In m-clark/mixedup: Miscellaneous functions for mixed models

knitr::opts_chunk$set(
  echo      = TRUE,
  message   = TRUE,
  warning   = FALSE,
  error     = FALSE,
  collapse  = TRUE,
  comment   = NA,
  R.options = list(width = 220),
  dev.args  = list(bg = 'transparent'),
  dev       = 'png',
  fig.align = 'center',
  out.width = '75%',
  fig.asp   = .75,
  cache.rebuild = FALSE,
  cache         = FALSE
)

brms_model <- mixedup:::brms_model

Intro

This is pretty printing of results of a mixed model. Correlations of random and fixed effects are turned off by default, as the focus of interest is usually the variance components and fixed effects themselves. 95% confidence intervals are provided by default.

Examples

library(lme4)
library(glmmTMB)
library(nlme)
library(brms)
library(mgcv)


lmer_model <-
  lmer(Reaction ~ Days + (1 + Days | Subject), data = sleepstudy)

lme_model  <-
  lme(Reaction ~ Days, random = ~ 1 + Days | Subject, data = sleepstudy)

tmb_model  <-
  glmmTMB(Reaction ~ Days + (1 + Days | Subject), data = sleepstudy)

# brms_model <-
#   brm(Reaction ~ Days + (1 + Days | Subject), 
#       data = sleepstudy,
#       cores = 4,
#       refresh = -1,
#       verbose = FALSE
#   )

# this is akin to (1 | Subject) + (0 + Days | Subject) in lme4
mgcv_model <-
  gam(
    Reaction ~  Days +
      s(Subject, bs = 're') +
      s(Days, Subject, bs = 're'),
    data   = lme4::sleepstudy,
    method = 'REML'
  )

library(mixedup)

summarize_model(lmer_model)


summarize_model(lme_model)


summarize_model(tmb_model)


summarize_model(brms_model)


summarize_model(mgcv_model)

Options

summarize_model(
  lmer_model,
  ci     = FALSE,
  cor_re = TRUE,
  cor_fe = TRUE,
  digits = 3
)


summarise_model(lmer_model, ci = FALSE)

Get Convergence

Often a model estimated by lme4 needs a bit of help for convergence. At present this function only works for lme4, but then, that's mostly where you're going to have the problem. This function will start with your current model and run it long enough to the point that if you still have problems, you probably really have data or model issues, and not just an estimation problem. See this for details.

# for some reason this example apparently only works interactively, not when knit
ss2 = sleepstudy
ss2$Days = ss2$Days * 10

lmer_not_converged <- lmer(
  Reaction ~ Days + (Days|Subject),
  data = ss2
)

# lmer_not_converged

lmer_converged <- converge_it(lmer_not_converged) # final result is a converged model

lmer_converged