Winda

Statistical Rethinking (2023)

Contents · 39

  1. 01-Science Before StatisticsExplains why causal assumptions underpin statistical modeling, introduces interventions and counterfactuals, and uses DAGs to guide scientific reasoning and analysis.
  2. 01-Science Before StatisticsIntroduces a science-centered statistics course focused on causal models, causal inference, and connecting data to scientific questions. Outlines the course’s core themes: DAGs, Golems, and OWLs.
  3. 02-Garden of Forking DataExplains Bayesian estimation using globe tosses to estimate Earth's water coverage, covering generative models, estimator design, testing, and uncertainty summaries.
  4. 02-Garden of Forking DataIntroduces generative models, estimands, and Bayesian reasoning through a globe-tossing example, showing how sample likelihoods inform plausible water proportions.
  5. 03-Geocentric ModelsHow epicyclic models predict planetary retrograde motion despite misstating solar system structure, and the history of Ceres’s discovery.
  6. 03-Geocentric ModelsIntroduces linear regression through the height–weight example, covering generative models, Gaussian variation, causal assumptions, and the workflow from question to analysis.
  7. 04-Categories and CurvesCategorical predictors and spline-based curves in linear models, with causal diagrams, estimands, and posterior postprocessing.
  8. 04-Categories and CurvesUsing categorical predictors, causal diagrams, and statistical models to estimate total and direct causal effects.
  9. 05-Elemental ConfoundsIntroduces threats to causal inference, spurious correlations, and the four elemental confounds, beginning with the fork and common causes.
  10. 05-Elemental ConfoundsExplains how common causes create confounding, using causal diagrams and U.S. marriage and divorce rates to show how stratification, regression, standardization, and prior predictive checks address it.
  11. 06-Good and Bad ControlsCausal diagrams and confounding, including how randomization removes confounding and how the do-operator represents interventions.
  12. 06-Good and Bad ControlsExplains causal diagrams, backdoor paths, and adjustment sets for estimating causal effects, including how inappropriate controls can introduce bias.
  13. 07-OverfittingEstimation challenges with finite data, balancing model simplicity and accuracy, and defining useful predictions.
  14. 07-OverfittingExplains overfitting, cross-validation, and predictive accuracy, then shows how regularizing priors and information criteria such as PSIS and WAIC help select models.
  15. 08-MCMCBayesian computation with random numbers, measurement error, and latent-variable models for interpreting scores and evaluations.
  16. 09-Modeling EventsUses the Berkeley admissions data to examine gender discrimination, causal pathways through department choice, and how to model event counts and admission probabilities.
  17. 09-Modeling EventsIntroduces generalized linear models for event data through UC Berkeley graduate admissions, gender, department, and mediation analysis.
  18. 10-Counts and Confounds
  19. 10-Counts and ConfoundsGeneralized linear models for count outcomes, plus confounding and collider bias in admissions and citation-network examples.
  20. 11-Ordered CategoriesModels ordered categorical responses with cumulative logits, using cutpoints and predictors to analyze trolley-problem judgments.
  21. 11-Ordered CategoriesIntroduces problem-solving heuristics and ethical dilemmas, using the trolley problem to examine how competing principles shape judgments.
  22. 12-Multilevel ModelsIntroduces multilevel models, showing how shared population-level information across stories or participants improves learning and reduces overfitting compared with separate categorical estimates.
  23. 12-Multilevel ModelsExplains multilevel models as a way to share information across groups and adaptively regularize estimates, using repeated observations and reedfrog survival to illustrate hierarchical priors and partial pooling.
  24. 13-Multilevel AdventuresExplains how multilevel modeling branches from foundational partial pooling into specialized methods, and offers ways to navigate the remaining course topics.
  25. 13-Multilevel AdventuresExplains multilevel models, varying effects and partial pooling, including district-level contraceptive-use analysis and practical modeling challenges.
  26. 14-Correlated FeaturesExplains multivariate normal priors for correlated group features and partial pooling, using Bangladesh contraceptive-use data and a Berkeley admissions example.
  27. 14-Correlated FeaturesModels correlated varying intercepts and slopes with multivariate priors, using the Bangladesh contraception data to estimate district-level effects.
  28. 15-Social NetworksModeling reciprocal food transfers among Nicaraguan households, separating generalized giving from social ties and household characteristics.
  29. 15-Social NetworksExplores reciprocity and generalized giving in household food-sharing networks, using causal diagrams and generative models to analyze directed ties and their dependencies.
  30. 16-Gaussian ProcessesModels spatial confounding in Oceanic island toolkits and examines how population size relates to technological complexity.
  31. 16-Gaussian ProcessesIntroduces Gaussian processes and covariance kernels, using island distances to model spatial variation in Oceanic tool counts. Covers kernel choices, priors, and a Poisson model for the data.
  32. 17-MeasurementConditional probability, Bayesian reasoning, and a pancake puzzle illustrate how to reason about mismeasurement and misclassification.
  33. 17-MeasurementExplains measurement error and misclassification, including how biased proxies can distort effects and how causal graphs and Bayesian models account for uncertain measurements.
  34. 18-Missing DataMissing-data mechanisms, causal models, and when to drop incomplete cases or use imputation.
  35. 18-Missing DataExplains MCAR, MAR, and MNAR missingness, their effects on bias, and Bayesian imputation and marginalization, with examples in discrete data and phylogenetic regression.
  36. 19-Generalized Linear MadnessMechanistic modeling of human weight from height using body volume, proportional shape, and density.
  37. 19-Generalized Linear MadnessExplains generalized linear and mixed models through scientific modeling, prior selection, scaling, simulation, and model fit.
  38. 20-HoroscopesExplains how statistics without scientific models resembles divination: vague inputs yield vague advice, while scientific models enable stronger conclusions.
  39. 20-HoroscopesStatistical research planning and workflow, including estimands, causal theory, sampling and analysis plans, preregistration, documentation, version control, testing, and code review.Statistical research planning and workflow, from estimands and causal assumptions to sampling, analysis, preregistration, documentation, version control, testing, and review.