Home Semiconductors20

Where Did Exxon Valdez Happen? The Location They Don't Want You To See!

Mar 12, 2026 • minute read

Contents

You’ve heard of the Exxon Valdez oil spill—the catastrophic 1989 disaster that dumped 11 million gallons of crude oil into Alaska’s Prince William Sound. But here’s the question few ask: Where did it really happen? Not just the coordinates, but the conceptual location in the world of economic research. The answer lies in a method so powerful it’s reshaping how we measure policy impacts: Difference-in-Differences (DID). Yet, this acronym sparks confusion. To an economist, “DID” means a rigorous quasi-experimental design. To a psychologist, it means Dissociative Identity Disorder. This article untangles the knots. We’ll explore the econometric DID through the lens of real-world events like Exxon Valdez, decode its assumptions, and reveal why its misuse is now a crisis in economics. By the end, you’ll understand not just where Exxon Valdez happened, but how we isolate its true impact from the noise of time.

What Exactly is Difference-in-Differences (DID)?

Imagine you want to know if a new law reduced crime. You could compare crime rates before and after the law in the city that passed it. But what if crime was already falling nationwide due to economic growth? That’s a time trend—a change that would have happened anyway. To isolate the law’s effect, you need a control group: a similar city that didn’t pass the law. You then compare the difference in crime changes between the two cities over time. That’s the “difference-in-difference.” DID’s name literally describes its core logic: the double subtraction of averages to cancel out confounding trends.

The brilliance of DID is its simplicity in handling unobservable factors. As key insights reveal, the method works because it “从全部效应中剔除‘时间趋势’…此时我们需要一个控制组去衡量这一‘时间趋势’” (removes the influence of ‘time trends’ from the total effect… we need a control group to measure this ‘time trend’). These time trends include everything from seasonal cycles to broad economic shifts—factors you can’t fully list or measure. By assuming the control group would have followed the same trend as the treated group in the absence of the policy, DID creates a credible counterfactual. The Exxon Valdez spill is a perfect case: to assess its long-term economic impact on Alaskan fisheries, researchers might compare fishing communities in Alaska (treated) to similar ones in British Columbia (control), examining outcomes before and after 1989.

The Core Assumption: Parallel Trends (And Why It’s Untestable)

The entire DID edifice rests on a single, fragile assumption: parallel trends. This means that in the pre-treatment period, the average outcome for the treatment and control groups would have followed parallel paths had the intervention not occurred. Crucially, this assumption is fundamentally untestable after the intervention. You can only check if trends were parallel before the policy—a necessary but not sufficient condition. As the guidelines note, the “反事实基础与不可直接检验” (counterfactual foundation and not directly testable) nature of parallel trends is a philosophical and practical hurdle.

Consider Exxon Valdez again. Suppose we want to study its effect on tourism revenue. We pick a control region with similar pre-1989 tourism trends to Alaska. But what if Alaska’s tourism was about to surge due to a hidden, upcoming marketing campaign? The parallel trends assumption would be violated, and our DID estimate would be biased. This is where event study designs (a DID extension) help. They plot the dynamic treatment effects over many leads and lags. If pre-treatment coefficients are statistically zero and the treatment effect jumps only after the event, it bolsters confidence. However, as the push for rigor states, journals now demand explicit tests and discussions of this assumption. You must justify your control group choice with subject-matter knowledge, not just statistical convenience.

Why DID Doesn’t “Solve” Endogeneity (And What Does)

A common misconception is that DID magically fixes all endogeneity problems. This is false. As clarified, “双重差分法作为一种计量模型，其本身不解决内生性问题…本质上仍然依赖于干预或政策冲击本身的外生性” (DID, as an econometric model, does not solve endogeneity problems… it fundamentally relies on the exogeneity of the intervention or policy shock itself). DID controls for time-invariant unobservables (via group fixed effects) and common time shocks (via time fixed effects). But it cannot fix time-varying confounders that differentially affect treatment and control groups after the intervention.

For example, if the Exxon Valdez spill coincided with a federal law that only affected Alaska (e.g., new safety regulations for oil tankers in state waters), the DID estimate would conflate the spill’s effect with the law’s. To address this, researchers sometimes combine DID with Instrumental Variables (IV). The IV must satisfy relevance (correlated with treatment) and exogeneity (affects outcome only through treatment). Finding a valid IV is notoriously hard. The key takeaway: DID is a design, not a cure-all. Its validity hinges on the exogenous nature of the policy shock—meaning the timing and targeting of the intervention are as-good-as random with respect to potential outcomes. If policymakers target regions based on need or political pressure, DID estimates can be severely biased.

The Rise (and Risk) of DID in Modern Economics

DID has exploded in popularity. 当本科生开始大面积使用 DID 进行研究时, 对经济学研究这到底意味着什么？ (When undergraduates start widely using DID for research, what does this mean for economics?) It means democratization and danger. Tools like did and eventstudyinteract in Stata/R have lowered technical barriers. But this leads to “plug-and-play” empirics where researchers apply DID without deeply interrogating its assumptions. Studies show that a majority of published DID papers in top journals now use two-way fixed effects (TWFE) models with staggered treatment adoption—a setting where standard DID can produce severely biased estimates if treatment effects are heterogeneous.

The economics community is responding. The Journal of Economics and others now require authors to conduct “placebo tests” (e.g., fake treatment dates) and “event study” graphs to pre-validate parallel trends. There’s also a push toward alternative estimators like Callaway and Sant’Anna (2021) or Sun and Abraham (2021) that handle staggered adoption better. The trend is clear: DID is here to stay, but its application is becoming more nuanced and demanding. For a budding economist, learning DID isn’t just about running a regression; it’s about understanding the causal identification strategy behind the numbers.

Multi-Period DID: A Practical Step-by-Step Guide

Most real-world policies roll out over time, not in one big bang. 多期数据DID操作指南 (Multi-period DID operational guide) becomes essential. Here’s a systematic approach:

Define Groups and Periods: Identify treatment and control units. Map out exactly when each unit first receives the treatment (the “adoption cohort”).
Choose the Estimator: For classic two-group/two-period DID, use a simple interaction term. For staggered adoption, avoid the naive TWFE regression. Instead, use an estimator that accounts for effect heterogeneity, such as the group-time average treatment effects framework.
Run an Event Study: Plot coefficients for leads and lags relative to first treatment. This checks pre-trends and shows dynamic effects. Use eventstudyinteract in Stata or did in R.
Aggregate to an Overall Effect: Sum the group-time effects using appropriate weights (e.g., never-treated groups as controls) to get an Average Treatment Effect on the Treated (ATT).
Conduct Robustness Checks: (a) Placebo Tests: Assign fake treatment dates. (b) Alternative Control Groups: Use only never-treated units. (c) Subgroup Analyses: See if effects differ by region or time.

Example: Studying the impact of the Exxon Valdez Oil Pollution Act (passed in 1990) on oil spill liability across states. Some states had similar laws pre-1990 (early adopters), others adopted post-1990. A multi-period DID would compare changes in liability insurance premiums for oil transporters in states with and without the law, accounting for when each state enacted its own version.

Clearing the Confusion: DID the Method vs. DID the Disorder

Here’s where the acronym collision becomes critical. 精神分裂症英文名叫 Schizophrenia ，俗称的人格分裂症或者说分离性身份识别障碍的英文全称 Dissociative Identity Disorder（DID） (Schizophrenia is the English name for psychosis; the so-called split personality or dissociative identity disorder’s full English name is DID). This is a completely different field. The mental health condition Dissociative Identity Disorder (DID) involves the presence of two or more distinct personality states. It is not related to the econometric method.

多重人格、DID系统在现实中真的极少吗？ (Are multiple personalities, DID systems, really rare in reality?) Clinically, diagnosed DID is rare (prevalence ~1-2% of the population, with higher rates in clinical settings). 国内对多重人格的定义仍很模糊 (China’s definition of multiple personality remains vague), often conflating it with schizophrenia. Research on DID’s neurobiology is nascent but suggests “双侧CA1亚区体积的减少是DID患者分离性遗忘的生物标志物” (bilateral CA1 subfield volume reduction is a biomarker for dissociative amnesia in DID patients). This is a medical topic, entirely separate from causal inference.

Feature	DID (Econometrics)	DID (Psychology)
Full Name	Difference-in-Differences	Dissociative Identity Disorder
Field	Econometrics, Economics, Epidemiology	Clinical Psychology, Psychiatry
Core Idea	A research design to estimate causal effects	A mental health condition involving identity fragmentation
Key Output	Average Treatment Effect (ATT)	Diagnosis, symptom management, therapeutic outcomes
“Parallel Trends”	A critical, untestable assumption	Not applicable
Typical Data	Panel data (units over time)	Clinical interviews, psychological assessments

When writing or reading, context is everything. If someone says “I’m using DID,” ask: “In a regression or in a therapy session?”

Conclusion: The Real Location of Exxon Valdez in Research

The Exxon Valdez spill physically happened in Prince William Sound, Alaska. But conceptually, it lives forever in the difference-in-differences framework. It’s a classic case study for teaching how to separate a shock’s effect from the relentless march of time. This method’s power—and peril—lies in its reliance on the parallel trends assumption and the exogeneity of the shock. As DID usage surges, so does the responsibility to apply it with rigor, not just routine. The location they don’t want you to see isn’t a hidden cove in Alaska; it’s the lazy application of a sophisticated tool without checking its foundational bricks. Whether you’re an undergraduate writing your first paper or a senior researcher, remember: DID is a lens for causal vision, not a magic wand. Wield it with care, question your control group, and never stop testing your assumptions. The true impact of any event—from an oil spill to a new law—is only as credible as the counterfactual you build.

“Blindingly stupid” Global warming did not cause Exxon Valdez spill – CFACT

Exxon Valdez Spill

Exxon Valdez Oil Spill