Disentangling the Layers

Abstract

Observed economic and social outcomes are essential evidence, but they are not self-explanatory. Similar levels of output, adoption, employment, compliance, or institutional performance can arise from very different combinations of technology, policy, institutions, and values. Analysis therefore has to move beyond the visible result and identify the structure that produced it.

This essay sets out a fit-for-purpose methodology for that task. Economic activity forms the analytical core. Technology becomes embedded in production and investment, while policy, institutions, and values condition the choices, constraints, incentives, expectations, and legitimacy surrounding that activity. At any moment, an activity operates under one realized configuration of these layers. Because the layers may evolve at different speeds, periods of mismatch are often as important as the equilibria on either side. The goal is not to build one model containing everything. It is to select the smallest analytical structure capable of revealing the relevant mechanism, and to endogenize additional layers only when their feedback is essential to the question.

1. Observed Outcomes Are Not Self-Explanatory

Economic and social analysis begins with what can be observed. Output rises or falls. A technology spreads. A tax changes behavior. Employment remains stable. An institution performs differently across countries. A social group responds differently from another. A political arrangement survives, adapts, or breaks.

These outcomes establish the puzzle. They do not by themselves explain it.

Stable output may reflect healthy productive capacity. It may also reflect temporary compensation, accumulated capital, external support, intensified use of a hidden stock, or a technology that preserves current production while reducing the human participation needed to reproduce future capability. Similar formal institutions may perform differently because enforcement, local capacity, expectations, or legitimacy differ. A common policy may produce heterogeneous effects because individuals differ in beliefs, preferences, resources, authority, or exposure.

The purpose of aggregation is to make patterns visible. Aggregate data can reveal that something systematic is happening and help establish the scale, direction, and persistence of a change. The mistake is not aggregation itself. The mistake is treating the aggregate relationship as the complete explanation.

Observed data are evidence of the system’s behavior, not a complete account of its structure.

Our method therefore begins with a second set of questions. What activity produced the observed outcome? Under what technological, policy, institutional, and social conditions did it occur? Which of those conditions changed? Which relationships remained approximately stable? What mechanism could generate the pattern, and what evidence would distinguish that mechanism from plausible alternatives?

Disentangling is the movement from the outcome back toward those underlying relationships.

2. Economic Activity at the Core

We place economic activity at the analytical core. This includes production, investment, exchange, allocation, work, consumption, organization, learning, maintenance, and capability formation. It is where agents make choices, combine resources, respond to incentives, and generate the outcomes that later appear in statistics and public debate.

Placing economic activity at the core does not mean that every social outcome is reducible to a narrow economic calculation. It means that the surrounding layers become analytically concrete when we identify how they alter choices, constraints, production possibilities, information, coordination, or expected returns.

Technology occupies a distinctive position. Once adopted, it becomes part of economic activity itself. It may enter as a machine, software system, production process, body of knowledge, organizational routine, or new feasible task. Acquiring it usually requires investment. Using it may require complementary capital, training, maintenance, and organizational redesign. Technology can therefore be analyzed as a productive input, a form of capital, a multiplier of existing inputs, a substitute for some activities, or a complement to others.

Technology can also arrive as a shock to the current configuration. It may change productivity without changing the structure of the activity. It may create a distinct production mode, such as human-only versus AI-assisted work. At a larger scale, it may alter the relevant relationships themselves by changing who performs the task, who holds decision authority, what information is available, or whether current production still generates learning and future capability.

Policy, institutions, and values condition this core from different directions. Policy affects prices, transfers, permissions, enforcement, and operational rules. Institutions define more durable arrangements of rights, authority, responsibility, ownership, enforcement, and dispute resolution. Values and norms originate in individual beliefs, preferences, identities, and expectations, but become socially consequential through their distribution and aggregation.¹

We use these categories to locate a mechanism. We do not allow them to dictate the analytical method in advance.

A tax rate may be treated as a continuous policy parameter. A change in the tax base may alter the structure of the choice problem. An institution may define a different rule of the game, while an aspect of institutional operation, such as enforcement intensity or court delay, may vary continuously within a stable regime. Values may enter a choice model as measurable preference weights, beliefs, or social costs. They may also be represented as distributions when heterogeneity within a society matters more than the average.

The central question is not what label belongs to the variable. It is what the variable does to the economic activity being studied.

3. One Realized Configuration at a Time

Every economic activity takes place under one realized configuration of the relevant layers at a given time. A firm operates with a particular technology, under a particular policy environment, within a particular institutional structure, and among people with a particular distribution of beliefs and expectations.

The configuration may be contested, unstable, or changing. It does not have to represent a permanent equilibrium. It is simply the environment within which current choices are being made.

This observation provides a practical starting point for research. We first identify the activity and the configuration surrounding it. We then determine which element is changing and which elements can reasonably be treated as fixed over the horizon of the question.

That last step is an analytical claim, not a neutral convenience. A comparative static is meaningful only if the relationships held constant are sufficiently stable for the proposed change. A model that varies a tax rate may reasonably preserve the surrounding legal and administrative system. A model of a new property-rights regime cannot automatically assume that the same choices, expectations, and enforcement relationships remain intact.

The same applies to technology. A gradual improvement in a familiar production process may be represented as a change in a productivity measure. A new technological architecture may require a categorical comparison. A technology that changes the division of labor, the information structure, or the reproduction of skill may require different equations on either side of the change.

Defining the realized configuration does not mean deriving the entire history that produced it. Often the right research strategy is to declare the relevant technological, policy, institutional, and value conditions, analyze behavior within that setting, and compare the result with another meaningful configuration. The transition between the two can be studied separately when it is itself the substantive question.

This modular approach is less ambitious than a universal model of coevolution. It is often more transparent about what the analysis can and cannot establish.

4. Different Layers, Different Clocks

The layers of a system need not adjust at the same speed.²

Economic activity can respond quickly to changes in prices, demand, finance, and opportunity. Technological capability can improve rapidly at the frontier, although adoption and complementary investment may take much longer. Policy can sometimes change through a single administrative or legislative act, but implementation may lag. Institutions are often slower because they are embedded in law, authority, organizational routines, and political coalitions. Values and norms can persist through habits, identities, family transmission, and cohort replacement.

This is a characteristic ordering, not a universal law. Policy can change abruptly. Institutions can collapse during a crisis. Norms can shift rapidly after a shock. A technology may exist for decades before complementary systems make it economically important. The relevant ordering has to be established for the domain being studied.

The nested structure on our homepage represents the modal configuration for many technology-driven questions in the program: the economic core changes first, policy and governance respond later, institutional reconstruction takes longer, and values or public legitimacy may move slowest. It should be read as a common pattern, not as a fixed chronology imposed on every case.

What matters is the possibility of mismatch.

A technology may change production before policy responds. A policy may change before organizations can implement it. Formal institutions may change before people trust or understand the new arrangement. Public dignity may remain attached to an occupation or life path after its material foundation has weakened. Conversely, social values may change before formal institutions recognize the new equilibrium.

These intervals are not merely noise around a final state. They can generate distinct economic and political outcomes. Some mismatches close. Some persist. Some create path dependence, hysteresis, or a threshold beyond which the old arrangement cannot be restored.

The Dislocated Middle develops one such case. A fast-moving material payoff frontier can weaken the economic return to established life paths before formal institutions and public systems of dignity can recognize and legitimate replacements. The result is not simply lower income. It is a population stranded between an old route that remains honorable and a new economic reality that no longer rewards it.

The broader lesson is methodological. When layers move at different speeds, the rate and sequence of adjustment can be part of the mechanism itself.

5. Fit-for-Purpose Analysis

A layered view does not require one grand model of the complete system. It requires a disciplined choice of analytical form.

When a measurable margin changes within a sufficiently stable relationship, a continuous parameter may be appropriate. Tax rates, subsidy levels, enforcement probabilities, financing costs, and some measures of technological efficiency often fit this form.

When the environment has distinct versions, a categorical representation may be more honest. A technology can be absent or present. A firm can operate under two governance systems. A policy can apply under one eligibility rule or another. Cultures or social groups can be compared through different distributions of measured beliefs and preferences.

Sometimes a categorical indicator selects different values within a common equation. In other cases, the rule of the game changes enough that the equations themselves must differ. An “if this, otherwise” structure is not an analytical failure. It can be the clearest way to represent a binding threshold, a corner solution, a new authority structure, or a different production process.

Heterogeneity may also be the central object. Values and norms are aggregates of individual beliefs, but the aggregate mean is not always sufficient. Behavior may depend on the variance, polarization, tail mass, or correlation of beliefs with income, education, generation, capability, or institutional position. A cross-cultural comparison may focus on differences between distributions. A within-society analysis may ask why the same policy or technology produces different choices across groups.

Dynamic analysis is needed when adjustment paths matter. Technology may evolve as a time-dependent state. Capital accumulates. Learning occurs through practice. Institutional expectations adapt. Cohorts replace one another. A shock may have a different short-run and long-run interpretation.

The correct structure depends on the mechanism, not on the prestige of the method.

Our guiding rule is simple: select the smallest structure capable of representing the mechanism relevant to the question. Complexity is justified when it changes the explanation or the observable implications. It is not justified merely because another layer could, in principle, be added.

6. Selective Endogenization

No layer is inherently exogenous.

Production can change future capability through learning by doing.³ Economic incentives can shape research investment and technological progress. Firms and organized groups can influence policy through lobbying. Repeated failures can generate institutional reform. Behavior can change beliefs and norms. Social values can alter policy coalitions, and policy can alter the experiences through which values are transmitted.

The question is not whether these feedbacks exist. The question is whether a particular feedback must be modeled to explain the phenomenon at hand.

This is the principle of selective endogenization.

A layer should be endogenized when its response is part of the mechanism or when treating it as fixed would overturn the conclusion. It can remain a declared condition when the paper asks a narrower, conditional question.

The distinction is important because endogenizing several layers at once imposes a rapidly increasing burden. Each layer needs a behavioral rule, an adjustment law, measurable state variables, and a defensible relationship to the other layers. A mathematically coherent system can still become empirically weak if the most important variables are latent or only loosely proxied.

An earlier modeling exercise in this program made values, practices, and institutions coevolve. That exercise captured an important possibility: who adopts can change what is rewarded, what is rewarded can change institutions, and institutions can reshape which types become influential. It also made the empirical problem clear. Practices are often observable, institutional dimensions require imperfect proxies, and value orientations are slow and latent. A unified dynamical system can clarify theoretical feedback while becoming difficult to identify and measure in a specific application.

The lesson is not to abandon endogeneity. It is to earn it.

7. From a General Architecture to a Tractable Mechanism

The development of Masked Erosion illustrates the fit-for-purpose approach.

The broader research question concerned layered systems in which fast visible output could diverge from slower processes of reproduction and institutional adaptation. One possible strategy was to model production, reproduction, policy, institutions, and values together. That would have preserved the full architecture, but it would also have required assumptions about how every layer adjusted and how each latent object could be measured.

The paper instead isolated one tractable mechanism. Current performance depended on an underlying stock and on the intensity with which that stock was used. Future stock depended positively on the inherited stock and negatively on persistent intensity. This produced a horizon at which current performance changed from an informative signal of future health into a misleading one.

Technology and institutions remained essential, but they did not have to be fully endogenized. Technology could alter the production-reproduction mapping. Institutions could continue applying measurement and reward rules developed under the previous mapping. The analytical core established the signal problem; alternative regimes interpreted why it emerged and why correction could lag.

This division of labor made the paper narrower and stronger. It did not explain every institutional origin or transition. It established one result that could travel across fisheries, organizational capital, infrastructure, safety systems, and AI-assisted knowledge work.

Our experience suggests a general research sequence. Begin with the substantive puzzle. Identify the layers and their characteristic horizons. Determine which relationship generates the observed pattern. Formalize that relationship at the minimum necessary scale. Add feedbacks only when they are required for the result. Then return to the broader layers for interpretation, empirical design, and policy relevance.

A framework organizes the inquiry. It should not force every inquiry into the same model.

8. Implications for Policy and Institutional Analysis

For policy analysis, the first task is to classify the intervention. Does it adjust a measurable margin within the existing configuration, or does it change authority, information, feasible actions, responsibility, or another rule of the game? The first case may support comparative statics. The second may require a different constraint set, game, or institutional regime.

Institutions often define the structure of interaction, but they also contain operating margins. Enforcement intensity, transparency, administrative capacity, court delay, and monitoring frequency can vary within an otherwise stable design. We therefore distinguish institutional redesign from variation in how an institution operates.

Values and norms can also enter economic analysis in more than one form. Measurable beliefs and preferences may enter individual choice models, while their distributions can support comparisons across societies and groups. Their interaction with economic position may explain heterogeneous responses. When values shape legitimacy or determine whether an arrangement is socially feasible, they may instead operate as a boundary condition on the institutional regime.⁴

This approach clarifies what kind of intervention is being proposed, what relationships are assumed to remain stable, which responses may unfold on different horizons, and where measurement should look for early evidence of mismatch. It also disciplines policy claims: a result derived within one configuration should not automatically be treated as invariant after the intervention changes that configuration.⁵

9. Strengths, Limits, and Guardrails

The principal strength of the layered approach is disciplined flexibility. It can accommodate formal models, comparative statics, heterogeneous-agent analysis, institutional comparison, historical evidence, and case studies without insisting that every question use the same structure.

It also reduces false precision. Institutions and values do not have to be assigned arbitrary coefficients merely to complete a simulation. They can be modeled more directly when measurement supports it, treated as alternative regimes when they change the rules, or held fixed when the analysis is explicitly conditional.

The approach has limits. Fit-for-purpose modeling requires judgment. Researchers may disagree about which relationships are stable, whether a change is parametric or structural, and which layer must be endogenized. Regime comparisons can describe different systems more easily than they explain the political or social transition between them. Latent values and informal institutions remain difficult to measure even when the conceptual role is clear.

There is also a danger of convenience. A researcher could invoke an unobserved layer or a regime change whenever the original model fails. That would turn flexibility into insulation from evidence.

The guardrail is falsifiability. Layers and regimes should be specified before they are used to explain the result. Each proposed mechanism should generate observable implications, distinguish itself from alternatives, and identify evidence that would weaken it. A hidden layer cannot function as a post hoc rescue device.

The framework is strongest when it clarifies the boundary of a claim. It should tell the reader not only what is modeled, but what is being held fixed, why that choice is reasonable, and when the conclusion should no longer be expected to hold.

10. Disentangling Before Explaining

Economic and social systems are not flat collections of variables. They are layered arrangements in which economic activity operates under particular technological, policy, institutional, and social conditions. These conditions influence one another, but they do not necessarily move together or enter every problem in the same form.

Observed outcomes remain the starting point. They show where the puzzle is. The next task is to disentangle the configuration that produced them, identify the changing layer and the channel through which it reaches the economic core, and choose an analytical structure proportionate to the mechanism.

Some questions require a continuous parameter. Some require categorical environments or different equations. Some depend on heterogeneous distributions. Some require dynamics, feedback, or endogenous transition. No single method belongs permanently to technology, policy, institutions, or values.

Our purpose is not to model everything at once. It is to see enough of the structure to explain the outcome without mistaking the visible pattern for its cause.

That is the methodological commitment of the research program: use observed outcomes as evidence, disentangle the layered conditions beneath them, and select a fit-for-purpose analytical structure to reveal the mechanism and structural pattern at work.

Endnotes

1. Douglass C. North, “Institutions,” Journal of Economic Perspectives 5, no. 1 (1991): 97–112. North defines institutions as formal and informal constraints that structure interaction and emphasizes their incremental evolution through time. Back
2. Oliver E. Williamson, “The New Institutional Economics: Taking Stock, Looking Ahead,” Journal of Economic Literature 38, no. 3 (2000): 595–613. Williamson distinguishes levels of social analysis and separates the institutional environment from governance and resource allocation. Back
3. Kenneth J. Arrow, “The Economic Implications of Learning by Doing,” Review of Economic Studies 29, no. 3 (1962): 155–173. Learning by doing is the classic example of a productive activity changing a future state that might otherwise be treated as external. Back
4. Alberto Alesina and Paola Giuliano, “Culture and Institutions,” Journal of Economic Literature 53, no. 4 (2015): 898–944. Their review emphasizes both the economic relevance of measurable cultural traits and the two-way relationship between culture and institutions. Back
5. Robert E. Lucas Jr., “Econometric Policy Evaluation: A Critique,” Carnegie-Rochester Conference Series on Public Policy 1 (1976): 19–46. The methodological relevance here is the warning that behavioral relationships estimated under one policy rule may not remain invariant when the rule changes. Back