dmvsnorm & mvst in fMultivar

[Yohan Chalabi]

>>>> "AA" == Adelchi Azzalini <azzalini at stat.unipd.it>
>>>> on Tue, 26 Aug 2008 10:44:16 +0200

   AA> Dear Colleagues,
   AA> 
   AA> I am writing about some functions incorporated into your package
   AA> "fMultivar", specifically dmvsnorm & mvst, which originate from 
   AA> packages 'sn'. In principle, it is welcome that your package
   AA> incorporates some routines from my package. Unfortunately, this
   AA> operation has been done improperly, and the effect is seriously
   AA> misleading for the user.
   AA> 
   AA> In the documentation of the above-mentioned functions, one reads:
   AA> 
   AA> mu, Omega, alpha, df: [*mvsnorm][*mvst] - 
   AA>            the model parameters: 
   AA>            'mu' a vector of mean values, one for each column, 
   AA>            'Omega' the covariance matrix, 
   AA>            'alpha' the skewness vector, and 
   AA>            'df' the number of degrees of freedom which is a measure for
   AA>           the fatness of the tails (excess kurtosis). 
   AA> 
   AA> and in the code mu is set equal to xi, which is a location parameter,
   AA> but _not_ the mean, as the documentation wrongly indicates. In fact 
   AA> in the skew-t case the mean may even not exist. Similarly, Omega is 
   AA> described as "the covariance matrix", which suggests that this is
   AA> the covariance matrix of the distribution, but it is not true.
   AA> Omega is _a_ covariance matrix, which is related to the covariance
   AA> matrix (when this one exists) but it is different. For the
   AA> parametrization based on mean and variance, see later.
   AA> 
   AA> In addition, in the code of dmvsnorm, for the one-dimensional case
   AA> there is the assignment
   AA> 
   AA>  ans = dsn(x, location = xi[1], scale = as.vector(Omega)[1], 
   AA>            shape = alpha[1])
   AA> 
   AA> which is wrong since in this case the scale is the square root of
   AA> Omega. Similar errors exist in pmvsnorm, dmvst, pmvst.
   AA> 
   AA> Please note that the above remarks do not attempt to be exhaustive,
   AA> since I have not done a full scan of your code and documentation.
   AA> These are just the problems which I have spotted after a quick
   AA> browsing of your package.
   AA> 
   AA> I take this opportunity to draw your attention to this note:
   AA>    http://tango.stat.unipd.it/SN/announce2.html
   AA> When the new "version 1.xx" will be available, it will include
   AA> a parametrization based on mean/variance.
   AA> 
   AA> With best wishes,
   AA> 
   AA> Adelchi Azzalini

Dear Adelchi,

Thank you for your message. Your comment is very helpful for us because we
are in the process to clean some packages and hopefully we will fix the
issues you raised.

Best regards,
Yohan

-- 
PhD student
Swiss Federal Institute of Technology
Zurich

www.ethz.ch