testing SDI

[vicawoo]

You really think a continuity correction would result in triple the error? Come on, everyone knows that binomial distributions approach continuous distributions at extreme values, and 3 standard deviation's is getting there. The majority of the error will come not from the discontinuity itself, but the fact that you should be starting at the equivalent of 35.5 instead of 36 in the continuous approximation, and given the ratio of .63 to .37 and 36 to 29, you're not going to get triple the error. Use some common sense.

 

[PieceOfMind]

You really think a continuity correction would result in triple the error? Come on, everyone knows that binomial distributions approach continuous distributions at extreme values, and 3 standard deviation's is getting there. The majority of the error will come not from the discontinuity itself, but the fact that you should be starting at the equivalent of 35.5 instead of 36 in the continuous approximation, and given the ratio of .63 to .37 and 36 to 29, you're not going to get triple the error. Use some common sense.

What are you talking about? I already corrected myself, saying I was meant to get 0.0019 as you did. I didn't comment on the accuracy of the normal approx to the binomial distribution. Regardless of my use of common sense or not, I commented because I believed your figure to be inaccurate, which it was, and mine was too.

Personally, I don't consider the intimate details of the normal approximation to the binomial distribution to be common sense, though I do find it interesting.

 

[BaronVonCP]

this thread needs to die. I play poker, and when you run thousands and thousands of an event, 'bad luck' will happen. If the % chance of something to occur is greater than zero, it will happen eventually.

i think its more of a psychological thing. I mean, when you win a battle at 90% odds, no one gets exited about their great fortune. But lose one of those battles, and its off to steam and question the RNG.

But anyways, just had to express my views on the subject. Time to play some holiday civ.

 

[vicawoo]

this thread needs to die. I play poker, and when you run thousands and thousands of an event, 'bad luck' will happen. If the % chance of something to occur is greater than zero, it will happen eventually.

i think its more of a psychological thing. I mean, when you win a battle at 90% odds, no one gets exited about their great fortune. But lose one of those battles, and its off to steam and question the RNG.

But anyways, just had to express my views on the subject. Time to play some holiday civ.

There's huge flaws in trying to apply qualitative arguments (lots of samples means probabilities are meaningless) to a quantitative situation. Taking experiences in poker, where you can easily play more hands in one night than you will play games of civ in an entire year. That's valid for discounting 7 misses in a row, which I believe I've ignored completely, not a one turn sample of 65. Generalizing an argument out of its context is for philosophy, not mathematics, and it's quite offensive to someone who's making mathematical arguments.

There was a low probability event which far outweighed its sample size (let's do a common sense correction, take the sample of "reportable" events in which half the missiles are intercepted in a decent sample size, do P(A|B), it still bumps in just under 90%). You could a) let a curious person do 80% of the work in checking (thanks besst for the code by the way) or b) flame his efforts while doing nothing. Being a mathematician, I'm used to people's eyes glazing over whenever basic algebra or arithmetic is used, but that extra little effort to insult instead of ignore the thread puts me in the holiday spirit.

---

I tested games in which SDI was built, in case the game was doing something strange like cheating to build multiple SDI, and for misleading interface.

 

[BaronVonCP]

well if you think that running a 100 sample size test is going to give you enough evidence to make quantitative arguments, we are just going to disagree.

If someone can prove that SDI does not work as described, we'd have something to talk about.

All this semantic arguing and whatnot just seems pointless. I have a math degree too. Nothing to get excited about.