Próba reprezentatywna – potrzeba i propozycja definicji

A ‘representative sample’ is one of few terms used in statistical inference which do not have a precise and commonly recognised definition. The need for such a definition is less nagging in sciences and natural sciences, where empirical experimental research has exploratory objectives, and thus rarely requires obtaining representative samples. On the other hand, descriptive and confirmatory objectives of social sciences and economics necessitate samples on the basis of which it would be possible to make generalisations of relevant estimates for a given population and assess their errors. The aim of this paper is to discuss the role of a representative sample in scientific and statisti-cal inference, and additionally to propose its definition and show its specificity in comparison to non-random samples, whose representativeness is limited to some specified variables only. In the proposed definition, the author focuses on the consistency of the sample structure with the structure of the population and the possibility of making inference in which estimates obtained from the sample have numerical error values assigned to them. It is also argued that samples which show convergence of the distributions of only some characteristics with their population distributions should be called ‘representative samples in terms of specified variables’ instead of just ‘representative samples’. Moreover, it is demonstrated that the category of a random sample which is inseparably linked to statistical inference does not need any equivalents.