Bayesian Augmentation by Chained Equations (BACE) or BACE with phylogeny (Phylo-BACE)

Bayesian Augmentation by Chained Equations (BACE) is a package that allows users to used chained equations to impute missing data for incomplete comparative datasets using phylogenetic information (or more correctly, Phylo-BACE).

It is currently able to impute missing data for continuous, categorical, and binary variables with variable imputation model structures. Phylogenetic information is included as a random effect in all the imputation models. It uses Bayesian models to estimate missing values across the data.

Phylo-BACE is currently under active development and should not be used until further testing has been done.

Installation

You can install the development version of BACE from GitHub with:

# install.packages("devtools")
devtools::install_github("daniel1noble/BACE")

Tutorial

A tutorial for using BACE can be found at: https://daniel1noble.github.io/BACE/

Future Development

• multiple random effect levels (phylogeny and species) to be used to imputing missing data, including random slopes.
• more flexibility in the random effect structure for different variables. Feed list in the ran_phylo_form and evaluate it differently. Probably easier to re-construct function as a list with length based on types. Then, if only one formula provided, all elements get the same, if a list, they are all unique. Name by response_var then add within functions and loops a call to ran_phylo. Then, turn off ginverse or evaluate so it is an identity matrix so that argument can remain consistent in MCMCglmm. We'll need to adjust the formula construction functions too because MCMCGlmm sets up random slopes differently (e.g., ~us(1 + Time):Chick - from coursenotes.) To make things more general in the random effect formula build we could always use us() functions in MCMCglmm to set up unstructured variance-covariance matrices, even for intercept only models.
• nitt, burnin, and thin arguments to be variable for different variable types. Probably provide a list or vector. Need checks to ensure these are the same length as the number of variable types.
• end model simulations to account for imputation uncertainty - sampling from the posterior
• options for parallelization of imputation of variables during each run.
• Prior B options for categorical variables - Gelman prior or flat prior. Improve mixing.
• Convergence checks across runs - need to store full predictions for each run, much like data.
• auto-restart if convergence not reached or, user can specify number of runs.

list("~1|phylo + ~1 + x1|species", "~1|species")