Between-Within Correlations-CodePudding

DATA = data.frame("GROUP" = sort(rep(1:4, 200)),
                  "TYPE" = rep(1:2, 400),
                  "TIME" = rep(100:101, 400),
                  "SCORE" = sample(1:100,r=T,800))

Cheers all,

I have 'DATA' and wish to estimation the CORRELATION VALUES of SCORE at each TIME and SCORE and TYPE combination BETWEEN AND WITHIN GROUP in this way:

CodePudding user response：

I am assuming you want to compute the correlation between groups 1-2, 1-3, 1-4 and so on for each combination of TIME and TYPE. Here's an approach:

# create the dataset
set.seed(123)
df <- data.frame("group" = sort(rep(1:4, 200)),
                 "type" = rep(1:2, 400),
                 "time" = rep(100:101, 400),
                 "score" = sample(1:100,r=T,800))
library(tidyr)
library(purrr)
library(data.table)

# another dataset to filter combinations 
# (G1G2 is same G2G1, so remove G2G1)
df2 <- combn(4, 2) %>% t %>% 
    as_tibble() %>% 
    rename(group1 = V1, group2 = V2) %>% 
    mutate(value = TRUE)

df %>% 
    
    # add identifiers per group
    group_by(time, type, group) %>% 
    mutate(id = row_number()) %>% 
    ungroup() %>% 
    
    # nest data to get separate tibble for each
    # combination of time and type
    nest(data = -c(time, type)) %>% 
    
    # convert each data.frame to data.table
    mutate(dt = map(data, function(dt){
        setDT(dt)
        setkey(dt, id)
        dt
    })) %>% 
    
    # correlation between groups in R
    # refer answer below for more details
    # https://stackoverflow.com/a/26357667/15221658
    
    # cartesian join of dts
    mutate(dtj = map(dt, ~.[., allow.cartesian = TRUE])) %>% 
    
    # compute between group correlation
    mutate(cors = map(dtj, ~.[, list(cors = cor(score, i.score)), by = list(group, i.group)])) %>% 
    
    # unnest correlation object
    unnest(cors) %>% 
    
    # formatting for display
    select(type, time, group1 = group, group2 = i.group, correlation = cors) %>% 
    filter(group1 != group2) %>%
    arrange(time, group1, group2) %>% 

    # now use df2 since currently we have G1G2, and G2G1
    # which are both equal so remove G2G1
    left_join(df2, by = c("group1", "group2")) %>% 
    filter(value) %>% 
    select(-value)

# A tibble: 12 x 5
    type  time group1 group2 correlation
   <int> <int>  <int>  <int>       <dbl>
 1     1   100      1      2     0.121  
 2     1   100      1      3     0.0543 
 3     1   100      1      4    -0.0694 
 4     1   100      2      3    -0.104  
 5     1   100      2      4    -0.0479 
 6     1   100      3      4    -0.0365 
 7     2   101      1      2    -0.181  
 8     2   101      1      3    -0.0673 
 9     2   101      1      4     0.00765
10     2   101      2      3     0.0904 
11     2   101      2      4    -0.0126 
12     2   101      3      4    -0.154

CodePudding user response：

Here is an alternative approach:

library(data.table)
setDT(DATA, key = c("GROUP", "TYPE", "TIME"))[
  , CJ(TY = TYPE, TI = TIME, GA = GROUP, GB = GROUP, unique = TRUE)][GA < GB][
    , CORRELATIONSTYPE := paste0("G", GA, "G", GB)][
      , CORRELATION := cor(DATA[.(GA, TY, TI), SCORE], DATA[.(GB, TY, TI), SCORE]), by = .I][]

Key: <TY, TI, GA, GB>
       TY    TI    GA    GB CORRELATIONSTYPE CORRELATION
    <int> <int> <int> <int>           <char>       <num>
 1:     1   100     1     2             G1G2  0.12240303
 2:     1   100     1     3             G1G3 -0.03969913
 3:     1   100     1     4             G1G4 -0.09666560
 4:     1   100     2     3             G2G3 -0.02250806
 5:     1   100     2     4             G2G4 -0.07106922
 6:     1   100     3     4             G3G4 -0.13140392
 7:     1   101     1     2             G1G2          NA
 8:     1   101     1     3             G1G3          NA
 9:     1   101     1     4             G1G4          NA
10:     1   101     2     3             G2G3          NA
11:     1   101     2     4             G2G4          NA
12:     1   101     3     4             G3G4          NA
13:     2   100     1     2             G1G2          NA
14:     2   100     1     3             G1G3          NA
15:     2   100     1     4             G1G4          NA
16:     2   100     2     3             G2G3          NA
17:     2   100     2     4             G2G4          NA
18:     2   100     3     4             G3G4          NA
19:     2   101     1     2             G1G2 -0.03600460
20:     2   101     1     3             G1G3  0.03328082
21:     2   101     1     4             G1G4 -0.09641611
22:     2   101     2     3             G2G3 -0.06334729
23:     2   101     2     4             G2G4  0.06455382
24:     2   101     3     4             G3G4 -0.07121206
       TY    TI    GA    GB CORRELATIONSTYPE CORRELATION