Question

这是我第一次执行多重插补，我选择R包MICE是因为它具有处理不同类型数据的灵活性。我有一个非常大的数据集，其中包含100多个变量，因此我需要将模型限制为最重要的预测变量。

我发现了以下关于如何为插补模型选择预测变量的建议（经过稍微修改以适合下面的示例数据集，这是原始来源：https://stefvanbuuren.name/fimd/sec-knowledge.html）：

”假定数据由结果变量out，两个背景变量bck1和bck2，具有3个项a1-a3的标度a，具有4个项b1-b4的标度b以及所有变量包含缺失值组成。填写初始估算后，估算模型将采取以下步骤：

根据给定的bck1，bck2，a，b进行计算，其中a和b是a1-a3和b1-b4的总和分数；
发出bck1，bck2，a和b；
发出bck2，bck1，a和b；
对a1，bck，b和a2-a3进行罚款；
给出a2，bck，b和a1，a3-a3；
以同样的方式对a3进行攻击；
对给出的b1，bck，a和b2-b4进行惩罚，其中a是更新的总分评分；
以同样的方式对b2-b4进行攻击。”

我根据上述说明创建了一个示例数据集：

> out<-c(45,433,234,435, 234,234,243,243,243,24,244,242, 453,234,689)
> a1<-c(1,4,2,1,4,4,2,5,7,3,2,5,2,2,2)
> a2<-c(4,3,4,3,2,1,3,4,2,3,4,5,6,6,7)
> a3<-c(2,2,1,3,4,2,1,3,4,2,3,4,4,2,1)
> b1<-c(5,5,2,3,6,7,3,2,4,5,6,7,4,5,6)
> b2<-c(2,3,1,2,3,2,1,2,2,3,4,3,1,2,4)
> b3<-c(4,3,4,5,2,1,2,3,4,2,1,2,3,4,2)
> b4<-c(1,1,1,2,4,2,2,4,2,1,1,3,4,2,1)
> bck1<-factor(c(1,1,1,1,1,1,1,1,2,2,2,2,2,2,2), level=c(1,2), label = c('x','y'))
> bck2<-c(44,32,34,23,45,23,67,43,23,18,45,33,23,45,74)
> test<-data.frame(out,a1,a2,a3,b1,b2,b3,b4,bck1,bck2)
> #Produce NAs 
> library(missForest)
> test.na <- prodNA(test, noNA = 0.1)
> head(test.na)
  out a1 a2 a3 b1 b2 b3 b4 bck1 bck2
1  45  1  4  2  5  2  4  1    x   44
2 433  4  3 NA  5  3  3  1 <NA>   32
3 234 NA NA  1  2  1  4  1    x   34
4 435  1 NA  3  3  2 NA  2    x   23
5 234  4  2  4  6  3  2  4    x   45
6 234  4  1  2  7  2 NA  2    x   23

然后，我尝试按照推荐的方式对MI模型进行编码。如果我做对了，我会很不安全。使用此代码，R会在插补过程中更新标度总和吗？哪个变量首先出现在预测变量矩阵中是否重要？

> ini <- mice(cbind(test.na,a=NA, b =NA),max=0,print=FALSE)
Warning message:
Number of logged events: 2 
> meth <- ini$meth
> meth["a"] <- "~I(a1 + a2 + a3)"
> meth["b"] <- "~I(b1 + b2 + b3 + b4)"
> 
> pred<-ini$pred
> pred[c("out"),c("a1","a2", "a3","b1","b2","b3","b4")] <- 0
> pred[c("a1","a2","a3"),"a"] <- 0
> pred[c("out"),c("a","b")] <- 1
> pred[c("b1","b2","b3", "b4"),"b"] <- 0
> pred[c("a1","a2","a3"),c("b1","b2","b3", "b4")] <- 0
> pred[c("a1","a2","a3"),c("b")] <- 1
> pred[c("b1","b2","b3","b4"),c("a")] <- 1
> pred[c("bck1","bck2"),c("a1", "a2","a3","b1","b2","b3", "b4")] <- 0
> pred[c("bck1","bck2"),c("a", "b")] <- 1
> pred
     out a1 a2 a3 b1 b2 b3 b4 bck1 bck2 a b
out    0  0  0  0  0  0  0  0    1    1 1 1
a1     1  0  1  1  0  0  0  0    1    1 0 1
a2     1  1  0  1  0  0  0  0    1    1 0 1
a3     1  1  1  0  0  0  0  0    1    1 0 1
b1     1  1  1  1  0  1  1  1    1    1 1 0
b2     1  1  1  1  1  0  1  1    1    1 1 0
b3     1  1  1  1  1  1  0  1    1    1 1 0
b4     1  1  1  1  1  1  1  0    1    1 1 0
bck1   1  0  0  0  0  0  0  0    0    1 1 1
bck2   1  0  0  0  0  0  0  0    1    0 1 1
a      0  0  0  0  0  0  0  0    0    0 0 0
b      0  0  0  0  0  0  0  0    0    0 0 0
> 
> impdat<-mice(cbind(test.na, a = NA, b = NA), pred=pred, meth = meth)

 iter imp variable
  1   1  out  a1  a2  a3  b1  b3  bck1  a  b
  1   2  out  a1  a2  a3  b1  b3  bck1  a  b
  1   3  out  a1  a2  a3  b1  b3  bck1  a  b
  1   4  out  a1  a2  a3  b1  b3  bck1  a  b
  1   5  out  a1  a2  a3  b1  b3  bck1  a  b
  2   1  out  a1  a2  a3  b1  b3  bck1  a  b
  2   2  out  a1  a2  a3  b1  b3  bck1  a  b
  2   3  out  a1  a2  a3  b1  b3  bck1  a  b
  2   4  out  a1  a2  a3  b1  b3  bck1  a  b
  2   5  out  a1  a2  a3  b1  b3  bck1  a  b
  3   1  out  a1  a2  a3  b1  b3  bck1  a  b
  3   2  out  a1  a2  a3  b1  b3  bck1  a  b
  3   3  out  a1  a2  a3  b1  b3  bck1  a  b
  3   4  out  a1  a2  a3  b1  b3  bck1  a  b
  3   5  out  a1  a2  a3  b1  b3  bck1  a  b
  4   1  out  a1  a2  a3  b1  b3  bck1  a  b
  4   2  out  a1  a2  a3  b1  b3  bck1  a  b
  4   3  out  a1  a2  a3  b1  b3  bck1  a  b
  4   4  out  a1  a2  a3  b1  b3  bck1  a  b
  4   5  out  a1  a2  a3  b1  b3  bck1  a  b
  5   1  out  a1  a2  a3  b1  b3  bck1  a  b
  5   2  out  a1  a2  a3  b1  b3  bck1  a  b
  5   3  out  a1  a2  a3  b1  b3  bck1  a  b
  5   4  out  a1  a2  a3  b1  b3  bck1  a  b
  5   5  out  a1  a2  a3  b1  b3  bck1  a  b
There were 11 warnings (use warnings() to see them)
> impdat
Class: mids
Number of multiple imputations:  5 
Imputation methods:
                    out                      a1                      a2 
                  "pmm"                   "pmm"                   "pmm" 
                     a3                      b1                      b2 
                  "pmm"                   "pmm"                      "" 
                     b3                      b4                    bck1 
                  "pmm"                      ""                "logreg" 
                   bck2                       a                       b 
                     ""      "~I(a1 + a2 + a3)" "~I(b1 + b2 + b3 + b4)" 
PredictorMatrix:
    out a1 a2 a3 b1 b2 b3 b4 bck1 bck2 a b
out   0  0  0  0  0  0  0  0    1    1 1 1
a1    1  0  1  1  0  0  0  0    1    1 0 1
a2    1  1  0  1  0  0  0  0    1    1 0 1
a3    1  1  1  0  0  0  0  0    1    1 0 1
b1    1  1  1  1  0  1  1  1    1    1 1 0
b2    1  1  1  1  1  0  1  1    1    1 1 0

如何在MICE中执行多个商品级插补？

0 个答案: