*** UNDER CONSTRUCTION ***


Provided are the S-plus functions for the paper titled
 "Improving power and sensitivity in multinomial goodness-of-fit tests,"
 written by Basu, A., Surajit, R., Park, C. and Basu, S. (2000).
This paper can be obtained via internet at 
 http://www.stat.psu.edu/~cspark/m7.ps  (postscript)
 http://www.stat.psu.edu/~cspark/m7.pdf (Adobe pdf)

We assume that we throw
   n=# of balls into k=# of different boxes with p probability.

We have the following functions:

* support(n,k)
  generates the sample sample of multinomial (n,k)     

* rmultinom(n,size,size)
  n    = sample size
  size = vector of (positive integer) numbers of flips
         for which  the  Multinomial  distribution  measures
         the number of each outcomes.
  prob = vector (matrix) of probabilities of each outcome.
         when matrix if nrow(prob) > 1, random vector for 
         each row vector of prob matrix is returned.
         If n > nrow(prob),
            repeat prob matrix from 1st row vector. 
         if sum(prob) for each row is not 1,
            then normalize by itself
  <output> : (n,k) matrix

* dmultinom(x,prob) 
  the density at x of multinomial dist. 

* critical.value(stat, prob, alpha)
  gives critical values of size alpha proposed by
     Read and Cressie (1988, p.77)
  output: cv, alpha1, alpha2

* rpower(stat, prob, cv, alpha, alpha1, alpha2):
     randomized power of  size alpha test

* PD()
  generalized Power Divergence (combined and penalized version of PD)
  #- a=tuning parameter
  #- h=penalty  
  #- a.in, a.out = in- and out-lier tuning parameter  
  For ordinary  PD, use   PD(d,f,a).
  For penalized PD, use   PD(d,f,a,h).
  For combined  PD, use   PD(d,f,a.in,a.out).
  For combined and penalized PD, use   PD(d,f, a.in, a.out, h).

* PBHM()
  mixture of Pearson Chisquare and blended weight Hellinger distance
  #- h=penalty  
  For ordinary  PBHM, use   PBHM(d,f,a).
  For penalized PBHM, use   PBHM(d,f,a,h).