JWileymisc (version 0.3.1)

acfByID: Estimate the effective sample size from longitudinal data

Description

This function estimates the (approximate) effective sample size.

Usage

acfByID(xvar, timevar, idvar, data, lag.max = 10L,
  na.function = c("na.approx", "na.spline", "na.locf"), ...)

Arguments

xvar

A character string giving the variable name of the variable to calculate autocorrelations on.

timevar

A character string giving the variable name of the time variable.

idvar

A character string giving the variable name of the ID variable. Can be missing if only one time series provided, in which case one will be created.

data

A data.table containing the variables used in the formula. This is a required argument. If a data.frame, it will silently coerce to a data.table. If not a data.table or data.frame, it will attempt to coerce, with a message.

lag.max

An integer of the maximum lag to estimate. Must be equal to or greater than the number of observations for all IDs in the dataset.

na.function

A character string giving the name of the function to use to address any missing data. Functions come from the zoo package, and must be one of: “na.approx”, “na.spline”, “na.locf”.

...

Additional arguments passed to zoo.

Value

A data.table of the estimated autocorrelations by ID and lag

References

For details, see Campbell, M. K., Mollison, J., & Grimshaw, J. M. (2001). Cluster trials in implementation research: estimation of intracluster correlation coefficients and sample size. Statistics in Medicine, 20(3), 391-399.

Examples

# NOT RUN {
## example 1
dat <- data.table(
  x = sin(1:30),
  time = 1:30,
  id = 1)
acfByID("x", "time", "id", data = dat)

## example 2
dat2 <- data.table(
  x = c(sin(1:30), sin((1:30)/10)),
  time = c(1:30, 1:30),
  id = rep(1:2, each = 30))
dat2$x[4] <- NA

res <- acfByID("x", "time", "id", data = dat2, na.function = "na.approx")

ggplot(res, aes(factor(Lag), AutoCorrelation)) +
  geom_boxplot()

## clean up
rm(dat, dat2, res)
# }