Bayesian dynamic models
kDGLM provides tools for sequential Bayesian
analysis of dynamic generalized linear models. It supports
filtering and smoothing, forecasting, monitoring, simulation, and
feed-forward interventions, with model-building components for trends,
seasonality, regressions, autoregressive terms, transfer functions, and
discount-based evolution.
The current package supports several univariate and multivariate response families, including Normal, bivariate Normal, Poisson, Gamma, and Multinomial models, and it can jointly model multiple time series. Its recursive structure is designed so that inference can be updated as the series grows rather than rebuilding the analysis from the beginning.
Longitudinal surrogate evaluation
OnlineSurr implements methods for surrogate
evaluation when both the primary outcome and the surrogate are measured
longitudinally. It fits marginal and surrogate-adjusted
Gaussian state-space models and uses them to estimate treatment effects
over time and the proportion of treatment effect explained by the
surrogate.
The package provides local and cumulative PTE summaries, bootstrap-based uncertainty quantification, plotting and summary methods, and a test for whether surrogate strength is homogeneous over time. It accompanies the methodological work on jointly longitudinal outcomes and surrogate markers.
