Moving average of previous three values in R

In the zoo package there is a function called rollmean, which enables you to make moving averages. The rollmean(x,3) will take the previous, current and next value (ie 4, 6 and 2) in the table below. This is shown in the second column.

x   rollmean    ma3
4       
6   4.0 
2   4.3 
5   3.0         4.0
2   6.3         4.3
12  6.0         3.0
4   6.0         6.3
2               6.0

I would like to get the same job done, but by averaging out the previous 3 values in the fourth row. This is displayed in the third column. Can anybody tell me the name of the function that will help to accomplish this?


You can use rollmean, but set align='right'. Or you could use rollmeanr, which has align='right' as the default.

ma3 <- rollmeanr(x[,1],3,fill=NA)

...but you would still need to lag the result. Another solution is to use rollapply with a list for the width argument:

ma3 <- rollapplyr(x[,1],list(-(3:1)),mean,fill=NA)

I struggled searching for a simple function for moving averages that had some flexibility to do what I needed. I finally wrote a couple functions extending the one based on the filter function which rinni gives above in the comment (but which itself won't work because it will include the current observation in the 3 period average).

  1. Moving average function that includes the current observation

    mav <- function(x,n){filter(x,rep(1/n,n), sides=1)} 
    
  2. Moving average function that does not include the current observation

    mavback <- function(x,n){
      a<-mav(x,1)
      b<-mav(x,(n+1))
      c<-(1/n)*((n+1)*b - a)
      return(c)
    }
    
  3. Backward looking moving average function, not including current obs, based on [h2] readings starting [h1] periods back

    mavback1<-function(x,h1,h2){
      a<-mavback(x,h1)
      b<-mavback(x,h1-h2)
      c<-(1/h2)*(h1*a -(h1-h2)*b)
      return(c)
    }