This function tests for homogeneity across groups of means and variances of
user-defined log-contrasts. Groups can be defined by either zero/unobserved data patterns or by a grouping
factor in fully observed zero-free data sets.
Usage
lcTest(X, label = 0, groups = NULL, lc = NULL, method = c("parametric",
"nonparametric"), b = 1000)
Unique label (numeric or character) used to denote zero or unobserved data in X (label = 0, default).
groups
Grouping factor in fully observed zero-free data sets (groups = NULL, default).
lc
User-defined log-contrast (see details below).
method
Approach used for mean and variance homogeneity testing (method = "parametric", default).
b
Number of bootstrap resamples used by permutation test (b = 1000, default).
Value
Test p-values for log-contrast means and variances.
Details
Homogeneity of log-contrast means and variances across groups is tested using either parametric or non-parametric tests. When
method = "parametric", ordinary analysis of variance and Bartlett's tests are used. Alternatively,
Kruskal-Wallis and Fligner-Killen tests are used instead when method = "nonparametric". The results of a permutation test of homogeneity of variation
arrays based on total weighted squared relative errors are also provided (see zVarArrayTest for more details).
The log-contrast is specified by the lc argument using a vector of codes 1, -1 and 0 for components
in the numerator, denominator and omitted respectively.
References
Palarea-Albaladejo J. and Martin-Fernandez JA. zCompositions -- R package for multivariate imputation of
left-censored data under a compositional approach. Chemometrics and Intelligence Laboratory Systems 2015; 143: 85-96.
# NOT RUN {data(Water)
zPatterns(Water, label = 0)
# Test of homogeneity in log-contrast Potassium/Arsenic*CalciumlcTest(Water, label = 0, lc = c(1,-1,-1,0))
# }