Kenichiro McAlinn
Several models and forecasters can provide useful but incomplete information about the same outcome. Bayesian predictive synthesis combines their predictive distributions while learning about biases, calibration, and dependence, allowing the contribution of each source to change over time. Related synthesis methods support causal inference by combining treatment-effect estimates, propensity-score models, and outcome models.
An intervention can change both outcomes and the way they develop over time. Dynamic causal inference studies these changes through potential outcome trajectories, with explicit assumptions for separating intervention effects from other sources of variation in a time series.
Statistical methods involve choices about what to learn, how to evaluate it, and which actions to take. Decision theory makes these choices explicit, connecting cross-validation, prediction, and generalized Bayes to the consequences and uncertainty that matter for a stated purpose.
Statistical conclusions depend on how probabilities, models, and agents are interpreted. Philosophy of statistics studies what these representations mean, how several imperfect models can be used together, and what warrants moving from a coherent calculation to a prediction or decision.