The statistical methods, theory, and algorithms for regression with
Abundance-Based Constraints (ABCs) are under active development. As
such, the current implementation of lmabc is not
exhaustive. This vignette will serve as a running list of limitations of
lmabc. Please email the package maintainer with questions
or suggestions.
lmabc throws an error for missing factor levels
Suppose we have two categorical variables, k1 and
k2, with levels a, b, and
c and uu, vv, respectively. If
either k1 or k2 has zero observations in at
least one level, then a model of the form y ~ k1 + k2 + ...
will throw an error. Similarly, if an interaction is included,
y ~ k1 + k2 + k1:k2 + ..., but a joint category (e.g.,
k1 = a and k2 = uu) has zero observations,
then lmabc will throw an error.
This behavior is different from lm, which instead
removes each unobserved level from the model and returns an
NA for its coefficient.
Missing implementations of generic methods
We implemented many of the common generic methods associated with
linear regression models. However, our implementation is not exhaustive,
especially for non-base packages. lmabc objects only have
class “lmabc”, so methods only implemented for class “lm” will fail
unless the generic method has a default.
We will implement additional generic methods throughout the development process.