Hi everybody, I have three questions to ask us:
a) R incorporates a function for the Non-central T
distribution which unfortunately and, as you know, is
not available in Splus 4.5. In
http://www.stats.ox.ac.uk/pub/Swin I found the Don
MacQueen?s noncent.zip but when I run it in Splus 4.5
the following error message appears: "Error in
.Fortran ("vectnc",: "VECTNC" is not a symbol in the
load table". May be I did not installed it correctly
or (as I suppose) it is incompatible with this version
of Splus. I looked in the directory for Splus 6.0
http://www.stats.ox.ac.uk/pub/MASS3/Winlibs but the
update of this function is not there. Do you know of
some alternative function for the Non-central T in
Splus 4.5 or how to solve the problem with
noncent.zip?
b) I wanted to apply the overall test for
coincidental regressions (see Zar, pag. 304), whose
statistic is:
F = ((SSt-SSp)/2(K-1))/( SSp/DFp) , which follows a F
2(k-1), p, where SSt is the total sum of squares, SSp
the pooled sum of squares, DFp the pooled degrees of
freedom and k the number of regressions compared. In
analogy with the ANOVA approach, I suppose that the
non centrality parameter of the F for this test is:
DFp(SSt-SSp)/ SSp, but I am not sure. Could you
confirm it?.
c) Finally, I also wanted to apply a multivariate
parametric mean comparisons test (the parametric
analogous of Friedman?s), whose statistic is: F (n-2)/p(AE(X))
(INV(ASA?))(AE(X))? which follows a F
p, n-2, where S is the variance-covariance matrix of p
x p dimension, A is the identity matrix, E(X) the
sample means matrix and n is the sample size. I was
told that the ncp of the non centrality Fp,n-2 for
this test is: D (ASA) D?, where D = (E(X1)-E(X2),
E(X2)-E(X3)), but I am not absolutely sure. Could you
confirm it?.
I wait for your responses,
Best wishes,
Jens Krann,
Biologist. Email: jenskrann at yahoo.com
___________________________________________________
Yahoo! Messenger - Nueva versi?n GRATIS
Super Webcam, voz, caritas animadas, y m?s...