Non-stationary time series and the robustness of circadian rhythms

The results of the various analyses allow the inference that, after high-frequency noise is filtered out, the value of the periodogram's Q(P) statistic reflects the extent of stationarity of the time series, and the "robustness" of a circadian rhythm can serve as an index of the station parity of the rhythm.

Abstract

Circadian rhythms are regular oscillations in the value of behavioral and physiological variables of organisms that recur on a daily basis. The purpose of this study was to evaluate the extent of non-stationarity of circadian rhythms over several days, to determine how damaging is the violation of the assumption of stationarity in the analysis of circadian rhythms, and to formalize the concept of "rhythm robustness" as an index of oscillatory ("weak") stationarity. Simulated (computer-generated) and experimental data sets (rhythms of body temperature and running-wheel activity in several rodent species) were analysed. Tests of stationarity based on the variance of the daily means and the variance of the daily variances revealed that most experimental data sets are not stationary. Analysis of linear trends indicated that significant trends are rare in experimental data sets. Although the non-stationarity of the experimental data sets reduced the spectral energy of the Enright periodogram used to assess rhythmicity, detection of circadian rhythmicity was not prevented in any of the rhythmic data sets. The results of the various analyses allow the inference that, after high-frequency noise is filtered out, the value of the periodogram's Q(P) statistic reflects the extent of stationarity of the time series. Thus, the "robustness" of a circadian rhythm (i.e. the magnitude of the empirical Q(P) value as compared to the Q(P) value associated with a perfectly rhythmic time series) can serve as an index of the stationarity of the rhythm.