问题
I want to analyse the default data set in R (mtcars data set). I am interested in creating column of correlation coefficients according to the below rule. Correlation coefficient of only first three observations ((i.e., row 1,2,3)) between "mpg" and "wt", then leaving the first row, calculate again correlation coefficient between next three observations (i.e., row 2,3,4) between mpg and wt then leaving the first two rows, calculate again correlation coefficient between next three observations (i.e., row 3,4,5) between mpg and wt and so on till end. For example
cor(mtcars$mpg[c(1,2,3)],mtcars$wt[c(1,2,3)])
cor(mtcars$mpg[c(2,3,4)],mtcars$wt[c(2,3,4)])
cor(mtcars$mpg[c(3,4,5)],mtcars$wt[c(3,4,5)]);
and so on. Can anyone help to how to automate this R code using loop etc.
Example, see how i need output, i have done it in excel but i need to do it in R.
回答1:
The value of cor(mtcars$mpg[c(1,2,3)],mtcars$wt[c(1,2,3)]) is -0.8884586; however, the first value in the Correlation column of the output image in the question is not that so there is some error in the image shown relative to the description of what is wanted. We will assume that the description is correct and the sample output is not.
Try a rolling apply, rollapply. It applies the function cor2 to a rolling window of width 3. align = "left" means it uses the current row and the next 2 rows so that the NA values appear at the end as in the image in the question. fill = NA causes it to generate NA values for the last 2 elements since there are not 3 more elements for those.
library(zoo)
mtcars2 <- mtcars[c("mpg", "wt")]
cor2 <- function(x) cor(x[, 1], x[, 2])
transform(mtcars2, cor = rollapply(mtcars2, 3, cor2, by.column = FALSE,
align = "left", fill = NA))
giving:
mpg wt cor
Mazda RX4 21.0 2.620 -0.88845855
Mazda RX4 Wag 21.0 2.875 -0.82589964
Datsun 710 22.8 2.320 -0.87097656
Hornet 4 Drive 21.4 3.215 -0.99520846
Hornet Sportabout 18.7 3.440 -0.99985063
Valiant 18.1 3.460 -0.99534538
Duster 360 14.3 3.570 -0.97267882
Merc 240D 24.4 3.190 -0.90784130
Merc 230 22.8 3.150 -0.96247218
Merc 280 19.2 3.440 -0.86602540
Merc 280C 17.8 3.440 -0.99308187
Merc 450SE 16.4 4.070 -0.05428913
Merc 450SL 17.3 3.730 -0.96311366
Merc 450SLC 15.2 3.780 -0.99534934
Cadillac Fleetwood 10.4 5.250 0.05301502
Lincoln Continental 10.4 5.424 -0.98658763
Chrysler Imperial 14.7 5.345 -0.96899291
Fiat 128 32.4 2.200 0.44730718
Honda Civic 30.4 1.615 -0.86317499
Toyota Corolla 33.9 1.835 -0.94182141
Toyota Corona 21.5 2.465 -0.99341821
Dodge Challenger 15.5 3.520 -0.94720046
AMC Javelin 15.2 3.435 0.21168794
Camaro Z28 13.3 3.840 -0.90670560
Pontiac Firebird 19.2 3.845 -0.99864434
Fiat X1-9 27.3 1.935 -0.99939736
Porsche 914-2 26.0 2.140 -0.99630829
Lotus Europa 30.4 1.513 -0.99962223
Ford Pantera L 15.8 3.170 -0.93453339
Ferrari Dino 19.7 2.770 -0.96372018
Maserati Bora 15.0 3.570 NA
Volvo 142E 21.4 2.780 NA
Also see this SO post which is similar except in a data.table context: Rolling correlation with data.table
回答2:
It's not clear to me why you want to calculate what looks to me like a rolling correlation within a 3 row/observation window, but you could do something like this in base R:
x <- lapply(seq(1, nrow(mtcars) - 2), function(x) seq(x, x + 2))
Here x is a list containing as entries the rows/observations based on which we calculate the correlation.
df <- do.call(rbind, lapply(x, function(x) cor(mtcars$mpg[x], mtcars$wt[x])))
df;
# [,1]
#[1,] -0.88845855
#[2,] -0.82589964
#[3,] -0.87097656
#[4,] -0.99520846
#[5,] -0.99985063
#[6,] -0.99534538
#[7,] -0.97267882
#[8,] -0.90784130
#[9,] -0.96247218
#[10,] -0.86602540
#[11,] -0.99308187
#[12,] -0.05428913
#[13,] -0.96311366
#[14,] -0.99534934
#[15,] 0.05301502
#[16,] -0.98658763
#[17,] -0.96899291
#[18,] 0.44730718
#[19,] -0.86317499
#[20,] -0.94182141
#[21,] -0.99341821
#[22,] -0.94720046
#[23,] 0.21168794
#[24,] -0.90670560
#[25,] -0.99864434
#[26,] -0.99939736
#[27,] -0.99630829
#[28,] -0.99962223
#[29,] -0.93453339
#[30,] -0.96372018
来源：https://stackoverflow.com/questions/49587940/series-of-correlation-coefficient-calculation