Causal Inference Approaches

Reduced Form Approach

• Based on randomized controlled experiments.
• Uses control groups to create counterfactuals to address causal inference problems.
• Requires identification of a treatment group and a control group.
  • Can be identified through experiments or natural experiments.
  • Confounders are adjusted using econometrics to derive average treatment effects.
• Limitations:
  • Not always applicable, especially if the treatment is a unique historical event.

Structural Approach

• Utilizes models (structural models) to derive counterfactuals.
  • Historians: use verbal reasoning and historical records.
  • Scientists/economists: use mathematical models.

Examples

1. Historians & JFK Assassination

   • Analyze historical records, speeches, and decisions to model potential alternative decisions by JFK.
   • No treatment/control group.

2. Climate Change

   • Construct climate models based on available data.
   • Re-run models to simulate climate without human activity to assess human impact.

3. Macroeconomics

   • Model economies to understand the impact of unique or unprecedented events (e.g., stimulus programs).
   • Simulate scenarios without certain policies to determine causal impacts.
   • Example: Analyzing the impact of a stimulus program post-Great Recession or COVID recession.

4. Healthcare Reform

   • Create a model of the healthcare market to predict effects of new, untested reforms.

Comparison of Approaches

• Reduced Form Approach

  • Relies on actual data and experiments.
  • Key Question: Have all confounders been accounted for?

• Structural Approach

  • Relies on models and simulations.
  • Key Question: Is the model rich enough to create a believable counterfactual?

Integration of Approaches

• Often used together to complement each other.
• Reduced form evidence can inform structural model construction.
• Each has its own set of advantages and limitations.