Generating a lagged time series cross sectional variable in R

I am a new R user. I have a time series cross sectional dataset and, a开发者_开发知识库lthough I have found ways to lag time series data in R, I have not found a way to create lagged time-series cross sectional variables so that I can use them in my analysis.

---

Here's how you could use the lag() function with zoo (and panel series data):

> library(plm)
> library(zoo)
> data("Produc")
> dnow <- pdata.frame(Produc)
> x.Date <- as.Date(paste(rownames(t(as.matrix(dnow$pcap))), "-01-01", sep=""))
> x <- zoo(t(as.matrix(dnow$pcap)), x.Date)
> x[1:3,1:3]
            ALABAMA  ARIZONA ARKANSAS
1970-01-01 15032.67 10148.42  7613.26
1971-01-01 15501.94 10560.54  7982.03
1972-01-01 15972.41 10977.53  8309.01

Lag forward by 1:

> lag(x[1:3,1:3],1)
            ALABAMA  ARIZONA ARKANSAS
1970-01-01 15501.94 10560.54  7982.03
1971-01-01 15972.41 10977.53  8309.01

Lag backward by 1:

> lag(x[1:3,1:3],k=-1)
            ALABAMA  ARIZONA ARKANSAS
1971-01-01 15032.67 10148.42  7613.26
1972-01-01 15501.94 10560.54  7982.03

As Dirk mentioned, be careful with the meaning of lag in the different time series packages. Notice how xts treats this differently:

> lag(as.xts(x[1:3,1:3]),k=1)
            ALABAMA  ARIZONA ARKANSAS
1970-01-01       NA       NA       NA
1971-01-01 15032.67 10148.42  7613.26
1972-01-01 15501.94 10560.54  7982.03

---

For cross-sectional time-series data the package plm is very useful. It has a lag function that takes into account the panel nature of the data.

library(plm)
data("Produc", package="plm")
dnow <- pdata.frame(Produc)
head(lag(dnow$pcap,1))
             ALABAMA-1970 ALABAMA-1971 ALABAMA-1972 ALABAMA-1973 ALABAMA-1974 
          NA     15032.67     15501.94     15972.41     16406.26     16762.67 

One problem with the package is that using with (or within or transform) gives you the wrong answer.

head(with(dnow, lag(pcap,1)))
15032.67 15501.94 15972.41 16406.26 16762.67 17316.26

So be careful.