元分析的数据准备.使用元数据

Question

为了计算影响大小并对连续结果(d或g)的二分预测器进行元分析，需要为每项研究提供一个由平均值、sd和样本大小组成的数据。

我试图编写一些代码，这些代码将根据原始数据创建所需的数据。这将意味着每项研究都不必手工完成这一过程。

示例原始数据集

Study <- c("andrew", "andrew", "andrew", "andrew", "peters", "peters", "peters", "jess", "jess", "jess")
Score = c(100, 308, 584, 241, 241, 111, 431, 123, 321, 411)
Sex = c(1, 1, 1, 2, 2, 1, 2, 2, 1, 1)
data = cbind(Score, Sex, Study)
data

 >     Score Sex Study   
 > [1,] "100" "1" "andrew"
 > [2,] "308" "1" "andrew"
 > [3,] "584" "1" "andrew"
 > [4,] "241" "2" "andrew"
 > [5,] "241" "2" "peters"
 > [6,] "111" "1" "peters"
 > [7,] "431" "2" "peters"
 > [8,] "123" "2" "jess"  
 > [9,] "321" "1" "jess"  
> [10,] "411" "1" "jess" 

我怎样才能把它变成下面的元数据文件，用于按性别和学习来划分数据？

Study       MeanMale   MeanFemale   SDMale    SDfemale    NrowsMale    NrowsFemale
andrew         X           X          X          X            X             X
peters         X           X          X          X            X             X
jess           X           X          X          X            X             X

我可以想象，在Splitdata中使用describeBy、statsBy或Splitdata是可行的，但将其转换成所需的格式是很麻烦的。下一个目标将是引入一个年度专栏，例如，

Study <- c("andrew", "andrew", "andrew", "andrew", "peters", "peters", "peters", "jess", "jess", "jess") 
Score = c(100, 308, 584, 241, 241, 111, 431, 123, 321, 411)
Sex = c(1, 1, 1, 2, 2, 1, 2, 2, 1, 1) 
Year = (1992, 1992, 1992, 1992, 1988, 1988, 1988, 1977, 1977, 1977) 
data = cbind(Study, Year, Score, Sex) 

若要生成以下data.frame

Study      Year  MeanMale   MeanFemale   SDMale    SDfemale    NrowsMale    NrowsFemale
andrew     1992    X           X          X          X            X             X
peters     1988    X           X          X          X            X             X
jess       1977    X           X          X          X            X             X

Answer 1

我们可以使用data.table的开发版本，即v1.9.5。安装开发版本的说明是here。

我们将“data.frame”转换为“data.table”(setDT(data))，按“性别”和“学习”进行分组，得到mean、sd和.N (nrows)，并使用dcast (可以使用多个value.var列)重新塑造“long”到“wide”格式。

library(data.table)#v1.9.5+
dcast(setDT(data)[, list(Mean= mean(Score), SD= sd(Score), Nrows=.N), 
.(Sex, Study)], Study~ c('Male', 'Female')[Sex], 
          value.var=c('Mean', 'SD', 'Nrows'))
#     Study Female_Mean Male_Mean Female_SD   Male_SD Female_Nrows Male_Nrows
#1: andrew         241  330.6667        NA 242.79484            1          3
#2:   jess         123  366.0000        NA  63.63961            1          2
#3: peters         336  111.0000  134.3503        NA            2          1

编辑

在@Arun的注释中，来自data.table的data.table也接受多个函数。

dcast(setDT(data), Study ~ c('Male', 'Female')[Sex],
        fun.agg=list(mean, sd, length), value.var="Score")
#    Study Female_mean_Score Male_mean_Score Female_sd_Score Male_sd_Score
#1: andrew               241        330.6667              NA     242.79484
#2:   jess               123        366.0000              NA      63.63961
#3: peters               336        111.0000        134.3503            NA
#   Female_length_Score Male_length_Score
#1:                   1                 3
#2:                   1                 2
#3:                   2                 1

或者我们可以使用reshape从base R获得mean，sd，nrow使用aggregate。

d1 <- do.call(data.frame,aggregate(Score~., transform(data, Sex=c('Male',
 'Female')[Sex]), FUN=function(x) c(Mean=mean(x), SD=sd(x), Nrows=length(x))))

reshape(d1, idvar='Study', timevar='Sex', direction='wide')
#  Study Score.Mean.Female Score.SD.Female Score.Nrows.Female Score.Mean.Male
#1 andrew               241              NA                  1        330.6667
#3   jess               123              NA                  1        366.0000
#5 peters               336        134.3503                  2        111.0000
#  Score.SD.Male Score.Nrows.Male
#1     242.79484                3
#3      63.63961                2
#5            NA                1

数据

data <- data.frame(Score, Sex, Study)

Answer 2

这对于dplyr和reshape2来说是非常直接的。我们将性别转换为一个命名的因素，使用变异按组获得SD和样本大小，然后熔化和转换数据，按组获得具有良好的变量名称的方法：

require(reshape2); require(dplyr)

data$Sex <- factor(data$Sex, levels = c(1, 2), labels = c('Male', 'Female'))
data <- mutate(group_by(data, Study), SD = sd(Score), Nrow = length(Score))
data <- melt(data, id.vars = c('Study', 'Sex'))
data$value <- as.numeric(data$value)
dcast(data, Study ~ variable + Sex, mean, na.rm = TRUE)