data test;
input a b c ;
cards;
1 2 3
1 . 3
. 2 3
;
run;
data test1;
set test;
array arr1 _numeric_;
do over arr1;
if arr1=. then arr1=0;end;
run;

data test2(drop=i);
set test;
array arr2 _numeric_;
do i=1 to dim(arr2);
if arr2(i)=. then arr2(i)=0;end;
run;

[此贴子已经被作者于2007-4-10 15:52:29编辑过]

如果你用累加方法计算,缺失值是自动转为0的。在proc summary等过程中，如果用class分组计算的话，缺失值自动会被剔除出分析过程，除非使用了missing选项。

可以用where语句处理吧？

Proc means automatically uses nonmissing values in x1 to calculate mean of x1, or sum of x1, you don't need to use where statement.

proc means data=yourdata n mean sum;

var x1 x2;

run;

If you use the following 'where' statement, it is the same as above (SAS default!):

proc means;

var x1 x2;

where x1 ne . or x2 ne .;

run;

But If you use 'And' in where statement:

proc means;

var x1 x2;

where x1 ne . and x2 ne . ;

run;

The data you are using to do proc means will be limited to the only records having both x1 and x2 nonmissing values, say you are missing 5 records in x1 and 100 records in x2, 95 to 100 extra records for x1 will be deleted due to x2 is missing. The means of x1, x2 will be different than previous results.

In general, I am agree with windwater's method by using array in datastep. But if you only need to calculate mean or sum of x1, x2, you don't need to replace the missing values. Sometimes if you replaced a lot of missing values by 0, you pushed your data towards 0.

[此贴子已经被作者于2007-4-11 22:52:47编辑过]

不同的目的还是有不同的处理方法.比如说观测了三次想看一个平均值.如果不把"."转为0,最后的分母就会是2而不是3.类似的情况还很多.进行数据步转换的好处在于可以预先根据你的分析目的将数据集整理好.

data a;
input x;
cards;
1
.
2
;
proc means data=a;run;
data b;
input y;
cards;
1
0
2
;
proc means data=b;run;