A generic SAS program for computing the power of the test Ho:tau(i) = 0
vs. Ha:tau(i) ~= 0 varying both Sum(tau(i)^2)/sigma^2 and its
coefficient (which is usually the block size) in the non-centrality
parameter.  You may have to adjust the bounds for both these variables
in order to get a good-looking plot.  Note that you may have to include
an extra variable in order to use this code for the BIBD (Don uses N, k
and a in his code).  This algorithm will have to be modified for the
unbalanced CRD (which may be why Don's power analysis used
contrasts...).


data power; 
file 'z:\stat706\power.dat';              /* File name of choice */
alpha=  ;                                 /* Enter alpha */
a= ;                                      /* Enter number of treatments */
am1=a-1;                                  /* Numerator df */
do s02= to  by  ;                         /* s02=Sum(tau(i)^2)/sigma^2 */
do n=  to  by  ;                          /* n=coefficient of s02 in nc=n*s02 */     
nc=n*s02                                  /* nc=lambda */
nu=    ;                                  /* nu=error df as function of */
                                          /* a and t */
fcrit=finv(1-alpha,am1,nu); 
power=1-probf(fcrit,am1,nu,nc); 
put n s02 power;                         /* n may not be satisfactory */
                                         /* output for BIBD--could add */
                                         /* b as function of a and n */
                                         /* and put b in the output file */

end; 
end; 
run;



After the data is stored in an external file, it can be plotted in

Minitab or Splus (or SAS/Graph for that matter). I think contour plots

are the most effective static 3-dimensional plots for power analysis.