Loading field map
A researcher wants to know whether a job-training program raises earnings. The data show that participants earn more than non-participants, but participants were also more motivated before the program started. How much of the earnings difference is caused by the training, and how much by pre-existing motivation? This is the core problem of microeconometrics: extracting credible causal inferences from individual-level data when the people being studied make their own choices. Two broad methodological schools have emerged to address this problem, each with a different answer to the question of what makes an empirical finding credible.
Structural microeconometrics, which took shape in the 1970s and remains a vigorous tradition, treats economic theory as the primary source of identification. The researcher begins by writing down a formal model of individual behavior—a utility function, a budget constraint, a dynamic optimization problem—and then uses data to estimate the deep parameters of that model: preferences, costs, expectations, and constraints. The central commitment is that the model itself, not a particular research design, tells the analyst which features of the data are informative about cause and effect.
Daniel McFadden's work on discrete choice in the 1970s exemplifies this approach. McFadden modeled a commuter's choice among travel modes as a utility-maximizing decision with random taste shocks. By assuming the shocks followed an extreme-value distribution, he derived the conditional logit model, which allowed the estimation of how travel time, cost, and other attributes influenced mode choice. The model's parameters had a direct economic interpretation: the value of time, the elasticity of demand. More importantly, the model could be used to predict behavior under new conditions—a new subway line, a congestion toll—that had never been observed. This capacity for counterfactual prediction is the hallmark of structural microeconometrics.
John Rust's 1987 model of bus engine replacement pushed the approach further into dynamic settings. Rust observed that a bus depot manager decided when to replace a bus engine based on current mileage and expected future costs. He estimated the manager's discount factor and the cost of replacement by solving the dynamic programming problem the manager faced. The structural parameters allowed Rust to simulate how replacement behavior would change under different maintenance policies. The method required strong assumptions—that the manager's expectations were rational, that the cost structure was stable—but it delivered a quantitative answer to a policy question that a reduced-form regression could not address.
Structural microeconometrics is defined by its reliance on theory-based identification. The causal parameter is not a simple treatment effect but a structural parameter embedded in a behavioral model. Functional form assumptions are explicit and often parametric, chosen for computational tractability and economic interpretability. The approach is most dominant in industrial organization, where researchers estimate demand systems, entry games, and auction models, and in labor economics for dynamic models of human capital and life-cycle behavior.
Design-based microeconometrics emerged in the 1990s as a direct response to the perceived fragility of theory-driven identification. If the structural model is misspecified, the estimated parameters may be economically meaningless. The design-based school argued that credible inference should come from the research design—the way data are generated—rather than from a theoretical model. The core intellectual engine is the Potential Outcomes Framework, formalized by Donald Rubin and extended by James Heckman, Joshua Angrist, Guido Imbens, and others.
In the potential outcomes framework, each individual has two potential outcomes: one under treatment and one under control. The causal effect for that individual is the difference between these two potential outcomes. The fundamental problem of causal inference is that we never observe both potential outcomes for the same person. Design-based microeconometrics solves this problem by finding or creating a comparison group that is credible because of how it was formed, not because of a model of behavior.
The most influential design-based methods include randomized experiments, difference-in-differences, regression discontinuity, and instrumental variables. In each case, the source of identification is a feature of the assignment process—a lottery, a cutoff rule, a policy change—that mimics randomization. Angrist and Krueger's 1991 study of returns to schooling used quarter of birth as an instrument for educational attainment, exploiting the fact that compulsory schooling laws create a natural experiment. The identifying assumption—that quarter of birth affects earnings only through education—was controversial, but the logic of the argument was transparent and testable.
Design-based microeconometrics prioritizes internal validity: the ability to credibly estimate a causal effect for the specific population and setting under study. The causal parameter is typically a local average treatment effect (LATE) for the subpopulation whose treatment status is affected by the instrument or cutoff. Functional form assumptions are minimal; the emphasis is on nonparametric or semiparametric methods that impose little structure beyond the design. The approach has been most influential in labor economics, development economics, and public economics, where researchers often evaluate specific programs or policies.
The two frameworks differ fundamentally on the source of identification. Structural microeconometrics identifies parameters through the theoretical model: the model's assumptions about preferences, technology, and expectations pin down what the data can reveal. Design-based microeconometrics identifies causal effects through the research design: the assignment mechanism, not a behavioral model, provides the identifying variation.
They also differ on the role of functional form assumptions. Structural models typically require parametric or semiparametric assumptions for computational feasibility and economic interpretation. Design-based methods strive to minimize such assumptions, relying instead on design features like discontinuity thresholds or instrument exogeneity. A structural model might assume a specific utility function; a design-based study might assume only that the density of the running variable is continuous at the cutoff.
The definition of the causal parameter differs sharply. In structural microeconometrics, the parameter of interest is a deep structural parameter—a preference parameter, a technology parameter—that is invariant to policy changes. In design-based microeconometrics, the parameter is a treatment effect for a specific subpopulation, often local to the instrument or cutoff. The structural parameter supports broad counterfactual predictions; the design-based parameter supports internally valid inference for a specific intervention.
External validity is a point of tension. Structural models are built for extrapolation: they can predict outcomes under new policies because the estimated parameters are supposed to be invariant. Design-based estimates are internally valid but may not generalize to other populations or settings. A structural model of demand can predict the effect of a new tax; a design-based estimate of a job-training program may only apply to the specific site and year studied.
Today, structural and design-based microeconometrics coexist as living traditions, each with its own domain of dominance. Structural methods are standard in industrial organization, where researchers need to predict the effects of mergers, entry, and regulation. Design-based methods dominate labor and development economics, where the focus is on program evaluation and causal inference from natural experiments.
There is also a growing zone of synthesis. Researchers increasingly use design-based methods to validate structural models, or embed design-based estimates within structural frameworks to improve external validity. For example, a structural model of consumer demand might be estimated using variation from a regression discontinuity design, combining the internal validity of the design with the counterfactual capacity of the model. This synthesis is the defining frontier of the subfield.
What the two frameworks agree on is that credible inference requires explicit assumptions and transparent reporting. Both traditions have pushed the field toward greater rigor in specification testing, sensitivity analysis, and replication. Where they disagree is on the role of theory: whether the model or the design should be the primary source of identification. This disagreement is not a sign of weakness but a productive tension that drives the field forward. A student entering microeconometrics today must understand both frameworks, because the best applied work often draws on the strengths of each.