We propose an approach for analyzing signals with long-range correlations b
y decomposing the signal increment series into magnitude and sign series an
d analyzing their scaling properties. We show that signals with identical l
ong-range correlations can exhibit different time organization for the magn
itude and sign. We find that the magnitude series relates to the nonlinear
properties of the original time series, while the sign series relates to th
e linear properties. We apply our approach to the heartbeat interval series
and find that the magnitude series is long-range correlated, while the sig
n series is anticorrelated and that both magnitude and sign series may have
clinical applications.