The question
How can the same causal influence produce different outcomes across people, contexts or times?
Definition
The same causal influence can produce different outcomes when the systems or conditions in which it operates differ. Identifying a common cause therefore does not imply a uniform consequence.
Suppose a learning intervention has been shown to improve later performance on average.
That does not mean it produces the same effect under every condition. Its causal effect may be larger in some circumstances, smaller in others, or close to zero elsewhere. Under some conditions, effects can even differ in direction.
What remains the same is the focal causal influence, not the entire causal situation surrounding it.
This distinction matters because identifying something as a cause does not imply that it carries one fixed consequence wherever it appears.
A cause can be real without having a uniform consequence.
A causal influence does not contain one fixed outcome
When evidence supports a causal effect, it tells us that the focal influence makes a difference relative to a relevant comparison under specified conditions.
It does not automatically tell us that the difference will be identical elsewhere.
A causal influence might have a substantial effect under one set of conditions and a much smaller effect under another. Under still other conditions, its effect might be negligible.
The more precise question is therefore not simply:
Does X cause Y?
It is also:
What effect does X have under these conditions?
This extends the distinction established in Correlation, Prediction, Causation and Mechanism. Establishing a causal effect is one epistemic achievement. Establishing whether that effect remains similar across people, contexts and times is another.
The two should not be confused.
The same influence can operate in different causal configurations
The title of this concept is Same Cause, Different Outcomes.
More precisely, many cases involve the same focal causal influence operating within different causal configurations.
A simplified representation is:
X + Configuration A → Y₁
X + Configuration B → Y₂
The focal influence X is present in both cases. The surrounding causally relevant conditions are not necessarily the same.
Those conditions might include differences in:
prior history;
current state;
capacity;
environmental conditions;
other causal influences;
Depending on the phenomenon, an enabling condition might alter what can occur, a constraint might narrow possible responses, or another causal influence might modify the effect.
The consequence of X can therefore depend on what else is true when X operates.
This does not make X irrelevant as a cause. It means that a causal influence operates within a wider configuration, and its effect can depend on that configuration.
Different effects can take different forms
For a specified outcome, causal effects can differ in magnitude.
An intervention might have a larger causal effect under one set of conditions and a smaller effect under another. In some cases, the effect may be close to zero under particular conditions. Under some causal structures, effects can even differ in direction.
The broader consequences of a common influence can also diverge in other ways.
They may differ in:
when they emerge;
how long they persist;
how they are expressed;
which downstream outcomes eventually develop.
Variation can occur between people, but it need not reflect only stable differences between individuals. The causally relevant conditions surrounding an influence can also change within the same person across contexts and times.
Developmental history adds another dimension. The consequences of an influence may depend partly on what preceded it or when in a developmental trajectory it occurred.
Developmental science uses the term multifinality for a related pattern in which similar developmental beginnings can lead to different developmental outcomes. Multifinality is not identical to heterogeneous causal effects, but it illustrates an important form of one-to-many development.
Similar beginnings need not produce identical developmental endpoints.
Variation does not explain itself
Suppose appropriate causal evidence indicates that an intervention has a larger effect under one set of conditions than another.
We have learned something important:
the causal effect varies.
We have not yet learned why.
A measured characteristic might mark conditions under which the effect differs. Another causal influence might interact with the intervention. A different underlying process might account for the pattern.
Evidence of heterogeneity alone cannot decide among these possibilities.
This is where the distinction from Mediators, Moderators and Mechanisms becomes important.
A moderator describes when, where or for whom an effect differs. A mechanism concerns the process through which an effect is produced.
Neither follows automatically from the observation that causal effects vary.
One possible source of heterogeneity is causal interaction: the effect of X may depend on another causal factor. But evidence that causal effects vary does not by itself establish which interaction, if any, produced that variation.
Variation in effects is something to explain; it is not itself the explanation.
Average effect does not mean uniform effect
Now return to the learning intervention.
Suppose the evidence supports a positive causal effect on performance, on average, in a defined population.
That average can remain scientifically meaningful even when the causal effect differs across identifiable conditions or groups and observed responses differ among individuals.
An average causal effect summarises a population-level contrast. It does not claim that every person experiences an identical causal effect.
This distinction is important because averages compress variation.
The estimated causal effect might be larger under one condition and smaller under another while the overall average still describes a genuine population-level causal pattern.
Neither level invalidates the other.
The population average does not erase heterogeneity.
Heterogeneity does not make the population average meaningless.
Both of these statements can therefore be true:
The intervention has a positive causal effect on average.
and:
The size of that causal effect differs across some conditions.
There is no contradiction.
Average effect does not mean uniform effect.
Heterogeneity does not eliminate causal regularity
Recognising heterogeneous causal effects can produce another mistake: assuming that causation in human systems is therefore so variable that useful generalisation becomes impossible.
That does not follow.
Variation in causal effects can itself be structured.
An influence may produce broadly similar effects across many circumstances while differing under particular conditions. Effects may vary systematically with features of the environment, developmental history or other causal influences.
Nor must every causal effect display substantial heterogeneity. Under strongly constraining conditions, the range of possible responses may narrow considerably.
General patterns and heterogeneous effects can therefore coexist.
A causal finding can generalise to a population or range of conditions without requiring an identical causal effect in every case.
The important empirical questions become:
How much does the effect vary?
Under which conditions does it vary?
What explains that variation?
Recognising heterogeneity does not weaken causal science. It makes the scope of a causal claim more precise.
Same cause, different outcomes
Different Causes, Same Outcome established one side of the causal problem:
different causal pathways can converge on similar outcomes.
This concept establishes the complement:
the same focal causal influence can contribute to different outcomes.
Together, they reveal why causes and outcomes should not be assumed to map onto one another one-to-one.
A shared outcome does not establish a shared causal pathway.
A shared causal influence does not establish a shared consequence.
Neither principle means that causal explanation fails. They show why explanations of complex human outcomes often require more than identifying an isolated input and observing an endpoint.
Identifying a cause can therefore open another question rather than close the inquiry:
Under what conditions does this causal influence produce which consequence?
A cause can be real without having a uniform consequence.
A common causal influence does not guarantee a common outcome.