Chapter 1 Lecture 01: Golems, Owls and DAGs

Golems:

Clay robots

Powerful

No wisdom or foresight

Dangerous

1.1 Limitation of Statistical models

  • Incredibly limiting

  • Focus on rejecting null hypotheses instead of research hypotheses

  • Relationship between hypothesis and test not clear

  • Industrial framework

image

1.2 Null models rarely unique

1.3 Hypotheses and Models

  • Research requires more than than tiny null robots

  • Precise process model(s)

  • Statistical model (procedure, golem) justifed by implications of process model(s) and question (estimand)

1.4 Owls

“How to draw owls” is a metaphor that this couse will program the models by hands

1.4.1 Why drawing the Bayesian Owl

  • Understand what you are doing

  • Document you work, reduce error

  • Respectable scientific workflow

1.4.2 Five steps of

  1. Theoretical estimand

  2. Scientific causal models

  3. Use 1 & 2 to build statistical models

  4. Simulate from 2 to validate 3 yields 1

  5. Analyze real data

1.4.3 Advantages of Bayesian Owl

  • Bayesian approach is permissive, flexible

  • Expresss uncertanty at all levels

  • Director solutions for measurement error, missing data

  • Focus on scientific modeling

1.5 DAGs (Directed Acylcic Graphs)

Bayes vs Frequentism

1.5.1 Science before statistics

For statistcal modesl to produce scientic insight, they require additional scoienctific modesl

The reasons for a statistical analuysis are not found in the data themselves, but rather in the causes of the data.

The causes of the data cannot be extracted from the data alone. No causes in; no causes out.

1.5.2 Causal inferences

More than association between variables

  • Causal inference is prediction of intervention

Know a cause -> predict the consequences of an intervention

Causal inferences is imputation of missing observations

Know a cause -> construct unobserved counterfactural outcomes.

Each letter is a type of measure.

Arrows mean the causes

1.5.3 Tips

Which control variables

Absolute not safe to add everything

How to test the casual models

With more scientic knowlege, can do more

  • Golems: Brainless, powerful statistical models

  • Owls: Documented, objective procedures

  • DAGs: Transparent scientifc assumptions to justify scientifc efort expose it to useful critique connect theories to golems