missing value

shiyiming · 发表于 2004-3-20 07:47:54

怎样利用SAS来处理丢失的数据，当然是只不删除记录条的前提下，不做数据的0输入处理或平均值输入处理，有没有其他的更好更有效的处理方法，恳请高手给出详细的指点。谢谢。

shiyiming · 发表于 2004-3-20 10:27:05

可试一试使用预测的方法

shiyiming · 发表于 2004-3-22 18:19:31

到底是用何预测的方法呢，请明示

shiyiming · 发表于 2004-3-22 21:58:42

范围太大了，还是要根据你的数据和常识经验出发。

shiyiming · 发表于 2004-3-26 10:54:01

one of the predictive imputation method:

suppose in dataset A, variable V1 has missing values, and there are some other variables as V2, V3, V4 ... are not missing, then u can build a regression model as

V1=a1+a2*V2+a3*V3+a4*V4+...

After that, use the estimated value for V1 instead of original V1, then there will be no missing values.

You should be careful when u use this method. Make sure it make sense to use V2, V3, V4 to estimate V1.

shiyiming · 发表于 2004-3-28 12:10:14

hannaqiu:
               hello!
   thank you for your sugestion,but I think your way must make sure the
model is  reliable firstly,because the regression model is not always suit for the variables;so I think your way only be used in the simple condition.not be used often .
   I want to give some ideal here,we can first use "cluster" to analysis the data with missing value, then we can input mean value of each group
for the missing value.
what do you think about my ideal ,thank you for your any opions.

shiyiming · 发表于 2004-3-28 13:45:18

我觉得不应该有多复杂。

1. 根据散点图或者经验，看数据趋势，预测曲线类型；

2. 拟合曲线方程，用预测值代替缺失值即可；

3. 偷懒的话，考虑“data fit”软件。

shiyiming · 发表于 2004-3-28 22:09:33

当然可以用cluster来做imputation,不过既然你问的是predictive imputation,我就告诉你怎么做了,用cluster遇到的问题其实是一样的,你又如何保证你的cluster是稳定的呢?

		自动登录	找回密码
密码			立即注册

missing value

missing value

预测的方法

missing value

Re