# Introduction

UNDER CONSTRUCTION

broom.mixed is a spinoff of the broom package. The goal of broom is to bring the modeling process into a “tidy”(TM) workflow, in particular by providing standardized verbs that provide information on

• tidy: estimates, standard errors, confidence intervals, etc.
• augment: residuals, fitted values, influence measures, etc.
• glance: whole-model summaries: AIC, R-squared, etc.

broom.mixed aims to provide these methods for as many packages and model types in the R ecosystem as possible. These methods have been separated from those in the main broom package because there are issues that need to be dealt with for these models (e.g. different types of parameters: fixed, random-effects parameters, conditional modes/BLUPs of random effects, etc.) that are not especially relevant to the broader universe of models that broom deals with.

# Mixed-model-specific issues

## Terminology

• the upper-level parameters that describe the distribution of random variables (variance, covariance, precision, standard deviation, or correlation) are called random-effect parameters (ran_pars in the effects argument when tidying)
• the values that describe the deviation of the observations in a group level from the population-level effect (which could be posterior means or medians, conditional modes, or BLUPs depending on the model) are called random-effect values (ran_vals in the effects argument when tidying)
• the parameters that describe the population-level effects of (categorical and continuous) predictor variables are called fixed effects (fixed in the effects argument when tidying)
• the categorical variable (factor) that identifies which group or cluster an observation belongs to is called a grouping variable (group column in tidy() output)
• the particular level of a factor that specifies which level of the grouping variable an observation belongs to is called a group level (level column in tidy() output)
• the categorical or continuous predictor variables that control the expected value (i.e., enter into the linear predictor for some part of the model) are called terms (term column in tidy() output); note that unlike in base broom, the term column may have duplicated values, because the same term may enter multiple model components (e.g. zero-inflated and conditional models; models for more than one response; fixed effects and random effects)

## Time-consuming computations

Some kinds of computations needed for mixed model summaries are computationally expensive, e.g. likelihood profiles or parametric bootstrapping. In this case broom.mixed may offer an option for passing a pre-computed object to tidy(), eg. the profile argument in the tidy.merMod (lmer/glmer) method.

# Capabilities

package object tidy glance augment effects.fixed effects.ran_vals effects.ran_pars effects.ran_coefs confint…Ww.ald. confint..profile. confint..boot. component.zi component.disp covstruct
lme4 glmer y y y y y y y NA y NA NA NA ?
lme4 lmer y y y y y y y NA y NA NA NA ?
nlme lme y y y y y y y NA n NA NA ? ?
nlme gls y y y y NA NA NA NA n NA NA ? ?
nlme nlme y y y y y n y NA n NA NA ? ?
glmmTMB glmmTMB y y y y y y n NA NA y y ?
glmmADMB glmmadmb y y y y y y n NA NA y ? ?
brms brmsfit y y y y y y n NA NA y ? ?
rstanarm stanreg y y y y y y n NA NA NA NA ?
MCMCglmm MCMCglmm y y y y y y n NA NA ? ? ?
TMB TMB y n n n n n n NA NA NA NA ?
INLA n n n n n n n NA NA ? ? ?