The use of ln s2 in dispersion analysis has two advantages: (i) It maps positive values into real values; exponentiating a fitted model to ln s2 always results in a positive estimate of the variance....

The use of ln s2 in dispersion analysis has two advantages:

(i) It maps positive values into real values; exponentiating a fitted model to

ln s2 always results in a positive estimate of the variance.

(ii) ln s2 = ln 𝜎2 + ln[𝜒2

n−1∕(n − 1)] allows ln 𝜎2 to be modeled separately.