corrects an error in family selection in the current.model function, which caused potentially erroneous results for gof = “RMSPE”.
more correctly handles datasets with missing values. Previous version potentially produced erroneous values for gof = “RMSPE”. Now for all model forms, observations with any missing values in y or xcan are deleted prior to analysis (and a warning is given that the observations have been deleted)
params, a list of parameters used in the analysis, added as a further output item added to the hier.part function
corrects a number of minor coding irregularities as required by CRAN.
expands the model families that can be used for hierarchical partitioning. All families accepted by glm are now valid, as well as “beta” (beta regression using betareg from the betareg package) and “ordinal”. This last option is not strictly a family, but is a flag to use polr (from the MASS package) for ordered logistic or probit regression.
the hier.part(), all.regs() and rand.hp() functions now accept additional arguments that can be passed to glm(), betareg() or polr(). The link argument can be used to specify the desired link for betareg() and polr().