The Intercept Is Where Causality Lives: A Structural Turn in Economics
Economics happens in the intercept
Intercept (n.): The constant term in a regression equation, representing the predicted outcome when regressors are zero. In economics, it captures the underlying baseline that shapes what is possible before any policy or intervention is applied.
Introduction
Two recent papers, read together, reveal a foundational problem in how economics understands causality.
Carlos Chávez’s Why Macro Never Had a Credibility Revolution argues that macroeconomics never experienced the credibility revolution that transformed microeconomics, not because it lacked identification strategies, but because its models abstract from how the economic environment responds to policy. VoxDev’s The Missing Intercept Problem shows why micro-level causal estimates fail to scale: the statistical “baseline” assumed to be fixed in experiments shifts systematically when policies expand.
Individually, these papers critique different failures – macro misspecification and the micro-macro disconnect. Together, they expose a common root cause: economics repeatedly holds fixed, in its models, the very structures that move in reality when policy changes.
That structure is what econometric models label the intercept – and it is where causality actually lives.
The Intercept Is the System
In standard regressions, the intercept is treated as a nuisance parameter – a baseline level of outcomes to be controlled for or differenced away. But in reality, it represents the inherited configuration of institutions, infrastructure, state capacity, political coalitions, spatial networks, and macroeconomic regimes within which policy operates.
These features are not passive background conditions. They determine which constraints bind, which responses are feasible, and how agents and institutions react when policy changes. When a policy scales, it alters this configuration. Prices adjust, administrative capacity is tested, political incentives shift, and fiscal and monetary authorities respond. These changes are not statistical noise. They are the core of the causal process.
Both macro and micro approaches abstract away from this structural evolution, misplacing causality.
Why Average Effects Mislead
This problem is most visible in cross-country analysis, where estimated coefficients– βs – are interpreted as general policy effects. For non-specialists: β represents the average effect of a policy on an outcome across diverse economies. The difficulty is that these averages are taken across economies with fundamentally different histories, institutions, and capacities. No country operates at the mean of these incompatible systems. As a result, the estimated effect often describes no actual case.
Development trajectories are path-dependent and non-ergodic (systems in which time averages and cross-sectional averages diverge), ie, systems where history matters, permanently. Early choices – colonial institutions, infrastructure placement, patterns of state formation – generate feedback loops that persist for decades. These inherited structures shape which constraints bind and which policies can succeed. This is why, as Dani Rodrik and Ricardo Hausmann have long emphasized, development hinges on identifying and relaxing binding constraints in specific contexts rather than importing “best practices” or average coefficients. The dominant empirical strategy averages across these differences instead of explaining them.
The same failure persists even when economists abandon cross-country comparisons and move to tightly controlled experiments.
Scaling and Non-Invariance
Randomized controlled trials improve internal validity, but they do not resolve the problem of scale. Local Average Treatment Effects (LATE) are identified under particular prices, institutional arrangements, and implementation conditions. When a policy expands, those conditions change. Labor markets adjust. Input bottlenecks emerge. Administrative limits bind. Political coalitions reorganize. Macroeconomic policy reacts. These responses are not implementation failures; they are structural reactions to scale. They are absorbed into what econometric models label “intercept.”
This is the dominant source of non-invariance in development policy. The instability does not primarily lie in β – the behavioral response estimated locally – but in the environment in which that response occurs. This might resemble the Lucas Critique, but its focus is inverted. Lucas warned that agents' behavioral rules (β) change when policy regimes change. The problem here is that the structural environment itself (the intercept) changes when policy scales. When this environment shifts, estimates that assume it is fixed lose their relevance.
This instability at scale connects directly to the “superpopulation” critique below: even if countries were comparable, scaling alters the structural environment itself.
From Superpopulations to Structure
This argument connects to the fundamental “superpopulation” problem in development economics (Ballinger, 2011). Researchers routinely treat countries as if they were random draws from an infinite and varying population – a statistical fiction that undermines inference from the outset.
The intercept problem reveals a deeper failure that persists even if one grants this assumption. Scaling policy changes the structural environment itself. Even a perfectly estimated parameter can’t be extrapolated when the system it depends on evolves.
Taken together, these critiques point to the same conclusion: economic outcomes are generated by historically specific systems, not by stable parameters acting on interchangeable units.
A Practical Implication: Diagnosing What Moves
Recent work by Hausmann and collaborators on binding constraints and economic complexity provides a concrete way forward. Rather than asking which policies work on average, this approach asks which specific constraint, in a given inherited structure, prevents reconfiguration.
In these terms, the intercept is not shifted all at once. It moves when particular constraints – logistics, energy, skills, finance, institutional capacity – are relaxed in sequence, changing the environment in which subsequent policies operate. This reframes development policy as a process of structural transformation rather than parameter optimization.
Conclusion
Economics struggles with credibility when it treats the intercept as noise. The intercept is the evolving structure – history, institutions, geography, state capacity, and macro response – that determines whether policy effects materialize, persist, or dissipate.
Countries do not develop by importing coefficients from cross-country regressions. They develop by transforming inherited structures, relaxing binding constraints to open new feasible trajectories. That process lives in the intercept.
In economics, the intercept is not where the error term hides.
It is where causality lives.
FIN
Cited:
Ballinger, Clint, 2011. “Why Inferential Statistics are Inappropriate for Development Studies and How the Same Data Can Be Better Used.” Link
Chávez, Carlos. 2024. “Why macro never had a credibility revolution.” Dec 30. Link
Moll, B. and Hanney, O., 2025,. The ‘missing intercept’ problem with going from micro to macro. VoxDev Oct 4 Link
Really sharp reframing of where the real action is in development economics. The insight that the intercept isn't noise but the evolving structural system itself is kinda obvious once stated but somehow gets burried in most empirical work. I've seen this exact pattern in smaller scale contexts where pilot programs work beautifully and then fall apart at scale, not because the intervention failed but because the enviroment it operated in fundamentally changed.