magComp 0.0.7
Improve exploration of composition factor signs by sampling unrestricted working loadings and jointly aligning loading columns and factor scores to positive anchors. This preserves the existing model, half-normal diagonal priors, and upper-anchor zeros for group, spatial, and space-time factors in leading and conditional allocation blocks.
Add a Harvard Forest inventory article with a mapped census, five model specifications, and a 1,000-layout repeated-sampling assessment. The data, analysis code, and compact results are available with the article.
magComp 0.0.6
Initial public release.
- Fit magnitude and composition models together or separately, using a negative-binomial model for total counts and multinomial allocation among one or two categorical variables.
- Specify separate covariates for magnitude, leading allocation, and conditional allocation. Include group, spatial, or space-time effects in intercepts and slopes.
- Represent dependence among composition categories with shared latent factors and smooth coefficients across ordered classes.
- Use GP or NNGP terms with exponential, Matérn, separable space-time, multiresolution, or Gneiting covariance functions. Covariance parameters have automatic defaults and can be estimated or fixed individually.
- Inspect posterior draws and chain diagnostics through
samples(),summary(), andplot(), including combined coefficients and effects. - Obtain fitted values and predictions for expected quantities or simulated observations. Aggregate density draws to reporting areas with
standSummary(), and assess models with posterior predictive checks and cross-validation. - Follow worked Quarto articles with downloadable data, from individual components to models with group, spatial, and space-time effects.
