Hi

One small change / addition to my earlier posting

James M. Clark
Professor of Psychology
204-786-9757
204-774-4134 Fax
[EMAIL PROTECTED]

>>> "Jim Clark" <[EMAIL PROTECTED]> 01-Jun-08 1:28:26 AM >>>
To illustrate, the third row selects cases above 700 (2 SDs above the mean), 
which amounts to about 16% of the 100,000 scores.  The sample r of .4 produces 
a population rho of .70, quite close to the actual value.  With extreme 
selection (bottom row, SAT > 800, .13% of scores), the formula appears to 
overcorrect.

   sdx      r   sigx  sigx2   rho1   rho2   rho3   rho4   rho5
100.00  .7100 100.00  10000  .7100  .7100  .7100  .7100  .7100
59.000  .5200 100.00  10000  .7181  .7181  .7181  .7181  .7181
44.000  .4000 100.00  10000  .7042  .7042  .7042  .7042  .7042
33.000  .3100 100.00  10000  .7029  .7029  .7029  .7029  .7029
22.000  .2800 100.00  10000  .7984  .7984  .7984  .7984  .7984

JC:
Rather than over-correction at SAT>800, estimate becomes more variable with 
extreme selection (or relatively small n), and I happened to report a high 
value.  I ran the simulation 10 times with SAT>800 and obtained a mean correct 
r of .716 (close to expected value) but with SD corrected r = .131.  Five 
simulations for SAT > 700 produced mean corrected r of .702 with SD corrected r 
= .023.

Take care
Jim


---
To make changes to your subscription contact:

Bill Southerly ([EMAIL PROTECTED])

Reply via email to