Variational Mixture Models for Clustering Categorical Data

A variational Bayesian finite mixture model for the clustering of categorical data, and can implement variable selection and semi-supervised outcome guiding if desired. Incorporates an option to perform model averaging over multiple initialisations to reduce the effects of local optima and improve the automatic estimation of the true number of clusters. For further details, see the paper by Rao and Kirk (2024) .

Hierarchical Modeling and Frequency Method Checking on Overdispersed Gaussian, Poisson, and Binomial Data

We utilize approximate Bayesian machinery to fit two-level conjugate hierarchical models on overdispersed Gaussian, Poisson, and Binomial data and evaluates whether the resulting approximate Bayesian interval estimates for random effects meet the nominal confidence levels via frequency coverage evaluation. The data that Rgbp assumes comprise observed sufficient statistic for each random effect, such as an average or a proportion of each group, without population-level data. The approximate Bayesian tool equipped with the adjustment for density maximization produces approximate point and interval estimates for model parameters including second-level variance component, regression coefficients, and random effect. For the Binomial data, the package provides an option to produce posterior samples of all the model parameters via the acceptance-rejection method. The package provides a quick way to evaluate coverage rates of the resultant Bayesian interval estimates for random effects via a parametric bootstrapping, which we call frequency method checking.

Simulate, Evaluate, and Analyze Dose Finding Trials with Bayesian MCPMod

Bayesian MCPMod (Fleischer et al. (2022) ) is an innovative method that improves the traditional MCPMod by systematically incorporating historical data, such as previous placebo group data. This R package offers functions for simulating, analyzing, and evaluating Bayesian MCPMod trials with normally distributed endpoints. It enables the assessment of trial designs incorporating historical data across various true dose-response relationships and sample sizes. Robust mixture prior distributions, such as those derived with the Meta-Analytic-Predictive approach (Schmidli et al. (2014) ), can be specified for each dose group. Resulting mixture posterior distributions are used in the Bayesian Multiple Comparison Procedure and modeling steps. The modeling step also includes a weighted model averaging approach (Pinheiro et al. (2014) ). Estimated dose-response relationships can be bootstrapped and visualized.

A Shiny Application for End-to-End Bayesian Decision Network Analysis and Web-Deployment

A Shiny application for learning Bayesian Decision Networks from data. This package can be used for probabilistic reasoning (in the observational setting), causal inference (in the presence of interventions) and learning policy decisions (in Decision Network setting). Functionalities include end-to-end implementations for data-preprocessing, structure-learning, exact inference, approximate inference, extending the learned structure to Decision Networks and policy optimization using statistically rigorous methods such as bootstraps, resampling, ensemble-averaging and cross-validation. In addition to Bayesian Decision Networks, it also features correlation networks, community-detection, graph visualizations, graph exports and web-deployment of the learned models as Shiny dashboards.

Neural AutoRegressive Fractionally Integrated Moving Average Model

Methods and tools for forecasting univariate time series using the NARFIMA (Neural AutoRegressive Fractionally Integrated Moving Average) model. It combines neural networks with fractional differencing to capture both nonlinear patterns and long-term dependencies. The NARFIMA model supports seasonal adjustment, Box-Cox transformations, optional exogenous variables, and the computation of prediction intervals. In addition to the NARFIMA model, this package provides alternative forecasting models including NARIMA (Neural ARIMA), NBSTS (Neural Bayesian Structural Time Series), and NNaive (Neural Naive) for performance comparison across different modeling approaches. The methods are based on algorithms introduced by Chakraborty et al. (2025) .

Variance Estimation of FMT Method (Fully Moderated T-Statistic)

The FMT method computes posterior residual variances to be used in the denominator of a moderated t-statistic from a linear model analysis of gene expression data. It is an extension of the moderated t-statistic originally proposed by Smyth (2004) . LOESS local regression and empirical Bayesian method are used to estimate gene specific prior degrees of freedom and prior variance based on average gene intensity levels. The posterior residual variance in the denominator is a weighted average of prior and residual variance and the weights are prior degrees of freedom and residual variance degrees of freedom. The degrees of freedom of the moderated t-statistic is simply the sum of prior and residual variance degrees of freedom.

Hierarchical Bayesian Small Area Estimation

Functions to compute small area estimates based on a basic area or unit-level model. The model is fit using restricted maximum likelihood, or in a hierarchical Bayesian way. In the latter case numerical integration is used to average over the posterior density for the between-area variance. The output includes the model fit, small area estimates and corresponding mean squared errors, as well as some model selection measures. Additional functions provide means to compute aggregate estimates and mean squared errors, to minimally adjust the small area estimates to benchmarks at a higher aggregation level, and to graphically compare different sets of small area estimates.

Sensitivity Assessment to Unmeasured Confounding with Multiple Treatments

A sensitivity analysis approach for unmeasured confounding in observational data with multiple treatments and a binary outcome. This approach derives the general bias formula and provides adjusted causal effect estimates in response to various assumptions about the degree of unmeasured confounding. Nested multiple imputation is embedded within the Bayesian framework to integrate uncertainty about the sensitivity parameters and sampling variability. Bayesian Additive Regression Model (BART) is used for outcome modeling. The causal estimands are the conditional average treatment effects (CATE) based on the risk difference. For more details, see paper: Hu L et al. (2020) A flexible sensitivity analysis approach for unmeasured confounding with multiple treatments and a binary outcome with application to SEER-Medicare lung cancer data .

Estimating Speakers of Texts

Estimates the authors or speakers of texts. Methods developed in Huang, Perry, and Spirling (2020) . The model is built on a Bayesian framework in which the distinctiveness of each speaker is defined by how different, on average, the speaker's terms are to everyone else in the corpus of texts. An optional cross-validation method is implemented to select the subset of terms that generate the most accurate speaker predictions. Once a set of terms is selected, the model can be estimated. Speaker distinctiveness and term influence can be recovered from parameters in the model using package functions. Once fitted, the model can be used to predict authorship of new texts.

Bayesian Regression Models using 'Stan'

Fit Bayesian generalized (non-)linear multivariate multilevel models using 'Stan' for full Bayesian inference. A wide range of distributions and link functions are supported, allowing users to fit -- among others -- linear, robust linear, count data, survival, response times, ordinal, zero-inflated, hurdle, and even self-defined mixture models all in a multilevel context. Further modeling options include both theory-driven and data-driven non-linear terms, auto-correlation structures, censoring and truncation, meta-analytic standard errors, and quite a few more. In addition, all parameters of the response distribution can be predicted in order to perform distributional regression. Prior specifications are flexible and explicitly encourage users to apply prior distributions that actually reflect their prior knowledge. Models can easily be evaluated and compared using several methods assessing posterior or prior predictions. References: Bürkner (2017) ; Bürkner (2018) ; Bürkner (2021) ; Carpenter et al. (2017) .