<<目次へ

1要因の分散分析　対応がない場合（CRデザイン）�T　[データ数が異なるケース]

プログラム

anova.CR2 <- function(dat){
        p <- ncol(dat)
        Aj <- c()
        nj <- c()
        for(i in 1:p){
                Aj[i] <- sum(dat[,i][!is.na(dat[,i])])
                nj[i] <- length(dat[,i][!is.na(dat[,i])])
        }
        bar.Aj <- apply(dat, 2, mean)
        G = sum(dat[!is.na(dat)])
        N = sum(nj)

        X = G^2 / N
        AS = sum(dat[!is.na(dat)]^2)
        A = sum(Aj^2 / nj)

        SSa = A - X
        SSwc = AS - A
        SSt = AS - X

        DFa = p - 1
        DFwc = N - p
        DFt = N - 1

        MSa = SSa / DFa
        MSwc = SSwc / DFwc

        Fa = MSa / MSwc

        Pa = 1 - pf(Fa, DFa, DFwc)

        result <- matrix(c(
                SSa,  DFa,  MSa,  Fa, Pa,
                SSwc, DFwc, MSwc, NA, NA,
                SSt,  DFt,  NA,   NA, NA), ncol=5, byrow=T)
        rownames(result) <- c("主効果", "誤差", "総")
        colnames(result) <- c("SS", "DF", "MS", "F-value", "p-value")
        result
}

使用法

　データ形式は参考文献と同じにしてください。NAは欠損値を表しています。

> dat
     [,1] [,2] [,3] [,4]
[1,]    9    5   13   11
[2,]    7    6    8   13
[3,]    8    3   13   14
[4,]    8    6   12   16
[5,]   12    7   14   12
[6,]   11   10   16   NA
[7,]    8   NA   10   NA
[8,]   13   NA   NA   NA
> anova.CR2(dat)
[1] 76 37 86 66
[1] 8 6 7 5
[1] 2700.962
[1] 2995
[1] 2877.938
             SS DF        MS  F-value      p-value
主効果 176.9766  3 58.992186 11.08668 0.0001227255
誤差   117.0619 22  5.320996       NA           NA
総     294.0385 25        NA       NA           NA

参考文献

森・吉田『心理学のためのデータ解析テクニカルブック』北大路書房　p90-91