| Superordinate objective (category) | Significance tests |
| Method | Bootstrap hypothesis test |
| Description + literature | Bootstrapping is a resampling method. On basis of a single sample, the test is repeated on parametrically or nonparametrically obtained resamples, and a distribution of the test statistic determined (null distribution) Ususally the test statistic obtained on the sample is compared with the null distribution. This leads to the test significance by comparing the sample-statistic with quantiles of the null distribution. Climate Time Series Analysis, Mudelsee, 2010, pp. 91-94 |
| Useful for (parameter, time resolution) | You use bootstrap methods when the theoretical distribution of the statistic of interest is unknown or if no parametric method is available. |
| Requirements for application | Determination of the null distribution depends on the problem at hand. It is important that the determination of the null distribution has to preserve the original properties of the data generating process (e.g., autocorrelation). For that aim, bootstrap adaptations, such as block bootstrap resampling, can be employed. |
| Result/interpretation | Significance of a hypothesis test (which guides you whether or not to accept the null hypothesis). |
| Assessment | On the one hand, this method delivers results that are independent of strong assumptions (positive aspect), but the implementation and application may be difficult (negative). Depending on concept of the bootstrap method and the sample size, this method may be rather computing-intensive. |
| Example/publication | Signifikance test of correlations |
| Contact/project | Oliver Krüger, Helmholtz-Zentrum Geesthacht, Institut für Küstenforschung, Oliver.krueger@hzg.de |