[R] mean as a condition of an effect?

[John Christie]

	I got a letter from a reviewer about a recent paper and it seems like 
he is asking for something that is... inappropriate.  But, I am not 
certain that the whole argument would be.
	The dv in the study is reaction time.  And, one of my hypotheses is 
that fast people will tend to show a postitive effect across the 
independent variable (A1>A2) while slow people will show a negative 
effect (A1<A2).  The editor suggested that I perform an analysis that 
looked at the interaction between my mean subject RT and my independent 
variable A on RT  (rt ~ A:subjectrt).  But, isn't this wrong given that 
one of the independent variables is necessarily related to the 
dependent variable?  I don't have anything against finding this 
interaction in principle.  But, it seems to me that performing the 
median split I had originally planned and looking at the effect 
direction on either side is more appropriate.
	Opinions?