rtn function

Truncated Normal Distribution RNG

Sample from Truncated Normal distributions

rtn(.mean = rep(0, 1), .sd = rep(1, length(.mean)), .low = rep(-Inf,
  length(.mean)), .high = rep(Inf, length(.mean)), .checks = TRUE)

Arguments

• .mean: Length K vector with the means of the K Normal distributions prior to truncation
• .sd: Length K vector with the standard deviations of the K Normal distributions prior to truncation
• .low: Length K vector with the lower truncation bound of the K Normal distributions prior to truncation
• .high: Length K vector with the upper truncation bound of the K Normal distributions prior to truncation
• .checks: Length 1 logical vector indicating whether to perform checks (safer) or not (faster) on the input parameters

Returns

A length K vector of expectations corresponding to the Truncated Normal distributions. NAs are returned (with a warning) for invalid parameter values.

Details

The special values of -Inf and Inf are valid values in the .low and .high arguments, respectively. The implementation is from Robert (1995). The computation is written in Rcpp-based C++ code, but respects R's RNG state. The draws from this function are reproducible because it respects R's RNG state. Draws using this algorithm (whether implemented in R code or C++) will be the same if seeded correctly. However, you should not expect these draws to match those from another algorithm.

Examples

set.seed(1)
rtn(0, 1, -Inf, Inf) # single draw from a single distribution

## [1] -0.6264538

set.seed(1)
rtn(0, 1, -Inf, Inf) # again, because it respects the RNG state

## [1] -0.6264538

rtn(rep(0, 3),
    rep(1, 3),
    rep(-Inf, 3),
    rep(Inf, 3)
    ) # multiple draws from a single distribution

## [1]  0.1836433 -0.8356286  1.5952808

rtn(c(0, 0),
    c(1, 1),
    c(-Inf, 5),
    c(1, Inf)
    ) # multiple draws, each from a different distribution
## [1] 0.3295078 5.3917301

References

Robert, Christian P. ``Simulation of truncated normal variables''. Statistics and Computing 5.2 (1995): 121-125. http://dx.doi.org/10.1007/BF00143942

Author(s)

Jonathan Olmsted