A.4  SAS file for CMH analysis of Chinese smoking data in Table 3.3

data cmh;
input center smoke cancer count @@;
cards;
1 1 1 126  1 1 2 100   1 2 1  35  1 2 2  61
2 1 1 908  2 1 2 688   2 2 1 497  2 2 2 807
3 1 1 913  3 1 2 747   3 2 1 336  3 2 2 598
4 1 1 235  4 1 2 172   4 2 1  58  4 2 2 121
5 1 1 402  5 1 2 308   5 2 1 121  5 2 2 215
6 1 1 182  6 1 2 156   6 2 1  72  6 2 2  98
7 1 1  60  7 1 2  99   7 2 1  11  7 2 2  43
8 1 1 104  8 1 2  89   8 2 1  21  8 2 2  36
;
proc freq; weight count;
   tables center*smoke*cancer / cmh;
run;




        The SAS System           15:02 Tuesday, May 18, 1999   9

               SUMMARY STATISTICS FOR SMOKE BY CANCER
                       CONTROLLING FOR CENTER


     Cochran-Mantel-Haenszel Statistics (Based on Table Scores)

   Statistic   Alternative Hypothesis    DF       Value      Prob
   --------------------------------------------------------------
      1        Nonzero Correlation        1     280.138     0.001
      2        Row Mean Scores Differ     1     280.138     0.001
      3        General Association        1     280.138     0.001



         Estimates of the Common Relative Risk (Row1/Row2)
                                                        95%
   Type of Study    Method              Value   Confidence Bounds
   --------------------------------------------------------------
   Case-Control     Mantel-Haenszel     2.174      1.985    2.382
     (Odds Ratio)   Logit               2.173      1.983    2.382

   Cohort           Mantel-Haenszel     1.519      1.447    1.595
     (Col1 Risk)    Logit               1.513      1.436    1.594

   Cohort           Mantel-Haenszel     0.700      0.671    0.730
     (Col2 Risk)    Logit               0.701      0.673    0.730

   The confidence bounds for the M-H estimates are test-based.



         Breslow-Day Test for Homogeneity of the Odds Ratios

   Chi-Square =   5.200           DF =   7           Prob = 0.636


   Total Sample Size = 8419