This new version (2.3) of Statistics::Descriptive contains:

- Code provided by Andrea Spinelli to prevent division by zero and to
  make consistent return values for undefined behavior.

- Andrea also wrote a great test bench to check out the division
  by zero.  We'll be extending it to check for other things, too.

- Code to calculate the percentile of the data as defined by RFC2330
  Thanks to Warren Matthews for his contribution.

- A bug fix for the calculation of frequency distributions.  Thanks to
  Nick Tolli for alerting me to this.

- Slight changes to the Makefile to make the ActiveState people happy.
  Hey, if they're going to make Perl easy for Windows users then I
  can add a few lines of text to a Makefile for them.

Documentation was also added to show how to turn the return value
of least_squares_fit into a hash.  If returning a hash is something
that many of you users want then I'll add it to the next release.

List of methods by class:

Statistics::Descriptive::Sparse
----------------------------------
add_data
count
mean
sum
variance
pseudo_variance
min
max
mindex
maxdex
standard_deviation
sample_range

Statistics::Descriptive::Full
----------------------------------
All methods above and:
get_data
sort_data
presorted
percentile
median
trimmed_mean
harmonic_mean
mode
geometric_mean
frequency_distribution
least_squares_fit

Ideas for ongoing development:

- re: Georg Fuellen
  A function to define the "maxima" of a distribution.  In other words,
  return an array that contains all data that falls within a given
  confidence interval.

- Move source to C and XS for speed improvements. (Interface to R or S?)

- Move to PDL for a unified math solution and maybe take advantage of
  their compact storage.

- Methods for covariance and correlation.

- An interface to Jon Orwant's Statistics::ChiSquare module to
  provide a method for determining how random data is (for a uniform
  distribution) or how well is fits another distribution (for
  non-uniform distributions like normal(gaussian), log-normal,
  rayleigh, etc).

  The major issue I'm concerned about is the updating of someone else's
  code and future extensibility.  Now, major paranoia should be obvious
  since Jon hasn't changed Statistics::ChiSquare for a long, long
  time.  Still, a wrapper module that inherits from several other
  modules is appealing.  Perhaps it will be called Statistics::Bundle.