刚刚又搜了一下，看到这篇帖子有启发https://bbs.pinggu.org/thread-1303263-1-1.html

网上有一个叫winsor的MACRO，你一搜就能找到。

先找到临界点：
proc univariate data=sashelp.class noprint;
   var height;
   output out=pct pctlpts =95 5 pctlpre = height ;
run;proc print data=pct;run;
再进行条件判别：
data class;
   set sashelp.class;
   if 51.3<height<72 ;/*临界值*/
run;

我想两倍的standard deviation是不是可以用来界定这个sample得上下限，所以我想试试下面的语句：
proc sql;
select *,
std(a) as stdA,
mean(a) as meanA,
from b
where a le max and a ge min;
quit;
meanA-2*stdA as min,
meanA+2*stdA as max

搜到了，谢谢