An Introduction to Multitaper Time-Series Analysis

Advances in instrumentation made over the last century now allows many disciplines to collect data in the form of time series and so, in principle, study dynamics of their problems. However, because the data date from the era of brass and mahogany galvanometers, inferences made from such data are often poor, or contradictory, as with Q 0957. Similarly, the continuing use of periodograms, known by Rayleigh (1903) to give inconsistent estimates, is another scientific embarrassment. Multitaper estimates of the spectrum, Thomson (1982) were introduced a century after Stokes (1879) suggestion of the periodogram, and solve many of the problems encountered in estimating spectra. In this talk, I review multitaper estimates of the spectrum and related functions such as coherence and tests for periodic components. Jackknife error estimates and problems with missing data are included, with examples.