Please forgive a general stats question:
I am doing a regression of the change in one variable on the change in
a second variable adjusted for the initial value of the dependent
variable:
delta y=f(y1,delta x)
where:
delta y = y2-y1 (i.e. the post treatment value of y minus the pre
treatment value of y)
y1= the pre treatment value of the dependent variable
delta x = x2-x1 (i.e. post treatment x minus pre treatment x)
A review of a plot of the residuals vs. the predicted values shows that
the residuals are large when predicted y in small, and the residuals
small when predicted y is large. (N.B. This is the opposite of the usual
finding where the residuals increase as predicted values increases, a
problem that is often solved by taking the square root of the dependent
variable. The residuals are not "<" shaped, but rather ">" shaped.)
Can someone suggest a transformation that might make the residuals
better behaved?
Thanks
John
P.S. Can anyone suggest a statistical listserve that could address
statistical question?
John Sorkin MD, PhD
Chief, Biostatistics and Informatics
Baltimore VA Medical Center GRECC and
University of Maryland School of Medicine Claude Pepper OAIC
410-605-7119
john@grecc.umaryland.edu
|