Bootstraps, Een Nieuw Statistisch Alternatief
E.D.J. Schils |
Instituut voor Toegepaste Taalwetenschap en Methodologie Katholieke Universiteit Nijmegen
The bootstrap is a Monte Carlo method for the approximation of the sampling error of a statistic. The bootstrap method estimates this standard error on the basis of the repeated calculation of the statistic at hand in each of a great number of so-called bootstrap samples, i.e. samples with replacement from a probability distribution which exactly mirrors the empirical relative frequency distribution. The method is useful when there is no analytical sampling theory available for the statistic at hand, or when violation of underlying assumptions precludes the application of an available sampling theory.
This paper uses an analytically well-known problem as the context for the presentation of the method, viz. the sampling distribution of the arithmetic mean. The method is then applied to an investigation of language loss using an unfamiliar research design.
Article language: Dutch