我将使用R函数qtukey(1-alpha,means,df)来制作家庭式CI。
例如,R函数qtukey(1-0.05,nmeans = 4,df = 16)给出了临界值 Ť ü ķ ë ÿ0.05 ,4 ,16= 4.046093。
给定对象间设计,其中k = 4组,则5 * k = 20个样本大小,例如(5-1)* k = 16 df 中号小号Ë[R [R Ø [R,
Ťü ķ ê ÿķ ,dF= M一个Xj=1,2,…,k{zj}−Minj=1,2,…,k{zj}χ2df/df−−−−−−√=Rangej=1,2,…,k{Mj−μjσM}SEM/σM=Rangej=1,2,…,k{Mj−μj}SEM=Max1≤j1,j2≤k{∣∣(Mj1−μj1)−(Mj2−μj2)∣∣}SEM=Max1≤j1,j2≤k{∣∣(Mj1−Mj2)−(μj1−μj2)∣∣}SEM
The radius of family-wise 1-α CIs is SEM×tukeyα,4,16=MSError5−−−−−−√×tukeyα,4,16 because--
{Tukeyk,df≤tukey0.05,4,16}={Max1≤j1,j2≤k{∣∣(Mj1−Mj2)−(μj1−μj2)∣∣}SEM≤tukey.05,4,16}=∩1≤j1,j2≤k{∣∣(Mj1−Mj2)−(μj1−μj2)∣∣≤SEM×tukey.05,4,16}
Given a within-subject design with k=4 levels, 17 sample size e.g. (17-1)=16 df for MSError, and Xi,j=(μj+vi)+εi,j=X˜i,j+εi,j, the radius of family-wise (1-α) CIs is MSError/17−−−−−−−−−√×tukeyα,4,16 because--
Tukeyk,df=Maxj=1,2,…,k{zj}−Minj=1,2,…,k{zj}χ2df/df−−−−−−√=Rangej=1,2,…,k{Mean1≤i≤n{X˜i,j+εi,j}−Mean1≤i≤n{X˜i,j}σMean1≤i≤n{εi,j}}σ^Mean1≤i≤n{εi,j}/σMean1≤i≤n{εi,j}=Rangej=1,2,…,k{Mj−(μj+Mean1≤i≤n{vi})}σ^Mean1≤i≤n{εi,j}=Rangej=1,2,…,k{Mj−μj}MSError/n−−−−−−−−−√=Max1≤j1,j2≤k{∣∣(Mj1−Mj2)−(μj1−μj2)∣∣}MSError/n−−−−−−−−−√