Lecture 3 Logic and Causal Models

Gang He

February 11, 2025

Recap lecture 2

  • Types and applications of evaluation
  • Exploratory evaluation
  • High Line Park
  • Cultural responsive evaluation
  • Case: Jobs Plus in NYC

Today’s agenda

  • Theory of change
  • Logic model
  • Counterfactuals and causal models
  • DAG: Directed Acyclic Graphs
  • Northwest Housing Alternatives

Theory of change

flowchart LR
  A(IRA) --> B(Subsidies)
  B --> C(Investment)
  B --> D(Jobs)
  C --> D
  C --> E(Domestic Manufacturing)
  D --> E(Domestic Manufacturing)

Elements of a program

Inputs:

Things that go into an activity; money, people, time, etc.

Activities:

Actions that convert inputs to outputs; things that the program does

Outputs:

Tangible goods and services produced by activities; you have control over these

Outcomes:

What happens when the target population uses the outputs; you don’t have control over these

Inputs → Activities → Outputs → Outcomes → Final outcomes

Logic model

flowchart LR
  A(Inputs) --> B(Activities) --> C(Outputs) --> D(Outcomes)

P is for “Produce”

C is for “Change”

Build a logic model for Jobs Plus in NYC

Capturing the wedege

Counterfactual

“The counterfactual is what would have happened—what the outcome (Y) would have been for a program participant—in the absence of the program (P).”

“Since we cannot directly observe the counterfactual, we must estimate it.”

A valid comparison group

  • has the same characteristics, on average, as the treatment group in the absence of the program;
  • remains unaffected by the program; and
  • would react to the program in the same way as the treatment group, if given the program.

Two counterfeit estimates of the counterfactual

  • Before-and-after comparisons (also known as pre-post or refl exive comparisons) compare the outcomes of the same group before and after participating in a program.
  • Challenge: Variables and conditions change
  • Enrolled-and-nonenrolled (or self-selected) comparisons compare the outcomes of a group that chooses to participate in a program with those of a group that chooses not to participate.
  • Challenge: Selection bias

Causal models

  • Instrumental variables
  • Randomized controlled trail (RCT)
  • Regressional discontinuity (RD)
  • Diference in difference (DiD)
  • Matching

Data

  • Experimental: you have control over which units receive treatment
  • Observational: you do not have control over which units receive treatment

Natural experiments

Real experiments could be

  • High costs
  • Infeasible
  • Unethical

Stories

  • Dell and Querubin (2018)

Directed Acyclic Graphs (DAG)

Directed: Each node has an arrow that points to another node

Acyclic: You can’t cycle back to a node (and arrows only have one direction)

Graph: It’s a graph

Draw a DAG

Step 1: List variables

Step 2: Simplify

Step 3: Connect arrows

Step 4: Use logic and math to determine which nodes and arrows to measure

What is the causal effect of an additional year of education on earnings?

Causal identification

A causal effect is identified if the association between treatment and outcome is propertly stripped and isolated.

  • Arrows in a DAG transmit associations
  • We can redirect and control those paths by “adjusting” or “conditioning”

Three types of associations

Confounding: Common cause

Causation: Mediation

Collision: Endogeneity/Selection

Confounding example

What’s the relations between money and win margin?

Money \(\rightarrow\) Win
Money \(\leftarrow\) Quality \(\rightarrow\) Win
Quality is a backdoor

Solution:

  • Find the part of campaign money that is explained by quality, remove it. This is the residual part of money.

  • Find the part of win margin that is explained by quality, remove it. This is the residual part of win margin.

  • Find the relationship between the residual part of money and residual part of win margin. This is the causal effect.

Causasion example

Should you control job connections?

  • Avoid overcontrolling

Causasion example

Should you control job connections?

  • Avoid overcontrolling

Colliders example

Do programming skills reduce social skills?

Hired by a tech company inadvertently connected the two.

Colliders example

Height is unrelated to basketball skill among NBA players

  • Colliders can create fake causal effects

  • Colliders can hide real causal effects

Counterfacture, intervention, and effects

  • Control backdoors
  • Average treatment effect
  • Sub groups (age, race, ethnicity, income, etc.)

References

Dell, Melissa, and Pablo Querubin. 2018. “Nation Building Through Foreign Intervention: Evidence from Discontinuities in Military Strategies.” The Quarterly Journal of Economics 133 (2): 701–64. https://doi.org/10.1093/qje/qjx037.
Gertler, Paul J., Sebastian Martinez, Patrick Premand, Laura B. Rawlings, and Christel M. J. Vermeersch. 2016. Impact Evaluation in Practice, Second Edition. Washington, DC: World Bank. https://doi.org/10.1596/978-1-4648-0779-4.