Re: Deriving Statistical Distribution (Mathematical Expression) of Sample Data.

On May 7, 10:13 pm, Russell <Russell.Mar...@xxxxxxx> wrote:

Am I correct in understanding that what you're after is a
method where the mathematical form of the distribution
is determined from the data, not just the values of the
parameters of a predetermined functional form? If so,
the only thing I know of which might be applicable is
kernel density estimation, but even that requires that
you choose a form for the kernel (at least AFAIK).

As many books covering kernel density estimation state, the estimated
density is fairly insensitive to the FORM of the kernel (Gaussian and
Epanechnikov are common choices) but IS sensitive to the width of the
kernel used.
Research has been done on "bandwidth selection", but simply trying
various widths and judging the tradeoff between bias and variance "by
eye" may work ok.

.