Full factorial 2^k & 3^k experiment.


 
 data dat1;
 infile 'C:\usr\dir\namedat.csv' dlm=',';
 input y A B C;
 
 If you have to write the data set or if you have to make the design
 on your own then go for this :

 

  proc factex;
   factors C B A/nlev=3;
   size design=27;        option-> fraction=2 
   model res=max;         option -> model estimate = (A|B|C@2)
			            model res=5 
   blocks nblocks=3;
   examine C A(3);
   output out=deslay blockname=block
                A nvals=(0 1 2)
                B nvals=(0 1 2)
                C cvals=('a' 'b' 'c');     Option-> designrep=3
  run; 

 
 

 Once you have the dataset, you have to analyse the data and get some outputs.
 Answer the following questions..

 1) R-squared,
 2) model-anova, parameter estimates, 
 3) Hypothesis tests.. 
 4) Contrast estimates.. 
 5) Residual Analysis..
 6) important factors..
 7) profile plots & interaction plots.
    If interaction is present then go for Contour plot.
 8) tests for violation of assumptions
    normality, independence, equality of variance.



 
  ods output Estimates =estime;
  ods listing close;

  proc glm data=dat1 outstat=ssout;
   class A B C;
   model y=A|B|C/ss1 solution clparm;  "solution" gives the parameter estimates
   means A B C/hovtest=levene;          Option-> hovtest=bartlett
   output out=resout r=resid p=predict;
   estimate 'A' A -1 1/divisor=1;        Main effect of A   
   estimate 'AB' A*B 1 -1 -1 1/divisor=2; 
   contrast 'A' A -1 1;
  run;
 

 ods listing;
 proc print data=estime;
 run; 

 Following code gives you the regression coefficients
 proc glm data=dat1;
  model y=A|B|C;
 run;


 Residual plots

 proc plot data=resout;
  plot resid*(predict A B C)="*"/vref=0;
 run;

 proc univariate data=resout noprint;
  probplot residual/normal(mu=est sigma=est);
  inset normal;
 run;
 Important Factors' Selection

 title 'Normal probability plot';
 proc univariate data=estime noprint;
  probplot Estimate/normal(mu=est sigma=est);
  inset normal;
 run;

 proc plot data=ssout;
  plot SS*_SOURCE_=_TYPE_;
 run;

  data estim1;
    set estim;
	absestim=abs(Estimate);
   run;

 title 'Half normal probability plot';
 proc univariate data=estim1 noprint;
  probplot absestim/normal(mu=est sigma=est);
  inset normal;
 run;



 Main effects plots

 proc means data=proce noprint;
   class A;
   var yield;
   output out=outa mean=avga;
  run;

  data meanout;
   merge outa outb outc outd;
  run;

  proc means data=proce noprint;
   class B;
   var yield;
   output out=outb mean=avgb;
  run;

  title 'Main effects plots';
  symbol i=join h=1 v=dot;
  legend1 label=none;
  proc gplot data=meanout;
   plot avga*A avgb*B avgc*C avgd*D/overlay legend=legend1;
  run;


Interaction Plots 

 proc means data=silic;
  class A B;
  var resis;
  output out=outab mean=avgab;
 run;

 title 'Interaction plot';
 symbol i=join v=dot h=1;
 proc gplot data=outab;
  plot avgab*A=B;
 run;

Contour Plots

  proc g3grid data=resout2 out=gridout;
   grid A*B=predict2/naxis1=31 naxis2=31 spline;
  run;

   proc gcontour data=gridoutab;
    plot A*B=predict2/grid autolabel;
   run;

 Surface Plot

   proc g3d data=gridoutab;
    plot A*B=predict2/grid autolabel;
   run;