Principled methods for the imputation of zeros, left-censored and missing data in
compositional data sets (Palarea-Albaladejo and Martin-Fernandez (2015)
Imputation of Zeros, Nondetects and Missing Data in Compositional Data Sets
Alternative imputation by multRepl in lrEM and lrDA of censoring patterns with only one observed component.
Fixed problem that caused error when a censoring pattern consisted of a single sample.
Minor bugs in error messages fixed.
cmultRepl: label argument added (default label = 0).
cmultRepl function: bayesian-multiplicative replacement of compositional count zeros included.
lrDA function: data augmentation algorithm, including multiple imputation, to replace left-censored values.
zPatterns function: summarises the patterns of unobserved values (censored, nondetects, ...) in a data set and generates a vector of labels.
General revision and optimisation. Documentation improvements.
multLN: parameter estimates based on the normal distribution on the positive real line. Random imputation based on truncated normal instead of rejection method.
Re-scaling to preserve ratios that leaves absolute observed values unaltered when working with non-closed data now implemented for all the methods (continuous data).
The replacement methods include an argument 'label' allowing the user to enter the label (special character, number, ...) denoting unobserved value in a data set.
Error handling introduced.
New lrEM function: replaces previous alrEM function and implements both ordinary and robust EM-based imputation methods. New arguments allow to specify the method to obtain initial estimates and the convergence criterion.