Against the statistical submission: a note on the significance tests
Section: Teaching
Authors
1Alina Benavides Rodríguez, 2Luis Carlos Silva Ayçaguer
Position
1Especialista de Primer Grado de Bioestadística, Dirección Provincial de Salud, Villa Clara.2Investigador Titular, Vicerrectoría de Investigación y Postgrado, ISCM/H.
Contact address
Vicerrectoría de Investigación y Postgrado.Instituto Superior de Ciencias Médicas de la Habana (ISCM/H).Edificio Ramón Paz
Abstract
The significance tests, as they are presently known, appeared in the 1940s; theyare the result of the combination of two rather incompatible theories. In spiteof their conceptual and practical limitations, they have been generallyconsidered as a single and coherent approach to make statistical inferences. Bymeans of an artificial example, which at the same time is quite typical in theframe of medical research, this article illustrates one of the most importantlimitations of the significance tests. Basically, that the rejection of the nullhypothesis could always be possible, provided that a large enough sample sizeis used. Other drawbacks are also mentioned: not to take into account thebackground knowledge and to use the findings as a way to take clinicaldecisions. It is explained that it is possible to avoid such an approach in theinvestigation. Some alternatives are presented; the most simple of them consistin the description of the findings by means of confidence intervals. The most complex is the bayesian theory which seems a very promising approach.
Keywords:
statistic; the significance tests; hypothesisconfidence intervals
Versión en Español
Título:
Contra la sumisión estadística: un apunte sobre las pruebas de significación