get a better idea of the range of RNR per 1000 shots. I am shooting 1000 shots, and then recording my crit and such, as well as blocked shots, and then I am clearing Recount and starting a new record. This way we can observe the minute differences, as well as giving us the option of stopping a test midway and doing it in parts.

Also, as long as we keep the parameters such as spec, hit and crit the same, then we can keep expanding a test sample until we basically get no evidence of RNR being effective and corrupting our returns.

One of the other things I am doing is going to great pains to be EXACTLY hit capped, including re-gemming. This way, for now, I can eliminate the possibility of that extra 0.1 to 0.2 percent hit some how effecting my returns.



I have considered this as well, so in the future I am going to run a test at approximately the same crit rating I have in my previous test, but with ~ 11 percent hit from gear. I can achieve this by switching out my crit trinkets for hit trinkets, then using lethal shots to replace the lost crit. Lethal shots has already been proven as a non factor in crit returns. I have also considered, since I can craft some gear running a expertise test to see if it reduces the amount of blocked damage shots, and that in turn replaces my crit loss. I don't see expertise as a solution to this regardless, because itemizing for 5 percent expertise to gain back 1 - 2 percent crit seems like its not really a worthwhile trade off.



This is pretty much what I am getting at by observing blocked damage shots, and I explained it in the above responses to Rosamonde. It occurs to me we should run test on the level 80 dummy at appropriate hit cap, and see if we get the same amount of blocked damage shots as well. This would be a good way to try to determine if the blocked damage is a static value, or something that increases as level difference increases, thus meaning we suffer from more crit depression on higher level mobs, such as raid bosses.

If resilience was a factor, then we would see a set amount of crit reduction regardless of crit percentage, and we would also see a reduction in the crit damage averages compared to normal hits, something I observed as well and found no evidence of occurring. As an example, in Rosamonde's test, her hit average was 564, while here crit average was 1172, so she was actually getting crits at an average of 207.8 percent, and any significant amount of resilience would have reduced this percentage, I assume.