[Paper Review] Fitting Linear Mixed-Effects Models Using lme4

Douglas M. Bates, Martin Mächler+2 more|arXiv (Cornell University)|Jun 23, 2014

Data Analysis with R4 references2,577 citations

TL;DR

This paper presents the lme4 package in R for fitting linear mixed-effects models using maximum likelihood or restricted maximum likelihood (REML) estimation. It details the computational framework, including model formulation, sparse matrix methods, Cholesky decomposition, and penalized least squares optimization, enabling efficient and flexible analysis of complex hierarchical data with random effects.

ABSTRACT

(Uploaded by Plazi for the Bat Literature Project) Maximum likelihood or restricted maximum likelihood (REML) estimates of the parameters in linear mixed-effects models can be determined using the lmer function in the lme4 package for R. As for most model-fitting functions in R, the model is described in an lmer call by a formula, in this case including both fixed- and random-effects terms. The formula and data together determine a numerical representation of the model from which the profiled deviance or the profiled REML criterion can be evaluated as a function of some of the model parameters. The appropriate criterion is optimized, using one of the constrained optimization functions in R, to provide the parameter estimates. We describe the structure of the model, the steps in evaluating the profiled deviance or REML criterion, and the structure of classes or types that represents such a model. Sufficient detail is included to allow specialization of these structures by users who wish to write functions to fit specialized linear mixed models, such as models incorporating pedigrees or smoothing splines, that are not easily expressible in the formula language used by lmer.

Motivation & Objective

• To provide a robust, efficient, and extensible framework for fitting linear mixed-effects models in R.
• To address computational challenges in fitting mixed models on moderately large datasets (10^4–10^6 observations).
• To support flexible model specification beyond standard formula syntax via modular design and extensibility hooks.
• To maintain backward compatibility while enabling future enhancements through modular architecture and experimental branches.
• To document and standardize the computational pipeline for linear mixed models, supporting both end-users and developers.

Proposed method

• Formulates linear mixed models using a joint distribution for response Y and random effects B, with conditional normality given B.
• Employs a relative covariance factor Λθ to parameterize the random effects variance-covariance matrix Σθ = σ²ΛθΛᵀθ.
• Uses sparse matrix representations for Z and Λθ to improve computational efficiency, especially for large, sparse designs.
• Applies Cholesky decomposition to solve the penalized least squares (PLS) problem derived from the profiled REML or log-likelihood criterion.
• Optimizes the profiled criterion using constrained optimization routines in R, with parameter estimation via iterative refinement.
• Provides modular functions for formula parsing, criterion evaluation, optimization, and model inference, enabling extensibility and customization.

Experimental results

Research questions

• RQ1 How can linear mixed-effects models be efficiently and flexibly implemented in R using modern sparse matrix techniques?
• RQ2 What computational strategies enable scalable estimation of mixed models on large datasets with complex random effects structures?
• RQ3 How can the lme4 framework support models beyond standard formula syntax, such as those with pedigrees or smoothing splines?
• RQ4 What are the key differences between the current PLS formulation in lme4 and earlier approaches like Henderson’s mixed-model equations?
• RQ5 How can the framework support advanced inference, including profile likelihood, parametric bootstrapping, and posterior predictive simulation?

Key findings

• The lme4 package successfully implements REML and ML estimation for linear mixed models using efficient sparse matrix computations and Cholesky-based PLS solving.
• The sleep deprivation study example shows fixed effects of 251.4 ms (intercept) and 10.47 ms/day (slope), with random effects standard deviations of 24.74 ms and 5.92 ms/day.
• Profile likelihood plots (zeta, density, and pairs) reveal that the sampling distribution of the random effects parameters is approximately normal, with minimal nonlinearity in the profile traces.
• The modular design of lme4 allows for extensibility, with hooks used by packages like gamm4 and blme to extend model capabilities.
• The framework supports flexible prediction and simulation via the `fitted`, `predict`, and `simulate` methods, with options to condition on random effects or set them to zero.
• The implementation is stable and scalable, supporting datasets with up to 10^6 observations, and forms the foundation for related tools like MixedModels in Julia and the flexLambda branch of lme4.