Bandwidth Selection by Pilot Estimation of Derivatives

Uses the method of Sheather & Jones (1991) to select the bandwidth of a Gaussian kernel density estimator.

width.SJ(x, nb = 1000, lower, upper, method = c("ste", "dpi"))

Arguments

x

a numeric vector

nb

number of bins to use.

upper, lower

range over which to search for solution if method = "ste".

method

Either "ste" ("solve-the-equation") or "dpi" ("direct plug-in").

Value

a bandwidth.

Note

A faster version for large n (thousands) is available in R \(\ge\) 3.4.0 as part of bw.SJ: quadruple its value for comparability with this version.

References

Sheather, S. J. and Jones, M. C. (1991) A reliable data-based bandwidth selection method for kernel density estimation. Journal of the Royal Statistical Society series B 53, 683–690.

Scott, D. W. (1992) Multivariate Density Estimation: Theory, Practice, and Visualization. Wiley.

Wand, M. P. and Jones, M. C. (1995) Kernel Smoothing. Chapman & Hall.

See also

ucv, bcv, density

Examples

width.SJ(geyser$duration, method = "dpi")
#> [1] 0.57479
width.SJ(geyser$duration)
#> [1] 0.36052

width.SJ(galaxies, method = "dpi")
#> [1] 3256.2
width.SJ(galaxies)
#> [1] 2566.4