Generating a moving sum variable in R

Question

I suspect this is a somewhat simple question with multiple solutions, but I'm still a bit of a novice in R and an exhaustive search didn't yield answers that spoke well to what I'm wanting to do.

I'm trying to create, for lack of better term, "moving sums" for a variable in my data frame. These would be 3-year and 5-year sums, lagged one year. So, a 5-year sum for an observation in 1986 would be the sum of all previous observations in 1981, 1982, 1983, 1984, and 1985. Here is an example of what I would like to do, where the sum variable is the sum of all x in the five years prior to the observation year.

country     year      x      x5yrsum
  A         1980      9        NA
  A         1981      3        NA
  A         1982      5        NA
  A         1983      6        NA
  A         1984      9        NA
  A         1985      7        32
  A         1986      9        30
  A         1987      4        36

  .....................

  B         1990      0        NA
  B         1991      4        NA
  B         1992      2        NA
  B         1993      6        NA
  B         1994      3        NA
  B         1995      7        15
  B         1996      0        22

This is unbalanced panel data. I suspect ddply would be appropriate, but I wouldn't know the exact coding for it.

Any input would be appreciated.

Answer 1

You can use filter in ddply (or any other function implementing the "split-apply-combine" approach):

library(plyr)
ddply(DF, .(country), transform, 
          x5yrsum2 = as.numeric(filter(x,c(0,rep(1,5)),sides=1)))

#    country year x x5yrsum x5yrsum2
# 1        A 1980 9      NA       NA
# 2        A 1981 3      NA       NA
# 3        A 1982 5      NA       NA
# 4        A 1983 6      NA       NA
# 5        A 1984 9      NA       NA
# 6        A 1985 7      32       32
# 7        A 1986 9      30       30
# 8        A 1987 4      36       36
# 9        B 1990 0      NA       NA
# 10       B 1991 4      NA       NA
# 11       B 1992 2      NA       NA
# 12       B 1993 6      NA       NA
# 13       B 1994 3      NA       NA
# 14       B 1995 7      15       15
# 15       B 1996 0      22       22

Answer 2

If DF is the input three-column data frame then use ave with rollapplyr from zoo. Note that we use a width of k+1 and then drop the k+1st element from the sum so that the current value of x is excluded and only the remaining k values are summed:

library(zoo)

k <- 5
roll <- function(x) rollapplyr(x, k+1, function(x) sum(x[-k-1]), fill = NA)
transform(DF, xSyrsum = ave(x, country, FUN = roll))

which gives:

   country year x xSyrsum
1        A 1980 9      NA
2        A 1981 3      NA
3        A 1982 5      NA
4        A 1983 6      NA
5        A 1984 9      NA
6        A 1985 7      32
7        A 1986 9      30
8        A 1987 4      36
9        B 1990 0      NA
10       B 1991 4      NA
11       B 1992 2      NA
12       B 1993 6      NA
13       B 1994 3      NA
14       B 1995 7      15
15       B 1996 0      22

Answer 3

you can also use filter of standard packages (stats) to do moving sum:

ms=function(x,n=5) as.numeric(stats::filter(x,rep(1, n),method="convolution",sides=1))
x=c(1,2,3,4,5,6,7,8,9)
ms(x,5)
NA NA NA NA 15 20 25 30 35

To do a 1-lag, insert NA at the begining and take the number of elements (or lines):

ms.1lag=c(NA,ms(x,5))[1:length(x)]
cbind(x,ms.1lag)
x ms.1lag
[1,] 1      NA
[2,] 2      NA
[3,] 3      NA
[4,] 4      NA
[5,] 5      NA
[6,] 6      15
[7,] 7      20
[8,] 8      25
[9,] 9      30

If you use this frequently,

ms=function(x,n=5,lag=0)
  c(rep(NA,lag),
    as.numeric(stats::filter(x,rep(1, n),method="convolution",sides=1)))[1:length(x)]
cbind(x,ms5.1=ms(x,5,1))
      x ms5.1
 [1,] 1    NA
 [2,] 2    NA
 [3,] 3    NA
 [4,] 4    NA
 [5,] 5    NA
 [6,] 6    15
 [7,] 7    20
 [8,] 8    25
 [9,] 9    30