/* Let's look at collinearity diagnostics using the Hald data, in SAS.
   Note that we get the same results as in the R code, which is a GOOD THING. */

options linesize=78 ovp nocenter nodate;
data hald;
  input ta ts taf ds heat;
  sum = ta + ts + taf + ds;
 /*  label ta        = '% tricalcium aluminate'
        ts        = '% tricalcium silicate'
        taf       = '% tetracalcium alumino ferrite'
        ds        = '%  dicalcium silicate'
        heat      = 'heat (calories /g cement)'; */
  ;
  datalines;
  7.0000000      26.000000      6.0000000      60.000000      78.500000    
  1.0000000      29.000000      15.000000      52.000000      74.300000    
  11.000000      56.000000      8.0000000      20.000000      104.30000    
  11.000000      31.000000      8.0000000      47.000000      87.600000    
  7.0000000      52.000000      6.0000000      33.000000      95.900000    
  11.000000      55.000000      9.0000000      22.000000      109.20000    
  3.0000000      71.000000      17.000000      6.0000000      102.70000    
  1.0000000      31.000000      22.000000      44.000000      72.500000    
  2.0000000      54.000000      18.000000      22.000000      93.100000    
  21.000000      47.000000      4.0000000      26.000000      115.90000    
  1.0000000      40.000000      23.000000      34.000000      83.800000    
  11.000000      66.000000      9.0000000      12.000000      113.30000    
  10.000000      68.000000      8.0000000      12.000000      109.40000    
  ;
proc print data = hald;
run;

proc gplot data = hald;
   plot heat* (ta ts taf ds);
run;

proc corr data = hald;
   var   ta ts taf ds heat;
run;

proc reg data=hald;
   model heat = ta ts /r influence xpx i collin collinoint tol vif;
run;

/* Look at multicollinearity in full model */

proc reg data = hald  ;
   model heat = ta ts taf ds / r influence xpx i ss1 ss2 pcorr1 pcorr2
                               partial collin collinoint tol vif;
run;