
---------- Forwarded message ----------
Date: Tue, 4 Mar 2008 08:52:47 -0800
From: Jan de Leeuw <deleeuw@stat.ucla.edu>
To: Kurt Hornik <Kurt.Hornik@wu-wien.ac.at>,
     Achim Zeileis <Achim.Zeileis@wu-wien.ac.at>
Subject: JSS 312 (coin)

Accept subject to minor revisions. AE still has to spprove next version.
---------------------------------------

?

Die Reports waren sinnigerweise in .doc Format, grausam. Deshalb haengen
sie unten in plain text an. Sind beide recht harmlos und haben recht wenig
Substanz, ein paar brauchbare Kommentare sind aber auch dabei.

lg,
Z

-- Referee A -------------
This is an excellent paper and appropriate for publication on the Journal
of Statistical Software as it stands.

However I am a bit concerned about the extent to which it goes over the
same ground as the authors' previous paper in the American Statistician.
The main advance appears to be that the authors have dropped the
requirement for observations to be iid, but only to the extent of the
fairly trivial introduction of a blocking factor.

-- Referee B -------------
The coin package implements, computationally, the theoretical ideas of the
1999 Strasser and Weber paper that the authors reference.  This make coin
an important and useful package, and one for which it will be useful to
the documentation that this paper provides.  I have no criticisms that are
of major consequence.

Queries:
1. Section 2.2 is pretty much a copy of what appears in the Amercian
   Statistician paper (Hothorn, Hornik and Zeileis, 2006).  Is it
   important to repeat the theory here, or would summary comments be
   adequate, referring to the American Statistician paper for further
   details?
2. Sections 2.4 and 4.2 may be heavy going for anyone who does not have
   some reasonable level of familiarity with S$ classes and methods.
   Should some of this go into an appendix?

The paper can be made more useful to users who are not deeply into R
internals by including one or two examples of the use of
independence_test(), either as part of the introduction, or immediately
following the introduction.  Reference might usefully be made, at this
point in the paper, to examples in the help page.

There is repeated reference to "Monte Carlo procedures".  These surely
involve random sampling from the relevant permutation distribution.  The
term "Monte Carlo" seems unnecessarily vague.

In a few places the wording can be improved:
1. Insert "the" as follows: P.5, l.-9 "in the case"; p.11, l.-17 "at the C
   level"; p.14, l.-17 and l.-16 "in the package" (or else omit
   "package").
2. 1st line of Section 2.2: Replace "to derive" by "deriving", or (maybe
   better) omit "suggest to".


