Page 97 - Contributed Paper Session (CPS) - Volume 2
P. 97

CPS1442 Uzuke C.A. et al.
                       r          ( ( P − ( 1− P +  − P − )) (P +  − P + ) 2
                                                   c
                                                  
            =   P + P − ( 1− P +  − P − ) c         1 = j  c  j
                                
                                            2
                                                                              2
                                                                  +
               + P + ( 1− P +  − P − ) (P j −  − P − ) + P + P −   ( ( P j +  − P + ) (P j −  − P − )) )
                                 1 = j                1 = j
            This when further simplified yield the test statistic
                        r
              2  =
                     −
                     ( − P
                 P + P 1  +  − P − )
               −    −  c  +   +  2  +    +  c  −   −  2   +  −  c  +   +  −   −  
                       
                                                               
                                            
                                      ( − P
                                                        2
                                                       + P
                                                                                 
               ( (  P 1 − P  )) (P j  − P  ) ( + P 1  )) (P j  − P  ) ( P  ) ( ( P j  − P  )(P j  − P  ))(12)
                      = j 1                 = j 1              = j 1            
            which under the null hypothesis approximately has a chi-square distribution
            with 2(c-1) degrees of freedom for sufficiently large sample r and c and may
            be used to test the null hypothesis for equation 1
                                                                 2
            Ho is rejected at the  level of significance if   2    1−  ( 2 ;  c −  ) 1   otherwise Ho is
            accepted.
            Note that if no observations are exactly equal to the common median M then
             0                               +    −
              j  =  0  so that from equation 4  j +   j  = 1 for all j = 1, 2, …, c. hence also
                                           +
                                  −
             P +  + P −  = 1 so that  P =1 − P =1 −  P =  q . In this case equation 12 is easily
            seen to reduce to
                    c      −   2    c
                                          2
                  r  P j  − P    r  P −  rc P  2
                                         j
                    = 
              2  =  j 1  −  −     =   j 1 =                                        (13)
                        P q              P q
                                   +
                      +
            Where  P =    P  and P =  P ,  q =1 −  P a  similar  test  statistic  may  also  be
                     j
                           j
            developed to test for the equality of the effects of r levels of the row factor A
            by redefining equations 5 and 6 appropriately for the levels of factor A. note
            that although the levels of factor A may have different medians  M    (I = 1, 2,
                                                                             i
            …, r) from the medians  M (j = 1, 2, …, c) of factor B, they will nevertheless
                                       j
            have the same common median M since they are one data set. Therefore the
            null hypothesis to be tested for the r levels of factor A would be
             Ho =  M =  M =...  =  M =  M                                            (14)
                          
                                    
                     
                    1     2         r
                                                         +
                                                                   −
            To help test this null hypothesis we simply let t ,t  and t  be respectively the
                                                            0
                                                         i
                                                                   i
                                                            i
            number  of  1s,  0s  and  -1s  in  the  ith  level  of  factor  A  consistent  with  the
                                                +
                                                              −
            specifications  in  equation2  and  p ,  p  and  p  be  the  corresponding
                                                     0
                                                              i
                                                i
                                                     i
            probabilities.  Clearly  the  total  number  of  1s,  0s,  and  -1s  as  well  as  their
                            +
                                        −
                                0
            corresponding p ,  p  and  p  remain unchanged.
                                                                86 | I S I   W S C   2 0 1 9
   92   93   94   95   96   97   98   99   100   101   102