與mice函數(shù)的流程極為相似
模擬練習(xí)數(shù)據(jù)
set.seed(8888)
id <- rep(1:15,each=10)#id號，1-15重復(fù)10次
time<-rep(1:10,15)#時間（天）1-10重復(fù)15次
map.raw <- abs((round(rnorm(150,mean = 50,sd = 25))))#隨機(jī)取150個符合正態(tài)分布的平均動脈壓
map<-round(ifelse(map.raw>=30,map.raw,map.raw+50))#保證沒有小于30的平均動脈壓
lac<-round(3-map*0.06+rnorm(150,mean = 0,sd = 0.4)-0.4*time*time+4.3*time,1)#乳酸值
lac.miss.tag<-rbinom(150,1,0.3)#取30%的缺失值
lac<-ifelse(lac.miss.tag==1,NA,lac)
age<-rep(round(abs(rnorm(15,mean = 65,sd = 19))),each=10)
data<- data.frame(id,time,age,map,lac)
abs(x)取x的絕對值
round（x）取x的整數(shù)
aggr(data,numbers = TRUE, prop= FALSE)
進(jìn)行插補(bǔ)
A <- amelia(data,m = 5,ts = "time",cs = "id")
查看插補(bǔ)效果
tscsPlot(A,cs = c(3,4,5,6),var = "lac")
下圖紅點是插補(bǔ)的值难述，紅線是其置信范圍
可見置信范圍較大，插補(bǔ)的效果不好
考慮時間因素邓厕，重新插補(bǔ)
A2<- amelia(data,m = 5,ts = "time",cs = "id",polytime = 2)
polytime#取0到3之間的整數(shù)昔园，表示應(yīng)在插補(bǔ)模型中包含多項式的冪以說明時間的影響。 設(shè)置為0將指示恒定水平阿浓，將1指示線性時間效應(yīng)他嚷，將2指示平方效應(yīng)，而將3指示立方時間效應(yīng)芭毙。
可見插補(bǔ)效果很好
tscsPlot(A2,cs = c(3,4,5,6),var = "lac")
考慮缺失值前后數(shù)值影響
lags參數(shù)和leads參數(shù)
A3<- amelia(data,m = 5,ts = "time",cs = "id",lags = "lac",leads = "lac")
效果也不好
tscsPlot(A3,cs = c(3,4,5,6),var = "lac")
加入先驗信息
有時候根據(jù)文獻(xiàn)和經(jīng)驗筋蓖，某一個缺失變量具有一定的取值范圍
如lac均值為3的患者容易存活，而且其標(biāo)準(zhǔn)差為1.2
使用prior參數(shù)
You can provide Amelia with informational priors about the missing observations in your data. To specify priors, pass a four or five column matrix to the priors argument with each row specifying a different priors as such:
one.prior <- c(row, column, mean,standard deviation)
or,
one.prior <- c(row, column, minimum, maximum, confidence).
假設(shè)第4例患者存活退敦，我們認(rèn)為其乳酸水平符合經(jīng)驗
data[data$id==4,]
lac.prior<-matrix(c(31,32,37,38,5,5,5,5,3,3,3,3,1.2,1.2,1.2,1.2),nrow = 4,ncol = 4)
A4<- amelia(data,m = 5,ts = "time",cs = "id",priors = lac.prior)
插補(bǔ)值診斷：插補(bǔ)值是否合理
opar <- par(no.readonly=TRUE)
par(mfrow=c(2,2))
compare.density(A,var = "lac",main = "1")
compare.density(A2,var = "lac",main = "2")
compare.density(A3,var = "lac",main = "3")
compare.density(A4,var = "lac",main = "4")
par(opar)
當(dāng)紅藍(lán)兩條曲線重合程度越多時粘咖，證明插補(bǔ)越合理
opar <- par(no.readonly=TRUE)
par(mfrow=c(2,2))
overimpute(A,var = "lac",main = "1")
overimpute(A2,var = "lac",main = "2")
overimpute(A3,var = "lac",main = "3")
overimpute(A4,var = "lac",main = "4")
par(opar)
當(dāng)點越靠近斜線周圍時，插補(bǔ)越合理
