Trend tests: a tendency to resampling

• Main Author: Ramos, Maria do Rosário
• Publication Date: 2014
• Other Authors: Clara, Cordeiro
• Language: eng
• Source: Repositórios Científicos de Acesso Aberto de Portugal (RCAAP)
• Download full: http://hdl.handle.net/10400.2/14725
• topic: Trend tests; Time series; Sieve bootstrap; Sub sampling; Serial correlation; Missing values

Trend analysis is an important problem in time series. Many studies have been developed to investigate this issue, with special attention to its application to environmental and hydrological time series. The presence of autocorrelation and missing observations affects the significance and power of trend tests, parametric or non-parametric. This study assesses the performance of two trend tests, t-test and the Mann-Kendall through an appropriate resampling technique. A new procedure based onsubsampling is proposed in order to assure good statistical properties of these tests. A comparison was established between this new approach and others already developed, such as bootstrap-based tests. In order to evaluate the performance of the new method, a simulation study is conducted considering a set of underlying slopes, different values of autocorrelation and different fractions of randomly missing data. The order of autocorrelation structure is estimated by the best fitting model obtained through the Akaike information criterion. Inspection of the data to detect missing observations is required, before applying the trend tests. In case of missing observations, their estimation and replace is performed by an imputation method available in software.