Re: [Gretl-users] skewness and kurtosis
On Mon, 30 May 2011, Riccardo (Jack) Lucchetti wrote:
On Mon, 30 May 2011, Sven Schreiber wrote:
>
> Just to be clear: There are the new (not-yet-approved) functions
> skewness() and kurtosis(). The function package would bind these and
> other functions together, essentially adding another API to those
> functions.
>
> IMO functions and commands serve different purposes, there is no real
> competition between them and they are not substitutes.
>
> I don't have anything against additional function packages, but I would
> be against adding this to gretl (=its namespace) by default, it's just
> redundant.
I agree. Besides, it's not quite clear to me how the proposed package is
meant to be used. [...]
Sorry for the delayed response; my internet access is still
intermittent. But here's my 2 cents worth (not intended to be
definitive). I agree with Jack's earlier point about keeping the
number of built-in functions down to a reasonable number. To that
end, I think I'd prefer to have a built-in "moments" function that
gives a matrix with mean, variance, skewness and (excess?)
kurtosis, and not bother with separate skewness and kurtosis
functions.
Rationale: mean and standard deviation (and/or variance) are
wanted so often that it makes sense to have specific functions for
those purposes. But skewness and kurtosis will be wanted much less
often, and when people want those they probably want the first two
moments too, so why not just one function for the "fancy" stuff?
Comments on some other comments: (1) the median: there's already a
quantile() function. (2) summary --matrix: the summary command
produces a lot of stuff that people probably don't need in matrix
output, and if we included all those columns it would make it
difficult to find what you really want -- as opposed to just the
first four moments, which is easy to parse.
Allin.