I believe combat is rigged in this game and it ruins it for me

[DrSpike]

Hmm, lots of junk here, largely from people who posted to correct others 'misconceptions'.

If as the first post suggests it has been tested with 50 iterations with a true 50/50 situation and the AI won 80% of the fights then clearly there is a problem somewhere.

A string of odd results can occur by chance, as several have posted, but anyone with sufficient skills to crank through the binomial calculations can see that winning 40 out of 50 in independent bernoulli trials with fixed probability of 0.5 is incredibly unlikely, certainly unlikely enough to warrant further investigation.

A control is unnecessary in this case, as there is a clear counterfactual to use in the distribution (both mean and variance moments) that occurs under Bernoulli trials with probability .5 carried out 50 times.

I really don't understand the hostility some have showed against the original poster. If his test is fair (which it might not be, for some reason we don't know) then he has identifed a problem, period.

Clearly Soren has no incentive to hide anything. However the possibility of a problem with the random seed or even an unintended combat bug can't be ruled out. Nor, of course, can the fact that the original poster did not conduct a fair test by accident.

[Lansing]

Originally posted by DrSpike
Hmm, lots of junk here, largely from people who posted to correct others 'misconceptions'.

If as the first post suggests it has been tested with 50 iterations with a true 50/50 situation and the AI won 80% of the fights then clearly there is a problem somewhere.

A string of odd results can occur by chance, as several have posted, but anyone with sufficient skills to crank through the binomial calculations can see that winning 40 out of 50 in independent bernoulli trials with fixed probability of 0.5 is incredibly unlikely, certainly unlikely enough to warrant further investigation.

A control is unnecessary in this case, as there is a clear counterfactual to use in the distribution (both mean and variance moments) that occurs under Bernoulli trials with probability .5 carried out 50 times.

I really don't understand the hostility some have showed against the original poster. If his test is fair (which it might not be, for some reason we don't know) then he has identifed a problem, period.

Clearly Soren has no incentive to hide anything. However the possibility of a problem with the random seed or even an unintended combat bug can't be ruled out. Nor, of course, can the fact that the original poster did not conduct a fair test by accident.

That's a great post, seriously.

It's possible I have had the weirdest run of results, or that the tests were not fair, though I took pains to make them so. I was getting extremely frustrated with poor combat results and wanted to find out what was going on. I still don't know, but I do know I am uninterested in playing the game in its current form, as I enjoy wars a lot more than diplomacy.

[Soren Johnson]

Originally posted by Lansing


That's nice, but it won't change the results that are happening to people right now. Has it crossed your mind that something you didn't intend in the code might be happening to players?

If you have a save where you can display outrageous combat results quickly over a large sample size (I would recommend using the WorldBuilder) then I would be happy to take a look at it.

[Lansing]

Originally posted by Soren Johnson Firaxis


If you have a save where you can display outrageous combat results quickly over a large sample size (I would recommend using the WorldBuilder) then I would be happy to take a look at it.

Fantastic. I'll give it some time when I have some to give.

Hopefully I do a series of trials again and they come out "normal" though.

[Fried-Psitalon]

Originally posted by Lansing


That's nice, but it won't change the results that are happening to people right now. Has it crossed your mind that something you didn't intend in the code might be happening to players?

Your arrogance is somewhere between astounding and dismaying. Has it crossed your mind that your tests may be invalid or poorly designed? That your issue may be psychological?

That, just maybe, 30-some-odd posters and the man who coded the game might be right, and you might be wrong?

Silly me. Of course not.

For someone who is telling other people to relax, you're sure hanging on to your point with both hands, with your face turning purple.

[Fried-Psitalon]

Originally posted by DrSpike
I really don't understand the hostility some have showed against the original poster.

Then you missed his dismissive attitude towards other posters and his willingness to display arrogance towards the game's designer while displaying total unwillingness to have his "testing" duplicated or explained or even repeated by him, which is pretty standard procedure when claiming a statiscal proof?

Originally posted by DrSpike
Nor, of course, can the fact that the original poster did not conduct a fair test by accident.

Or on purpose, with the intent of simply drawing attention and activity to himself. That's far from unheard-of around here, too.

[Lansing]

Originally posted by Fried-Psitalon


Your arrogance is somewhere between astounding and dismaying. Has it crossed your mind that your tests may be invalid or poorly designed? That your issue may be psychological?

That, just maybe, 30-some-odd posters and the man who coded the game might be right, and you might be wrong?

Silly me. Of course not.

For someone who is telling other people to relax, you're sure hanging on to your point with both hands, with your face turning purple.

What the hell are you talking about? I said I'd send in more test data to Soren when I have the time (for some of us work and life get in the way of civ occassionally).

I've also explained my tests and the results.

Relax buddy, it's just a game.

[DrSpike]

Originally posted by Fried-Psitalon


Then you missed his dismissive attitude towards other posters and his willingness to display arrogance towards the game's designer while displaying total unwillingness to have his "testing" duplicated or explained or even repeated by him, which is pretty standard procedure when claiming a statiscal proof?

Well tbh I'd say these occured after people had posted to dismiss his original post on dubious grounds. Reread the whole thread. As for Soren, I'm sure he is being honest with us, why wouldn't he, but it wouldn't be the first time that a patch had unintended side effects.

I repeat, if the original results are sound, then there is a problem. If. The longer it goes on without anyone else verifying them then the more I'll think it's a prank.

[Lansing]

Originally posted by DrSpike


Well tbh I'd say these occured after people had posted to dismiss his original post on dubious grounds. Reread the whole thread. As for Soren, I'm sure he is being honest with us, why wouldn't he, but it wouldn't be the first time that a patch had unintended side effects.

I repeat, if the original results are sound, then there is a problem. If. The longer it goes on without anyone else verifying them then the more I'll think it's a prank.

It's not a prank. It would be fantastic if someone else who has the time could do some tests. I'd be *happy* to be wrong about this.

[Mergle]

Well, the OP appeared to describe a proper test which had given surprising results, and I was disappointed that there were so many replies which didn't address this - just brought up the usual perception issue, which is true for a real game, but not for the test described.

So it concerned me enough to try a test out. Quickly generated a random map at monarch level and plonked down 120 of my destroyers (meant it to be 100 but lost count!) and 120 of saladin's with WB. No coastal squares. Attacked with all my destroyers... and the result was: 55 wins out of 120.

Well, that looks as near to 50% as I could have expected. Certainly if there was any pro-AI bias, I would have expected much more skew. It's only one datapoint, but then so is the original post. And I'm going to attach my save (before combat and after) so you can verify [never tried this before so may not work). I note the OP has no means of proving his results as yet.

So until someone comes up with clear evidence, there seems no reason to think there is any bias in the combat system

http://apolyton.net/upload/view.php?file=860_TestingBC-4000.Civ4SavedGame

http://apolyton.net/upload/view.php?file=860_TestingaBC-4000.Civ4SavedGame

[vee4473]

One little question: It was asked before, I think, about the level on which the OP's test was done. Noble is the only advertised "equal" level.

So, how can you test combat strengths on Monarch and be surprised if the AI is programmed to have an advantage?

[Urban Ranger]

Originally posted by DrSpike
A string of odd results can occur by chance, as several have posted, but anyone with sufficient skills to crank through the binomial calculations can see that winning 40 out of 50 in independent bernoulli trials with fixed probability of 0.5 is incredibly unlikely, certainly unlikely enough to warrant further investigation.

Not necessarily. Since the original poster did not mention his methodology in performing his experiment, and it is highly likely that he is not a trained scientist (neither are you ), there is no assurance that he actually recorded what happened.

Excuse me for being cynical, but I have seen too many reports of anomalies or "miracles" that later turned out to have perfectly natural causes or are even common occurances for me to take this seriously.

I am not saying he's lying. I am saying he doesn't have sufficient training to interpret what happened.

Originally posted by DrSpike
A control is unnecessary in this case, as there is a clear counterfactual to use in the distribution (both mean and variance moments) that occurs under Bernoulli trials with probability .5 carried out 50 times.

As I said, you are not a trained scientist. A control is always, always necessarily in any experiment, regardless of how trivial it is.

The claim is that the program rigs against human players (when they play against the computer). Thus, you must rule out the possibility that the same kind of events also happen when a human plays against another human.

Originally posted by DrSpike
I really don't understand the hostility some have showed against the original poster. If his test is fair (which it might not be, for some reason we don't know) then he has identifed a problem, period.

See my sig

I did not see him performed a test, let alone a controlled test. He was upset because he lost some battles that he thought he should have won.

That's why I asked for it. To prove a point he must set up a controlled experiment, explain his methodology, and publish the results. However, the original poster appears to be uninterested in such an undertaking. Instead, he simply restates his results in different ways.

[vee4473]

Originally posted by vee4473
One little question: It was asked before, I think, about the level on which the OP's test was done. Noble is the only advertised "equal" level.

So, how can you test combat strengths on Monarch and be surprised if the AI is programmed to have an advantage?

Along with my above comment, I would like to register as one of those who have had...ummm, unlikely losses, but at the same time they were much, much, much less frequent than probable victories.

[Urban Ranger]

Originally posted by vee4473
One little question: It was asked before, I think, about the level on which the OP's test was done. Noble is the only advertised "equal" level.

So, how can you test combat strengths on Monarch and be surprised if the AI is programmed to have an advantage?

As I understand it, combat resolution is not skewed at any level, with the exception of fighting against barbarians (low levels for the human player, all levels for computer players).

[Vlad Antlerkov]

Inspired by Mergle, I did my own tests. 1.52, Asoka vs. Bismarck, Noble, destroyers on open water. 200 combats; for the first 100, I recorded the HP remaining on the winner (and for the second 100, I got lazy). The final tally was 98 for the player, 102 for the AI-- close to 50/50. (For the first 100, it was 50/50 exactly.)

The HP remaining after each battle:

Player, 24 hp: 5 (30 damage inflicted)
Player, 18 hp: 9 (108 damage inflicted)
Player, 12 hp: 14 (252 damage inflicted)
Player, 6 hp: 22 (528 damage inflicted)
AI, 6 hp: 20 (480 damage inflicted)
AI, 12 hp: 12 (216 damage inflicted)
AI, 18 hp: 15 (180 damage inflicted)
AI, 24 hp: 6 (36 damage inflicted)

Total damage inflicted by the human: 2412
Total damage inflicted by the AI: 2418

Since each round of combat has the loser lose 6 hp (in this case), this means that the human won 402 rounds of combat, the AI 404.

And the save: http://apolyton.net/upload/view.php?....Civ4SavedGame