Computer‐Intensive Methods

‘Resampling techniques’ and ‘computer-intensive methods’ refer to all methods in which the observed data are used to generate a reference distribution by means of randomization, which is then used to assess the significance of a statistic calculated from the observed (not randomized) data.

Abstract

Abstract With recent advances in the power of personal computers, reference distributions of statistics are now often generated using computer‐intensive methods. Several terms are erroneously used interchangeably when referring to such computer‐intensive statistical methods, including ‘resampling techniques’, ‘Monte Carlo’, ‘permutation’, ‘randomization’ and ‘bootstrap’. These techniques are not interchangeable but are fundamentally different in terms of their statistical mechanics. The generic terms ‘resampling techniques’ and ‘computer‐intensive methods’ refer to all methods in which the observed data are used to generate a reference distribution by means of randomization. This reference distribution is then used to assess the significance of a statistic calculated from the observed (not randomized) data. Significance is evaluated under the assumption that the statistic computed using the observed data is sampled from the reference distribution generated with a randomization technique.