a****t 发帖数: 720 | 1 i only know little about statistics.
we have a conjecture in number theory(some kind of property follow some distribution) hard to prove,
then i did a computer experiment to test it. and got lot of data (around 10^4).
from the graph(normalized data), there are several distributions(include our target
distribution) fit that graph pretty good. so how can i analysis those data, to say our
target one is the true one? thanks | l*******l 发帖数: 204 | 2 Kolmogorov-Smirnov Test.
However statistical test do NOT prove that the distribution fits. All it can
do is to prove that the distribution does NOT fit. | S***e 发帖数: 108 | 3 Residual plots?
Compare AIC or BIC? | p****o 发帖数: 1340 | 4 a straightforward and useful way is to use qq-plot.
distribution) hard to prove,
10^4).
our target
data, to say our
【在 a****t 的大作中提到】 : i only know little about statistics. : we have a conjecture in number theory(some kind of property follow some distribution) hard to prove, : then i did a computer experiment to test it. and got lot of data (around 10^4). : from the graph(normalized data), there are several distributions(include our target : distribution) fit that graph pretty good. so how can i analysis those data, to say our : target one is the true one? thanks
| y**t 发帖数: 205 | 5 If you cannot find a good test to check which distribution is better, you
can either compare CDF or moments.
Derive your theory CDF/moments then compare with empirical CDF/moments.
distribution) hard to prove,
10^4).
our target
data, to say our
【在 a****t 的大作中提到】 : i only know little about statistics. : we have a conjecture in number theory(some kind of property follow some distribution) hard to prove, : then i did a computer experiment to test it. and got lot of data (around 10^4). : from the graph(normalized data), there are several distributions(include our target : distribution) fit that graph pretty good. so how can i analysis those data, to say our : target one is the true one? thanks
|
|