The detection of discontinuities in a regression curve or surface.
This function uses a comparison of left and right handed nonparametric regression curves to assess the evidence for the presence of one or more discontinuities in a regression curve or surface. A hypothesis test is carried out, under the assumption that the errors in the data are approximately normally distributed. A graphical indication of the locations where the evidence for a discontinuity is strongest is also available.
sm.discontinuity(x, y, h, hd, ...)
x |
a vector or two-column matrix of covariate values. |
y |
a vector of responses observed at the covariate locations. |
h |
a smoothing parameter to be used in the construction of the nonparametric
regression estimates. A normal kernel
function is used and |
hd |
a smoothing parameter to be used in smoothing the differences of the
left and right sided nonparametric regression estimates. A normal kernel
function is used and |
... |
other optional parameters are passed to the |
The reference below describes the statistical methods used in the function. There are minor differences in somecomputational details of the implementation.
Currently duplicated rows of x
cause a difficulty in the two covariate case. Duplicated rows should be removed.
a list containing the following items
p |
the p-value for the test of the null hypothesis that no discontinuities are present. |
sigma |
the estimated standard deviation of the errors. |
eval.points |
the evaluation points of the nonparametric
regression estimates. When |
st.diff |
a vector or matrix of standardised differences between the left and right sided estimators at the evaluation points. |
diffmat |
when |
angle |
when |
h |
the principal smoothing parameter. |
hd |
the smoothing parameter used for double-smoothing (see the reference below). |
a plot on the current graphical device is produced, unless the option
display="none"
is set.
Bowman, A.W., Pope, A. and Ismail, B. (2006). Detecting discontinuities in nonparametric regression curves and surfaces. Statistics \& Computing, 16, 377–390.
par(mfrow = c(3, 2)) with(nile, { sm.discontinuity(Year, Volume, hd = 0) sm.discontinuity(Year, Volume) ind <- (Year > 1898) plot(Year, Volume) h <- h.select(Year, Volume) sm.regression(Year[!ind], Volume[!ind], h, add = TRUE) sm.regression(Year[ ind], Volume[ ind], h, add = TRUE) hvec <- 1:15 p <- numeric(0) for (h in hvec) { result <- sm.discontinuity(Year, Volume, h, display = "none", verbose = 0) p <- c(p, result$p) } plot(hvec, p, type = "l", ylim = c(0, max(p)), xlab = "h") lines(range(hvec), c(0.05, 0.05), lty = 2) }) with(trawl, { Position <- cbind(Longitude, Latitude) ind <- (Longitude < 143.8) # Remove a repeated point which causes difficulty with sm.discontinuity ind[54] <- FALSE sm.regression(Position[ind,], Score1[ind], theta = 35, phi = 30) sm.discontinuity(Position[ind,], Score1[ind], col = "blue") }) par(mfrow = c(1, 1)) # The following example takes longer to run. # Alternative values for nside are 32 and 64. # Alternative values of yjump are 1 and 0.5. # nside <- 16 # yjump <- 2 # x1 <- seq(0, 1, length = nside) # x2 <- seq(0, 1, length = nside) # x <- expand.grid(x1, x2) # x <- cbind(x1 = x[, 1], x2 = x[, 2]) # y <- rnorm(nside * nside) # ind <- (sqrt((x[, 1] - 0.5)^2 + (x[, 2] - 0.5)^2) <= 0.25) # y[ind] <- y[ind] + yjump # image(x1, x2, matrix(y, ncol = nside)) # sm.discontinuity(x, y, df = 20, add = TRUE)
Please choose more modern alternatives, such as Google Chrome or Mozilla Firefox.