Assume a clinical trial, two groups A and B, equal random allocation to the 
two groups, with a quantitative outcome Y.

     Assume that Y is measured at baseline and then one year later.

     There are three approaches to analysis that might be considered:

     1.  Let the dependent variable be delta(Y), i.e., Y1 - Y0, where Y0 is
         the baseline level and Y1 is the year 1 level.  The regression equation
         is

                   Y1 - Y0 = b0 + b1 * group + e,

         where e ~ N(0, sigma^2), group = 1 (group A) or 0 (group B),
         and b0 and b1 are unknown coefficients.  The parameter of most
         interest is b1; the null hypothesis is b1 = 0.

     2.  Let the dependent variable be Y1.  The regression equation is

                   Y1 = b0 + b1 * group + e,

         where as before e ~ N(0, sigma^2), and b0 and b1 are unknown, and
         the parameter of interest is b1; null hypothesis again is b1 = 0.

     3.  As in 2., but the baseline value is included in the regression as
         a covariate:

                   Y1 = b0 + b1 * group + b2 * Y0 + e.

     4.  As in 1., but again the baseline level is included as a covariate,
         that is,

                   Y1 - Y0 = b0 + b1 * group + b2 * Y0 + e.


     The question is: which of these 4 approaches to analysis has the
greatest power?  Can you demonstrate that with simulation studies?