Abstract

Classical spectral analysis is usually badly suited for the detection of periodicities in
biological or biomedical time series because of the shortness and low density of the series of data.
Mathematical study of these series thus requires particular methods adapted to the nature of the data.
The well-known Cosinor test and the periodogram method are compared from that point of view.
An extensive description of both methods is given. Comparisons between assumptions, application fields,
estimates provided, reliability of the results are made. The Cosinor test requires more initial assumptions,
appears as an extremely particular case of the periodogram method, and may lead to wrong interpretation of the data.
The periodogram procedure is more profitable and safer but involves more computations. The Cosinor test may avoid
unnecessary calculations when the phenomenon has one single, a priori known, natural frequency.