Accuracy While Firing On the Move (OR FAST) in CMBO

[opfor6]

I agree with Treeburst. At 500 yds/m it should be very difficult to hit even if the ground is nice and level. Anybody that has tried to go cross country in anything but a M-1 or a Leo 2 at 20 MPH should know how uneven unprepared ground is. You spend most of your time trying not to bang against the inside of the vehicle. We could also get into barrel whip, cant, obscuration of the target due to muzzle blast and crew quality.

It was mentioned early in the thread that a gunner could tell the driver to slow down or take it easy for a second so he could take a shot. This is a very valid point, a short stable period would allow a gunner to get a shot off. This period could be anthing from a slowing to finding a less bumpy path. I don't think it is very easy to model that in the CM graphics engine but could be accounted for in the formulas for crew quality. I have not checked the stats on this but I think the Stuart has a fast turret and Gyro. These are a very powerful combo. It allows for a quicker lay on target during those stable moments. Combine this with a fairly accurate gun and the somewhat lax spotting rules, bada bing, you have a nice platform.

In my earlier post about TD tactics I tried to point out that tank hunters spend alot of time moving from one firing position to another. This is true on the attack as much as the defense. If there are no more worthy positions then a charge might be in order. Sending rounds down range as fast as possible whether on the move or stationary is preferrable to being passive and hoping the enemy doesn't pop you.

I have not run any tests to prove or disprove the theories here. I only know what I have done and what has happened in games I have played. If you are someone that likes to send tanks out on fairly unrealistic moves to take a chance on a flank shot go for it.

[Battlefront.com]

Treeburst/Opfor,

I've asked Charles what he thinks of Treeburst's findings. I agree with Opfor. If Charles didn't intend momentary halts/slowdowns while on Fast, I'd say the 31% is too high at 400-500m range. It is worth a looksee in any case

I also totally agree with Opfor about the tactics thing. For example, my test was done under "lab conditions", and therefore can be seen as some sort of best case result. But as I said, once I gave the Panthers ammo and/or cover the whole set of results changes. So I guess I would be curious to see what Treeburst can come up with, using the same scenario, but tossing in some common variables like terrain or armed targets.

Steve

[Treeburst155]

My test was also done under lab conditions, but even for lab conditions I think the Stuart hit percentage is too high.

I will run tests giving the targets some ammo to fight back with. I will run the test until the Stuart has managed 100 shots. Since the Stuart will probably be killed often I will have to run it many times.

Edit: The Stuart doesn't live long enough to test, even against conscripts. LOL!

Here's my point. In an actual CM battle a Stuart has a 31% chance of hitting an average size stationary enemy vehicle at 500 meters while fast moving over clear terrain for AS LONG AS HE LIVES.

If the Stuart lives long enough to get off one round the chances are 31% at 500 meters that he will hit that enemy AFV somewhere WITH THAT ONE ROUND. This is the kind of behavior that started this thread, I'm sure. I've seen that 31% chance pay off too many times, probably 31% of the times the situation comes up.

Thanks, Steve, for mentioning this to Charles. I really appreciate it, whether the game is tweaked or not. It's still a great game and you guys are great for listening to our gripes.

[ 07-28-2001: Message edited by: Treeburst155 ]

[Battlefront.com]

Treeburst:

<BLOCKQUOTE>quote:</font><HR>My test was also done under lab conditions, but even for lab conditions I think the Stuart hit percentage is too high.<HR></BLOCKQUOTE>

I am inclined to agree. I'm just curious to see how much a bit more "realistic" surroundings might impact the test results. It could help identify, or rule out, factors which might be making the hit percentage too high.

<BLOCKQUOTE>quote:</font><HR>I will run tests giving the targets some ammo to fight back with. I will run the test until the Stuart has managed 100 shots. Since the Stuart will probably be killed often I will have to run it many times.<HR></BLOCKQUOTE>

Thanks. I'm curious to see what you turn up.

Steve

[Treeburst155]

Steve, see my edit to my previous post; and get some sleep time in this week.

[Slapdragon]

Treeburst, I always test with no ammo in the target tanks, and some wau yp keep them stationary on level ground (like making a lake with an island.(

I am running tests right now on this.

[Slapdragon]

I can directly confirm Treeburst's observation with my own. On a test field that included three tracks designed to shield a Stuart as a turn starts then allow it to run the length of the track, in clear weather, on flat ground, firing at a disarmed Tiger tank, Stuarts hit 32 percent of the time at 450-500 meters. Crews were regular.

This seems high from discussion in the Hellcat oral histories, where the Cats used their ability to pump a dozen round down range or more a minute when they were firing at speed to make hits, knowing that tests of gyros after the war had them only adding around a 3/.10 chance to hit under perfect conditions.

Likely firing on the fast move is over modelled.

If anyone wants my test range, let me know and I will e-mail it.

[Treeburst155]

That extra 1% you got was probably because of the larger profile of the Tiger compared to my PzIVs. My victims were getting off some smoke in the last 30 seconds too which caused the Stuart to lose LOS and the targetting line briefly. I figure that just cut down on the average ROF. The hit percentage is still the same.

[Slapdragon]

<BLOCKQUOTE>quote:</font><HR>Originally posted by Treeburst155:

That extra 1% you got was probably because of the larger profile of the Tiger compared to my PzIVs. My victims were getting off some smoke in the last 30 seconds too which caused the Stuart to lose LOS and the targetting line briefly. I figure that just cut down on the average ROF. The hit percentage is still the same.<HR></BLOCKQUOTE>

Not even that really, when you do a test with less than ten thousand runs (which no one can really do), you have an error rate because you could have rolled all snake eyes 30 times in a row. Your tests was a great indicator, but not knowing how many times you ran it, I ran one that resulted in 128 shots which is high enough that the error starts getting low. That it compares with your tests just shows you are on the right trail, but if I had gotten say 27% or 35% it would have still been "in the ball park." What really would not have been was if I had gotten 8%. That would indicate that somehow our tests where different, and that was causing your high accuracy.

I am going to rerun the test with 75 armed Shermans and 75 armed cromwells to see what the change is. I already know that the Cromwell does not hit very well at that range and fats move, while the Sherman did not hit at all in my last tests, but I have a different board now.

[ 07-28-2001: Message edited by: Slapdragon ]

[Treeburst155]

I ran my test until I had a total of 107 shots fired by the Stuart (the end of ten runs). The little guy hit 33 out of 107 with 5 track hits counted in the 33 hits.

[gunnergoz]

Forgive me if this has already been discussed, but has anyone considered the effect of approach angle between the moving tank and its target? This "angle off" would in my thinking be a major element in any equation attempting to simulate to-hit probabilities.

Simply put, the closer the target is to dead ahead of the firing tank, the less turret traverse has to be factored in as a variable. And the more that the target approaches a 90 degree angle off, the faster the turret has to be slewed to keep up with the target.

So my question is...how are all our little simulations factoring this issue in? If we are using different approach angles for these tests, then the results are difficult to equate to one another.

One suggestion...run the test with the approach angle being 90 degrees off the target, i.e. the firing tank is forced to display maximum variation of turret traverse throughout the exercise.

Alternately, run the test with the tank having a constant angle on the target (i.e. "circling the target" like movie injuns circling the wagons, and firing all the while.)

I'll bet the circling tank gets a lot more hits, or should in my admittedly fuzzy understanding of this sort of thing.

Any of this make sense? If it doesn't, have another glass of wine or a beer and it will.

(You math/geometry/trig/calculus wizzes out there will have to help me out if my logic is faulty here and explain just how it is...)

[Treeburst155]

My Stuart was running laterally to his victims. His first shots were fired with turret about 45 degrees to right. His last shots were fired at about 135 degrees. Lots of tracking involved. No running straight at the targets. The turret moved this far in one minute of firing. Roughly 1.5 degrees per second on the average.

[ 07-29-2001: Message edited by: Treeburst155 ]

[Treeburst155]

A regular Stuart crew moving fast on open terrain lateral to his targets (turret 60-120 degrees)at a range of 215-245 meters hits what he aims at 43.3% of the time. They hit 45 times out of 104 shots with one track hit included in my latest test. Them boys is good!!! Imagine what a crack crew could do.

[Claymore]

Bump

[Triumvir]

Dredging up horrors from the depths of my memory, I seem to recall in my Stat 101 class an algorithm for determining the minimum population size necessary to prove a hypothesis at some confidence level. Can anyone refresh my memory so that I can see whether it can apply to CM and thus make our tests a little less anecdotal?

[Treeburst155]

Triumvir,

You don't need Stats 101 for this. The more test shots you fire the lower your margin of error. A stats person could probably tell you the margin of error for the tests run above but they are within +/- 5 percent I'll bet. I would hardly call 104 shots anecdotal.

[ 07-30-2001: Message edited by: Treeburst155 ]

[Slapdragon]

Standard Error = Z Score sub alpha divided by 2 times sample standard deviation divided by the square root of the number of samples you took. Z score is determined from a chart by how much chance of an error you want to allow (1 in 20 is considered best for most things) and the figure you get is the range away from the mean at which you would not be surprised to find the answer.

For eample, the data set that follows is the number of shots needed to achieve a hit at x range.

3

2

4

5

6

8

8

2

4

8

5

3

2

2

1

4

5

3

5

8

7

5

2

3

2

1

2

3

8

7

Using that formula above we get mean of 4.2667 shots, a standard deviation of 2.3183 and a standard error of .4233 for the mean. That means we are 95% sure that it takes from 3.4 and 5.1 shots on the average to achieve a hit at the range we were firing at. If someone else gets a mean of 5, then we are right on the money.

I will have to look up the formula for reversing this, since I usually test normal curve items with either 100 or 400 individuals in the population. Also note if you are testing something that has a yes / no answer, the test is different.

[ 07-30-2001: Message edited by: Slapdragon ]

[imported_Wildman]

Treeburst,

Send me your data, and I'll find the standard dev, etc. IF you want.

[Slapdragon]

You can send me your data and I will run it into SPSS or SAS for the full treatment. If you put your data into an Excel file, either as a list of numbers (75 hit out of 100 on first shot, 65 on second) or as a list of how many shots it took to hit, I will figure error for that particular test.

[Triumvir]

Treeburst, I don't mean to call your runs anecdotal. But there is some algorithm I recall that pins down exactly what your minimum population size should be; if we can find that, then we can say with certainty that something appears wrong and make a good case for fixing it.

Slapdragon, what you posted is almost exactly what I was talking about, except that if I remember correctly, you can also fix the minimum size so that we can determine if the set is sufficiently large to make predictions from. I honestly can't recall what the average samples should be; it's been a _long_ time since Econometrics or Stat 101.

[Slapdragon]

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

Treeburst, I don't mean to call your runs anecdotal. But there is some algorithm I recall that pins down exactly what your minimum population size should be; if we can find that, then we can say with certainty that something appears wrong and make a good case for fixing it.

Slapdragon, what you posted is almost exactly what I was talking about, except that if I remember correctly, you can also fix the minimum size so that we can determine if the set is sufficiently large to make predictions from. I honestly can't recall what the average samples should be; it's been a _long_ time since Econometrics or Stat 101.<HR></BLOCKQUOTE>

Well, turning that around, n (the number of test runs to create a set error at a confidence interval of 95%) equals 1.96 time .05/2 times mean divided by the margin of error.

[Battlefront.com]

Folks,

Charles got back to me and reported this:

<BLOCKQUOTE>quote:</font><HR>Moving fire accuracy is based on page 59 of Jentz's _Tank Combat in North Africa_. The Brits did a test firing a 2pdr gun from one of their moving tanks at an 8ft x 8ft target.

When moving broadside to the target at 650 yards:

Speed Rounds fired Hits Percent

10 mph 19 4 21

15 mph 10 3 33

20 mph 6 2 33

When moving head-on toward the target from 900 to 600 yards at 10mph, 14% hits (# of rounds fired not specified).

These numbers are higher than you might have expected. The 2pdr was fired from the gunner's shoulder (!) and had no gyrostabilizer. So I think the gyro-equipped Stuart, firing at shorter range in your example (about 580 yds on average) compared to this one (about 700 yards on average), ought to be able to do even better.

We can tweak things in CMBB if necessary but I don't think the moving accuracy is too far off the mark in CM1.<HR></BLOCKQUOTE>

While this is not an exact match to situations we have been describing above, I think it does show that the accuracy for this small calibre gun was better than one would expect.

Steve

[Treeburst155]

Well, that's good enough for me. Case closed as far as I'm concerned. Thanks again, Steve, for checking that out with Charles.

[Triumvir]

Likewise, Steve... although I must confess to being a bit shocked at the results. They seem much higher than one would expect... proving that gut feelings can be nasty treacherous things.

[Battlefront.com]

You will also notice that accuracy was BETTER at faster speeds than at slower. I totally forgot about this, but a bunch of physics/engineering weenies had a big discussion about this here some time ago. Basically, if the suspension system of the vehicle was even decent, on moderately bumpy terrain a faster vehicle would actually bounce around less because the vehicle would more or less "fly" over rough spots instead of pitching the vehicle up and down over them.

IIRC the factors involved were speed, weight, suspension system, and how bumpy the terrain was. For the British tests that Jenz quotes, I would imagine that it was ideal for seeing an improvement from going fast as opposed to slow.

Steve