Statistische proefopzetten: bewijs en detectie

Statistics: proof and detection The validity of the conclusions drawn from a statistical analysis depends to a great extent on its historical development. If the experiment has been designed carefully in order to find the answer to questions specified in advance by means of experimental and statistical methods agreed upon before the observations are available, the validity of the conclusions is only restricted by the level(s) of significance used. If, on the other hand, the observations have given rise to new questions, which are then investigated on the basis of these observations themselves or if, during the analysis the statistical methods are changed in order to adap them to the observations, the validity of the conclusions may be impaired. In statistical practice the latter situation seems to be the rule and the former the exception; mixtures of both also occur very often. In this paper the importance of distinguishing between these two methods of using statistics is stressed. Whereas in the first case one might speak of a “statistical pro of”, the second method — however useful it may be — has the character of “statistical detection” and should strictly speaking always be followed by a corroborative experiment which meets the conditions of a statistical proof. When writing a statistical report one should provide the reader with sufficient historical data to enable him to judge the validity of the results for himself.