An improved statistical methodology to estimate and analyze impedances and transfer functions.

Estimating the mathematical relationship between pulsatile time series (e.g., pressure and flow) is an effective technique for studying dynamic systems. The frequency-domain relationship between time series, often calculated as an impedance (pressure/flow), is known more generally as a frequency-response or transfer function (output/input). Current statistical methods for transfer function analysis 1) assume erroneously that repeated observations on a subject are independent, 2) have limited statistical value and power, or 3) are restricted to use in single subjects rather than in an entire sample. This paper develops a regression model for transfer function analysis that corrects each of these deficiencies. Spectral densities of the input and output time series and the cross-spectral density between them are first estimated from discrete Fourier transforms and then used to obtain regression estimates of the transfer function. Statistical comparisons of the transfer function estimates use a test statistic that is distributed as chi2. Confidence intervals for amplitude and phase can also be calculated. By correctly modeling repeated observations on each subject, this improved statistical approach to transfer function estimation and analysis permits the simultaneous analysis of data from all subjects in a sample, improves the power of the transfer function model, and has broad relevance to the study of dynamic physiological systems.