How Do You Avoid Combining Descriptive and Causal Evidence as Though They Answer the Same Question?

What if some studies describe what is happening while others estimate what would happen if something changed?

A review can contain studies that look closely related because they examine the same population, exposure, intervention, or outcome. Yet some may simply estimate what exists in a population: prevalence, incidence, frequencies, distributions, patterns, or differences between groups. Others attempt to estimate what would happen under a specified intervention or exposure compared with an alternative.

Those studies can belong in the same broader evidence base, but they do not necessarily answer the same question.

The problem arises when synthesis erases that distinction. A high prevalence of an outcome does not tell you whether a particular intervention will reduce it. A difference between groups does not by itself tell you what would happen if people moved from one condition to another. Describing the world and estimating the consequences of changing it are related scientific tasks, but they require different interpretations.

Keep descriptive and causal questions separate throughout the synthesis

The distinction begins with what the study is trying to estimate

Descriptive evidence tells you about the world as observed

A descriptive question seeks to characterize some feature of a population. It might ask how common an outcome is, how it is distributed, whether its frequency has changed over time, or how that frequency differs across groups.

For example, a study might estimate the percentage of university students using generative AI for coursework, the incidence of academic integrity cases during a semester, or the distribution of digital literacy scores among first-year students.

These are substantive research questions in their own right. Descriptive research should not be treated as merely preliminary or inferior causal research. Its validity still depends on issues such as defining the target population appropriately, obtaining representative observations when required, measuring variables adequately, and handling missing data and selection processes appropriately.

Causal evidence asks about a contrast between possible conditions

A causal question asks something different. It concerns the outcome that would occur under one intervention or exposure condition compared with the outcome that would occur under another condition.

Instead of asking how many students use generative AI, for example, a causal study might ask whether allowing a particular form of AI-assisted feedback changes students' writing performance compared with conventional feedback.

This distinction is more fundamental than whether the study is cross-sectional, longitudinal, experimental, or observational. Study design matters, but the first question for synthesis is what the study was intended to estimate.

A comparison between groups is not automatically a causal question

This is where synthesis can become surprisingly slippery. Researchers sometimes see two groups in a paper and assume that the study has estimated an effect.

Suppose a study reports that students who use an AI tutoring system have higher grades than students who do not. That difference describes an observed contrast. It does not automatically estimate what would happen if otherwise comparable students were made to use the system rather than not use it.

The groups may differ in prior achievement, motivation, course selection, access to technology, instructor practices, or other characteristics. A causal interpretation requires a design and analysis capable of supporting the relevant counterfactual contrast.

This overlaps with the broader problem of distinguishing association from causation during evidence synthesis, but descriptive evidence creates an additional issue: some studies were never designed to estimate an exposure-outcome effect in the first place.

The same variable can play different roles in different questions

Two studies may measure exactly the same variables yet target different estimands.

The variables alone therefore cannot tell you which evidence category a study belongs to. You need the research question, estimand, design, analysis, and intended interpretation.

Do not make descriptive evidence answer an intervention question

Suppose descriptive studies show that an undesirable outcome is especially common among a particular group. That finding may help identify the scale and distribution of a problem. It does not establish which intervention will change that outcome.

Similarly, observing that an outcome became more common after a policy was introduced does not necessarily establish that the policy caused the increase. Time trends can be valuable descriptive evidence, but causal attribution requires consideration of what would have happened in the policy's absence.

Do not make causal evidence answer a descriptive prevalence question either

The mistake can also run in the opposite direction. A randomized trial may provide strong evidence about the effect of an intervention while providing poor evidence about how common an outcome is in the broader population.

Trial participants may have been selected according to restrictive eligibility criteria. Their baseline risk may differ from that of the population whose prevalence you want to estimate. Randomization strengthens the internal comparison between treatment groups, but it does not automatically make the trial sample representative of a wider population.

Evidence quality is therefore question-specific. A study can be highly informative for one estimand and unsuitable for another.

Organize the review around evidence functions

If both descriptive and causal evidence are relevant, one useful strategy is to organize them according to what they contribute.

Descriptive evidence may establish the magnitude, distribution, or trajectory of the problem. Causal evidence may then address whether a particular intervention changes the outcome. The evidence streams can inform the same broader decision without being merged into a single inferential statement.

This separation becomes particularly important when causal studies themselves use different comparison groups, because even studies that all estimate effects may not estimate the same effect.

Do not pool simply because the numbers can be converted to a common format

Statistical compatibility is not conceptual compatibility. Software may allow several estimates to be entered into the same analysis, but that does not mean the resulting summary answers a coherent research question.

Before pooling, define the target quantity you want the summary estimate to represent. Then ask whether each study actually estimates that quantity, or something sufficiently close to it that combination is scientifically defensible.

If not, separate synthesis is often more informative than a single pooled number. A review does not become weaker because it reports two clearly interpreted bodies of evidence rather than one ambiguous average.

How should descriptive and causal studies be handled in the same review?

The synthesis is richer because the evidence streams complement one another without being made interchangeable. Description tells you what problem you are looking at. Causal evidence can help answer what might change it.

Common ways descriptive findings acquire causal meanings they never earned

Label the question before you classify the evidence

During data extraction, record what each result is intended to estimate. A simple “study design” column is not enough. Include the target population, outcome, relevant exposure or intervention, comparator where applicable, timing, and intended inferential claim.

Your conclusions should mirror that structure. Use descriptive language for descriptive results and causal language only where warranted. That discipline may seem almost pedantic until one verb changes “was more common among” into “was caused by.” Academic prose has started larger arguments over less.

Before synthesizing descriptive and causal evidence, check:

Common questions about descriptive and causal evidence

Preserve the question each piece of evidence was designed to answer

How to Cite This Guide

This guide is intended to be read, shared, and used in research, teaching, and academic work. If you draw on its ideas, explanations, or other content, please acknowledge the source by citing the guide. Doing so gives appropriate credit and helps your readers locate the original resource.