Sample Size & Power - TOST

[안재형]

# CV는 %대신 100을 곱하지않은 값으로 사용.

# correlation이 높으면 적은 sample size가 필요하다. corr=0으로 최악의 경우를 제시.

proc power;
  pairedmeans test=equiv_ratio
  lower=.8
  upper=1.25
  meanratio=1 1.2
  corr=0
  cv=.1 .2 .3
  npergroup=.
  power= .8 .9;
run;

### 결과 

                                       Fixed Scenario Elements

                               Distribution                  Lognormal
                               Method                            Exact
                               Lower Equivalence Bound      0.8
                               Upper Equivalence Bound     1.25
                               Correlation                           0
                               Alpha                                0.05

                                           Computed N Pairs

                                     Geo
                                    Mean            Nominal    Actual      N
                          Index   Ratio    CV      Power     Power    Pairs

                              1      1.0     0.1        0.8     0.901        6
                              2      1.0     0.1        0.9     0.901        6
                              3      1.0     0.2        0.8     0.802       15
                              4      1.0     0.2        0.9     0.910       19
                              5      1.0     0.3        0.8     0.816       32
                              6      1.0     0.3        0.9     0.901       39
                              7      1.2     0.1        0.8     0.804       76
                              8      1.2     0.1        0.9     0.901      104
                              9      1.2     0.2        0.8     0.801      293
                             10      1.2     0.2        0.9     0.900      405
                             11      1.2     0.3        0.8     0.800      641
                             12      1.2     0.3        0.9     0.900      888