Reproducible Ecological Generalized Linear Mixed Model Pipelines

Fits and compares generalized linear mixed models for multiple ecological responses and environmental predictors. The package supports additive and temporal-interaction candidate models, AICc model selection, likelihood-ratio tests, coefficient extraction, Nakagawa R-squared, simulation-based diagnostics, figures, and spreadsheet exports. Model selection follows Burnham and Anderson (2002, ISBN:9780387953649); marginal and conditional R-squared follow Nakagawa and Schielzeth (2013) .

ecoGLMM

ecoGLMM implements a reproducible candidate-model pipeline for ecological generalized linear mixed models. It was created from the ecoacoustic analysis workflow developed by André Felipe Carneiro dos Santos at the Graduate Program in Animal Biology, Universidade Federal de Pernambuco (UFPE).

Authors

André Felipe Carneiro dos Santos — author and maintainer — ORCID: 0009-0000-8596-1975
Bruna Martins Bezerra — author — ORCID: 0000-0003-3039-121X
Bárbara Lins Caldas de Moraes — author — ORCID: 0000-0002-1804-9828

Version 0.1.3 fits one additive and one temporal-interaction model for each response–predictor combination, compares models using AICc, performs nested likelihood-ratio tests, extracts coefficients and Nakagawa R², runs DHARMa diagnostics, creates figures, and exports results.

Installation from the source archive

install.packages("ecoGLMM_0.1.3.tar.gz", repos = NULL, type = "source")

Installation from GitHub

install.packages("remotes")
remotes::install_github("andre-fcsantos/ecoGLMM")

Documentation

The package includes two complementary resources:

A generic editable template installed at inst/examples/ecoGLMM_template.R.
A step-by-step vignette covering data structure, model configuration, diagnostics, figures, and exports.

After installation, open the vignette with:

vignette("getting-started", package = "ecoGLMM")

Copy the complete script to the current working directory with:

script <- system.file("examples", "ecoGLMM_template.R", package = "ecoGLMM")
file.copy(script, "ecoGLMM_template.R")

Development dependencies can also be installed manually:

install.packages(c(
  "glmmTMB", "MuMIn", "performance", "DHARMa", "ggeffects", "ggplot2",
  "writexl"
))

Ecoacoustic example

library(ecoGLMM)

index_config <- data.frame(
  response = c("NDSI_beta", "AEI", "ACT", "BI", "ACI", "EVN", "ADI"),
  family = c("beta", "beta", "beta", "gamma", "gaussian", "gaussian",
             "gaussian"),
  label = c("NDSI", "AEI", "ACT", "BI", "ACI", "EVN", "ADI")
)

analysis_data <- merge(acoustic_data, environmental_data,
                       by = "ponto", all.x = TRUE)
analysis_data$site <- analysis_data$ponto
analysis_data$period <- analysis_data$periodo
analysis_data <- prepare_acoustic_indices(analysis_data)

fit <- run_ecoglmm(
  data = analysis_data,
  config = index_config,
  predictors = c("forest", "urban", "proximity", "watercourses"),
  period = "period",
  group = "site",
  period_levels = c("T-0", "T-1", "T-2", "T-3", "T-4")
)

fit
summary(fit)

diagnostics <- diagnose_models(fit, nsim = 1000, seed = 123)
plot_effect(fit, "NDSI_beta")
plot_selection(fit, "NDSI_beta", metric = "delta")
plot_coefficients(fit)

export_ecoglmm(fit, "results", diagnostics = diagnostics)

The columns R2_marginal and R2_conditional are calculated with MuMIn::r.squaredGLMM() to reproduce the original analytical pipeline. The alternative results from performance::r2_nakagawa() are retained in columns ending in _performance.

Important statistical safeguards

Candidate models are fitted to one shared complete-case dataset by default, ensuring that AICc values are based on identical observations.
AICc comparison stops if models have different sample sizes.
Likelihood-ratio tests are only performed for nested additive and interaction models belonging to the same predictor.
Diagnostic status reflects the actual DHARMa test results; graphical residual inspection is still required.
Only continuous environmental predictors are z-standardized. Period contrasts are not described as standardized coefficients.