Date:         Thu, 22 Jan 2009 17:47:31 -0500
Reply-To:     "P. Cristian Gugiu" <crisgugiu@YAHOO.COM>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         "P. Cristian Gugiu" <crisgugiu@YAHOO.COM>
Subject:      Re: Matched-Group Design

Okay, I just generated 5000 bootstrap samples and analyzed them with each
of the three proposed methods. Some interesting results. All three methods
failed to reject H0: treatment group has an impact. The p-values for the
group variable were similar in the first two methods although the second
method found a stronger effect for the covariate (Pre_Systolic). The third
method seems to be the most powerful, unless I am misinterpretting the
results.

Cristian


Method 1: Proc GLM
Obs    _SOURCE_        DF       SS         F         Prob
  1     ERROR           21    4949.01     .          .
  2     GROUP            1     376.33    1.87309    0.38280
  3     MATCHID         22    5388.42    1.15557    0.47513
  4     PRE_SYSTOLIC     1    1929.07    8.94759    0.04430


Method 2: Proc Mixed
                        Num     Den
Obs    Effect            DF      DF     FValue     ProbF
  1     GROUP             1      21       1.84    0.3708
  2     PRE_SYSTOLIC      1      21      13.58    0.0034


Method 3: Proc Mixed of paired differences
        Num     Den
Obs      DF      DF     FValue     ProbF
  1       1      22       5.97    0.1032