Reshaping repeated measures data in R wide to long

2019-06-10 01:56发布

问题:

I need to convert a "wide" dataframe of annually repeated measures on individuals into "long" format so that I can model it like lm(y_year2 ~ x_year1) as well as lm(z_year2 ~ y_year2)

I can get it into the format I want "by hand" but cannot get figure out how to melt/dcast it into the shape I want

Below I've illustrated what I'm doing with some simulated data

The dataframe is like this in wide format, one individual per line

ID  SITE    L_03  M_03  R_03  L_04  M_04  R_04  L_05  M_05  R_05
1   forest    X     a   YES     Y     b   YES     Z     c   NO
2   forest    ... 

I'd like it in LONG format:

ID  SITE    L_year1  L_year2  M_year1  M_year2  R_year1   R_year2   year1  year2
1   forest      Z       Y       a         b       YES       YES       03    04
1   forest      Y       Z       b         c       YES       NO        04    05
2   forest      ...  
2   forest      ...

Some Simulated data: L and M are numeric (length & mass), R is a Yes/No factor (reproductive), 3 years of repeated measurements (2003-2005)

    ID <- 1:10; SITE <- c(rep("forest",3), rep("swamp",3), rep("field",4))
    L_03 <- round(rnorm(10, 100, 1),3) ; M_03 <- round((10 + L_03*0.25 + rnorm(10, 0, 1)), 3)
    R_03 <- sample(c("Yes", "No"), 10, replace = TRUE) ; L_04 <- round((2 + L_03*1.25 + rnorm(10, 1,10)), 3) 
    M_04 <- round((10 + L_04*0.25 + rnorm(10, 0,10)), 3) ;R_04 <- sample(c("Yes", "No"), 10, replace = TRUE)
    L_05 <- round((2 + L_04*1.25 + rnorm(10, 1,10)),3) ; M_05 <- round((10 + L_05*0.25 + abs(rnorm(10, 0,10))),3)
    R_05 <- sample(c("Yes", "No"), 10, replace = TRUE); rm_data <- data.frame(ID, SITE, L_03, M_03, R_03, L_04, M_04,R_04, L_05, M_05, R_05)

Approach 1: My ad hoc reshaping "by hand" with rbind 1st, make subset with 2003 & 2004 data, then another w/ 2004 & 2005

rm_data1 <- cbind(rm_data[ ,c(1,2,3:5, 6:8)], rep(2003,10), rep(2004,10))
rm_data2 <- cbind(rm_data[ ,c(1,2,6:8, 9:11)],rep(2004,10), rep(2005,10))
names(rm_data1)[3:10]<- c("L1", "M1", "R1", "L2", "M2", "R2", "yr1", "yr2")
names(rm_data2)[3:10]<- c("L1", "M1", "R1", "L2", "M2", "R2", "yr1", "yr2")
data3 <- rbind(rm_data1, rm_data2)

Approach 2?: I'd like to do this with reshape/melt/dcast. I can't figure out if I can use dcast directly on the wide dataframe or, once I melt it, how to dcast it into the format I want.

library(reshape2)
rm_measure_vars <- c("L_03", "M_03", "R_03", "L_04", "M_04","R_04", "L_05", "M_05", "R_05")
rm_data_melt <-  melt(data = rm_data, id.vars = c("ID", "SITE"), measure.vars = rm_measure_vars, value.name = "data")

I add a designator of the year the measurement was taken to the melted data

obs_year <- gsub("(.*)([0-9]{2})", "\\2", rm_data_melt$variable)
rm_data_melt <- cbind(rm_data_melt, obs_year)

The dcast seems like it should be something like this, but this is not yet what I need

dcast(data = rm_data_melt, formula = ID + SITE + obs_year ~ variable)
   ID   SITE obs_year    L_03   M_03 R_03    L_04   M_04 R_04    L_05   M_05 R_05
1   1 forest       03   99.96 35.364   No    <NA>   <NA> <NA>    <NA>   <NA> <NA>
2   1 forest       04    <NA>   <NA> <NA> 129.595 47.256  Yes    <NA>   <NA> <NA>
3   1 forest       05    <NA>   <NA> <NA>    <NA>   <NA> <NA> 177.607 58.204  Yes

Any suggestions would be greatly appreciated

回答1:

I gave it some try. The reshape is the easy part. The rest needs some semi-manual handling, I believe. The following should give you what you want.

output <- reshape(rm_data, idvar=c("ID","SITE"), varying=3:11, 
                v.names=c("L_","M_","R_"), direction="long")
output$time <- output$time + 2    # to get the year
names(output)[3:6] <- c("year1", "L_year1", "M_year1", "R_year1")
output$year2 <- output$year1+1
rownames(output) <- c()

sapply(output[,4:6], function(x) {
  i <- ncol(output)+1
  output[,i] <<- x[c(2:length(x), NA)]
  names(output)[i] <<- sub("1","2",names(output)[i-4])
})

output <- output[,c(1,2,4,8,5,9,6,10,3,7)]    # rearrange columns as necessary

Hope this helps!



标签: r reshape