How The World Works | Second-Order Effects

A bridge shortens the journey.

That is the first consequence.

Then people change where they live.

Businesses change where they locate. Land values change. Traffic redistributes. New developments become viable. Public transport demand changes. Another road becomes the bottleneck. A place that felt distant becomes ordinary commuting territory.

The bridge did more than shorten one journey.

It changed decisions made because the journey became shorter.

Those are second-order effects.

They are what happens after the thing you were looking at has already happened.

Quick Read

A second-order effect is a consequence caused by an initial consequence rather than directly by the original action.

If A causes B, and B changes behaviour or system conditions in a way that causes C, then B is the first-order effect and C is a second-order effect.

A → B → C

The chain can continue to third-order effects and beyond.

Second-order effects can arise through:

behavioural adaptation;
substitution;
feedback;
new incentives;
displacement to another place or time;
rebound after efficiency improves;
changed expectations;
capacity moving to a new bottleneck;
new complementary investments;
institutional responses.

OECD systemic-thinking work repeatedly emphasises that interventions in interconnected systems can create unintended consequences, spillovers and trade-offs elsewhere, which is precisely why policy analysis must go beyond the immediate target.

The core question is:

What will people and systems do after the first effect changes their world?

The One-Sentence Answer

Second-order effects work because interventions change conditions, incentives or information, and those changed conditions generate new behaviour whose consequences may reinforce, offset, relocate or exceed the original effect.

First-Order Thinking

First-order thinking asks:

If we do X, what happens next?

This is necessary.

A tax raises the cost of an activity. A new road increases capacity. A scholarship lowers the price of education to the recipient. Automation reduces the labour required for a task. A school adds homework. A company cuts a fee.

Those direct effects matter.

The trouble begins when analysis stops there.

Second-Order Thinking

Second-order thinking asks another question:

And then what?

The price changed. What substitutes appear?

The road became faster. Do people make more trips?

The subsidy increased demand. Does supply expand or does price rise?

The metric became important. Do people begin optimising the metric rather than the underlying objective?

The technology reduced unit cost. Does total consumption increase?

The new safety rule reduced one risk. Did users create a workaround that introduces another?

Second-order thinking treats the receiver as an actor, not a passive endpoint.

The Second-Order Chain

intervention → direct effect → changed incentives / constraints / information → adaptation → indirect effect → new state → further adaptation

The chain matters because the direct effect changes the system that receives the next round of behaviour.

This is also why second-order effects are often nonlinear.

The response is not simply the first effect multiplied by time.

Second-Order Effects Are Not the Same as Externalities

An externality is a cost or benefit that spills outside the original decision or transaction and is not fully reflected in the decision-maker’s incentives.

A second-order effect is defined by causal distance, not by who receives it.

A second-order effect can fall on the original actor.

For example, a household buys a more efficient appliance. Electricity cost falls. The household uses the appliance more often. That extra use is a second-order behavioural response even if no external party is involved.

If the increased energy use also creates pollution costs for others, those downstream consequences can be externalities too.

See How The World Works | Externalities.

Second-Order Effects Are Not the Same as Feedback

Feedback occurs when an output returns to affect the process that generated it.

A second-order effect can arise without a loop.

A new regulation changes product design; the changed design alters supplier demand. That is an indirect effect even if the supplier response never feeds back into the original regulation.

When the indirect effect does feed back, the chain becomes a loop and the dynamics can amplify or damp themselves.

See How Feedback Works.

Behavioural Adaptation: People Do Not Stay Still Inside Policy

A rule, price or technology changes the environment.

People respond.

This response is often the largest source of second-order effects.

Raise the price of one product and consumers substitute another. Add a waiting period and some users leave the process. Introduce a target and workers prioritise measurable tasks. Increase road capacity and travel patterns change. Provide insurance and risk behaviour may change.

A model that treats behaviour as fixed can therefore get the engineering right and the world wrong.

Substitution Is a Second-Order Engine

An intervention makes A more expensive or difficult.

Users move toward B.

The original analysis may have predicted lower use of A and stopped there.

The system outcome depends on what B is.

If B is safer, the policy works better than the first-order estimate.

If B creates another harm, the benefit is smaller.

If B is unavailable, users may bear the full cost instead of switching.

See How The World Works | Substitution.

Rebound: Efficiency Changes Demand

Suppose a technology uses half as much energy per unit of service.

First-order prediction: energy use falls by half.

But the service is now cheaper to use.

People may use more of it.

Total energy use still may fall, but by less than the engineering calculation suggested. In some contexts the behavioural response can be large.

This is rebound.

The lesson is broader than energy:

When you make an activity cheaper, faster or easier, expect more of the activity.

Displacement: Solve It Here, Find It There

A city restricts parking in one district.

Parking pressure falls there.

Drivers begin parking in neighbouring streets.

The original location improves. The wider system may improve less.

This is displacement.

Crime prevention, environmental rules, traffic restrictions, school admissions, advertising regulation and many other interventions can shift behaviour across boundaries rather than eliminate it.

Always widen the observation window after a local success.

Temporal Displacement: The Problem Moves Into Tomorrow

Not all displacement is spatial.

Maintenance can be postponed. Debt payments can be deferred. Students can delay revision. Inventory can be run down. Staff can work overtime.

Today’s problem appears solved.

Tomorrow inherits a larger one.

Second-order thinking therefore needs time horizons.

A measure can improve this quarter by consuming next quarter’s capacity.

The Metric Response

Measure a system and people learn what matters.

Reward the measure and the measure becomes a target.

That changes behaviour.

A school judged narrowly by one examination may shift teaching time toward tested content. A call centre rewarded for short calls may shorten calls in ways that reduce resolution quality. A hospital judged on one waiting-time metric may change queue classification.

The first-order effect is increased attention to the metric.

The second-order effect can be gaming, neglect of unmeasured work or redesign of the workflow around the indicator.

This does not mean metrics are bad.

It means metrics enter the system they measure.

Safety Rules Create Adaptation Too

A new safety rule reduces one hazard.

Users may change behaviour because they feel more protected, or create workarounds if the rule adds too much friction.

The eventual risk therefore depends on both the direct protective effect and the behavioural response.

Good safety engineering observes the complete system after introducing the control.

It does not assume compliance and behaviour remain fixed.

The Capacity Response

Increase capacity at one bottleneck.

Throughput rises.

Then another bottleneck appears.

The first-order effect is relief.

The second-order effect is a change in where constraint lives.

This is why successful interventions can create new complaints. A hospital expands diagnostic capacity and now treatment queues become more visible. A school improves reading and discovers writing is the next limiting skill. A factory accelerates production and logistics becomes the bottleneck.

The problem did not necessarily move because the intervention failed.

It moved because the intervention worked.

Success Creates Its Own Second-Order Problems

A popular service attracts more users.

More users create congestion.

A safe city attracts residents and investment.

Success raises land prices and infrastructure demand.

A strong university attracts applications.

Selection becomes more competitive.

A low-friction platform grows.

Growth creates moderation, governance and reliability problems that did not exist at small scale.

Success changes the environment in which success must continue.

Failure Creates Second-Order Effects Too

A service fails.

Users lose trust.

They create backups, switch providers or avoid the service.

Revenue falls.

Maintenance budgets shrink.

Reliability can fall further.

The first failure becomes a new condition that changes future probability of failure.

This is where second-order effects become feedback and path dependence.

Expectations Are Second-Order Machinery

People do not only react to what happened.

They react to what the event makes them expect next.

A central bank announcement changes expectations before any physical change occurs. A company’s price cut can make customers wait for future discounts. A school’s new grading policy can change study behaviour before the first grade is issued.

The first-order event changes the model inside people’s heads.

The changed model produces second-order behaviour.

Second-Order Effects and Nonlinearity

Indirect responses bend the curve.

An intervention that initially produces a proportional effect may later trigger substitution, rebound or saturation.

That is why the final outcome can be much larger or smaller than the direct engineering effect.

See How The World Works | Nonlinearity.

Second-Order Effects and Complementarity

Adding one input can increase demand for its complements.

Build a rail line and station-area development becomes more valuable. Adopt electric vehicles and charging infrastructure becomes more valuable. Introduce AI tools and verification, data governance and training become more valuable.

The first-order adoption of A creates a second-order investment response in B.

See How The World Works | Complementarity.

Second-Order Effects and Path Dependence

Repeated indirect effects can harden an intervention into a path.

A new technology changes behaviour. Changed behaviour encourages complementary investment. Complementary investment increases switching cost. The technology becomes standard.

The original choice is now surrounded by consequences that reinforce continuation.

See How The World Works | Path Dependence.

Second-Order Effects and Latency

Indirect consequences often arrive later than the first effect.

This makes them easy to miss.

A policy is declared successful after six months because the direct metric improved. Three years later, adaptation and displacement reveal a different total outcome.

Evaluation windows therefore need to match the causal timescale.

See How The World Works | Latency.

Second-Order Effects and Irreversibility

The most dangerous indirect consequence is one that becomes irreversible before we notice it.

A platform design changes social behaviour slowly until a norm becomes entrenched. Infrastructure redirects development until reversal becomes extremely expensive. An environmental change triggers ecological loss that cannot be recreated.

When second-order effects are slow and irreversible, monitoring needs to begin early.

See How The World Works | Irreversibility.

Second-Order Effects and Information Asymmetry

Different actors see different parts of the downstream chain.

A manufacturer sees process changes. Customers see product outcomes. Regulators see aggregated harms. Workers see workload. Investors see financial metrics.

No one observer automatically holds the full second-order map.

This is why system evaluation needs multiple receivers and data sources.

See How The World Works | Information Asymmetry.

Second-Order Effects in Education

A teacher introduces more homework to raise practice volume.

First-order effect: students complete more questions.

Second-order possibilities:

students become more fluent;
students rush and practise errors;
sleep falls;
parents begin helping more;
copying increases;
time shifts away from another subject;
students learn to game the completion requirement;
teachers receive more marking than they can return quickly.

The first-order metric—questions completed—cannot tell us which second-order world emerged.

We have to observe the learner.

The Tuition Example

A parent sees a weak examination score and increases tuition hours.

First-order effect: more instructional time.

But if the original problem was fatigue, anxiety, missing sleep or overloaded scheduling, the extra hours can worsen the learner’s total state.

If the problem was an upstream misconception and the extra time is diagnostic and targeted, the effect can be strongly positive.

The intervention is not the outcome.

The mechanism determines the downstream chain.

The AI Example

AI lowers the cost of producing text.

First-order effect: more text can be produced faster.

Second-order effects can include:

more total content;
higher demand for verification;
changes in what writing skill is valuable;
more spam and low-cost persuasion;
new expectations for response speed;
greater value of trusted provenance;
changes in school assessment design;
new complementary software and workflows.

The technology cannot be understood only by benchmarking output quality.

Lower production cost changes the surrounding ecology.

The Queue Example

A service improves and waiting time falls.

More people begin using it.

Demand rises.

Waiting time partially returns.

The first-order capacity gain created a second-order demand response.

This does not prove the capacity improvement was useless. More people may now receive value.

It means the correct metric cannot be waiting time alone.

The Price-Control Example

A policy lowers the allowed price of a scarce good.

First-order effect: buyers who obtain the good pay less.

Second-order questions include whether suppliers reduce quantity, quality changes, waiting queues appear, rationing shifts to another mechanism or informal markets emerge.

This is why economic analysis follows incentives after the first price change.

The point is not ideological.

It is causal.

The Disclosure Example

A regulator requires more disclosure.

First-order effect: more information becomes available.

Second-order effects depend on usability.

Companies may simplify behaviour to look better on the disclosed measure. Consumers may ignore the information if volume becomes overwhelming. Intermediaries may emerge to interpret it. Standardised disclosure may make comparison easier and intensify competition.

Disclosure changes the information environment, which changes behaviour.

The Infrastructure Example

Infrastructure has unusually long second-order chains because it changes the cost of future decisions.

A station changes accessibility.

Accessibility changes land value.

Land value changes development.

Development changes ridership.

Ridership changes service levels.

The original infrastructure and the later city begin reinforcing one another.

This is how second-order effects become path dependence.

Why Predictions Get Worse With Causal Distance

Each causal step adds uncertainty.

We may know reasonably well that a price rise reduces quantity demanded.

Predicting exactly which substitutes people adopt, how suppliers respond, what new technology appears and how institutions adapt is harder.

Second-order analysis should therefore avoid fake precision.

Use scenarios, ranges and sign predictions where exact magnitudes are not defensible.

The goal is not to predict every consequence.

It is to avoid pretending the first consequence is the last.

Known Unknowns and Unknown Unknowns

Some second-order effects are foreseeable.

Others are surprises.

Systems thinking cannot eliminate surprise.

It can design for correction.

Use pilots where possible. Preserve reversibility. Monitor adjacent outcomes. Maintain channels for frontline evidence. Revisit assumptions. Create stop conditions.

The response to uncertainty should not be omniscience theatre.

It should be adaptive architecture.

The Boundary Problem

Second-order effects disappear when the evaluation boundary is too small.

A company saves money but pushes unpaid work onto customers. A city improves one road while congestion worsens nearby. A school raises one subject score while another subject loses time. A platform reduces moderation cost while users carry more abuse risk.

Local success can coexist with system failure.

OECD systemic-thinking guidance emphasises exactly this problem: isolated solutions can create unintended consequences in interconnected areas.

The Time-Horizon Problem

A policy can look successful at three months and poor at three years.

Or the reverse.

Training can initially reduce productivity before raising it. Preventive maintenance costs money now and avoids failure later. Early-childhood education can have effects that appear slowly. Infrastructure can create development over decades.

Evaluation therefore needs multiple clocks.

Immediate receipt.

Medium-term adaptation.

Long-run structure.

The Second-Order Effects Audit

Define the intervention. What exactly changes first?
Identify the first-order effect. What happens directly?
Ask who adapts. Consumers, firms, workers, institutions, competitors?
Map new incentives. What became cheaper, harder, safer, riskier or more rewarding?
Find substitutes. What behaviour moves elsewhere?
Find complements. What new demand or investment becomes valuable?
Find displacement. Does the problem move across place, time or population?
Find rebound. Does efficiency increase total use?
Find feedback. Does the indirect effect return to amplify or damp the original process?
Find the new bottleneck. What constraint becomes visible after success?
Check metrics. Will people optimise the measure?
Widen the boundary. Who outside the target system is affected?
Extend the time horizon. What appears after adaptation?
Check irreversibility. Could an indirect effect become hard to reverse?
Monitor surprises. What signals would show the causal model was incomplete?
Preserve correction. Can the intervention be adjusted when second-order evidence returns?

When the Second-Order Lens Fails

There is a danger in endlessly asking “and then what?”

Every event has infinitely many downstream consequences if the boundary is wide enough.

Good second-order analysis therefore prioritises material, plausible and decision-relevant chains.

Do not invent remote scenarios merely to sound sophisticated.

Ask which indirect responses have credible mechanisms, meaningful magnitude and realistic time horizons.

Second-Order Thinking Is Not Cynicism

People sometimes use unintended consequences as an argument against doing anything.

“Every intervention has side effects, so leave the system alone.”

That is incomplete because inaction has second-order effects too.

Do nothing about a maintenance backlog and failures accumulate. Do nothing about a weak foundation and later learning becomes more expensive. Do nothing about congestion and location choices adapt around it. Do nothing about misinformation and trust may erode.

Baseline continuation is still a policy path.

Compare the downstream chains of action and inaction.

The Reversible Pilot

When second-order effects are uncertain, reversible experiments are valuable.

Introduce the change at limited scale.

Measure the direct target.

Measure adjacent systems.

Watch for adaptation.

Then decide whether to scale, redesign or stop.

This converts uncertainty about second-order effects into learning while preserving option value.

The Counterfactual After the Intervention

When an indirect effect appears, do not automatically attribute it to the intervention.

Ask what would have happened anyway.

Other trends may have changed at the same time. Competitors acted. Weather changed. Technology improved. Demographics moved.

Second-order thinking still needs causal discipline.

Otherwise every later event becomes part of a story that cannot be falsified.

How Second-Order Effects Connect to the Rest of the World

Nonlinearity: adaptation bends the final response away from simple proportionality.
Substitution: actors replace restricted or expensive routes.
Complementarity: one change can increase demand for supporting inputs.
Feedback: indirect effects can return and alter the original process.
Externalities: downstream consequences can fall outside the original transaction.
Latency: second-order effects often appear later than first-order results.
Path dependence: repeated indirect effects can harden into infrastructure and norms.
Irreversibility: slow indirect effects can close recovery routes before detection.
Information asymmetry: different actors see different parts of the causal chain.
Capacity: successful intervention can move the bottleneck.
Incentives: changed rewards create behavioural adaptation.
Monitoring: evaluation must observe the wider system after intervention.

Questions a Reader Can Now Ask

What is the direct effect?
What changes because of that effect?
Who will adapt?
What new incentive appears?
What substitute becomes attractive?
What complement becomes necessary?
Does the problem move somewhere else?
Does efficiency increase total use?
Does success reveal a new bottleneck?
Will people game the new metric?
Which consequences arrive later?
Which indirect effects could become irreversible?
What would happen under inaction?
How will we know our causal chain was wrong?

Frequently Asked Questions

What is a second-order effect?

It is an indirect consequence caused by the first consequence of an action. If A causes B and B causes C, C is a second-order effect relative to A.

Are second-order effects always unintended?

No. They can be anticipated and deliberately sought. Infrastructure is often built partly for the development it enables later, not only for the immediate travel-time reduction.

Are second-order effects the same as side effects?

Not exactly. A side effect can be direct. Second-order specifically refers to causal sequence: a consequence of a consequence.

Why are second-order effects hard to predict?

Because people adapt, multiple systems interact and each causal step adds uncertainty. Prediction becomes harder as causal distance increases.

What is the best practical habit?

After identifying the direct effect, ask “and then what?” at least once for every major actor, substitute, complement, bottleneck and time horizon affected by the change.

Research Basis and Further Reading

OECD, Systemic Thinking for Policy Making, on nonlinear system behaviour and assessing the consequences of interventions.
OECD-IIASA, Introduction to the Strategic Partnership, on unintended consequences, spillovers, trade-offs and the need to analyse interconnected systems.
OECD, Introducing Systems Thinking into Public Sector Institutions, on policy solutions generating unforeseen and unintended effects in changing contexts.

The Larger Idea

The first consequence is seductive because it is closest.

Build the bridge: journey shorter.

Add the rule: behaviour changes.

Cut the price: demand rises.

Automate the task: labour falls.

All true.

Then the world notices.

People move.

Markets adapt.

New bottlenecks form.

Substitutes appear.

Complements grow.

Expectations change.

The intervention returns to us through a world that has been altered by the intervention itself.

That is the habit this article is trying to build.

Never stop at what the change does. Ask what the changed world does next.

How The World Works | Second-Order Effects — What Happens After the First Consequence