partR2

partR2 partitions the coefficient of determination (R^2) in generalized linear mixed models (GLMMs) to quantify the unique contribution of fixed-effect predictors to model variance.

Key Features:

• Partitioning Variance: Quantifies part R^2 for fixed-effect predictors by iteratively removing predictors and measuring changes in the variance of the linear predictor to isolate unique contributions.
• Structure Coefficients: Calculates structure coefficients as correlations between predictors and fitted values to estimate each predictor's total contribution to prediction independent of other predictors.
• Inclusive R^2: Transforms structure coefficients into inclusive R^2 by squaring the structure coefficient and multiplying by the total R^2 to represent total variance explained by a predictor.
• Beta Weights: Reports standardized regression coefficients (beta weights) for interpretation and comparison across predictors.
• Parametric Bootstrapping: Uses parametric bootstrapping to generate confidence intervals for estimates.
• Interactions Handling: Accounts for complexity introduced by interactions among predictors when partitioning variance.

Scientific Applications:

• Variance decomposition in GLMMs: Applied to Gaussian and binomial GLMMs to partition R^2 among fixed-effect predictors.
• Biological and ecological studies: Used to assess and compare predictor contributions in biological and ecological research contexts where understanding predictor-specific variance is critical.

Methodology:

Fits generalized linear mixed models, iteratively removes predictors of interest to monitor changes in the variance of the linear predictor, computes structure coefficients as correlations between predictors and fitted values, transforms structure coefficients into inclusive R^2 (square × total R^2), reports standardized beta weights, uses parametric bootstrapping for confidence intervals, and addresses interactions among predictors.