Identifying naturally occurring treatment assignments or exogenous shocks is the cornerstone of a successful natural experiment or instrumental variable (IV) strategy. The goal is to find situations where a "treatment" is assigned in a way that is as good as random with respect to unobserved factors that also influence your outcome of interest. This "as if" randomization is what allows you to make causal claims.

Here's a guide to identifying such opportunities, broken down by common types of shocks/assignments:

Key Principle: Exogeneity ("As If Random")

Before diving into categories, remember the golden rule: the shock or treatment assignment must be exogenous to the outcome. This means it's not caused by the outcome, not correlated with unobserved factors that also cause the outcome, and ideally, not chosen by individuals/entities based on their anticipation of the outcome. You want situations that mimic a randomized controlled trial (RCT) as closely as possible.

Categories of Naturally Occurring Treatment Assignments / Exogenous Shocks:

1. Policy Changes & Regulatory Shifts

These are often the most fertile ground for natural experiments. Governments, agencies, or institutions implement new laws, regulations, or programs.

• How to identify: Look for specific dates, geographic boundaries, or demographic cutoffs for when a policy takes effect.


• Examples:


• Minimum Wage Increases: A state or municipality raises its minimum wage, while a neighboring, economically similar jurisdiction does not. The minimum wage increase is the "treatment."


• Health Care Reform: The Affordable Care Act (ACA) expanded Medicaid in some states but not others. State-level expansion status serves as a natural experiment for the effects of insurance on health, labor, etc.


• Environmental Regulations: New emissions standards applied to specific industries or regions (e.g., pollution controls for power plants in certain areas).


• Tax Changes: Introduction of a new tax (e.g., soda tax, carbon tax) in one jurisdiction but not others, or a tax rebate for a specific group.


• School Desegregation Orders: Judicial rulings that forced desegregation in some school districts but not others, offering a staggered treatment.


• Voting Law Changes: Introduction of voter ID laws, early voting, or online registration in different states at different times.

2. Geographic Boundaries & Administrative Jurisdictions

Policies, resources, or even natural conditions often vary sharply at administrative borders.

• How to identify: Compare outcomes for units (individuals, firms, land parcels) that are very close to a border but fall on opposite sides. The closer to the border, the more similar the units are likely to be in unobservable ways, except for the border-defined difference.


• Examples:


• State/County Borders: Towns straddling a state or county line where one side has a different speed limit, alcohol sales laws, abortion access, or property tax rate.


• School District Borders: Students living on opposite sides of a school district boundary, attending different schools due to residency rules.


• Electoral Districts: Voters assigned to different districts with varying levels of political competition or representation.


• Urban/Rural Designations: Eligibility for certain grants or services that depend on whether an area is officially designated as urban or rural.

3. Time-Based Events & Historical Shocks

Sudden, unexpected events that affect some groups or regions but not others, or at different times.

• How to identify: Look for specific dates of major events that had differential impacts.


• Examples:


• Natural Disasters: Earthquakes, hurricanes, floods, wildfires affecting specific geographic areas. The shock is the disaster; treatment is exposure.


• Economic Crises: A sudden financial collapse (e.g., 2008 crisis) that affects industries or regions with differing pre-existing exposures or vulnerabilities.


• Technological Innovations: The introduction of broadband internet, a new crop variety, or a new medical device in different regions or at different times.


• War/Conflict: Exogenous exposure to conflict or conscription for certain cohorts or regions.


• Mass Layoffs/Plant Closures: A major employer in a town suddenly closes, affecting local labor markets.

4. Rule-Based Assignments & Thresholds (Regression Discontinuity Design - RDD)

When treatment is assigned based on whether a continuous "forcing variable" crosses a specific, arbitrary threshold.

• How to identify: Look for any program, policy, or eligibility rule where there's a sharp cutoff.


• Examples:


• Age Cutoffs:


• Drinking age (effects on health, crime).


• Voting age (effects on political participation).


• School enrollment age (effects on educational attainment, labor market outcomes).


• Retirement age for social security benefits.


• Test Score Cutoffs: Students admitted to a selective school or program based on a minimum test score.


• Income/Poverty Lines: Eligibility for welfare programs, food stamps, or housing assistance based on income falling below a certain threshold.


• Geographic Proximity Thresholds: Eligibility for a program if you live within X miles of a facility.


• Firm Size Cutoffs: Regulations (e.g., environmental, labor) that apply only to firms above a certain number of employees.

5. Lotteries & Random Assignment (The Gold Standard, but Rare)

When treatment is explicitly assigned randomly, even if not by the researcher. This is the closest to an RCT.

• How to identify: Look for situations where explicit random draws determine participation or eligibility.


• Examples:


• School Lotteries: When charter schools or oversubscribed public schools use lotteries for admission.


• Visa Lotteries: Green Card lotteries or other immigration lotteries.


• Military Draft Lotteries: Historical examples where draft order was randomized.


• Jury Duty Selection: Random assignment to jury pools.


• Pilot Programs: Government agencies sometimes randomly select areas or individuals for pilot programs before wider rollout.

6. Instrumental Variables (IV) Specifics

While the above categories often provide the instrument, sometimes the instrument is a distinct variable that only affects the treatment, not the outcome directly.

• How to identify: Think about a variable that causes the endogenous treatment but is otherwise unrelated to the outcome. This requires a strong theoretical argument for the exclusion restriction (the instrument affects the outcome ONLY through the endogenous treatment).


• Examples:


• Weather as an Instrument:


• Rainfall as an IV for agricultural yields (treatment) on local economic development (outcome). Rainfall affects yield but shouldn't directly affect the economy beyond agriculture.


• Wind direction as an IV for exposure to air pollution (treatment) on health outcomes (outcome).


• Judicial Leniency: Random assignment of judges (with varying average leniency) as an IV for the effect of incarceration (treatment) on future employment (outcome).


• Proximity to Institutions (with caution): Proximity to a college as an IV for educational attainment (treatment) on wages (outcome), if one can argue proximity doesn't directly impact wages independent of education. (This is often debated and tricky to satisfy the exclusion restriction).


• Historical Accidents/Legacy: Persistence of historical institutions or boundaries that are now effectively random with respect to modern outcomes (e.g., colonial borders).

How to Evaluate a Potential Shock/Assignment:

Once you have an idea, critically evaluate its suitability:

• Exogeneity/Credibility: Can you make a strong argument that the treatment assignment is "as if random" or unrelated to unobserved determinants of the outcome? This is the most crucial step. What's the narrative of why it's exogenous?


• Mechanism: How does the shock cause the treatment? Is this link clear and plausible?


• Data Availability: Can you actually measure the shock/treatment, the control group, and the outcome variables with sufficient detail and over time?


• "As If" Randomization Check:


• For DiD: Are pre-treatment trends similar between treated and control groups? (Parallel Trends Assumption)


• For RDD: Are covariates smoothly distributed around the cutoff, with no jumps other than the treatment itself?


• For IV: Does the instrument correlate strongly with the endogenous treatment (first stage)?


• Exclusion Restriction (for IV): Is it truly plausible that your instrument only affects the outcome through the endogenous treatment, and not through any other channel? This is often the hardest assumption to defend.


• Placebo Tests: Can you conduct tests where you shouldn't see an effect (e.g., apply the "treatment" to an earlier period, or to a group that shouldn't be affected)?