Robust correlation coefficients.
The pbcor function computes the percentage bend correlation coefficient, wincor the Winsorized correlation, pball the percentage bend correlation matrix, winall the Winsorized correlation matrix.
pbcor(x, y = NULL, beta = 0.2, ci = FALSE, nboot = 500, alpha = 0.05, ...) pball(x, beta = 0.2, ...) wincor(x, y = NULL, tr = 0.2, ci = FALSE, nboot = 500, alpha = 0.05, ...) winall(x, tr = 0.2, ...)
x: a numeric vector, a matrix or a data frame.y: a second numeric vector (for correlation functions).beta: bending constant.tr: amount of Winsorization.ci: whether boostrap CI should be computed or not.nboot: number of bootstrap samples for CI computation.alpha: alpha level for CI computation....: currently ignored.It tested is whether the correlation coefficient equals 0 (null hypothesis) or not. Missing values are deleted pairwise. The tests are sensitive to heteroscedasticity. The test statistic H in pball tests the hypothesis that all correlations are equal to zero.
pbcor and wincor return an object of class "pbcor" containing:
cor: robust correlation coefficient
test: value of the test statistic
p.value: p-value
n: number of effective observations
cor_ci: bootstrap confidence interval
call: function call
pball and winall return an object of class "pball" containing:
pbcorm: robust correlation matrix
p.values: p-values
H: H-statistic
H.p.value: p-value H-statistic
cov: variance-covariance matrix
Wilcox, R. (2012). Introduction to Robust Estimation and Hypothesis Testing (3rd ed.). Elsevier.
twocor
x1 <- subset(hangover, subset = (group == "control" & time == 1))$symptoms x2 <- subset(hangover, subset = (group == "control" & time == 2))$symptoms pbcor(x1, x2) pbcor(x1, x2, beta = 0.1, ci = TRUE) wincor(x1, x2) wincor(x1, x2, tr = 0.1, ci = TRUE) require(reshape) hanglong <- subset(hangover, subset = group == "control") hangwide <- cast(hanglong, id ~ time, value = "symptoms")[,-1] pball(hangwide) winall(hangwide)