Surprise attack bonus - Page 5

rah · February 9, 2003, 04:44

YES, THERE IS A SURPRISE BONUS AGAINST A HUMAN PLAYER. We tested it early this morning. Thanks Baron O.

I'll post the results of the test later today after a little sleep.
The evidence will prove it beyond a shadow of doubt.

RAH

ottok · February 9, 2003, 08:32

But how big is calculated by level!

DrSpike · February 9, 2003, 08:54

Quote:

Originally posted by rah
YES, THERE IS A SURPRISE BONUS AGAINST A HUMAN PLAYER. We tested it early this morning. Thanks Baron O.

I'll post the results of the test later today after a little sleep.
The evidence will prove it beyond a shadow of doubt.

RAH

Good. Did you do 50 repititions? If so I'll do some trickery when you post the results and derive the most likely specification for the bonus.

Baron O · February 9, 2003, 12:22

dr spike

we ran 30 simulations
warrior vs warrior on surprise either total suprise and after treaty 30-0

warrior vs phalanx no surprise 0-30

on the various ends of the spectrums, rah has #'s on other permentations. but inital data on surprise warrior vs phahlax the warrior did win occsionally. most of the time on surprise it did at least take phalanx to red most of the time.

rah will post the results later. i just got up earlier than he did

Baron O · February 9, 2003, 12:24

dr spike

i was too tired to run 50 simulations

DrSpike · February 9, 2003, 12:51

Yeah well evidently 30 is enough to say there is a bonus, but pinning down the most likely specification for the bonus we could use a little more.

But I went ahead with 30......though bear in mind 30 observations is the bare minimum necessary to invoke the central limit theorem (which underpins my distibutional assumptions).

H0: bonus is 85%
H1: bonus is greater than 85%

The null was rejected in this test. For those interested I derived the proportion of expected wins under a bonus of 85% from the formulae in the combat thread, and derived the variance of the sample estimate with 30 repititions. I then normalised the estimated proportion and tested against a N(0,1) distribution.

H0: bonus is 90%
H1: bonus is greater than 90%

This time the null is just accepted.

Unfortunately given the extreme nature of the sample results I cannot go any further. With 50 repitions it would be clearer......we must consider now the possiblity that the bonus is 100%.

With 50 repititions a 50-0 score would lead to the second test above being rejected.......and just one loss would give me some additional statistical traction.

20 more please guys.

DrSpike · February 9, 2003, 12:54

Quote:

Originally posted by Baron O
but inital data on surprise warrior vs phahlax the warrior did win occsionally

More exact figures here would also yield some statistical traction. I await Rah's awakening with interest.

Baron O · February 9, 2003, 12:56

dr spike

i think you will have a netter perspective after rah posts the other data, i am game for 20 more but mrs rah probably won't let that happen today.

DrSpike · February 9, 2003, 12:59

Of course those with any statistical skill can probably see that if the bonus is 100% warrior-warrior can ultimately prove nothing, because the test statistic asymptotically approaches infinity as the expected proportion approaches 1!

/me worries what Rah's reaction will be if he is told warrior v warrior tests are inconclusive.

It is still worth going to 50 before we cross that bridge though.

Baron O · February 9, 2003, 13:10

dr spike

are you a mathamatician, or a closet math freak?

this is just for my curiosity

rah · February 9, 2003, 13:21

Test condition
30 repititions all attacks from and to hexs with no defensive bonus.

First set of fifteen rounds. Four attacks per round in the order below.
warrior on warrior surprise
15 wins 0 loses 4 vet promotions
warrior on warrior control
6 wins 9 loses 5 vet promotions
made peace
warrior on phalanx surprise
3 wins 12 loses 8 vet promotions
warrior on phalanx control
0 wins 15 loses 4 vet promotions

Second set of fifteen rounds. Four attacks per round in the order below.
warrior on phalanx surprise
6 wins 9 loses 8 vet promotions
warrior on phalanx control
0 wins 15 loses 6 vet promotions
made peacewarrior on warrior surpise
15 wins 0 loses 9 vet promotions
warrior on warrior control
5 wins 10 loses 8 vet promotions

It appears that after making peace the second surprise round has similar results to the initial surprise attacke.

SO warrior on warrior surprise 30 wins to ZERO loses.
Quite convincing proof of a surprise bonus, especially when you consider that in the control it was only 11 wins to 19 loses.

for the warrior on phalanx surprise it was 9 wins to 21 loses. Not that unsimilar to the control of warrior on warrior.

It would seem to have it be 30 to zero it would have to me more than 50% . It was the opposite of the control for warrior vs phalanx 0 wins to 30 loses. So it leads me to believe that it's closer to 100% , but I'll let the statisticians prove just what it is.

And when given time, I'll do more testing to add data so the actual amount of the bonus can be more acurately determined.

THERE IS A SURPRISE BONUS AGAINST HUMAN PLAYERS.

Thanks Baron O for you help.

RAH

DrSpike · February 9, 2003, 13:36

Quote:

Originally posted by Baron O
dr spike

are you a mathamatician, or a closet math freak?

this is just for my curiosity

Nah, I'm an academic economist, but I specialise in applied empirical work.

DrSpike · February 9, 2003, 14:05

Hmm the initial phalanx results suggest is lower than 100%, but a null of 'bonus is 100%' cannot be rejected given the sample size.

Based on the warrior data so far the range 90% - 100% is likely, though the phalanx results are indicating slightly lower. 20 more tests on warrior v warrior with surprise bonus and 20 more warrior v phalanx would pin things down further.

For practical play the conclusion is as we expected though: surprise attack when you can, and cry when you see you are going to get surprise attacked.

War4ever · February 9, 2003, 14:14

well done guys......well done!

rah · February 9, 2003, 15:01

Now that we've got a save, it won't take as long to get another 20 results. There's also a river near the area we selected so we can try some attacks on that too.

Baron O · February 9, 2003, 15:58

dr spike

so you are statisical junkie, like us lay folks

Ming · February 9, 2003, 15:59

Yeah... now it's time to figure out exactly what the bonus is. But good show guys... Thanks rah and barano...

Baron O · February 9, 2003, 16:03

rah

we need to introduce dr spike to baseball, for the stats he would pop his cookie, with the varieity and utter sheer weight of those #'s

DrSpike · February 9, 2003, 16:33

Quote:

Originally posted by Baron O
dr spike

so you are statisical junkie, like us lay folks

Hehe. Who would have thought techniques I teach for their application to the testing of competing economic models could be used for something far more important like civ?

DrSpike · February 9, 2003, 16:36

Quote:

Originally posted by Baron O
rah

we need to introduce dr spike to baseball, for the stats he would pop his cookie, with the varieity and utter sheer weight of those #'s

Baseball is not that big in the UK at all I'm afraid. I have to have some interest to apply my methods to problems on a Sunday afternoon of all times, so I don't think I'll be analysing baseball stats anytime soon.

War4ever · February 9, 2003, 23:27

spike here is your first line to learn about baseball

9 3 1 1 20 0 6 important digits....

this game is the first time a pitcher struck out 20 batters in a 9 inning game...

stats are as follows

9 innings pitched
3 hits allowed
1 runs allowed
1 earned runs/HR...depends on which stats book your reading , or if your looking at a boxscore in the paper
20 is the # of k's (strikeouts) in the game
0 is the walk total...

pretty impressive stats....since this time roger has done it again, 10 years later, and then kerry wood, and randy johnson have done it as well

i love stats....can quote baseball stats with the best of them.....

any questions just ask

Alexander's Horse · February 9, 2003, 23:35

Good work guys

I'm glad I asked.

But would the same principle apply with later units?

Baron O · February 10, 2003, 01:35

ah

i think it would, we did test with the units most likely to encounter each other early in game.

i remember a game against war where i attacked city on surpise with a horse ended up taking city. he had a fortified unit in city on river.

Alexander's Horse · February 10, 2003, 01:48

The implication of this would be save your surprise bonus for a special attack if you can - like seaborne attack on capital of enemy civ.

Scouse Gits · February 10, 2003, 03:39

Doc - doesn't the 'established fact' that civ uses integer arithmetic on eighths simplify your sums somewhat - which is more likely given the results a bonus of <= 0.875 or >= 1?
Or am I just muddying the already murky waters?

SG[1]

Rah - Baron O well done and thank you

Ming · February 10, 2003, 05:18

Quote:

Originally posted by Alexander's Horse
The implication of this would be save your surprise bonus for a special attack if you can - like seaborne attack on capital of enemy civ.

While you may want to save it for a special attack...
I think the real implication is that you want to take the surprise attack bonus FIRST, so that you are the one that gets it, and not your opponent. In many of our games, the first time you meet an opponent, you attack. That way, you get the advantage, and it's now over with. (unless you make peace with somebody) You don't want you opponent to have it. When somebody you haven't met ends a turn next to one of your cities early in the game, it's sometimes better to attack out, even if you only have a single defensive unit. Because if it's a one size city, it's very likely the city will be destroyed anyway if the surprise bonus is used against you. Also... if you see a unit end one square away from your city, and you know it will be a close battle under normal circumstances, you might want to move a warrior or other cheap unit out of the city, hoping he attacks it... and then using up the surprise bonus. That way, when he does locate the city, the battle will be a close one, instead of him still having the surprise bonus for the more important attack.

While many people in this thread never knew about the surprise bonus, or didn't believe it existed because they didn't play much MP... Almost every MP'er in our group already knew about it and planned accordingly. It's nice to see that it is now "proved" and all we really need to see is what that actual value is... but we have already seen what the implications are... since we knew about it already

DrSpike · February 10, 2003, 06:21

Quote:

Originally posted by Scouse Gits
Doc - doesn't the 'established fact' that civ uses integer arithmetic on eighths simplify your sums somewhat - which is more likely given the results a bonus of <= 0.875 or >= 1?
Or am I just muddying the already murky waters?

SG[1]

Rah - Baron O well done and thank you

Well you know SG I did think about this a lot.......and I'm not sure. If the program does indeed only use a bonus of either 87.5% or 100% it makes things easier for sure....but IIRC the combat thread didn't seem certain on this.

rah · February 10, 2003, 09:18

Yes, as ming said, it's not as important to get it as it is to deny it to your oponent when it could really hurt you. People new to the group always thought us a bit bloodthirsty since as soon as units meets, the first to have the opportunity to attack would. They would assume we're all war crazy not realizing that we're just trying to clear the surprise bonus. We've obviously known about it for years. Now there's a little proof. And a little more testing should tell us just how much.

RAH

-Jrabbit · February 10, 2003, 09:53

rah and Baron O --

Well done!

Great to see the obvious confirmed and (nearly) quantified.

Ming · February 10, 2003, 10:13

But rah... many of the group are warmongers