Economics 588: Advanced Econometrics

Advanced Econometric Methods for Causal Inference

Brigham Young University β€” Professor Brigham Frandsen, Notes and Project from Patrick Beal

Welcome

This site contains resources from Economics 588: Advanced Econometrics at Brigham Young University, including comprehensive course materials and a final research project on the integration of Directed Acyclic Graphs with traditional econometric methods.

πŸ“š Course Material

Comprehensive coverage of econometric estimation and inference, including OLS, instrumental variables, maximum likelihood estimation, quantile regression, and nonparametric methods. All estimators are unified through the extremum estimation framework and illustrated through the 6-step econometric process.

View Course Materials β†’

πŸ“Š Final Project

An in-depth analysis of how Directed Acyclic Graphs (DAGs) complement traditional econometrics for robust causal inference. This project explores collider bias, the frontdoor criterion, M-bias, and practical falsification tests for validating causal assumptions.

Explore the Project β†’

Key Findings

πŸ“Š

Collider Bias

Magnitude: Bias of -0.50 when true effect = 0
Lesson: Conditioning on colliders creates spurious correlation. DAG reasoning prevents this error.

πŸ”—

Frontdoor Criterion

Achievement: Identifies effects despite unmeasured confounding
Result: Bias reduction from 0.198 to 0.0001 vs. naive OLS

⚠️

M-Bias

Risk: Pre-treatment variables that are colliders induce bias
Magnitude: Downward bias of -0.068 when true effect = 0.4

βœ“

Falsification Tests

Innovation: DAG-implied conditional independence relationships are testable
Benefit: Validate causal model assumptions before formal estimation

The Integrated Framework

DAGs are not a replacement for econometricsβ€”they are complementary. This project presents a systematic five-step workflow integrating DAG-based identification with traditional econometric estimation.

1

Specify Causal Structure

Draw a DAG encoding your causal assumptions explicitly

2

Test Assumptions via D-Separation

Derive and validate conditional independence implications in the data

3

Identify Estimation Strategy

Apply backdoor, frontdoor, or do-calculus criteria to determine necessary covariates

4

Estimate Using Standard Methods

Implement OLS, IV, or other estimators with DAG-justified covariate specifications

5

Document the Analysis

Present causal model, assumptions, and identification strategy alongside results

Core Concepts

Directed Acyclic Graph (DAG)

A visual representation of causal assumptions where nodes are variables and arrows represent direct causal effects.

D-Separation

A criterion determining which variables are conditionally independent given a set of conditioning variables.

Backdoor Criterion

Identifies which variables must be controlled for to block non-causal paths from treatment to outcome.

Frontdoor Criterion

Enables identification despite unmeasured confounding when a complete mediator is observed.

Do-Calculus

A formal system of rules for transforming causal queries into identifiable statistical estimands.

Collider Bias

Spurious correlation created by conditioning on a variable affected by both treatment and outcome.

Key Takeaways

  • DAGs provide systematic rules for covariate selection, replacing ad-hoc intuition with algorithmic reasoning
  • Traditional guidance can increase bias in realistic causal structures (colliders, M-bias)
  • D-separation enables identification in novel causal structures (frontdoor, do-calculus)
  • DAG-implied assumptions are testable through conditional independence tests before estimation
  • Integration of DAGs and econometrics yields transparent, rigorous, and robust causal inference