Abstract
It is shown by two examples that the algorithm proposed by Chervin and Schneider (1976) is of little use in deciding, with a given risk, whether a GCM result differing from others (or nature) is caused by chance or by prescribed changes of some boundary values. What remains is that one can believe in a “significant” change if the rate of rejections of local null hypotheses is quite large (e.g., three times expectation or more). Additionally, the Chervin-Schneider algorithm can be used to gain a first guess.