A comparison of methods that characterize pulses in a time series

There is general recognition that some glands secrete hormones primarily as a series of pulses. One can generally classify the proposed methods of pulse identification and characterization as either (i) criterion‐based, that is, they use a criterion such as a test statistic to identify a rise and/or fall in hormone level within a moving window, or (ii) model‐based, that is, they specify a statistical model for the time‐varying portion of the signal and estimate its parameters. Using simulated data, we compare and contrast seven criterion‐based methods and three model‐based methods. The model‐based methods perform better in estimating the parameters of interest; they are most effective with the sampling rate chosen so that there are 3–5 samples taken during the half‐life of the hormone. At higher sampling rates the methods may overidentify pulses (false positives) and at lower sampling rates they may miss pulses (false negatives), both of which lead to biased estimates for the parameters of the signal.