lme4 (R)

Quick start

1. •Confirm install/version: packageVersion("lme4")
2. •Load: library(lme4)
3. •Find entrypoints: see references/help-index.txt and references/exports.txt

Common workflows

Fit a linear mixed model (lmer)

library(lme4)

m <- lmer(reaction ~ days + (1 + days | subject), data = sleepstudy)
summary(m)

Use REML = FALSE when comparing fixed effects via likelihood-ratio tests:

m0 <- lmer(reaction ~ 1 + (1 + days | subject), data = sleepstudy, REML = FALSE)
m1 <- lmer(reaction ~ days + (1 + days | subject), data = sleepstudy, REML = FALSE)
anova(m0, m1)

Fit a generalized linear mixed model (glmer)

library(lme4)

m <- glmer(
  cbind(incidence, size - incidence) ~ period + (1 | herd),
  data = cbpp,
  family = binomial
)
summary(m)

Specify common random-effects structures

• •Random intercept: (1 | group)
• •Random intercept + slope: (1 + x | group)
• •Remove intercept–slope correlation: (1 + x || group)
• •Nested: (1 | g1 / g2) (equivalent to (1 | g1) + (1 | g1:g2))
• •Crossed: (1 | subject) + (1 | item)

Extract effects, variance components, and model matrices

fixef(m)          # fixed effects
ranef(m)          # random effects (BLUPs)
coef(m)           # per-group conditional coefficients
VarCorr(m)        # variance/covariance of random effects
sigma(m)          # residual SD (LMMs)

getME(m, "X")     # fixed-effects model matrix
getME(m, "Z")     # random-effects model matrix
getME(m, "theta") # variance parameters (internal scale)

Predict with/without random effects

predict(m)                         # include fitted random effects
predict(m, re.form = NA)           # population-level (fixed effects only)
predict(m, newdata = nd, re.form = NA)
predict(m, newdata = nd, allow.new.levels = TRUE) # new groups in nd

Uncertainty and model comparison

• •Prefer confidence intervals via confint(m, method = "profile") or bootstrapping (bootMer) over “Wald SEs only” when it matters.
• •lme4 does not provide default p-values for lmer(); for tests consider likelihood-ratio tests (anova() with REML=FALSE), parametric bootstrap, or companion packages like lmerTest (if appropriate for your workflow).

Troubleshooting

"boundary (singular) fit"

1. •Confirm with isSingular(m, tol = 1e-4).
2. •Simplify random effects: drop slopes/correlations (use ||), remove weak terms, or use a more parsimonious grouping structure.
3. •Center/scale predictors (especially for random slopes) and check for near-zero variance components.

"Model failed to converge" / gradient warnings

1. •Center/scale predictors; check collinearity and separation (GLMMs).
2. •Increase iterations / change optimizer:

ctrl <- lmerControl(optimizer = "bobyqa", optCtrl = list(maxfun = 2e5))
m2 <- update(m, control = ctrl)

1. •Inspect summary(m2)$optinfo$conv and consider simplifying the random-effects structure.

GLMM fit issues (binomial/Poisson)

• •Check for (quasi-)separation, zero-inflation, and overly complex random-effects terms relative to data size.
• •Consider alternative families/links or specialized packages (e.g., zero-inflation) when diagnostics indicate misspecification.

References

• •references/package-description.txt: package metadata (installed)
• •references/exports.txt: exported symbols
• •references/help-index.txt: help index dump
• •references/vignettes.tsv: vignette names + titles (when present)