Magnitude-squared coherence

Coherence is the frequency-domain analogue of a correlation coefficient: at each frequency it is the cross-spectral density normalized by the two power spectral densities, and magnitude-squared coherence is the squared modulus of that, a number between 0 and 1 measuring the fraction of one signal's power at that frequency that is linearly predictable from the other. Values near one indicate a consistent amplitude and phase relationship across the estimation window.

Coherence is biased upward when estimated from few segments; with a single segment it is identically one at every frequency, which is meaningless. So the number of independent averages (segments or tapers) both sets the variance and defines the chance level against which a coherence value must be tested. Coherence also cannot, by itself, distinguish genuine interaction from a shared reference, a common input, or volume conduction.