你可以看看人家的讨论
http://cos.name/cn/topic/2470

ls的指导很到位，其解源于对回归理论的熟练

我的想法是：如果这两个系数可比的话。如果两个系数不那么相关的话
let d =beta1 -beta2, now test if d =0.
var(d) =var(beta1-beta2) =var(beta1) +var(beta2) -2*cov(beta1, beta2). 这三个值可以从估计的covariance 矩阵而来从而是可知的。然后
t =d/sqrt(var(d)), 得到t统计量，检测是否差异为零。
on the other hand, since beta1 and beta2 are not independent, can t value here be approximated by T distribution? if cov(beta1, beta2) is relevently small, sounds no problem here.
Another suggestions is to use bootstrapping. Unfortunately, i am still learning.
京剧

proc reg里面的test语句就可以啊。
proc reg data=a;
model y=x1 x2;
test x1-x2;
run;

我也是用ls的方法检验系数是否差异
test x1-x2=0