Questions and Answers

Routing Summary

Answered questions from the vault knowledge base. Each answer is cross-linked to its source notes and related concepts. Browse by topic below or search for keywords.

By Topic

Marketing Measurement / Adaptive Experimentation

Econometrics / Simulation-Based Estimation

Causal Inference / Identification

Research Methodology / Multiple Comparisons

Statistical Modeling / General

Bayesian vs. Frequentist Statistics

  • Q - Differences Between Frequentist and Bayesian Statistics — Core philosophical divide (probability as frequency vs. belief), confidence vs. credible intervals, priors, hierarchical models, model comparison
  • Q - Does Peeking Matter for a Bayesian — The posterior given the model is unaffected by optional stopping, but frequentist error of posterior-threshold rules, sign/magnitude error at a fixed truth and prior sensitivity are not protected; the mSPRT statistic is a Bayes factor; 7-step media-test protocol

Bayesian Computation, Calibration and Shared Machinery

  • Q - Four Meanings of Calibration — SBC calibrates the computation (prior-averaged), conformal is a finite-sample marginal-coverage theorem, forecast calibration is an empirical property, solver calibration is average-case, and ABM/persona ‘calibration’ just means fitting; each property needs a sharpness check
  • Q - The Kalman Filter Across BSTS State-Space Models and ODE Solvers — One algorithm — filter, RTS smoother, innovations likelihood = O(N) GP regression with a Gauss–Markov prior; BSTS predicts without updates for a counterfactual, ODE filters feed themselves zero residuals and read covariance as numerical error
  • Q - Variational Bounds Compared from the ELBO to EIG Estimators — Every objective is ‘intractable quantity = surrogate ± expected KL’: expectation under q gives reverse KL (under-disperses), under the joint gives forward KL (over-covers); NPE loss equals the Barber–Agakov bound up to prior entropy; 2×2 table places NMC, PCE, VNMC, ACE
  • Q - Partial Pooling Across Statistics and ML and When It Hurts — Shared mechanism: units as draws from a learned population with precision-weighted compromise (hierarchical Bayes, James–Stein, Gamma-Gamma/NBD); global forecasters and LLMs pool without an inspectable τ; six conditions under which pooling hurts

Machine Learning and AI Bridges

Recent Questions

QuestionDateKey Sources
Q - Comparing Geo-Test Estimators from TBR to Synthetic DiD2026-09-18Geo-Experiment Methodology - Overview, Geo-Experiment Design and Power Analysis, Time-Based Regression Estimator for Geo Experiments, TBR Design Sensitivity and the Stationarity Assumption
Q - Using Experiment Results as Priors in a Bayesian MMM2026-09-18Bayesian Media Mix Modeling - Overview, Bayesian Estimation and Priors for MMM, ROAS, mROAS, and Optimal Media Mix, Shape (Saturation) Effects
Q - How Adstock Breaks Switchback and Sequential Test Assumptions2026-09-18Switchback Experiment Design and Analysis, Interference and Marketplace Experiments, Always-Valid p-values and the mSPRT, Confidence Sequences
Q - Optimizing Media Spend on CLV with Delayed Feedback2026-09-18Customer Lifetime Value - Overview, Pareto-NBD Model, BG-NBD Model, Gamma-Gamma Model of Monetary Value
Q - Budget Allocation Under Power Laws from Chinchilla to Media Mix2026-09-18Neural Scaling Laws, Compute-Optimal Training (Chinchilla), ROAS, mROAS, and Optimal Media Mix, Shape (Saturation) Effects
Q - Covariate Adjustment for Precision vs Identification2026-09-18CUPED and Regression-Adjusted Variance Reduction, Logic of Regression Adjustment, Table 2 Fallacy, Nuisance Parameter Bias Simulation
Q - The Common Structure of Doubly-Robust Estimators2026-09-18Frequentist Causal Estimation, DML Estimators for ATE and the Interactive Model, Neyman Orthogonality, Doubly-Robust Estimands for ATT(g,t)
Q - Which Heterogeneous Treatment Effect Method Answers Which Question2026-09-18Causal Estimands, Metalearners for CATE, S-Learner, T-Learner and Minimax Rate
Q - A Unified View of Sensitivity to Assumption Violations2026-09-18Honest DiD - Sensitivity to Parallel Trends Violations, Pre-Trend Testing and Its Pitfalls, Sensitivity Analysis in Observational Studies, Plausible GMM - Overview
Q - Exchangeability and What Replaces It When It Fails2026-09-18Permutation Tests and Exact Inference, Fisher Randomization Test and the Sharp Null, Randomization Inference - Overview, Sharp vs Weak Null Hypotheses
Q - Choosing a Simulation-Based Inference Method for ABM Calibration2026-09-18Simulation-Based Estimation - Overview, Method of Simulated Moments, SMM Weighting Matrix and Inference, Indirect Inference
Q - Sample Splitting and Pre-registration as Cures for Forking Paths2026-09-18Garden of Forking Paths, Researcher Degrees of Freedom, Forking Paths and Bayesian Approaches, Prediction vs Postdiction
Q - Does Peeking Matter for a Bayesian2026-09-18The Peeking Problem and Optional Stopping, Always-Valid p-values and the mSPRT, Confidence Sequences, Garden of Forking Paths
Q - Four Meanings of Calibration2026-09-18Simulation-Based Calibration - Overview, Data-Averaged Posterior Self-Consistency, Rank Statistics and Uniformity, Interpreting SBC Histograms
Q - The Kalman Filter Across BSTS State-Space Models and ODE Solvers2026-09-18State-Space Models and the Kalman Filter - Overview, Linear-Gaussian State-Space Models, The Kalman Filter, The RTS Smoother
Q - Variational Bounds Compared from the ELBO to EIG Estimators2026-09-18The ELBO and KL Divergence Minimization, Mean-Field Family and Coordinate Ascent VI (CAVI), Normalizing Flows for Variational Inference, Reparameterization Trick and Variational Autoencoders
Q - Partial Pooling Across Statistics and ML and When It Hurts2026-09-18Hierarchical Models, Hierarchical Linear Models, Partial Pooling as Multiple Comparisons Correction, Empirical Bayes - Overview
Q - When Can LLM Silicon Samples Replace Consumer Data in an ABM2026-09-18LLM-Powered Agents - Overview, Silicon Samples and Algorithmic Fidelity, Validity, Bias and Calibration of LLM-Simulated Populations, Persona Mixture Calibration of LLM Agents
Q - In-Context Learning as Amortized Bayesian Inference2026-09-18In-Context Learning and Few-Shot Prompting, Autoregressive Language Modeling and Pretraining, Transformers and LLM Foundations - Overview, Neural Posterior Estimation (NPE)
Q - A Map of Sequential Decision Methods from Bandits to RLHF2026-09-18Multi-Armed Bandits and Thompson Sampling - Overview, Bernoulli Bandit and Thompson Sampling Algorithm, UCB and Greedy Algorithms for Bandits, Regret Bounds for Thompson Sampling
Q - Continuous Learning in Media Measurement with Interaction Effects2026-07-01Bayesian Media Mix Modeling - Overview, Sequential and Adaptive BED, Expected Information Gain, Horseshoe and Regularized Horseshoe Priors, Bayesian Optimisation
Q - Encoding a Geo-Holdout as a Bayesian Experimental Design and Computing Its EIG2026-07-01Expected Information Gain, Nested Estimation and Nested Monte Carlo, Bayesian Media Mix Modeling - Overview, Bayesian Structural Time-Series Model
Q - BED vs Bayesian Optimization vs Bandits for Media Experimentation2026-07-01Bayesian Optimisation, Acquisition Functions, The Global Optimisation Problem, Q- and A-learning - Overview
Q - Carryover Dynamics and the Timing of Sequential Media Experiments2026-07-01Carryover (Adstock) Functional Forms, ROAS, mROAS, and Optimal Media Mix, Linear-Gaussian State-Space Models, Bayesian Structural Time-Series Model
Q - Using SMM to Calibrate Agent Based Models2026-04-11Method of Simulated Moments, ABM Calibration Overview, Genetic Algorithm Calibration for ABM, SMM Weighting Matrix and Inference
Q - Uncovering Causal Estimates from Non-Experimental Data2026-04-10The Selection Problem, Instrumental Variables, Differences-in-Differences, Synthetic Control
Q - Differences Between Frequentist and Bayesian Statistics2026-04-09Probability and Bayesian Inference, Asymptotics and Frequentist Connections, Hierarchical Models
Q - Common Pitfalls in Statistical Modeling2026-04-09Spurious Association and Confounds, Garden of Forking Paths, Overfitting and Information Criteria, Model Checking
Q - Handling Multiple Comparisons When Selecting From Hundreds of Models2026-04-09Multiple Comparisons - Bayesian Perspective, Garden of Forking Paths, Partial Pooling as Multiple Comparisons Correction

All Questions