It is shown that the exponential distribution is a special case of the double exponential distribution. A condition is given which must be satisfied by the parameters of the double exponential distribution. It is shown that the maximum likelihood equations for the parameter estimates point out the existence of such conditions. A model of a phenomenon is exhibited whose statistical properties may be described by the double exponential distribution. The meaning of the transition described in this work is explained by using that model.
For balanced classifications and for unbalanced nested classifications, we give a simple rule which establishes a one-to-one correspondence between numbers of degrees of freedom and projection operators in the analysis of variance. The rule enables us to determine explicitly the matrix representations of appropriate sums of squares.