CODEWITHSUNDEEP
rdplyrplyrcross-correlation
Pulsar_534511
Pulsar_534511

Reputation: 98

acf function using group_by in R

I have a dataframe with 5 variables. These data are for several years and I have grouped them season wise. I want to compute the cross correlation and lag correlation among the 5 variables for every season. How can I do this using acf function in R? I found some examples but they are giving the correlations only between two variables using the 'cor' function or 'lag' function. Since I have 5 variables, I can use acf function which can give cross and lag correlations among all the variables, but I don't know how to use it with the group_by in dplyr package. I think there must be elegant way to do this in R. The dataframe looks like:

 Season   Res1      Res2       Res3      Res4      Res5
 summer   4.4336    4.8965    31.4385   -0.6288   -1.1579
 summer   2.5130    3.7541    -2.2947   12.4083   -0.6241
      .        .         .          .         .         .                         
      .        .         .          .         .         .

For example, I can compute the correlations using acf for the whole data. If I take the 5 variables as matrix Resdf then I can do it like this:

M<-acf(Resdf,lag.max =1,type ="correlation",plot=TRUE)

This will give me the cross correlation and lag-1 correlations among the 5 variables. I can extract the cross correlations as M0<-M$acf[1,,] and lag-1 correlations as M1<-M$acf[2,,] which will give the 5x5 matrices like this: 

>M0
           [,1]       [,2]       [,3]       [,4]       [,5]
[1,]  1.0000000  0.8606853  0.0500022 -0.3440501 -0.1709415
[2,]  0.8606853  1.0000000  0.2662694 -0.5228191 -0.2376250
[3,]  0.0500022  0.2662694  1.0000000 -0.5710574 -0.2005080
[4,] -0.3440501 -0.5228191 -0.5710574  1.0000000  0.2163159
[5,] -0.1709415 -0.2376250 -0.2005080  0.2163159  1.0000000

and lag-1 correlations as

> M1
            [,1]          [,2]       [,3]       [,4]        [,5]
[1,]  0.72688806  0.7648605807  0.2416748 -0.4725366 -0.24970773
[2,]  0.66442943  0.7413684874  0.3125458 -0.4918965 -0.25046233
[3,] -0.06882386  0.0002300747  0.2523668 -0.1015463 -0.01341474
[4,] -0.13060710 -0.2369795768 -0.3061068  0.4032776  0.12751785
[5,] -0.10527689 -0.1044584694 -0.1070397  0.1025203  0.33448922

Is there any way I can use acf in this way to get the correlation matrices for the 4 seasons?

Upvotes: 0

Views: 1903

Answers (1)

CPak
CPak

Reputation: 13591

Example data following your format:

set.seed(1)
df <- data.frame(Season=c(rep("spring",3),rep("summer",3)),
                 Res1=rnorm(6))

df1 <- df %>% mutate(Res2=Res1+(rnorm(6)*0.1),
                     Res3=Res1+(rnorm(6)*0.2),
                     Res4=Res1+(rnorm(6)*0.3),
                     Res5=Res1+(rnorm(6)*0.4))

Use tidyverse nest to perform 'complex' operations on a grouped data frame. I perform acf in the first mutate...map, and then extract acf[1,,] and acf[2,,] and convert to data frame in the second mutate...map:

library(tidyverse)
df2 <- df1 %>% 
          group_by(Season) %>%
          nest() %>% 
          mutate(data = map(data, ~acf(., lag.max=1, type="correlation", plot=F))) %>%
          mutate(data = map(data, ~as.data.frame(rbind(.x$acf[1,,], .x$acf[2,,])))) %>%
          unnest(data)

The first 10 lines of output:

   Season            V1           V2           V3           V4            V5
 1 spring  1.000000e+00  0.999926654  0.888928901  0.999945732  0.9501684141
 2 spring  9.999267e-01  1.000000000  0.894411297  0.999998566  0.9463231324
 3 spring  8.889289e-01  0.894411297  1.000000000  0.893652539  0.7018425064
 4 spring  9.999457e-01  0.999998566  0.893652539  1.000000000  0.9468691987
 5 spring  9.501684e-01  0.946323132  0.701842506  0.946869199  1.0000000000
 6 spring -6.415051e-01 -0.649989355 -0.892898812 -0.648808668 -0.3899507753
 7 spring -6.360042e-01 -0.644491958 -0.888124854 -0.643310737 -0.3846451323
 8 spring -3.639938e-01 -0.371690371 -0.615653299 -0.370617371 -0.1470652339
 9 spring -6.367791e-01 -0.645266390 -0.888800271 -0.644085234 -0.3853904576
10 spring -7.499137e-01 -0.757869871 -0.969595555 -0.756763981 -0.5063447715

summer follows in the full data frame. The first 5 rows of each season contain acf[1,,] and the following 5 rows contain acf[2,,]

Upvotes: 2

Related Questions