Hi!
I want to do a ks.test of some sample data, say "x", against a
theoretical distribution, e.g. a Weibull.
So suppose we have these data:
set.seed(1);
x <- rweibull( 200, 1.3, 8.7 );
1. Is it better to do a 1-sample or a 2-sample test?
ks.test( x, "pweibull", 1.3, 8.7 ); # 1-sample
ks.test( x, rweibull( 200, 1.3, 8.7 ); # 2-samples
2. If I perform a 2 sample test, what I thought to do was using some
kind of resampling from the theoretical distribution and then averages
all ks statistic obtained on each sampling:
n <- 1000; # number of resampling
mean <- 0; # KS statistic mean
sd <- 0; # KS statistc std-err
for ( k in 1:n )
{
ks <- ks.test( x, rweribull( 200, 1.3, 8.7 ) );
mean <- mean + ks.statistic;
sd <- sd + ks.statistic^2;
}
ks.mean <- ks.mean/n
ks.sd <- sqrt( (ks.sd - n*ks.mean^2)/(n-1) );
# Calculate p-value with Marsaglia K(n,d) function (used by R)
#p.value <- 1-K(200, ks.mean);
cat( paste( "KS statistic: ", ks.mean ) );
cat( paste( "Standard Error: ", ks.sd ) );
cat( paste( "p-value: ", p.value ) );
Has this any sense?
Any other critic/suggestion is appreciated.
Thank you very much!
-- Marco