Establishing causality, rather than mere correlation, is the central methodological challenge in applied economics. Over the past few decades, the field has developed a well-established toolkit of identification strategies, each suited to different data-generating contexts. Knowing which technique fits your setting — and being able to justify that choice explicitly — is often what separates a convincing paper from a merely suggestive one.
Instrumental Variables (IV)
IV estimation addresses endogeneity by using a variable that affects the outcome only through its effect on the endogenous regressor — never directly. The credibility of an IV design rests almost entirely on the exclusion restriction, which cannot be tested directly and must be argued persuasively based on institutional knowledge of the setting.
Difference-in-Differences (DiD)
DiD compares changes in outcomes over time between a treated group and a comparable untreated group, isolating the treatment effect under the key assumption of parallel trends — that both groups would have evolved similarly absent the treatment. Recent methodological advances have significantly refined DiD estimation in settings with staggered treatment timing across units, an area that saw substantial technical development in the past several years.
Regression Discontinuity Design (RDD)
RDD exploits a sharp cutoff in treatment assignment (an eligibility threshold, an exam score cutoff) to compare outcomes for units just above and just below the threshold, under the assumption that units near the cutoff are otherwise comparable. RDD offers strong internal validity but applies only locally, near the threshold — a limitation on generalizability worth addressing explicitly.
Matching Methods
Matching techniques (propensity score matching, coarsened exact matching) construct comparison groups with similar observable characteristics to the treated group, addressing selection on observables. Matching does not address selection on unobservables, a limitation that should be acknowledged directly rather than implied away.
Synthetic Control Methods
Particularly useful in settings with a single treated unit (a country, a region) and multiple potential comparison units, synthetic control constructs a weighted combination of untreated units that closely tracks the treated unit’s pre-treatment trajectory, providing a credible counterfactual for aggregate-level policy analysis.
Randomized Controlled Trials (RCTs)
Where feasible, RCTs remain the gold standard for causal identification, since random assignment directly addresses both observable and unobservable confounding by design. Field experiments have become increasingly common in applied economics, though feasibility, cost, and ethical constraints limit their applicability to many important questions.
Choosing the Right Technique
The right method depends entirely on your specific institutional setting: What variation exists in your data? Is there a plausible instrument, a policy discontinuity, a staggered rollout, or an opportunity for randomization? Strong applied papers typically justify their identification strategy based on genuine features of the setting, rather than reaching for whatever technique is currently fashionable in the literature.
What Reviewers Look For
Beyond correct implementation, reviewers scrutinize whether the paper’s key identifying assumption is explicitly stated, plausibly justified given the institutional context, and where possible, supported by falsification or placebo tests that would fail if the assumption didn’t hold.
For a widely used and rigorous reference on modern causal inference methods in economics, Scott Cunningham’s Causal Inference: The Mixtape is freely available online and frequently cited in graduate econometrics training.
Applying a causal inference strategy in your latest paper? Check the journal’s scope and submit through the paper submission page.