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

[Hauptman]

what i understand 1.5* did was this...


In 1.00, a half strength infantry unit @ 10 str was in actuality NOT half strength at all, because of the hidden factor of hitpoints. So an infantry unit at 10str had half his atack ability, AND half of his damage absorbtion, thus in fact a 10 str infantry was QUARTER strength.

A knight vs "half" str infantry would net the knight a 66% chance to win. as It only needed to win half the fights an infantry would.

So 1.54 (or 1.52, I'm in the same boat) they made units do damage based on their orriginal str.

So now a knight and 1/2 inf are 50/50. each has a 50-50 chance to win a fight, but while the inf has half the HPs it does twice the damage.


So what I think is happening, the modifiers (like +25% vs melee) used to be a HUGE bonus, whereas now they are much reduced. and a hurt unit, wich used to be very easy to take out, is now still quite deadly.

[LordShiva]

Originally posted by Krill
This beats MtGs' put down of Sava any day of the week...

Link? Or care to jog our memories?

[Blake]

Originally posted by Hauptman
what i understand 1.5* did was this...


In 1.00, a half strength infantry unit @ 10 str was in actuality NOT half strength at all, because of the hidden factor of hitpoints. So an infantry unit at 10str had half his atack ability, AND half of his damage absorbtion, thus in fact a 10 str infantry was QUARTER strength.

A knight vs "half" str infantry would net the knight a 66% chance to win. as It only needed to win half the fights an infantry would.

So 1.54 (or 1.52, I'm in the same boat) they made units do damage based on their orriginal str.

So now a knight and 1/2 inf are 50/50. each has a 50-50 chance to win a fight, but while the inf has half the HPs it does twice the damage.


So what I think is happening, the modifiers (like +25% vs melee) used to be a HUGE bonus, whereas now they are much reduced. and a hurt unit, wich used to be very easy to take out, is now still quite deadly.

I'm pretty sure that firepower still benefits from defensive bonuses and promotions. Saying it's based on the normal unmodified-by-anything strength is a minomar, it's simply calculated as if the unit has full health.

[Urban Ranger]

Originally posted by TheHateMale
Provided both units have the same strength/HP and the fight is to the death it should still stand. It's like flipping a coin 11 times and repeating the test 10 times. The number of results where you get more heads should be (roughly) the same as the number of times you get more tails.

Suppose a destroyer vs destroyer combat takes 5 rounds (because of the 0.2 multipler). So, the side that wins needs only to win 3:2.

Suppose you stage 50 fights in the WB and loses all ten. If each combat is resolved by a die roll losing all 50 is a 1/(0.5)^50 occurance, extremely rare. However, in a multiple round resolution system the other side needs to only get 50 more wins out of 250 rounds than you to win all of them. It suddenly becames much more likely.

[Urban Ranger]

Originally posted by DrSpike
Well I have a Ph.D in econometrics and have taught statistics at university level. You?

This just proves my point that you are not a trained scientist.

Originally posted by DrSpike
And insisting on a control in this case just indicates your ignorance, I'm afraid.

Appeal to Authority isn't going to fly.

[snoopy369]

A control would not be necessary, or possible, if he is testing the claim of "combat is not consistent with the statistics shown". "Control" is the method of comparing the norm with a single changed variable; in this case, he's comparing the statistics shown (the norm) with the changed variable (actually combatting). There's no way to limit the changed variable to one, as you can't test only one part of the combat... there's no meaningful control here.

If he's making the claim that "SP results are biased against the human", that's different. There he's arguing that the results vary based on who the Player and Enemy are. He'd want to compare either the results where the computer attacks him (as opposed to where he attacks the computer), or where two human players oppose each other, or preferably both.

I think 59/100 is not bad. I'd have to see 1) how many HP were left on both sides, and 2) whether this was reproducible on a regular basis, before I believed anything.

[killthejoe]

Originally posted by Arrian
This appears to be a new version of a very old debate: some players just don't like any randomness. The "my maceman on hill should always win!" comment is indicative of this. I'm *not* saying that's wrong, per se, although I disagree. Is that a flaw in the combat system, or just personal preference butting heads with game design?

I also used to pull my hair out over random events "ruining" my experience, as Arrian has mentioned. If anyone has ever played the boardgame Risk, you would know that you could play a perfect game and the roll of the dice could screw you in the end. But then a friend of mine once gave me the advice that, "Skill will prevail more often than not". What he meant was that if I was to play multiple number of games, I may not win every time since the dice might control my destiny, but I would win most of the time ...

The same is true in Civ. You may not win every battle, but if you're a good player, you will prevail in the end.

[rjmatsleepers]

Originally posted by DougM


Actually, a control is necessary in the dice rolling experiment. You have to know what the proper outcome for an unweighted die is in order to understand whether or not the weighted die has a different outcome.

Now, you could claim that you are comparing the outcome to an ideal die for which the outcome is always perfectly evenly distributed among the six outcomes: in which case, the model is your control.

Otherwise, you need to compare against the actual outcome of a real unweighted die, in which case the real unweighted die is your control.

Either way, your experiment always has a control.

We seem to be using the word control in different senses.

I can't believe any "scientist" would bother rolling "a real unweighted die" in order to compare it with a suspect die. If they did, the number of trials they would need to get even a crude approximation to the theoretical distribution would be very large.

On the other hand, comparison with a thought experiment using a platonic ideal dice seems to me to be an unusual usage for the word "control".

Standard statistical theory will tell you the distribution of expected results from a given number of independent trials. It is straight forward to test whether or not your actual result is within acceptable bounds. What you are prepared to accept depends on circumstances, but by convention, non-critical cases tend to take what is often described as a 95% confidence interval. Even so you will still make mistakes - accepting a hypothesis you should reject, or rejecting a hypothesis that is actually true. This is not what I would describe as a "control", but I suspect this is what you mean.

As far as "scientist" vs "statistician" is concerned, I wonder how well some practising scientists understand the theory of probability and statistics. It would be interesting to try them on the old chestnut about how many people you need in a room to make it more likely than not that two (or more) of them have the same birthday.

Anyway, to return to the original topic. The evidence is mounting that others are unable to replicate the OP's results. He may have made an error - sorry Lansing, but I have to consider that possibility. There may be some factor operating that he didn't know about, but others do. And there may be some factor operating that we didn't know about - sorry Soren and Urban Ranger, but I have to consider that possibility.

In the good old Civ II days, there was a long discussion of a sneak attack bonus in which a lot of contradictory data appeared. I believe that this eventually concluded that such a bonus only applied in certain circumstances. Is it possible that we have something like that here?

RJM at Sleeper's

[xxFlukexx]

Ahhh, ok, thanks for clearing that up Soren.

I'll tally those up for ya snoopy, I think both sides were equally damaged though, lots of 1s and 2s and 3s and 4s. Great thing about axeman vs axeman combat is that they hit for exaclty one point of damage. I don't think my results show that there is something wrong with the RNG or combat odds yet, further testing would be nice. But it does show one thing, in a large scale war, luck still plays a much bigger factor than it should. You can have fully healed troops with great promotions and on defensive terrain and still lose and have your game ruined.

Here's a question, name me one game where you can pit two identical units against each other, have them fight, but one unit beats the other unit without taking any damage. I think that luck should have a part but imagine if Starcraft worked like that......

[Mergle]

Originally posted by rjmatsleepers


We seem to be using the word control in different senses.

I can't believe any "scientist" would bother rolling "a real unweighted die" in order to compare it with a suspect die. If they did, the number of trials they would need to get even a crude approximation to the theoretical distribution would be very large.

On the other hand, comparison with a thought experiment using a platonic ideal dice seems to me to be an unusual usage for the word "control".

Standard statistical theory will tell you the distribution of expected results from a given number of independent trials. It is straight forward to test whether or not your actual result is within acceptable bounds. What you are prepared to accept depends on circumstances, but by convention, non-critical cases tend to take what is often described as a 95% confidence interval. Even so you will still make mistakes - accepting a hypothesis you should reject, or rejecting a hypothesis that is actually true. This is not what I would describe as a "control", but I suspect this is what you mean.

As far as "scientist" vs "statistician" is concerned, I wonder how well some practising scientists understand the theory of probability and statistics. It would be interesting to try them on the old chestnut about how many people you need in a room to make it more likely than not that two (or more) of them have the same birthday.

23. I don't like appealing to authority, but I have a degree in mathematics and statistics from Oxford University (the real life one) and I agree with Dr Spike and rjm... make of that what you will

Anyway, to return to the original topic. The evidence is mounting that others are unable to replicate the OP's results. He may have made an error - sorry Lansing, but I have to consider that possibility. There may be some factor operating that he didn't know about, but others do. And there may be some factor operating that we didn't know about - sorry Soren and Urban Ranger, but I have to consider that possibility.

In the good old Civ II days, there was a long discussion of a sneak attack bonus in which a lot of contradictory data appeared. I believe that this eventually concluded that such a bonus only applied in certain circumstances. Is it possible that we have something like that here?

RJM at Sleeper's

I agree. The difference I noticed between the OP test and the otyers was he used randonm seed & repeated the same attack. The rest of us just created lots of units. It could be that the "reset random seed" doesn't always work - thus creating a likelihood that saveandreload would still repeat previous result.

[TheHateMale]

Originally posted by Urban Ranger

Suppose a destroyer vs destroyer combat takes 5 rounds (because of the 0.2 multipler). So, the side that wins needs only to win 3:2.

Suppose you stage 50 fights in the WB and loses all ten. If each combat is resolved by a die roll losing all 50 is a 1/(0.5)^50 occurance, extremely rare. However, in a multiple round resolution system the other side needs to only get 50 more wins out of 250 rounds than you to win all of them. It suddenly becames much more likely.

Yeah, you're right. It is more complicated. I suppose the health of each unit would need to be set to 1 to do it properly. There'd probably need to be an extra n(C)r thingummy in there to model the number of rounds won to the number of combats won. Although the total number of combats wouldn't be static either... eurgh. I'm just going to leave it alone now.



Although I would point out that a scientist is just someone who applies the scientific method to research. It doesn't require much training. The training part seems to mostly deal with understanding the results and analysis, which is really tedious. You know, you can't spell analyst without...

[Mergle]

Originally posted by Urban Ranger



Suppose a destroyer vs destroyer combat takes 5 rounds (because of the 0.2 multipler). So, the side that wins needs only to win 3:2.

Suppose you stage 50 fights in the WB and loses all ten. If each combat is resolved by a die roll losing all 50 is a 1/(0.5)^50 occurance, extremely rare. However, in a multiple round resolution system the other side needs to only get 50 more wins out of 250 rounds than you to win all of them. It suddenly becames much more likely.

Only up to a point. The fact remains that in a destroyer vs destroyer combat, there is a 0.5 chance that the attacker wins, and a 0.5 chance that the defender wins -there will of course be a range of hitpoints left afterwards but that's irrelevant to this test. Therefore, it's a simple binomial distribution problem to work out the probability distribution of wins over a number of combats.

Then a quick 2-tailed test at 5% significance should tell us whether we have distribution from expected. I'll try to write this up in another post.

[Mergle]

OK, let's try this out.

So where H is number of human wins and n=number of combats we have

H ~ B(n,p)

Now, the combat odds claim to be "fair" so for identical units with no modifiers, we would expect p=0.5

H0: p=0.5 (null hypothesis)
H1: p!=0.5 (alternative hypothesis)

Note I'm using a two-tailed test - this will detect bias either towards or against the human.

I'll take 5% significance [i.e. reject null hypothesis if something happens that is less than 5% likely if the null hypothesis were true). I'm too lazy to do the exact work on this and can't find binomial tables, so I'll use the central limit theorem and approximate to the normal distribution (all distributions approx. to normal [Gaussian] if sample size big enough). This should be OK as I'm doing a rough cut.

http://www.bun.kyoto-u.ac.jp/~suchii/norm.approx.html

I'll take sample size 120 for this example as that's what I boneheadedly used in my test. This gives us mean 60 and standard deviation of squareroot(30)= 5.48 (check above link for details as to how you do this).

I got 55 wins. Thats 5/5.48=0.91 standard deviations from the mean.

Looking that up on the chart at

http://www.statsoft.com/textbook/sttable.html

I find 0.3186. I.e. there's a ~32% chance that we'd end up this close to (but below) the mean. That relates to an 18% chance that we'd get at least this much deviation below the mean by chance if the distribution were as the null hypothesis.

This is greater than our 5% significance, so we have no reason to reject the null hypothesis. My test gives no evidence that the combat system is biased.

I'll see if there's enough data in the OP to apply the test to that.

[Mergle]

Right, to take the first example in the OP, AI won 40/50. That's a value for h of 10, n=50.

We have mean=50*.5=25, standard deviation = squareroot(np(1-p))=sqroot(50*.25)=sqroot(12.5)=3.54

So h is (25-10)/3.54=4.24 s.ds away from the mean. That's off the chart and a highly significant result.

So by itself it would suggest bias in the combat system. Hence, if the OP were the only thing to go on, it would be highly suggestive of a bug (as we know design intent is p=0.5 from Soren).

However, the fact that we have been unable to replicate these results, and the lack of proof of them submitted, we must treat this information as unreliable unless and until someone is able to replicate it. Like I said, this doesn't mean the OP is lying - he may have made a subtle mistake in his test, his computer's RNG may not work properly, or there's an unrelated but (e.g. seed not properly being regenerated despite the option being checked - something the other tests are not vulnerable to).

[Lul Thyme]

Originally posted by Solver
Wow... some people don't know what odds mean.

If your unit has a 75% chance of winning, it doesn't mean that your units would win 75 out of 100 such battles. Winning 60 of 100 would be significantly more probable!

Sorry to burst your bubble
your first statement is right.
It doesnt mean you will win 75 out of 100.
But your second statement is false
winning 60 out of 100 is less probable.
75 out of 100 is the most probable although it's not that probable in itself.

I guess I know what you're trying to explain, but since this is usually such a sensitive subject, you have to be careful.