Jackknifing multiple-window spectra

This paper describes some aspects of resampling multiple-window estimates of spectra and related functions in a time-series setting. We show that the jackknife variance of a log spectrum estimate is lower than that obtained by computing the variance directly, and describe a hybrid form with better performance than either. We also give some details on jackknifing line frequency parameters. Finally, we describe a projection filter and an associated variance estimate that appears to be useful for exploratory work. Emphasis is on the jackknife, but the ideas apply to other resampling ideas.<
>