Hi Similar to Claudia's approach, I wrote an spss program to simulate regression to mean for my wife's clinical psych lectures. Define 10 groups on basis of rank of scores at time 1 (t1) and then compute t2 through t6 with specified reliability between successive times. Plotting means as function of time and rank at time 1 (the 10 groups) nicely shows convergence of extreme groups. Below is the program. Varying #r (the reliability between successive times) demonstrates that convergence to zero of extreme groups occurs more rapidly as reliability decreases.
input program. loop o = 1 to 10000. comp #r = .7071. comp t1 = rv.norm(0,1). comp t2 = t1*#r + rv.norm(0,1)*sqr(1-#r**2). comp t3 = t2*#r + rv.norm(0,1)*sqr(1-#r**2). comp t4 = t3*#r + rv.norm(0,1)*sqr(1-#r**2). comp t5 = t4*#r + rv.norm(0,1)*sqr(1-#r**2). comp t6 = t5*#r + rv.norm(0,1)*sqr(1-#r**2). end case. end loop. end file. end input program. rank t1 /ntiles(10) into group. glm t1 to t6 by group /wsf = time(6) /plot = profile(time*group). Take care Jim James M. Clark Professor of Psychology 204-786-9757 204-774-4134 Fax j.cl...@uwinnipeg.ca >>> "Claudia Stanny" <csta...@uwf.edu> 09-Feb-09 12:45:19 PM >>> I do a version of this as a demonstration of regression effects using a deck of cards or a random number generator on my calculator to "measure" achievement (and create a "deficient" group for treatment and a "high achieving" group for comparison based on the pretest measure) The deficient group always gets better in the post test (only takes a 30-second "treatment" that involves much waving of my hands) and the high achieving group shows some slippage. The effect of random error is more obvious in this demonstration. --- To make changes to your subscription contact: Bill Southerly (bsouthe...@frostburg.edu)