COMPARISON OF SEVERAL METHODS OF ASSESSMENT OF STATISTICAL SIGNIFICANCE OF LINEAR TRENDS

D. B. Kiktev and V. N. Kryzhov

Comparison of different methods of the assessment of the statistical significance of linear trends is performed. It is shown that traditional methods overestimate statistical significance at serial correlation exceeding 0.1. Accounting for the sample autocorrelation through a decrease in the number of degrees of freedom is not effective, and prewhitening is allowed only for samples of a certain correlation structure. The most realistic estimates of the trend significance are obtained using bootstrap methods requiring no assumptions of the sample distribution and autocorrelation structure.