Explanations Are Not Interventions

An organisation investigates why many of its employees are exhausted. The evidence points to several contributing conditions: chronic workload pressure, limited control over schedules and inadequate opportunities for recovery.

The organisation responds by offering a stress-management programme.

That programme may be valuable if its purpose is to improve coping or recovery. But evidence that it helps people manage stress would not establish that it has resolved the workload conditions identified in the original explanation. The explanation concerned one part of the system; the selected intervention acted primarily on another.

The explanation may therefore be well supported while the intervention remains insufficient for the broader claim made on its behalf.

This does not make explanations impractical. A good explanation can reveal causes, mechanisms and constraints. Intervention studies can then test those explanations, strengthen them or force their revision. The distinction concerns what each claim establishes—not a separation between the two forms of research.

Understanding why something happens and establishing what will change it are connected achievements. They are not interchangeable.

Understanding and changing are different questions

Several forms of inquiry are often compressed into one.

Description asks what is happening.

Prediction asks what is likely to happen.

Causal explanation asks what contributes to producing the outcome.

Intervention asks what happens when a specified action changes the system.

Each question can inform the next, but each contributes different evidence.

An explanation of employee exhaustion might identify workload, limited schedule control, inadequate recovery, workplace norms and individual circumstances. An intervention claim must also specify what action will be taken, what it is expected to alter, what it will be compared with, whose outcomes matter and over what period.

"Reduce stress" is not yet a defined intervention. Neither is "increase motivation" or "improve the culture." Each phrase can refer to different actions that reach different targets and carry different consequences.

Explanations can narrow the possibilities. They can reveal where attention should be directed, identify implausible actions and generate stronger intervention hypotheses. Intervention evidence, in turn, can test whether the proposed causal account survives deliberate change.

But explanatory plausibility is not intervention evidence.

A convincing account of why something happens remains a hypothesis about where and how to intervene until the proposed action is examined as an intervention.

A predictor is not automatically a lever

The gap is wider when the apparent explanation is actually a prediction.

Previous exhaustion, poor sleep, job role or low schedule control may help predict who is likely to struggle later. A combined risk score might forecast that outcome even more accurately.

The score can be useful. It may help direct attention or plan support. But lowering the number displayed by the score would change nothing by itself. Action must address some part of the underlying system—and the score does not automatically establish which part is causal or modifiable.

The same problem occurs whenever a measurement is mistaken for the process it represents. A survey response can predict an outcome without becoming an intervention target. A demographic characteristic may improve prediction while offering no defensible instruction for action. A consequence of a problem may forecast its continuation without causing that continuation.

One useful test is to imagine changing the recorded predictor while leaving the underlying conditions intact. If the outcome would remain unchanged, the predictor may be informative without functioning as a lever.

This does not diminish prediction. Prediction can help anticipate demand, identify elevated risk, allocate attention and plan investigation. It simply answers a different question.

Knowing where an outcome is likely does not necessarily tell us how to move it.

What started a pattern may not be what maintains it

Even a genuine cause may not identify what can change an outcome now.

Imagine that someone begins avoiding public speaking after a humiliating presentation. The event helps explain how the pattern started. Months later, the original event is over. The pattern might now be maintained by anticipatory arousal, repeated avoidance or the absence of experiences that challenge the person's expectations.

The historical event cannot be removed. Changing the present pattern requires evidence about what is operating now.

This distinction appears in many systems. Early workload pressure may contribute to the development of exhaustion. Later persistence may also involve continuing demands, inadequate recovery, changed expectations, reduced control or established organisational routines. Removing one original pressure may help without reversing everything that developed around it.

This is not a universal rule that causes of onset and causes of change must differ. Sometimes altering the initiating cause is both feasible and sufficient. The point is that history cannot settle the question in advance.

Prevention and reversal can therefore require different evidence. Preventing a process from beginning may involve one action. Changing an established process may involve another because the relevant relationships have adapted or reorganised.

A historical explanation can be accurate and useful without identifying the strongest present leverage point.

The missing links between mechanism and outcome

Suppose an explanation identifies a process that may help maintain the problem.

A mechanism is the process proposed to produce or sustain the outcome. A target is the part of that process an intervention is designed to alter.

Moving from mechanism to outcome still requires several links.

First, the proposed mechanism must be causally relevant rather than merely associated with the outcome.

Second, it—or a point influencing it—must be alterable in practice and permissible to alter.

Third, the selected intervention must reach the intended target.

Fourth, the target must change sufficiently and for long enough.

Fifth, that change must improve the outcome that matters.

Finally, the benefit must remain meaningful after burdens, risks and indirect consequences are considered.

A well-supported but partial explanation may fail to generate an intervention that completes this chain.

A stress-management programme may reach employees without changing workload. It may change a measure of coping without reducing exhaustion. It may improve the specified outcome for some participants while remaining inaccessible or burdensome to others.

An organisational intervention faces the same evidential demands. A workload policy may exist formally without changing daily practice. It may reduce one demand while creating another. It may alter workload while leaving other maintaining conditions intact.

Even target engagement is not the end of the chain. An intervention may change exactly what it was designed to change without improving the final outcome. The target may contribute too little, other pathways may compensate, or the selected outcome may fail to capture what matters most.

Reaching people is not the same as engaging the target. Engaging the target is not the same as improving the outcome. Improving one outcome is not necessarily the same as producing net benefit.

Every link introduces a question that must be supported well enough for the strength and consequences of the decision being made. Not every decision requires perfect evidence, but stronger claims and higher-stakes actions require stronger support.

Interventions enter systems, not diagrams

Explanations simplify. They select the relationships relevant to a question and leave other details outside the model. This is necessary: an explanation containing everything would explain very little clearly.

An intervention enters wider conditions.

It encounters people responsible for delivery, organisations deciding whether to adopt it and environments determining whether it can be used. It competes with existing priorities and incentives. Access is unequal. Resources are limited. People and institutions adapt.

An intervention may therefore produce little change because it was not delivered adequately or because it did not engage its intended target. The target may change without being sufficient to alter the outcome. Other processes may compensate. Effects may differ across people or settings. Benefits in one area may be accompanied by costs elsewhere.

A null outcome alone cannot identify which of these occurred.

Researchers therefore examine implementation as well as outcomes. Did relevant settings adopt the intervention? Did it reach the intended people? Was it delivered sufficiently? Could participants use it? Did it remain in place long enough to matter?

These questions help distinguish problems in implementation from problems in intervention design or theory. The distinction is useful but not always clean. If an intervention cannot be delivered reliably under the conditions in which it is intended to operate, implementation difficulty may reveal a limitation in the design itself.

This prevents two opposite errors. The first is declaring an explanation false when the intervention never altered the proposed target. The second is protecting the explanation from every negative result by claiming, without evidence, that implementation must have failed.

Both the intervention's theory and its operation in practice require scrutiny.

A plausible target is only one part of producing meaningful change under the conditions where an intervention will be used.

Effective action can precede complete explanation

If explanation cannot establish an intervention on its own, must action wait until every mechanism is understood?

No.

A well-designed comparative study can provide evidence that an intervention produced a better average outcome than its comparator in the studied population, setting and period. The estimate still carries uncertainty, and the result may not transfer unchanged elsewhere. But complete mechanistic knowledge is not required before comparative outcome evidence becomes informative.

Outcome evidence and mechanistic explanation are different achievements.

This is not blind trial and error. An intervention may draw on strong theory, prior evidence and a plausible route to benefit. Comparative testing contributes something those foundations cannot supply alone: evidence about what occurred after introducing the intervention rather than its alternative.

Mechanistic uncertainty still matters. Without understanding why an intervention produced its effects, it may be harder to identify the active component, anticipate failure or determine whether the result should transfer. Mechanistic knowledge can improve targeting, implementation, efficiency and safety.

Incomplete explanation does not erase a credible comparative effect.

The reverse is equally important. Improvement in the specified outcome does not prove the proposed mechanism. The intervention may have operated through another pathway. Failure to improve that outcome does not automatically falsify the explanation; the intervention may not have reached or engaged the intended target.

We should therefore reject both mechanistic perfectionism and mechanistic indifference.

We do not need to understand everything before acting. We do need evidence proportionate to the action and the claims made for it.

Build the bridge, then test it

Moving responsibly from explanation to intervention requires three conceptual tests.

First, is the explanation causal and relevant to the outcome as it exists now—not merely predictive or historical?

Second, does the proposed intervention reach and alter a defensible target?

Third, does that alteration produce sufficient net benefit for the relevant people under the conditions in which the intervention will be used?

These tests show why the word because cannot be converted directly into the word therefore.

Even evidence of benefit does not settle whether an intervention should be implemented. Decisions also depend on burden, harm, feasibility, acceptability, resources, equity and ethical limits. The capacity to change something does not by itself establish permission to change it.

Explanation identifies and disciplines possibilities. Intervention evidence estimates what a specified action does under studied conditions. Decision-making evaluates whether that action is justified in context.

The goal is not to choose between understanding and action. It is to let each improve the other without asking either to establish what only the other can show.

An explanation can tell us where to look—but only an intervention, tested as an intervention, can tell us what changing that part will do.