The question
Why can an intervention succeed at its intended target and still create other effects elsewhere?
The target improves, but the system changes too
Imagine installing an app blocker to reduce evening social-media use.
The intervention works according to its intended measure. During the blocked period, the selected applications become inaccessible, and the time recorded in them decreases.
But that is not necessarily the whole outcome.
The person might move to an unblocked application, browse from a laptop or concentrate more checking into the period before the block begins. The interruption might create frustration—or make space for reading, conversation or sleep.
These are possibilities, not established effects of app blockers. Determining whether any of them occurred would require observation and evidence.
The example reveals the central possibility: an intervention can succeed at its intended target while additional changes emerge elsewhere in the system.
Interventions enter systems
An intervention does not act on an isolated target.
Even a seemingly simple behaviour is connected to cues, rewards, available alternatives, relationships, time and the function the behaviour currently serves.
If evening social-media use provides social contact, fills boredom or postpones an uncomfortable task, blocking one pathway does not automatically remove those conditions. It changes how the person can respond to them.
Another activity may become more attractive. Time may be redistributed. The need previously met through the blocked behaviour may remain unresolved—or be met differently.
Research on complex interventions treats outcomes as potentially shaped by intervention mechanisms, implementation and context. Process-evaluation guidance therefore examines not only whether an intervention was delivered, but how participants and environments responded to it.
This does not mean every intervention produces important wider consequences. It means that the intended target is only one place where effects might appear.
Three questions about every effect
Intervention consequences can be organised through three separate questions.
Was the effect intended?
An intended effect is part of the intervention’s explicit aim.
An unintended effect is an additional consequence outside that primary aim.
Through which pathway did it arise?
A direct effect arises through the intervention’s primary target pathway.
An indirect effect arises through another changed behaviour, relationship, incentive or downstream pathway.
The distinction concerns pathway rather than speed. A direct effect can be delayed, while an indirect effect can appear quickly.
Was the effect beneficial or harmful?
An effect may be beneficial, harmful, neutral or mixed.
These dimensions should not be collapsed.
An unintended effect is not automatically harmful. A harmful effect is not automatically unintended. An indirect effect is not necessarily weaker than a direct one.
Nor is every failed outcome a side effect. If an intervention was never delivered as intended, that is an implementation problem. A side effect can occur even when implementation and target change both succeed.
Compensation, substitution and displacement
Several wider responses are often treated as though they were the same. They are not.
Compensation occurs when another response offsets part of the intended effect.
An exercise programme might increase structured activity while activity elsewhere decreases, reducing the total gain. But this is not automatic. Systematic reviews of physical-activity interventions have found heterogeneous and often inconclusive evidence for compensation. The evidence supports treating compensation as something to investigate, not a universal law of behaviour.
Substitution occurs when another behaviour or pathway replaces the targeted one.
If someone uses an unblocked application instead of a blocked one, that could represent substitution—but only if the increase is observed and linked to the restriction.
Displacement occurs when behaviour or burden moves to another time, place, target, person or system component.
Checking immediately before the block begins could represent temporal displacement. Work transferred from one employee to another could represent burden displacement.
Substitution identifies replacement. Displacement identifies relocation. The two can overlap, but they are not identical.
These categories organise what could be investigated. They do not establish that any of these responses occurred in the app-blocker example.
Rebound is a more specialised term whose meaning differs across fields. A return to an old pattern may be relapse rather than rebound. Calling it rebound requires a field-appropriate mechanism through which the intervention, its removal or an efficiency change contributed to the later increase.
Improvement here can create cost elsewhere
An intervention can improve one outcome while moving cost or work beyond the measured target.
Imagine a workplace introducing a strict response-time target to reduce how long customers wait.
Response time may improve. But complex cases might be rushed, corrective work might accumulate later or emotional burden might move to employees whose work is not represented in the target measure.
These consequences are illustrative, not inevitable. The response target might also improve coordination without creating meaningful additional burden.
The point is that local improvement and system-wide improvement are different claims.
To evaluate the wider result, we need to ask who receives the benefit, who performs additional work and which outcomes remain outside the measurement boundary.
Research on spillovers distinguishes effects on targeted outcomes from effects on other outcomes or people. An intervention can affect another outcome in the same person, the same outcome in another person, or both another outcome and another person. The INTENTS framework makes these different spillover locations explicit.
Side effects can be beneficial or mixed
“Side effect” often sounds like another word for harm. But unintended describes the effect’s relation to the original aim—not its value.
Some of the clearest positive spillovers appear in infectious-disease interventions. Treating or vaccinating targeted people can also benefit non-targeted people through reduced transmission.
A systematic review found its strongest evidence for health spillovers in vaccines and mass drug administration, while evidence for several social and resource-substitution pathways was weaker. This shows both that beneficial spillovers can occur and that evidence quality depends on the mechanism and domain.
Other consequences may be mixed.
An app blocker might create more time for sleep while reducing a valued form of social contact. A workplace change might benefit customers while increasing employee effort. The same consequence may help one group, burden another or change value over time.
The intended target cannot answer all these questions by itself.
What we measure determines what we see
Every evaluation draws boundaries.
It selects:
- which outcomes count;
- who is included;
- where effects are measured;
- how long observation continues.
If an evaluation measures only time in blocked applications, it may miss use on another device. If it measures only immediate customer response, it may miss corrective work later. A narrow boundary can also miss benefits that occur outside the intended outcome.
An average can conceal how observed benefits and costs were distributed across the measured population. Explaining why individual outcomes differ remains a separate question.
Timescale matters for the same reason.
Some effects are immediate. Others are delayed, temporary or cumulative. An early improvement may fade. Initial disruption may settle. Participants may adapt to the intervention, and the mechanisms operating after several weeks may differ from those operating on the first day.
An intervention can therefore appear successful or unsuccessful partly because of where the evaluator stops looking.
After does not always mean because of
Wider observation creates another risk: treating everything that follows an intervention as one of its effects.
An outcome may instead reflect:
- natural change;
- another event occurring at the same time;
- altered measurement;
- expectations;
- a return toward typical levels;
- random variation.
A plausible pathway helps identify what to investigate. It does not prove that the pathway produced the outcome.
Spillovers make causal evaluation especially difficult because one person’s exposure may affect another person’s outcome. Researchers therefore use designs and assumptions created specifically for situations in which treatment effects cross between connected people or units. Methodological work on spillovers shows why ordinary treated-versus-untreated comparisons can become insufficient when units influence one another.
The reader-facing principle is simpler:
An effect needs more than a convincing story.
Temporal order matters, but “after” does not automatically mean “because of.”
Behavioural change reorganises more than behaviour
Attempts to change habits often alter the conditions surrounding them.
Removing one response may change how a person responds to its cues or seeks the function it previously served. Another behaviour may replace it, but substitution should be demonstrated rather than assumed.
Relationships also respond to intervention.
A new boundary may reduce one person’s burden while changing expectations, responsibilities or uncertainty elsewhere. This does not make the boundary harmful. It means relational change can affect more than the person who initiates it.
Organisations respond through workflow, incentives and coordination. A new target can change what employees prioritise, what they report and which workarounds become useful. Again, these are possible pathways, not inevitable consequences.
The broader lesson is not that intervention always creates a new problem. It is that behavioural, relational and organisational targets participate in wider patterns. Changing the target can alter those patterns as well.
Intervene, observe, revise
Incomplete prediction does not make intervention futile.
It makes certainty limited and learning necessary.
Wider evaluation attends not only to whether the intended target changed, but also to what happened elsewhere, how consequences were distributed and whether effects persisted or emerged later. It also asks whether the available evidence supports attributing those consequences to the intervention.
Complex-intervention frameworks treat development, feasibility, evaluation and implementation as interacting phases rather than a one-way sequence. They support repeated attention to context, uncertainty and refinement as evidence accumulates.
Observation does not guarantee control. Measurement can miss outcomes, causal interpretation can be wrong and adjustment can create further consequences.
But intervention need not depend on complete foresight. It can remain responsive to what is learned.
A target effect is not the whole effect. Side effects are not automatically failures, and unintended effects are not automatically harmful. Their significance depends on what changed, through which pathway, for whom and over what period.
An intervention is not complete when it changes its target. Its meaning becomes clearer through what the wider system does next.