I've got the following design problem;
Observations on three (related) variables are to be made
on a patient at each of a sequence of visits to a clinic.
i.e. for patient 1, the resulting data looks like
VISIT 1 (x_11,x_12,x_13)
VISIT 2 (x_21,x_22,x_23)
. .
. .
. .
VISIT r (x_r1,x_r2,x_r3)
This process is to be carried out for a number
of patients that we can regard as exchangeable.
(the objective is to demonstrate that variable X_1
varies systematically with visit number, which is a
proxy for treatment)
The problem with the measurement procedure at any individual visit
is that the patient suffers "fatigue" and hence each successive measurement
is "less reliable" than the previous one; not quite sure yet how this
manifests itself, but let's say we systematically under-measure.
My question is: should the measurement of variables be carried out in
randomly selected order, or not ?
I realize that the nature of the loss of "reliability" is crucial,
but was wondering if anybody had any thoughts, or pointers.
(For what it's worth, I think you shouldn't randomize measurement order,
but should try to model the fatigue effect - the clinician thinks the opposite)
Thanks a lot,
David Stephens
Imperial College
[log in to unmask]
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|