To estimate changes in gamma power on behavioral time scales, a spectrogram was constructed for each LFP trace using the multitaper method (Mitra and Bokil, 2008; Percival and Walden, 1993). As described previously for neural data (DeCoteau et al., 2007; Pesaran et al., 2002) this method estimates the spectral power S(f) in a finite, sliding time window by averaging over Fourier transforms (discrete Fourier transforms evaluated using the Fast Fourier Transform algorithm on zero-padded data) obtained from each of a set of K orthogonal tapers applied to the data: (1)S(f)=1K∑k=1K |xk(f)|2 with (2)xk(f)=∑t=1T wt(k)xte−2πift where T is the number of points in the time window in which the spectrum is estimated, wt(k) are the tapers, and xt is the signal. The tapers are the first K functions that optimize spectral concentration (the tradeoff between broadband and narrowband bias inherent in spectral estimation), known as Slepians (Mitra and Bokil, 2008; Percival and Walden, 1993). We used the Chronux mtspecgramc function for MATLAB, with the following parameters: window size, 0.5 s; time step, 50 ms; five tapers; bandwidth 6 Hz. From the
