Why many studies promise explanation or causation even though their design can only describe or associate

Opening: the problem is often not the result, but the claim

One of the most persistent mistakes in empirical research is not that the data are fabricated, the analysis is sloppy, or the topic is unimportant. The deeper mistake is that the study claims to explain or identify a cause when its design only allows description or association. This happens across many fields. A researcher observes that two variables move together and writes as if one produces the other. Another identifies a difference between groups and treats it as evidence of an intervention effect, even though the groups were not made comparable by design. A third study uses language such as “impact,” “effect,” “driver,” or “determinant” even though the evidence supports nothing stronger than correlation. Antonakis et al. (2010) argue that researchers frequently make causal claims without meeting the design and estimation conditions needed to justify them.

This mistake matters because causal and explanatory claims are strong claims. They tell readers not merely that variables are related, but that one thing helps produce another, or that a study has identified why something happens. Shadish, Cook, and Campbell’s framework for experimental and quasi-experimental design treats causal inference as a design issue grounded in counterfactual reasoning, not as a rhetorical upgrade added after analysis. When researchers move from association to causation too quickly, they do not merely exaggerate. They alter the meaning of the study.

Why researchers commonly make this mistake

Researchers often make this mistake because causal language sounds stronger, more useful, and more publishable than descriptive or associational language. “X is associated with Y” sounds modest. “X affects Y” sounds important. That difference matters in academic writing, grant applications, policy briefs, and media communication. But the attractiveness of causal language creates pressure to say more than the design can carry. Antonakis et al. (2010) show that this is not a marginal problem. In their review, they found that many published studies failed to satisfy the conditions needed for valid causal claims.

A second reason is conceptual confusion. Researchers often blur four different levels of claim: descriptive, associational, explanatory, and causal. A descriptive claim says what is happening. An associational claim says variables co-vary. An explanatory claim proposes why a pattern may occur. A causal claim says that changing one thing would change another. These are not interchangeable. Hernán (2006) emphasizes that causal inference asks what would happen under different interventions or conditions, which is a stronger question than simply describing observed patterns. Rohrer (2018) likewise argues that observational correlations are often discussed too casually, without enough clarity about the assumptions required for causal interpretation.

A third reason is that some designs produce data that look convincing. Large samples, sophisticated models, longitudinal panels, or richly visualized trends can create an aura of explanatory strength. But methodological sophistication does not automatically turn a non-causal design into a causal one. Hernán and Robins’ causal-inference framework is especially useful because it keeps attention on the causal question, the study design, and the assumptions required for interpretation rather than on statistical complexity alone.

Dominant design context: mainly quantitative, but not only quantitative

This mistake appears most visibly in quantitative research because quantitative studies often use the language of prediction, modeling, intervention, and effect estimation. But the problem is broader than that. It also appears in cross-design work whenever researchers present evidence as explanatory or causal without the design logic needed to support that move. Even case study research can contribute to causal understanding under specific conditions, but only when the design is explicit about what kind of causal evidence it provides and what it does not. Green et al. (2022) argue that case studies can contribute to explanatory and causal reasoning, but only with clarity about mechanisms, context, and inferential limits.

The dominant logic here is M > RQ > RH. Methodology comes first because the central failure lies in the design’s inability to support the claim being made. Research question comes second because many studies ask causal questions without first asking whether their design can answer them. Hypothesis comes third because a hypothesis about effects or determinants may still be reasonable in substance, yet too strong relative to the evidence the design can generate.

Where the failure occurs in the RQ–RH–D–M chain

At the RQ level, the trouble often begins when researchers formulate a causal or explanatory question too early. They ask, “What is the effect of X on Y?” or “Why does X lead to Y?” before establishing whether the study design can support those questions. A cross-sectional survey, for example, may describe patterns at one point in time very well. It cannot automatically establish temporal order, eliminate confounding, or identify intervention effects. Hernán et al. (2024) propose six core questions for observational studies that aim to support causal claims, beginning with the causal question itself and continuing through design, assumptions, and interpretability.

At the M level, the problem becomes decisive. A design may be descriptive, cross-sectional, correlational, or observational without sufficient control over confounding, selection, reverse causation, or timing. Yet the discussion may still use causal language. Antonakis et al. (2010) emphasize that nonexperimental research can sometimes support causal interpretation, but only when specific design and estimation conditions are addressed. Shadish et al.’s framework similarly reminds us that causal inference depends on how the comparison is constructed and what counterfactual logic the design can sustain.

At the D level, the data may still be high quality in a narrow sense. They may be accurate, large, and carefully measured. But good data cannot answer a stronger question than the design permits. That is why the core problem is not data quality alone. It is design adequacy relative to the claim. At the RH level, hypotheses about “effects,” “drivers,” or “determinants” often sound sharper than the evidence can justify. The hypothesis then reinforces overclaiming rather than disciplined inference.

How this mistake distorts findings and conclusions

This mistake distorts research by making conclusions appear more decisive than they are.

In Psychology, a study may find that higher social media use is associated with higher anxiety and then conclude that social media causes anxiety. But the design may not rule out reverse causation, third variables, or reciprocal dynamics.

In Economics, a study may observe that households receiving microcredit show different financial outcomes and then talk as if microcredit produced those outcomes, even though selection into borrowing may already separate borrowers from nonborrowers in important ways.

In Political Science, a study may find that exposure to certain media is associated with lower trust in institutions and then treat that association as if it demonstrated media effects, despite the possibility that prior attitudes shaped media choice. These are not trivial wording issues. They change the meaning of the result. Rohrer (2018) and Hernán (2006) both emphasize that causal interpretation requires much more than observed co-variation.

In Ecology, the same issue appears when observational environmental patterns are described as if one factor has been shown to drive another without sufficient design support. Ecological systems are especially vulnerable to confounding, timing issues, and interacting causes. A study may identify a plausible association, but a causal conclusion requires stronger design reasoning than the pattern alone can provide.

More generally, when researchers confuse description with explanation or explanation with causation, they invite readers, reviewers, practitioners, and policymakers to believe that the study has established more than it actually has. Hernán (2021) argues that strengthening causal inference from observational data requires explicit attention to design and assumptions, not just stronger rhetoric.

How to avoid the mistake before collecting data

The best prevention is to decide early what level of claim the study is designed to support. Is it descriptive, associational, explanatory, or causal? That question should be answered before data collection begins. If the design cannot support causal inference, the research question and hypothesis should not be written as if it can. This is not weakness. It is design honesty.

A second preventive step is to ask what counterfactual comparison the study is trying to approximate. Shadish et al. treat causal design as fundamentally concerned with what would have happened under different conditions. If the design cannot plausibly construct or approximate such a comparison, the study should probably not claim causal effects. Hernán et al. (2024) recommend asking what quantity would answer the causal question, what assumptions are required, and whether a causal interpretation is tenable in principle and practice.

A third preventive step is linguistic discipline. Terms such as “effect,” “impact,” “determinant,” and “driver” should be used only when the design genuinely supports them. Sometimes a study can aim at explanation in a softer sense, proposing mechanisms, building theory, or clarifying processes, without claiming direct causal identification. Green et al. (2022) are helpful here because they show that explanatory contribution is not identical to strong causal proof.

What can still be repaired after data collection

After data collection, some repair is possible, but mostly at the level of claim, not at the level of the original causal ambition. The most honest repair is often to rewrite the study so that the question, hypothesis, and conclusion match what the design actually supports. A study initially framed as causal may need to be reframed as associational. A paper that claimed to identify determinants may need to say it identified correlates or patterns consistent with certain mechanisms. That may sound like retreat, but it is really a methodological correction.

In some cases, additional design elements or stronger assumptions may improve interpretation. Observational studies can sometimes support stronger causal reasoning through careful design, measurement, temporal structure, and causal-inference methods. Antonakis et al. (2010), Hernán (2006), and Hernán et al. (2024) all make clear that causal claims from nonexperimental data are not impossible, but they are demanding. What usually cannot be repaired is a fully unsupported causal conclusion built on a design that never addressed the key threats in the first place.

Brief cross-field illustrations

In Psychology, an adolescent survey may find that loneliness and screen time are related. That supports an association. It does not, by itself, prove that screen time causes loneliness.

In Economics, a study may observe that firms adopting a new management technology later become more productive. That pattern is interesting, but without stronger design logic it does not automatically prove that the technology caused the productivity gain.

In Political Science, researchers may find that regions with higher campaign exposure show higher turnout. That may support a descriptive or associational claim, but causal language requires a much stronger argument about selection, targeting, and comparability.

Across all these examples, the lesson is the same: the design must set the ceiling for the claim.

Short takeaway checklist

Before collecting data, ask:

• Is my question descriptive, associational, explanatory, or causal?
• Can my design support that level of claim?
• What alternative explanations or sources of bias would threaten a causal interpretation?
• If my results are strong, what is the strongest sentence I will actually be entitled to write?
• Am I designing the study for the claim I want, or merely hoping to use that claim later?

A good study does not become stronger because its language becomes stronger. It becomes stronger when its design earns the language it uses.

References

Antonakis, J., Bendahan, S., Jacquart, P., & Lalive, R. (2010). On making causal claims: A review and recommendations. The Leadership Quarterly, 21(6), 1086–1120. https://doi.org/10.1016/j.leaqua.2010.10.010

Green, J., Edgerton, J., Naftel, D., Shoub, K., & Cranmer, S. J. (2022). Case study research and causal inference. Perspectives on Politics, 20(1), 277–294. https://doi.org/10.1017/S1537592721000924

Hernán, M. A. (2006). Estimating causal effects from epidemiological data. Journal of Epidemiology and Community Health, 60(7), 578–586. https://doi.org/10.1136/jech.2004.029496

Dahabreh, I. J., Robertson, S. E., Tchetgen Tchetgen, E. J., Stuart, E. A., & Hernán, M. A. (2024). Causal inference about the effects of interventions from observational studies in medical journals. JAMA, 331(16), 1380–1387. https://doi.org/10.1001/jama.2024.3001

Rohrer, J. M. (2018). Thinking clearly about correlations and causation: Graphical causal models for observational data. Advances in Methods and Practices in Psychological Science, 1(1), 27–42. https://doi.org/10.1177/2515245917745629

Shadish, W. R., Cook, T. D., & Campbell, D. T. (2002). Experimental and quasi-experimental designs for generalized causal inference. Houghton Mifflin.