The Effect of Serial Correlation on Statistical Inferences Made with Resampling Procedures

Abstract Resampling procedures include hypothesis testing methods based on Permutation Procedures and interval estimation methods based on bootstrap procedures. The former are widely used in the analysis of climate experiments conducted with general circulation models (GCMs) and in the comparison of the simulated and observed climates. The latter are used less frequently than their flexibility and utility warrants. Both resampling techniques are powerful tools, which provide elegant means of overcoming fundamental statistical difficulties encountered in the analysis of observed and simulated climate data. Unfortunately, inference based on both resampling schemes are as sensitive to the effects of serial correlation as classical statistical methods. These tools must therefore be used with the same amount of caution as other statistical methods when it is suspected that the data might be serially correlated.