Within Cause Check
The Hidden Factor Behind the Pattern
A hidden third factor can make a cause look powerful even when it is only travelling with the real driver.
On this page
- Common causes and hidden drivers
- Reverse causation and selection effects
- Questions that expose weak causal stories
Page outline Jump by section
Introduction
A confounder is a hidden factor that makes one variable appear to cause another when, in reality, both are being influenced by something else. This is one of the most common reasons people overestimate the strength of a causal relationship. A striking correlation may still be genuine, but its size—or even its direction—can be distorted because an important third factor has not been taken into account. Observational studies are particularly vulnerable because people, organisations and environments rarely differ in only one respect. Modern causal inference therefore treats confounding as a central problem to solve rather than a minor statistical nuisance. [jech.bmj.com]jech.bmj.comCausal inference and effect estimation using observational databy E Igelström · 2022 · Cited by 113 — Fourth, we define and explain biase…
Understanding confounders is an essential thinking skill because many persuasive stories are built on visible patterns rather than fair comparisons. The critical question is not simply whether two things move together, but whether something else could be producing both.
The Hidden Factor Behind the Pattern
A confounder is a variable that influences both the supposed cause and the outcome without lying on the causal pathway between them. If it is ignored, the observed relationship can look much stronger—or sometimes weaker—than the true causal effect. [jech.bmj.com]jech.bmj.comCausal inference and effect estimation using observational databy E Igelström · 2022 · Cited by 113 — Fourth, we define and explain biase…
Imagine researchers observe that people who carry reusable water bottles are healthier than those who do not. It would be easy to conclude that carrying the bottle improves health. Yet health-consciousness could be the real driver. People who care more about their health may exercise more, eat better and also choose reusable bottles. The bottle becomes a marker of the underlying lifestyle rather than its cause.
The same mechanism appears across many fields:
- Wealthier neighbourhoods may have more parks and better health outcomes. Income, education and healthcare access may partly explain both.
- Students attending optional revision sessions often achieve higher grades, but motivation may influence both attendance and exam performance.
- Companies adopting new management software may outperform competitors because they were already better organised and more willing to invest in improvement.
In each case, the apparent cause travels alongside the real driver.
Common Causes and Hidden Drivers
Many confounders arise because one underlying factor affects several outcomes simultaneously. This creates a statistical association that looks causal even when neither observed variable directly influences the other.
A classic illustration involves ice cream sales and drowning deaths. Both tend to increase during warmer weather. Buying ice cream does not cause drowning, nor does drowning increase ice cream purchases. Temperature affects both independently, creating a misleading correlation.
In medicine, age is one of the most common confounders. Older people often take more medications and also experience more illnesses. If age is ignored, a harmless medication might appear associated with poor health simply because older patients are more likely both to receive the drug and to become ill.
Researchers therefore ask whether potential confounders existed before the suspected cause occurred. Variables measured after the exposure may instead represent consequences of the treatment rather than genuine confounders, making adjustment inappropriate. Modern causal frameworks often use directed acyclic graphs (DAGs)—causal diagrams showing assumed relationships—to identify which variables should and should not be controlled. [jech.bmj.com]jech.bmj.comCausal inference and effect estimation using observational databy E Igelström · 2022 · Cited by 113 — Fourth, we define and explain biase…
Importantly, controlling for every available variable is not automatically better. Adjusting for the wrong variables can introduce new biases, including collider bias, where conditioning on a common consequence of two variables creates a spurious association that did not previously exist. [jech.bmj.com]jech.bmj.comCausal inference and effect estimation using observational databy E Igelström · 2022 · Cited by 113 — Fourth, we define and explain biase…
Reverse Causation and Selection Effects
Not every misleading causal story is caused by confounding alone. Two related mechanisms frequently reinforce the illusion.
Reverse causation
Sometimes the apparent outcome actually influences the supposed cause.
For example, researchers might observe that people taking stronger pain medication report worse pain. The medication could seem ineffective or harmful. In reality, severe pain often leads doctors to prescribe stronger medication. The severity of illness influences treatment choice, producing an association that points in the opposite direction from the true causal pathway. [NCBI]ncbi.nlm.nih.govNCBIPrinciples of CausationStatPearls - NCBI Bookshelf - NIHby R Dhawan · 2024 · Cited by 2 — Causation refers to a process wherein an initial or inciting event (ex…
This is especially common in observational healthcare research because treatments are rarely assigned randomly.
Selection effects
Selection effects arise when the people or organisations being compared entered the study for reasons related to both the exposure and the outcome.
Suppose a leadership course is voluntary. Participants later perform better than non-participants. The improvement may partly reflect the fact that ambitious employees were more likely to enrol in the first place. The course may still help, but its apparent impact is inflated because the comparison groups differed before the intervention.
Selection problems also occur when analysing only survivors, successful firms, or accepted applicants. Restricting analysis to a selected group can manufacture relationships that do not represent the wider population. [jech.bmj.com]jech.bmj.comCausal inference and effect estimation using observational databy E Igelström · 2022 · Cited by 113 — Fourth, we define and explain biase…
Questions That Expose Weak Causal Stories
When confronted with an impressive correlation, a few disciplined questions often reveal whether confounding is a plausible explanation.
- What could influence both the proposed cause and the outcome? Look for common drivers such as age, income, motivation, geography or prior health.
- Were the groups already different before the exposure? Pre-existing differences frequently explain later outcomes.
- How were participants selected? Voluntary participation, referral patterns and drop-out can all distort comparisons.
- Could the direction run the other way? Consider whether the outcome might influence the exposure instead.
- Would the relationship remain after accounting for plausible confounders? Strong causal claims become more convincing when they persist across different analytical approaches and populations. [PMC]pmc.ncbi.nlm.nih.govAssessing causality in epidemiology: revisiting Bradford Hill to…by M Shimonovich · 2020 · Cited by 231 — Bradford Hill argued that…
These questions do not prove that a claim is wrong. Instead, they identify alternative explanations that deserve investigation before accepting a causal conclusion.
Why Randomisation Helps—but Does Not Solve Everything
Randomised experiments reduce confounding because chance assignment tends to balance both measured and unmeasured characteristics between groups before treatment begins. As a result, differences observed afterwards are less likely to be explained by hidden common causes. [NCBI]ncbi.nlm.nih.govNCBIPrinciples of CausationStatPearls - NCBI Bookshelf - NIHby R Dhawan · 2024 · Cited by 2 — Causation refers to a process wherein an initial or inciting event (ex…
Outside experiments, researchers attempt to reduce confounding using approaches such as matching comparable participants, stratifying analyses, statistical adjustment, instrumental-variable methods and carefully designed natural experiments. None of these methods automatically removes hidden bias, because they depend on assumptions about which confounders have been measured and how accurately they have been represented. [journals.publisso.de]journals.publisso.deCausal evidence in health decision makingby F Kühne · 2022 · Cited by 34 — Causal inference and health decision science are two methodolo…
The practical lesson is that confounding is rarely eliminated by a single statistical technique. Credible causal claims usually emerge from multiple studies, different methods and repeated attempts to rule out alternative explanations rather than from one striking correlation alone. [PMC]pmc.ncbi.nlm.nih.govAssessing causality in epidemiology: revisiting Bradford Hill to…by M Shimonovich · 2020 · Cited by 231 — Bradford Hill argued that…
Amazon book picks
Further Reading
Books and field guides related to The Hidden Factor Behind the Pattern. Use these as the next step if you want deeper reading beyond the article.
The Book of Why
Explains causation, confounding, and how to distinguish correlation from cause.
Calling Bullshit
Shows how to question misleading causal claims, statistics, and apparent patterns.
The Art of Statistics
Covers observational evidence, bias, confounding, and careful interpretation of data.
How to Lie with Statistics
Helps readers recognize misleading statistical relationships and weak causal arguments.
eBay marketplace picks
Marketplace Samples
Live-tested eBay searches with available results related to this page.
Endnotes
-
Source: jech.bmj.com
Link: https://jech.bmj.com/content/76/11/960Source snippet
Causal inference and effect estimation using observational databy E Igelström · 2022 · Cited by 113 — Fourth, we define and explain biase...
-
Source: ncbi.nlm.nih.gov
Title: NCBIPrinciples of [Causation]({{ ‘causation/’ | relative_url }})
Link: https://www.ncbi.nlm.nih.gov/books/NBK606119/Source snippet
StatPearls - NCBI Bookshelf - NIHby R Dhawan · 2024 · Cited by 2 — Causation refers to a process wherein an initial or inciting event (ex...
-
Source: pmc.ncbi.nlm.nih.gov
Link: https://pmc.ncbi.nlm.nih.gov/articles/PMC8882362/Source snippet
Note: Confounding and Causality in Observational...by C Horvat · 2021 · Cited by 17 — As an increasing number of Bradford Hill criteria...
-
Source: pmc.ncbi.nlm.nih.gov
Link: https://pmc.ncbi.nlm.nih.gov/articles/PMC8206235/Source snippet
Assessing causality in epidemiology: revisiting Bradford Hill to...by M Shimonovich · 2020 · Cited by 231 — Bradford Hill argued that...
-
Source: journals.publisso.de
Link: https://journals.publisso.de/en/journals/gms/volume20/000314Source snippet
Causal evidence in health decision makingby F Kühne · 2022 · Cited by 34 — Causal inference and health decision science are two methodolo...
-
Source: pmc.ncbi.nlm.nih.gov
Link: https://pmc.ncbi.nlm.nih.gov/articles/PMC12133282/Source snippet
evolved interpretation of Austin Bradford Hill's causal...by CR Lesko · 2024 · Cited by 5 — In 1965, Sir Austin Bradford Hill articulate...
-
Source: youtube.com
Title: Confounding Variables in Research: What They Are & How to Control Them
Link: https://www.youtube.com/watch?v=tL86VPitfC4Source snippet
Causal Inference - EXPLAINED...
-
Source: youtube.com
Title: Causal Inference
Link: https://www.youtube.com/watch?v=Od6oAz1Op2kSource snippet
Confounding Examples - Causal Inference...
-
Source: Wikipedia
Link: https://en.wikipedia.org/wiki/Confounding -
Source: pmc.ncbi.nlm.nih.gov
Link: https://pmc.ncbi.nlm.nih.gov/articles/PMC2706236/Source snippet
role of causal criteria in causal inferences: Bradford Hill's...by AC Ward · 2009 · Cited by 141 — Research in epidemiology and the heal...
-
Source: pmc.ncbi.nlm.nih.gov
Link: https://pmc.ncbi.nlm.nih.gov/articles/PMC524370/Source snippet
In his last few paragraphs, he offers an important...R...
-
Source: pmc.ncbi.nlm.nih.gov
Link: https://pmc.ncbi.nlm.nih.gov/articles/PMC4589117/Source snippet
the Bradford Hill criteria in the 21st century: how data...by KM Fedak · 2015 · Cited by 895 — In 1965, Sir Austin Bradford Hill publish...
Additional References
-
Source: merriam-webster.com
Link: https://www.merriam-webster.com/dictionary/causalSource snippet
CAUSAL Definition & Meaning1. expressing or indicating cause: causative a causal clause introduced by since 2. of, relating to, or const...
-
Source: healthknowledge.org.uk
Link: https://www.healthknowledge.org.uk/e-learning/epidemiology/practitioners/causation-epidemiology-association-causationSource snippet
Causation in epidemiology: association and causationAt the end of the session you should be able to differentiate between the concepts of...
-
Source: linkedin.com
Title: Bradford Hill: A tool for causal inference | William J
Link: https://www.linkedin.com/posts/williamjlee_great-post-i-think-about-bradford-hill-a-activity-7231625823621931008-KNNWSource snippet
Lee, JD...Aug 20, 2024 — There are several challenges to inferring causation from observational data. We really need data from multiple...
-
Source: Wikipedia
Title: Bradford Hill criteria
Link: https://en.wikipedia.org/wiki/Bradford_Hill_criteriaSource snippet
Bradford Hill criteriaThe Bradford Hill criteria, otherwise known as Hill's criteria for causation, are a group of nine principles tha...
-
Source: who-umc.org
Title: bradford hill criteria
Link: https://who-umc.org/signal-management/bradford-hill-criteria/Source snippet
3 Feb 2026 — The Bradford Hill criteria for causation were developed by Sir Austin Bradford Hill in 1965. They consist of nine conditions...
-
Source: youtube.com
Title: Correlation and Causation
Link: https://www.youtube.com/watch?v=cvwRt9aCNpkSource snippet
understanding the Bradford Hill...In this video we're going to be looking at reasons for correlation and it includes how we better under...
-
Source: onlinelibrary.wiley.com
Link: https://onlinelibrary.wiley.com/doi/10.1111/add.16329Source snippet
Bradford Hill's 'Environment and disease...27 Aug 2023 — Modernizing the Bradford Hill criteria for assessing causal relationships in ob...
-
Source: frontiersin.org
Link: https://www.frontiersin.org/journals/neurology/articles/10.3389/fneur.2022.938163/fullSource snippet
Applying the Bradford Hill Criteria for Causation to...by CJ Nowinski · 2022 · Cited by 189 — This article aims to explore the question...
-
Source: youtube.com
Title: Clearing Up Confounding
Link: https://www.youtube.com/watch?v=fjdb4ID7HVgSource snippet
Confounding Variables in Research: What They Are & How to Control Them...
-
Source: youtube.com
Title: Confounding Variables: Definition & Examples (3 Minute Explanation)
Link: https://www.youtube.com/watch?v=Wl5nSDTL66USource snippet
Clearing Up Confounding...
Topic Tree