transform on each burst. The Fourier tranforms of 16 bursts per channel were then ensemble averaged. Two values were extracted from each ensemble-averaged spec trum, the weighted sum of squares of the spectrum (WSS
- computed as the step frequency multiplied by the power of the spectrum at that frequency) and the sum of squares value of the spectrum (SSV -computed as the signal power at each frequency step) such that changes in MES amplitude (represented by SSV) and MES mean frequency (by definition, computed as the ratio WSS/SSV) over time could be monitored, and such that the resolution of the mean frequency value was not lim ited by the small epoch lengths. The validation of this process is fully described in McLean et al. (2000). For each sequential 1s of data recorded, the raw data from the previous 1s were overwritten by the next set of bursts of MES, and only the SSV and WSS were stored in the memory of the portable system. One estimate of SSV and WSS was computed for every 2 s of MES data over each 3-h recording period. This method of data reduction allowed for a prolonged MES recording period using four MES channels despite the system memory limita tions (512 kB). Immediately following each MES data collection, the SSV and WSS values were uploaded in ascii format to an IBM 133MHz Think-Pad laptop computer.