hayes model 3

가장 인기있지만 3박자가 맞아야 하는 조절된 조절이다. 첫박은 이론적으로 근거가 막강하게 가설을 잡아야 하고 둘째로 데이터 수집도 말끔하게 해야 하며 마지막으로 이상한 표본도 없거나 effect가 표본수 대비 충분해 줘야 하는 것 같다.

데이터 부터 만들어 보자.

x<-rnorm(100)
w1<-rnorm(100)+ x^2
w2<-rnorm(100)+ abs(x)
y<-rnorm(100, 0,1)  + w1*w2*x
co1<-rnorm(100)

d<-data.frame(x,w1,w2,y,co1)

 

boot3<-function(xxx,mmm, mmm2,yyy,d,bootnum){
  ###estimate a*m
  boot3_1<-function(xxx,mmm,mmm2, yyy,d){
    n<-sample(1:nrow(d),nrow(d),replace = T)
    nnk<-d[n,]
    nnk<-as.data.frame(nnk)
    k2<-lm(nnk[,yyy]~ nnk[,xxx]+nnk[,mmm] + nnk[,mmm2]+ nnk[,xxx]*nnk[,mmm] + nnk[,xxx]*nnk[,mmm2] + nnk[,xxx]*nnk[,mmm]*nnk[,mmm2], data=nnk)
    s2<-summary(k2)
    coem<-s2$coefficients
    eff<-as.data.frame(coem)
    eff1<-eff[nrow(eff)-2,1]
    eff2<-eff[nrow(eff)-1,1]
    eff3<-eff[nrow(eff),1]
    efff<-c(eff1, eff2, eff3)
    efff<-matrix(efff, ncol = 3)
    efff
  }
  k<-1
  l<-matrix(rep(NA,bootnum*3),ncol = 3)
  l<-as.data.frame(l)
  repeat{
    l[k,]<-boot3_1(xxx,mmm, mmm2,yyy,d)
    k<-k+1
    if(k>=bootnum+1) break
  }
  estimates<-list(l)
  ci1<-quantile(l[,1],probs = c(.001,0.01,0.05,0.10,0.90,0.95,0.99,.999))
  ci2<-quantile(l[,2],probs = c(.001,0.01,0.05,0.10,0.90,0.95,0.99,.999))
  ci3<-quantile(l[,3],probs = c(.001,0.01,0.05,0.10,0.90,0.95,0.99,.999))
  kmkmkmkm<-list(c(mean(l[,1]),sd(l[,1])),ci1, c(mean(l[,2]), sd(l[,2])),ci2,
                 c(mean(l[,3]),sd(l[,3])),ci3)
  names(kmkmkmkm)<-c("moderation_mean_BootSE_x*w1", "moderation_CI_x*w1","moderation_mean_BootSE_x*w2", "moderation_CI_x*w2", "moderation_mean_BootSE_x*w1*w2", "moderation_CI_x*w1*w2")
  kmkmkmkm
}

 

막 만든 데이터라 유의하지 않다.

boot3(1,2,3,4,d,1000)

$`moderation_mean_BootSE_x*w1`
[1] 0.9347755 0.1685584

$`moderation_CI_x*w1`
     0.1%        1%        5%       10%       90%       95%       99%     99.9% 
0.4957663 0.6142070 0.6843268 0.7289439 1.1602119 1.2217586 1.4037690 1.5743151 

$`moderation_mean_BootSE_x*w2`
[1] 0.007263357 0.059032701

$`moderation_CI_x*w2`
       0.1%          1%          5%         10%         90%         95%         99%       99.9% 
-0.24961595 -0.15938998 -0.09916184 -0.07266851  0.07595744  0.08879090  0.13960394  0.17953389 

$`moderation_mean_BootSE_x*w1*w2`
[1] -0.01454421  0.07065602

$`moderation_CI_x*w1*w2`
       0.1%          1%          5%         10%         90%         95%         99%       99.9% 
-0.31850929 -0.23549835 -0.15771775 -0.10560361  0.05762863  0.07463888  0.10941380  0.15503864