Jim Bjork asked a question regarding contrast test in 3dDeconvolve from an old thread (http://afni.nimh.nih.gov/afni/phorum/read.php?f=1&i=2051&t=2000):
When making a GLT contrast between event type A and event type B using the simple, single-line AUC matrix (0 0 0 0 -1 1 0 0, etc), how are results affected if event A is far more *numerous* than event B?
In other words, if the idealized waveform of A is, say, three times as complex than the waveform of B, is it necessary to triple the maximum peak height of the idealized B waveform to compensate?
===========
Jim,
This is really a good question. If your interest is in comparing the magnitude between the regressor coefficient (beta value or percent signal change) of event A and that of event B, then just simply assign 1 and -1 at the corresponding locations in the contrast vector/matrix with others being 0.
The fact that event A occurrs more often (i.e., 3 times more) than event B in the experiment should not change the way you make comparison of their magnitude, and the estimates of the beta's are always unbiased. Instead, it would only have effect on the sampling error of the regressor estimation. That is, since event A has more data available for estimating its beta value than event B, the estimation error for the beta of event A would be smaller than event B. In other words, it does not make sense that your estimate of the percent signal change would depend on how many time points avaible for its estimation.
Gang