Dear Gang
thank you very much for the quick answer. Upon further thought, however, I am pretty convinced that the two operations are not commutable. Try the following simulation:
Assume we have 3 timecourses:
aa = [80 166 16 29 78 104 126 17 7 39]
bb = [174 1 191 155 18 79 67 80 124 33]
cc = [96 17 89 27 69 194 24 96 233 43]
A percent normalization would yield:
aa.n = 100*aa/mean(aa)
bb.n = 100*bb/mean(bb)
cc.n = 100*cc/mean(cc)
The smoothing can be simplistically construed as the average of the 3 timecourses:
mm = (aa + bb + cc)/3
Now the two cases are:
a) percent normalization FOLLOWED by smoothing:
n.sm = (aa.n + bb.n + cc.n)/3
b) smoothing FOLLOWED by normalization:
sm.n = 100*mm/mean(mm)
The difference between n.sm and sm.n is not a null vector:
(n.sm - sm.n) =
[-2.4 15.9 -9.2 -4.6 4.9 1.2 8.9 -4.6 -11.4 1.2]
...but maybe these difference are negligible with real data values.
giuseppe