r - How can I create sum of previous days but ignore the value at time t? -

i have df column t. want create columns t2 , t3

• t2 gives sum of t-2 , t-3
• t3 gives sum of t-2, t-3 , t-4 etc.

this dataframe

year      t    t2   t3 19620101    1   na  na 19630102    2   na  na 19640103    3   3   na 19650104    4   5   6 19650104    5   7   9 19650104    6   9   12 

a different version of question here: how can create sum of previous days?

basically don't want use value @ time t in accumulation..

define roll function in terms of zoo's rollapplyr , use it. note rollapplyr width argument list(-seq(k)) means use indicated offsets. example, if k = 2 equals list(c(-1, -2)) means use values 1 , 2 prior.

library(zoo)  roll <- function(t, k) rollapplyr(t, list(-seq(k)), sum, fill = na) transform(df, t2 = roll(t, 2), t3 = roll(t, 3)) 

an alternative this:

roll <- function(t, k) rollsumr(t, k+1, fill = na) - t 

giving:

      year t t2 t3 1 19620101 1 na na 2 19630102 2 na na 3 19640103 3  3 na 4 19650104 4  5  6 5 19650104 5  7  9 6 19650104 6  9 12 

note: used df:

df <- structure(list(year = c(19620101l, 19630102l, 19640103l, 19650104l,  19650104l, 19650104l), t = 1:6), .names = c("year", "t"), row.names = c(na,  -6l), class = "data.frame")