最简单的方法，用软件产生随机数，再把数据按随机数大小排序（一般由小-大），取前面n个即可。

比如我有100000个观测值，100个变量，但是我依然要使用这100000个观测值，只是将变量减少，这样可以减小文件大小，stata就可以读取了。是不是用 use xx.dta option  ，这个option应该是什么呢？

据我所知不能这样。Stata一般读个3、4G的文件不会是很大问题啊，为什么非要这么做？实在是需要的话用SAS先处理一下吧。

我也是奇怪，以前读1个多G可以，现在这个不到3个G，有如下提示，望指点：

op. sys. refuses to provide memory
    Stata's data-storage memory manager has already allocated 1520m bytes and it just attempted to allocate
    another 16m bytes.  The operating system said no.  Perhaps you are running another memory-consuming task and
    the command will work later when the task completes.  Perhaps you are on a multiuser system that is
    especially busy and the command will work later when activity quiets down.  Perhaps a system administrator
    has put a limit on what you can allocate; see help memory.  Or perhaps that's all the memory your computer
    can allocate to Stata.
r(909);

谢谢各位，已找到相关命令，use 变量名 using 数据名，即可，附参考网站：
http://www.stata.com/support/faqs/windows/large-datasets-and-memory/