jeudi 29 juin 2017

Latin Hypercube Sampling from the Normal Distribution in R

I've been trying to generate random values of collinear variables in R, such that the resulting sets of random variable values are characterized by properties that closely resemble the historical means, standard deviations, skews, and kurtoses of each these variables. I'm basing my approach on a particular method I found published online. This method employs

Distributions[,i] <- sort(rnorm(N, mu, sigma))

to generate random values of a reasonably normally-distributed variable given specified mean and standard deviation. I've found that specifying the mean and standard deviation alone do little to preserve the skews and kurtoses of the variables as distributed in reality (they most closely resemble the normal distribution than any other type of distribution). For this reason, I'm looking to rather than sampling randomly from the normal distribution, do so with Latin hypercube, so as to direct the sampling to take more random values from the tails of the distribution.

Any idea how to go about this in R? Please share. Thanks for your help.




Aucun commentaire:

Enregistrer un commentaire