SAS上有,自己看看帮助应该就可以了.

BOX-COX可以在误差非正态场合，回归函数非线性，误差方差非齐性，观测值间不独立时均可使用~通过这个变换，线性回归中四条假设都满足了~

oh ,I see, that is right

This is TOTALLY wrong. Box–Cox transformation will solve everything? Actually topic of Box–Cox transformation is only one page or two out of one thousand pages of any textbook.

Usually we assume that the functional form is linear in predictors
for example,

y=a+bx+cz  {x, z} are predictors/cocariates/indepent variables/exogenous variables

This is because its simpleness and easiness of mathametical analysis. Functional space can be vary big!

But the theory or data may suggests the linear form may not properly justify the problem/question. For example, the cobb-douglas  function.

Why the power should be 1 not 0.5, 0.3 or 2.5  ?

The Box–Cox transformation is a mean to generize(expand) the linear model in a simple way. The power in Box–Cox transformation is a parameter and need to be estimated from data.

The transformed variable usually needs to be positive.

HTH

On the other hand, the dependent-variable transformation has comparatively limited realistic usage than expected considering its intractable practical meaning of the transformation itself.
We are often puzzled by the question, like “what is that, the square root of weight?”