[Eeglablist] ICA Sample Size
arno at salk.edu
Fri Nov 17 13:26:46 PST 2006
Not really knowing how to answer your question, I asked Tony about this
(Tony Bell, the creator of Infomax ICA). He said:
"His reasoning is flawed.
Briefly, ICA is estimating a whole host of
cross-node higher-order statistics in parallel.
Your friend should look up the sampling theory
for multivariate higher-order statistics, not
univariate 2nd order statistics. "
A. Debebe wrote:
> Dear All,
> The central limit theorem mentions: when n is
> sufficiently large, the smpling distribution of mean
> of the distribution is well approximated by a normal
> curve, even when the population distrbution is not
> itself normal. The central limit theorem can safely be
> applied if n exceeds 30.Refere, any elementary
> statistics book. This explains the magic number 30.
> Best regards,
> Debebe asefa, PhD
More information about the eeglablist