Supported Models and Production Scope

Purpose

mlmr is designed to make mixed-effects and multilevel models easier to specify, understand, teach, and report. The app is intentionally opinionated: it guides users through a declared modeling workflow while still exposing the underlying lme4 formula, reproducible R code, equations, diagnostics, and manuscript-ready outputs.

This article states the current software scope plainly so users know what is ready for routine use, what is advanced or experimental, and what remains future work.

Current Production Path

The strongest current workflow is a Gaussian mixed-effects model for nested educational, psychological, health, or organizational data. The app is especially well suited for:

• two-level nested models, such as students within schools;
• three-level nested models, such as students within schools within districts;
• random-intercept and random-slope models;
• correlated or independent random-effects structures;
• grand-mean and cluster-mean centering;
• fixed interactions and cross-level interaction-style models;
• APA-style tables, equations, Tau matrices, diagnostics, and reproducible exports.

The built-in HSB-style example is intentionally more complex than a minimal random-intercept model so users can see how centering, interactions, random slopes, and variance-covariance structures work together.

Supported Scope Table

The same scope table is available programmatically:

mlmr::mlm_supported_models()

This table is useful for instructors, analysts, and beta testers because it separates software support from statistical responsibility. A model can be available in the interface and still require careful evaluation of assumptions, sample size, convergence, and interpretation.

Advanced Workflows

mlmr includes advanced pathways for GLMMs and flexible random-effect structures because real research often moves beyond strict nesting. These workflows are valuable, but they require more careful checking:

• binomial, Poisson, negative binomial, and Gamma GLMMs should be checked for outcome coding, link function, overdispersion, and convergence;
• crossed random effects should be used only when the grouping variables truly represent non-nested sources of variation;
• longitudinal models can be represented through grouping variables and time predictors, but users should evaluate residual patterns and time structures;
• weighted multiple-membership models are not a current production feature.

What mlmr Does Not Replace

mlmr does not replace statistical judgment. It helps users express and report models more clearly, but users remain responsible for:

• deciding whether a mixed-effects model is appropriate for the design;
• checking whether grouping factors and predictor levels are correctly declared;
• choosing centering and contrast coding that match the research question;
• evaluating convergence, singular fits, residual patterns, and influential observations;
• confirming that manuscript language matches the fitted model.

Production Readiness Standard

For a model to be treated as production analysis, users should be able to:

1. reproduce the fitted model from the exported R code;
2. explain the centering, interactions, and random-effects choices;
3. inspect the fixed effects, variance components, ICCs, and Tau matrix;
4. review diagnostics and convergence messages;
5. export manuscript tables, equations, and software citations;
6. save a Quarto or script-based record of the full workflow.

That standard is the guiding target for mlmr development.