Back to the topic : Doubts in CM

[danyzn]

"What should we do if we do hit mine?"

"Well, Lieutenant, standard procedure is to jump 10 feet into the air, then scatter yourself over a wide area."

[dNorwood]

<BLOCKQUOTE>quote:</font><HR>Originally posted by Scipio:

Okay, here are the first testresults. The numbers are the numbers of shoots that were necessary to hit the target.

<HR></BLOCKQUOTE>

This looks cool - indulge me a bit and help me understand what's up...

<BLOCKQUOTE>quote:</font><HR>

Test 1:

Target 8 x M4 Sherman,

a) 3-4-2-2-3-9-1-15-4

6-5-4-3-1-7-5-3

c) 20-5-6-5-2-7-5

<HR></BLOCKQUOTE>

OK. First, each letter equals one run? Looking at , it took 6 shots to hit (and kill?) the first tank, 5 shots to hit and(and kill?) the second, etc. Eight numbers means each tank killed with the first hit. Nine numbers in a) means one of them lived through one hit? Only 7 numbers in c)? Twenty shots!? (There'd be a "do somefink" thread after that!!).

<BLOCKQUOTE>quote:</font><HR>

Test 2:

When a target were hit, do the gunners have a higher chance to hit the same target again?

Gun 88mm AA, Target 1 x Churchill VII, Distance to target 2000m, chance for a hit 10% (No lethal hit on target)

a)6-4-4-8-1-1-8-1-1-2-1-2-3-8

b)2-1-3-1-8-6-1-1-1-1-1-3-3-1-4-13

c)13-2-4-3-2-2-4-1-3-4-1-1-8-1

<HR></BLOCKQUOTE>

OK, this really gets to it. Number is number of shots until a hit, with a final kill at the end of the run?

I'd like to think about how to analyze this, (especially test two) so I'm not just wasting your time...(although I guess I'm being a bit dim - I want to make sure I understand before I spend any time on it..)

[Slapdragon]

dNorwood ,

If yoiu would like, I have an instrument that tests this also so you can look at it yourself. It lines up 16x pak 43 guns on one side, and 16 M4A3 on the other. Borg spotting was chosen. The tanks where constrained on a bridge, and "ditches were added to keep guns firing at their tank in question and not going for a side shot. In case firing went on more than 60 seconds, a group of high value units were stuck in holes at the side of the board to keep one side from surrendering.

With 160 tests, I found that:

1st round hit -- 25%

2nd round hit -- 33%

3rd round hit -- 70%

4th round hit -- 90%

Two problems, some tanks pop smoke (but seemed to still die, probably because I had them constrained on a bridge between two cliffs) so the smoke may skew things, and in this test the error grows greater for each round -- the 4th round hit is only based on 10 tanks compared to the first 160 and the second 118.

A warning about why my data and Scipio s is different -- our tests may be extremely different in units involved, method of imobilizing tanks, weather, etc.

My tests were done at 2000 meters, disarmed tanks, clear weather, regular units shooting and receiving fire. E-mail me for the test file if you would like it.

[ 07-20-2001: Message edited by: Slapdragon ]

[Scipio]

dNorwood

I hope my bad English won't cause problems

The numbers 1-8-9... means this :

1st shot hits

8th shot hits

9th shot hits...

In other words

The gun shot 1 time and hit. Then the gun misses for 7 times and the 8th shot is a hit again...

Indeed you shouldn't compare the results with Slapdragon's results, cause Slap doesn't answer the question (I hope I understand him correct. This is very difficult for me to explain in English, I guess he will correct me if I'm wwrong)

His results show:

If you shoot one time one target, you will hit it with a propability of 25%

If you shoot two times on the target, you will hit it once with a propability 33%

I have no doubts that his results are correct,but the problem is

Test 1: Several targets close together with the same distance to the gun (the greatest problem to aim with a gun is an unknown distance to target). If the first target was hit, the distance to the next target is known. Because of that, the probability to hit the second target with the first or second shot should be higher then in was to hit the first target with the first or second shot.

Test 2: if you have the same target hit once already (so the target datas necessary for the aiming are known), you have an much higher chance to hit it again, usually with the next or next but one shot.

[Slapdragon]

<BLOCKQUOTE>quote:</font><HR>Originally posted by Scipio:

dNorwood

I hope my bad English won't cause problems

The numbers 1-8-9... means this :

1st shot hits

8th shot hits

9th shot hits...

In other words

The gun shot 1 time and hit. Then the gun misses for 7 times and the 8th shot is a hit again...

Indeed you shouldn't compare the results with Slapdragon's results, cause Slap doesn't answer the question (I hope I understand him correct. This is very difficult for me to explain in English, I guess he will correct me if I'm wwrong)

His results show:

If you shoot one time one target, you will hit it with a propability of 25%

If you shoot two times on the target, you will hit it once with a propability 33%

I have no doubts that his results are correct,but the problem is

Test 1: Several targets close together with the same distance to the gun (the greatest problem to aim with a gun is an unknown distance to target). If the first target was hit, the distance to the next target is known. Because of that, the probability to hit the second target with the first or second shot should be higher then in was to hit the first target with the first or second shot.

Test 2: if you have the same target hit once already (so the target datas necessary for the aiming are known), you have an much higher chance to hit it again, usually with the next or next but one shot.<HR></BLOCKQUOTE>

You are partly right, but my esults are based on 1st shot probability, 2nd shot probability etc, but are based on percentage chance to hit of remaining units. That means on shot one I count the number of tanks hit (even if they were not killed) divide by the number of guns - tank firing combinations. On the second shot, I drop the gun / tank combinations that have hit from consideration, and just consider the guns that have missed the first shot. This allows me to only test 2nd shot hits. I do this through shot 4. This is the simplelist way to test if the second shot has a greater chance to hit than the first shot, as long as I can isolate variables. It also allows for easy stats to determine error, had it been a big issue (it was not a big issue except for shot 4, where shot three's results fall into the error range of shot 4's results)

So, when firing at a single target, the answer is a qualified yes that the chance of hitting goes up with each shot (qualified because I only tested one gun / tank combination in one setting, but because this is software rather than real world it is likely I also got to the engine in my test).

As for your tests, I also need to think about how to look at your data objectively. I have sent my instrument to dNorwood to use as a cross check on my data, and I am going to try and devise a test to see if I can look at your data.

As for your cluster of tanks, it raises the interesting question: does hitting at range carry over an advantage when firing at an adjacent target. I can think of some logical reasons why it would, and some logical reasons why it might not actually cause an effect that can be seen. On the one side, in the case of a line of tanks with a predictable interval, an AT gunner should get a bonus for the ranging shot similar to the bonus of an established ambush point. On the other hand, a set of tanks moving to attack, through dust and smoke, firing back at the gunners, I am not sure what sort of recording keeping they did in the heat of battle (as opposed to prepared ambush).

I will have to think on this and get back to you.

[WWB]

While I have done no scientific tests, I have noticed that your hit chance remains nearly the same if shooting at an adjacent target at range.

WWB

[Scipio]

<BLOCKQUOTE>quote:</font><HR>Originally posted by Slapdragon:

As for your cluster of tanks, it raises the interesting question: does hitting at range carry over an advantage when firing at an adjacent target. I can think of some logical reasons why it would, and some logical reasons why it might not actually cause an effect that can be seen. On the one side, in the case of a line of tanks with a predictable interval, an AT gunner should get a bonus for the ranging shot similar to the bonus of an established ambush point. On the other hand, a set of tanks moving to attack, through dust and smoke, firing back at the gunners, I am not sure what sort of recording keeping they did in the heat of battle (as opposed to prepared ambush).

I will have to think on this and get back to you.<HR></BLOCKQUOTE>

Of course that's a general problem of those tests - they can't be compared with battle conditions for many reasons, only to find out the gun accuracy on a test range. We can try to compare the CM test range with the results from real test ranges how they are shown on the 88-page (link posted somewhere above). Compared to that, the results are okay with 2 - 4 shots needed for the first hit. To bad that I can't find a table with results for the following shots.

[dNorwood]

<BLOCKQUOTE>quote:</font><HR>Originally posted by Slapdragon:

As for your tests, I also need to think about how to look at your data objectively.

<HR></BLOCKQUOTE>

This is pretty much where I am - thinking about how to think about this. Your reduction of data addresses the issue of eventually getting a hit, but I think the two important questions here are:

A) IF I hit a tank, is it easier to hit one next to it; and

2) IF I hit a tank, is it easier to hit it again.

Re A): Irregardless of whether or not it works that way in real life (or not), I suspect that it DOESN'T work that way in the engine - simply because there doesn't seem to be that level of "memory" in the units. It seems that a simple Mean Time Between Hits or Average/Standard deviation of raw data would be a sufficient means of data reduction on this score.

re 2): This to me is the more interesting issue. (I remember that it's been discussed before - no really! - but the search times out - if anybody can point to the threats[sic] I'd appreciate it.) And it seems that the data reduction is also more involved. This is what really got me interested - how do we make Scipio's data talk to us? (And I think it's the right experiment to do).

<BLOCKQUOTE>quote:</font><HR>

I have sent my instrument to dNorwood to use as a cross check on my data, and I am going to try and devise a test to see if I can look at your data.

<HR></BLOCKQUOTE>

Got it - thanks (incidentally, I'm at work so it'll be this weekend before I can do any experiments). Also, any ideas about data reduction would be helpful.

Scipio:

Did you save your experiment? I'd enjoy a look. You can email to

dnorwood@intrigueweb.com

if you care to.

[ 07-20-2001: Message edited by: dNorwood ]

[Slapdragon]

I can say with a great confidence that the answer to the question:

A) If a tank is hit is it easier to hit again?

Is yes, for the same reason that it becomes progressively easier to hit a tank that you miss. I tested this with Jumboes and 50mm AT guns and came up with a data set that I will post tonight or tomorrow.

If a tank is hit, will one at the same range be easier to hit?

My theory here is the answer is no, because I do not believe the AI has, and my testing has shown that this to be the case in other tests, a memory once it switches targets.

To test this, I am going to take my normal firing range and put 4 Shermans in front of each Pak 43. The Pak 43 has a 99.9% chance of killing one tank in 4 shots. I will let each gun fire at a tank until it hits, then see if shot 1 at tank 2 has a significant difference in to hit from shot 1 at tank 1, and so forth through tank 4. I will also test for significance in shot order to see if it continues to get better in progression despite the switched shot.

If the answer is no, the first tank is fired at with about a 25% first round chance to hit instead of something better than 50%,, and if it is a significant finding, then we can assume the AI keeps no memory of shot ranges for adjacent tanks.

The question then would become, should it, and if it should, in what situations would it happen. Finally, accepting that it cannot be coded into CM: BO and CM: BB, can it be coded into CM-II, and if so, how would that be done.

[Scipio]

<BLOCKQUOTE>quote:</font><HR>Originally posted by Slapdragon:

I can say with a great confidence that the answer to the question:

A) If a tank is hit is it easier to hit again?

Is yes, for the same reason that it becomes progressively easier to hit a tank that you miss. I tested this with Jumboes and 50mm AT guns and came up with a data set that I will post tonight or tomorrow.

If a tank is hit, will one at the same range be easier to hit?

My theory here is the answer is no, because I do not believe the AI has, and my testing has shown that this to be the case in other tests, a memory once it switches targets.

To test this, I am going to take my normal firing range and put 4 Shermans in front of each Pak 43. The Pak 43 has a 99.9% chance of killing one tank in 4 shots. I will let each gun fire at a tank until it hits, then see if shot 1 at tank 2 has a significant difference in to hit from shot 1 at tank 1, and so forth through tank 4. I will also test for significance in shot order to see if it continues to get better in progression despite the switched shot.

If the answer is no, the first tank is fired at with about a 25% first round chance to hit instead of something better than 50%,, and if it is a significant finding, then we can assume the AI keeps no memory of shot ranges for adjacent tanks.

The question then would become, should it, and if it should, in what situations would it happen. Finally, accepting that it cannot be coded into CM: BO and CM: BB, can it be coded into CM-II, and if so, how would that be done.<HR></BLOCKQUOTE>

Not necessary, Slap, that's what I tested in Test 1. 98% of the hits were deadly, so the gun switched to the next target. And you see the result.

Well, it's a difficult question if it's possible to get that level of realism without extraordinary efforts. This would also rise the question for using landmarks with known distance to the gun (a standard procedure in reality) etcetra... Of course it would be fine if we will see something like that...someday. But I guess I can wait for it. As it was already said - the conditions in a battle has also a great effect.

About question two : I guess that could be realized much easier, as for example with a bonus to the hit propability?

dNorwood, I've sended the files.

[dNorwood]

<BLOCKQUOTE>quote:</font><HR>Originally posted by Slapdragon:

I can say with a great confidence that the answer to the question:

A) If a tank is hit is it easier to hit again?

Is yes, for the same reason that it becomes progressively easier to hit a tank that you miss.

<HR></BLOCKQUOTE>

Scipio can confirm or deny, but I think his claim (one that I've heard before) is that a hit should DRAMATICALLY enhance follow on hits - more so than a simple near miss. I agree with your call for real evidence of this (no matter how reasonable an idea may be, the bottom line is "can you prove it"), but I think it's a different claim than that to which you attest ... to.

[dNorwood]

<BLOCKQUOTE>quote:</font><HR>Originally posted by dNorwood:

I agree with your call for real evidence of this <HR></BLOCKQUOTE>

as in - evidence that in real life a hit greatly increases the chance of a follow on hit.

[Slapdragon]

<BLOCKQUOTE>quote:</font><HR>Originally posted by dNorwood:

Scipio can confirm or deny, but I think his claim (one that I've heard before) is that a hit should DRAMATICALLY enhance follow on hits - more so than a simple near miss. I agree with your call for real evidence of this (no matter how reasonable an idea may be, the bottom line is "can you prove it"), but I think it's a different claim than that to which you attest ... to.<HR></BLOCKQUOTE>

This is how you can test the issue of subsequent hits:

Using my firing range, change all 88 Pak 43 to 50mm, and M4A3 to M4A3E2. This will allow you to make lots of hits at 2000 meters.

Run the turn in 5 second blurbs, and watch the guns. When a Jumbo is hit, track the chance of a hit afterwards. If the hit for a low powered weapon such as the 50mm at 2000 meters is fairly high, then I would say that it is modelled, even if it is not modelled specifically. The question here is how high should high be. I would be apt to say a hit chanche of above 75% for subsequent fires is high enough to make the whole issue moot -- in other words, it could be simulated, but the game handles it in a way that was not intened but works well enough.

[Slapdragon]

<BLOCKQUOTE>quote:</font><HR>Originally posted by dNorwood:

as in - evidence that in real life a hit greatly increases the chance of a follow on hit.<HR></BLOCKQUOTE>

Yes. I think this is not a given at all. Nor is the chance of hitting a unit next to another unit with an AP (I already have evidence that direct fire HE that getting range on one unit can be used to fire on other units easier).

[Vanir Ausf B]

CM does not model hits increasing to hit chance more than misses.

After about 5 shots, the to hit chance "maxes out" in CM. For example, Pak40 vs. Churchills max out at 41% to hit at 1600m. No matter how many times they fire at that same Churchill, they will never exceed 41% to hit.

Against a stationary target, a hit should almost always dramatically increase to hit chance of following shots. Most guess work in gunnery is in estimating range. When an AT gun (or tank) hits a stationary target, the range to that target is known. Hitting it again afterword is much easier.

Chance of first shot hits against any target near the first one would be higher as range is already known.

[Scipio]

<BLOCKQUOTE>quote:</font><HR>Originally posted by dNorwood:

Scipio can confirm or deny, but I think his claim (one that I've heard before) is that a hit should DRAMATICALLY enhance follow on hits - more so than a simple near miss. I agree with your call for real evidence of this (no matter how reasonable an idea may be, the bottom line is "can you prove it"), but I think it's a different claim than that to which you attest ... to.<HR></BLOCKQUOTE>

I confirm

[dNorwood]

First, my apologies for bringing this all back up...

<BLOCKQUOTE>quote:</font><HR>Originally posted by Slapdragon:

I can say with a great confidence that the answer to the question:

If a tank is hit, will one at the same range be easier to hit?

My theory here is the answer is no, because I do not believe the AI has, and my testing has shown that this to be the case in other tests, a memory once it switches targets.

<HR></BLOCKQUOTE>

I think I can confirm this answer(even though I suspect nobody doubts it...). I used Scipio's experiment (see above) except that I buttoned the Shermans and pointed them away from the 88.

I did 12 runs added to Scipio's 5 (that get's me ahead of Steve's "magic 14"). The median of number of shots to get a hit was:

5-4-6-4-4-5-5-3

(first is first tank, second is next tank, etc.).

I think you'd be hard pressed to argue any benefit. So, the next order of business would be to argue (in the scientific sense - as described by Slapdragon) that this is wrong and should be (eventually) fixed.

I did a large number of experiments with the second point (do hit's enhance follow on hits more than misses), and got results that look pretty much like Scipio's. But I'm still thinking about the analysis. It seems that this is a conditional probability question (is Prob. of hit GIVEN a previous hit -greater- than Probability of hit GIVEN no previous hit) and I need to find some way to automate the data grinding to answer it.