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

[Mergle]

OK, just done a test of 100 knights v 100 pikemen. Knights won 33 times.

The combat odds calculator showed 27.0%, so p=0.27

Mean is clearly 27, variance np(1-p)=100*0.27*0.73=19.71, so standard deviation is sqroot(19.72)=4.44.

so my test was (33-27)/4.44=1.35 s.d.s from the mean.

Approximating to the normal as in my previous test and looking this up, we have 0.4115, so that's a 9% chance of this skewed a result or more.

That's not statistically significant (for a 2-tailed 5% test, it'd need to be 2.5% or less, 1-tailed and it would be 5%).

This means we have no evidence from this test to reject the null hypothesis that p=0.27, i.e. no bias from what is given by the combat odds display.

I think that's it for me in this ballgame - the cumulative evidence seems clear that there's no bias in the combat system.

http://apolyton.net/upload/view.php?file=860_Testing2BC-4000.Civ4SavedGame

http://apolyton.net/upload/view.php?file=860_Testing2aBC-4000.Civ4SavedGame

[sabrewolf]

... a two-tailed 95% certainty Z-distribution test! i had that in statistics 1999 and when i was studying for my CFA last year. i was actually hoping that you would stop half way through your proof so i could brag about my knowledge

anyhow, well done! a sample size of 100 should be plenty and your H-0 has been rejected... so if soren's guarantee a few posts ago wasn't enough for anybody, this should be!


... unless of course mergle is a DL of soren and the test results are faked (never trust a statistic you did not fake yourself)

[SpencerH]

My 2 cents.

Its not unexpectedly losing a battle that is annoying, it's looking at the combat log and seeing 'streaks of hits' (as have been shown with the various screens) that pisses me off no end.

We had exactly the same argument with CIV3. Tests with large samples proved that the combat results came out 'as expected' based on the odds. That suggested that the RNG was 'streaky' (which is obviously' still the case in cIV since we can see the combat log).

[sabrewolf]

Originally posted by SpencerH
My 2 cents.

Its not unexpectedly losing a battle that is annoying, it's looking at the combat log and seeing 'streaks of hits' (as have been shown with the various screens) that pisses me off no end.

We had exactly the same argument with CIV3. Tests with large samples proved that the combat results came out 'as expected' based on the odds. That suggested that the RNG was 'streaky' (which is obviously' still the case in cIV since we can see the combat log).

the thing you call "streaky" is precicely the randomness that makes lucky-punches possible.
if you expect a 75% win chance to always be WWWL WWWL, the outcome would in fact be 100% victory. there must be streakyness to make the odds possible.

in civ3 "streakyness" had a bigger effect because you only had avg. ~4 hitpoints per unit, while in civ4 battles need more hits for fortune and less for preferable stats. that is why in civ4 a 10% strength difference can mean 20% better odds.

one thing i don't understand is why people who do not grasp (or do not want to grasp) probabilty theory keep insisting on rigged combat, unfairness and see for them unexplainable stuff as false.

[sabrewolf]

and unless somebody can actually PROVE things are wrong, they are not. in our western world we have to prove guilt, not innocence. in this thread it is the other way around. blamers and critics say it's wrong, while defenders prove it's not...

crazy

[WolfEyes]

and unless somebody can actually PROVE things are wrong, they are not. in our western world we have to prove guilt, not innocence. in this thread it is the other way around. blamers and critics say it's wrong, while defenders prove it's not...

I disagree, this isn't a criminal case, where innocence would (theoretically) be presumed. I'd say it's closer to a duel between opposing hypotheses.

[sabrewolf]

Originally posted by WolfEyes
I disagree, this isn't a criminal case, where innocence would (theoretically) be presumed. I'd say it's closer to a duel between opposing hypotheses.

you are correct that there are two different theories. but as i wrote earlier on, none of the people who say the game is biased towards the AI has actually based their theory on any practice. a 5-unit test just does not count... there must be 30 at least to be valid.
what i'm saying: initiate whatever hypothesis you want. but unless one can actually convincingly support it, do not insist it is true!

the opening post and several other posts are saying that the combat is rigged and the AI gets battle advantages. it may not be a criminal case, but such accusations are an insult towards soren and the other AI-involved programmers. in addition, anybody who contradicts since his post is in fact calling him a liar... which is an allegation.

[funat]

Combat is great.
Anyone saying different is nothing but a spoiled brat.

Buuuuh, I can't wipe out a civ of riflemen with my 20 armor stack.
Well, if that was possible, Croatia would cease to exist in the first days of summer 1991. As it had 0 (zero) tanks.

Now, when I unload a stack of armor on foreign continent, and they throw me back to the see, I fill happy. Cause that is what should be happening.

If you want to win a war, you have to plan. No more cakewalks with armor, because that just isn't possible in real life. Ask Americans.

A classic invasion of a foregn country includes hours of preparation, and all types of units now.

1. put spies to every enemy city
2. calculate turns they need to counter attack you on your attacking course from all angles of their empire
3. Begin hostilities by:
a) cuting and pillaging the roads with choppers and bombers
b) bombarding cities
c) destroying production with your spies 1 turn before a unit is complete

and all sorts of other stuff. Most important - know your enemy. Spies must be everywhere. No fog of war. Every single enemy unit must be seen and taken into calculation.

[SpencerH]

Originally posted by sabrewolf


the thing you call "streaky" is precicely the randomness that makes lucky-punches possible.
if you expect a 75% win chance to always be WWWL WWWL, the outcome would in fact be 100% victory. there must be streakyness to make the odds possible.

No, I expect that in any battle with equal type weapons (ie swords v spears v maces v clubs etc) that the chance to hit will be relatively equal and therefore both sides will usually take losses (hits). Some streaks of hits are reasonable (say 2-3 hits in a row). Streaks of 4-6 hits should be extremely unlikely but they will (and should) occur. My irritation is that they clearly occur too often. AFAIK this is a problem of RNG that are (by definition) not random and it seems to be a big problem for the RNG used by Firaxis.

in civ3 "streakyness" had a bigger effect because you only had avg. ~4 hitpoints per unit, while in civ4 battles need more hits for fortune and less for preferable stats. that is why in civ4 a 10% strength difference can mean 20% better odds.

To be honest, and in a general sense, I really dont see much improvement in combat from earlier civs compared to cIV ie I still spend too much time gnashing my teeth over improbable combat results.

On more than one occasion SM has said that civ is not a wargame (yet that is how it is most commonly played). Despite SM's comments over the years though, it seems to me that combat is much more prominant/important part of gameplay in cIV than in previous versions. In previous versions, for example, I would build 2-3 units/city (always upgrading). Now I find that even 6 modern units/city will often leave me in the middle of the 'power' scale and a target for more powerful civs. The development of a new/improved combat system also suggests that Firaxis has recognized the inadequacy of combat in previous versions of civ. What I wonder is why they have taken such an important first step (ie developed the better system) but continue to use poor combat concepts inherited from previous versions of the game? I think there is a huge potential with the cIV combat system but it still needs substantial improvement and fixing the streaks would be a good first step.

one thing i don't understand is why people who do not grasp (or do not want to grasp) probabilty theory keep insisting on rigged combat, unfairness and see for them unexplainable stuff as false.

I dont know, I accept the evidence. Maybe though, it's because it looks that way at first glance and hidden AI 'bonuses' are commonly used to provide human players with a challenge. Even with cIV, where Soren et al have been remarkably open, they have mentioned the previously unknown fact that there are AI bonuses at 'Noble level' (which is supposed to be equal). Given the potential for AI bonuses, its easy to accept that they are present as an explanation for bizarre results.

[sabrewolf]

Originally posted by SpencerH
No, I expect that in any battle with equal type weapons (ie swords v spears v maces v clubs etc) that the chance to hit will be relatively equal and therefore both sides will usually take losses (hits).

maybe i don't understand you correctly. but why should macemen and warriors have the same of striking? if you mean from realism point of view (you swing the blade the same fast as the club), then you may be right. but then a maceman hit should kill in the first go and a club wouldn't have big effect against him.
this is done by game design. it may not be the perfect solution, but not badly solved imho (better than in civ1-3 where only chance counted in regular combat and hitpoints were equal for all.

Some streaks of hits are reasonable (say 2-3 hits in a row). Streaks of 4-6 hits should be extremely unlikely but they will (and should) occur. My irritation is that they clearly occur too often. AFAIK this is a problem of RNG that are (by definition) not random and it seems to be a big problem for the RNG used by Firaxis.

it is true that a RNG "calculates" in random values and thereby it is not really random.
now let's assume that the chosen RNG algorithm produces more streaks than pure random would - if we'd have the formula (when the SDK is released?) we could confirm/contradict this. so assuming streakyness... this would have no effect on the computer having advantages as the human would equally often receive a lucky-streak right?

and also: for the distribution (binomial is actually more appropriate, but the normal distribution is the calculation base, i assume) also the so called "kurtosis" is important. this defines how the bell curve is formed. it is similar to the variance/st.dev but responsible for the shape.
a RNG that produces more streaks than pure randomness would represent a higher kurtosis value.

To be honest, and in a general sense, I really dont see much improvement in combat from earlier civs compared to cIV ie I still spend too much time gnashing my teeth over improbable combat results.

not much really changed... except what i mentioned above (with the "power" of each hit), which i think is a good move.

On more than one occasion SM has said that civ is not a wargame (yet that is how it is most commonly played). Despite SM's comments over the years though, it seems to me that combat is much more prominant/important part of gameplay in cIV than in previous versions. In previous versions, for example, I would build 2-3 units/city (always upgrading). Now I find that even 6 modern units/city will often leave me in the middle of the 'power' scale and a target for more powerful civs. The development of a new/improved combat system also suggests that Firaxis has recognized the inadequacy of combat in previous versions of civ. What I wonder is why they have taken such an important first step (ie developed the better system) but continue to use poor combat concepts inherited from previous versions of the game? I think there is a huge potential with the cIV combat system but it still needs substantial improvement and fixing the streaks would be a good first step.

this only proves that the AI is either more sophisticated, more aggressive or has more bonuses or a combination of all. but the combat model is not really influenced by the larger necessity of defence.

I dont know, I accept the evidence. Maybe though, it's because it looks that way at first glance and hidden AI 'bonuses' are commonly used to provide human players with a challenge. Even with cIV, where Soren et al have been remarkably open, they have mentioned the previously unknown fact that there are AI bonuses at 'Noble level' (which is supposed to be equal). Given the potential for AI bonuses, its easy to accept that they are present as an explanation for bizarre results.

i think it was never officially said that noble is exactly the human level. iirc sirian once said it is what is closest to the human player. probably the production and research benefits are the same as the human has (equivalent to the "prince" level in civ3). but the AI still gets a slight production boost in the beginning of a new city (preventing easy human rushes).

i've also seen the AI start with archery instead of hunting. again, i believe this is just to protect him from the ultra-early rushes...

[Myrddin]

Has anybody checked through the combat calculator?

Last night it showed 129% on one combat

[Blake]

That's a first strike chances bug with the odds calculator.

[DrSpike]

Originally posted by funat
Combat is great.
Anyone saying different is nothing but a spoiled brat.

Buuuuh, I can't wipe out a civ of riflemen with my 20 armor stack.
Well, if that was possible, Croatia would cease to exist in the first days of summer 1991. As it had 0 (zero) tanks.

You make me miss UR.

[eonwe]

I did some tests using world builder and created an ocean-map with 500 green battleships for me and 500 for Asoka who I was in war with.
The map is on monarch difficulty-level and option to generate new random seed on reload.
I ran it through only two times, each time attacking each one of Asoka's battleships with one of mine.
I didn't attack any wounded ships nor did I attack with a ship that had already attacked because I only played one round.
So every battleship had one chance to either win or lose.

Results were following: on the first time I was left with 227 battleships and on the second time with 218 battleships. I calculated probalilites to be left with X or less battleships using binomial model and calculating probability density functions using maple, with n = 500 and p=q=0.5. These resulted probalities of 0.027184509 and 0.002395787. Quite improbable one could say.

I didn't think this over too much but one thing one should think about is that is binomial model accurate for this? That is because I have vague feeling that it is too simple and that multiple combat rounds increase complete variance.

But anyways, if someone is interested to look at my scenario, they are attached below. Fighting combat can be quite tedious, be sure to put your scout and settler on wait mode so you don't accidentally end turn by pressing space at the wrong time and so lose your results which can be easily seen from military advisor.

[Edit: Numbers in this post are garbage, I checked wrong results]

[SpencerH]

Originally posted by sabrewolf

maybe i don't understand you correctly. but why should macemen and warriors have the same of striking? if you mean from realism point of view (you swing the blade the same fast as the club), then you may be right. but then a maceman hit should kill in the first go and a club wouldn't have big effect against him.

Exactly,

Step 1: does unit 1 hit unit 2 - guys swinging clubs have the same chance of hitting their opponents as guys swinging swords. OTOH guys swinging clubs have less chance of hitting their opponents than guys firing machine guns at them 500 yds away.

Step 2: what damage does said 'hit' have - a club will do less damage to a swordsman (in armour) than a sword will do to a caveman.

step 3: repeat until one unit has no more hit points

this is done by game design. it may not be the perfect solution, but not badly solved imho (better than in civ1-3 where only chance counted in regular combat and hitpoints were equal for all.

no argument from me

it is true that a RNG "calculates" in random values and thereby it is not really random.
now let's assume that the chosen RNG algorithm produces more streaks than pure random would - if we'd have the formula (when the SDK is released?) we could confirm/contradict this. so assuming streakyness... this would have no effect on the computer having advantages as the human would equally often receive a lucky-streak right?

I never said the AI had an advantage from the streakyness. I only said that it was annoying (and fixable).

and also: for the distribution (binomial is actually more appropriate, but the normal distribution is the calculation base, i assume) also the so called "kurtosis" is important. this defines how the bell curve is formed. it is similar to the variance/st.dev but responsible for the shape.
a RNG that produces more streaks than pure randomness would represent a higher kurtosis value.

Great. I'm familiar with the term but I generally use non-parametric tests myself to avoid the issue.

not much really changed... except what i mentioned above (with the "power" of each hit), which i think is a good move.

which was present in civ2

this only proves that the AI is either more sophisticated, more aggressive or has more bonuses or a combination of all. but the combat model is not really influenced by the larger necessity of defence.

I didnt say it was. I used my example of 'combat preparedness' ie my observation that there is a requirement to build many more units as evidence of an increased importance of combat (ie war) in cIV compared to previous versions.

i think it was never officially said that noble is exactly the human level. iirc sirian once said it is what is closest to the human player. probably the production and research benefits are the same as the human has (equivalent to the "prince" level in civ3). but the AI still gets a slight production boost in the beginning of a new city (preventing easy human rushes).

i've also seen the AI start with archery instead of hunting. again, i believe this is just to protect him from the ultra-early rushes...

You may be correct as to what Firaxis actually said but it doesnt change the perception by some that the AI gets bonuses (which we know it does) and that there may be combat bonuses that they are unaware of.