Correlation vs. Causation — Interpret results and distinguish correlation vs causation in genetic studies - Genetics

SECTION 1

Historical Context & Motivation

For centuries, people assumed that if two things happened together, one must cause the other. If a rooster crows every morning right before sunrise, does the rooster cause the sun to rise? Of course not — but this kind of mistake has been surprisingly common throughout the history of science. In genetics, the stakes are even higher. When scientists discover that a certain gene variant appears more often in people with a disease, it's tempting to conclude that the gene causes the disease. But that jump from observation to conclusion can lead researchers — and the public — astray.

The concepts of correlation (two things happening together in a pattern) and causation (one thing actually making the other happen) have been debated by scientists and philosophers for hundreds of years. Understanding the difference is one of the most important skills in modern genetics and medicine.

1747 • Lind's Scurvy Trial

James Lind ran one of the first controlled experiments, showing that citrus fruit prevented scurvy. He didn't just notice a correlation — he tested it, moving toward proving causation.

1900s • Rediscovery of Mendel's Laws

Mendel's pea plant experiments showed a causal link between inherited factors (genes) and traits. His controlled crosses proved that specific alleles cause specific phenotypes.

1950s • Smoking and Lung Cancer Debate

Scientists found a strong correlation between smoking and lung cancer, but tobacco companies argued it was only a correlation, not proof of causation. Decades of additional studies were needed to establish the causal link.

2003 • Human Genome Project Completed

Mapping the entire human genome launched thousands of studies linking gene variants to diseases. Many found correlations, but proving causation required much more work.

2010s–Present • GWAS and the Correlation Flood

Genome-Wide Association Studies (GWAS) scan millions of genetic markers. They find many correlations between gene variants and traits, but distinguishing true causes from coincidences remains a major challenge.

The big question that scientists still wrestle with today is this: when a genetic study finds a link between a gene and a trait or disease, how do we figure out whether the gene truly causes the trait, or whether the two just happen to appear together? This lesson will give you the tools to answer that question.

SECTION 2

Core Principles & Definitions

Before we can tell correlation and causation apart, we need to understand exactly what each term means and why confusing them is so dangerous — especially in genetics.

Correlation

A correlation exists when two variables change together in a predictable pattern. For example, people who carry a certain gene variant might also be taller on average. This doesn't mean the gene makes them tall — it only means the two things are statistically linked.

Causation

Causation means that one variable directly produces a change in another. If a specific mutation in a gene disrupts a protein that controls bone growth, and this disruption reliably leads to shorter stature, we can say the mutation causes the change in height.

Confounding Variables

A confounding variable (also called a confounder) is a hidden third factor that influences both variables you're studying. It can create the illusion of a direct link when there isn't one. For example, a gene variant might correlate with a disease only because both are more common in a certain population due to ancestry, not biology.

Reverse Causation

Reverse causation occurs when you think A causes B, but actually B causes A. In genetics, this is less common with DNA (since your DNA doesn't change because you get a disease), but it can happen with gene expression — a disease might turn certain genes on or off, making it look like those genes caused the disease.

Linkage Disequilibrium

Linkage disequilibrium is a genetics-specific source of false correlations. Two gene variants that sit close together on the same chromosome are inherited together. A variant near a disease-causing gene may appear correlated with the disease, even though it plays no role in causing it.

✦

KEY TAKEAWAY

Think of correlation like noticing that every time you bring an umbrella to school, you also get a good grade on a test. The umbrella doesn't cause the good grade. Maybe rainy days keep you indoors studying — that's the confounding variable. In genetics, just because a gene variant shows up alongside a trait doesn't mean the gene is responsible. You need controlled experiments to prove a cause-and-effect relationship.

SECTION 3

Visualizing Correlation vs. Causation

The diagram below shows three different relationships that can exist between a gene variant and a disease. Understanding these patterns is the key to interpreting genetic study results correctly.

Scenario 1 shows true causation: a gene variant breaks a protein, which leads to disease through a known mechanism. Scenario 2 shows confounding: a hidden factor like shared ancestry makes a gene and disease appear linked. Scenario 3 shows linkage disequilibrium: a harmless gene sits near the real culprit and gets "blamed" because they're inherited together.

Notice how in Scenario 1, there's a clear chain of events from gene to protein to disease. That's what real causation looks like — you can trace every step. In Scenarios 2 and 3, the gene variant and the disease appear together, so they show a correlation, but there's no direct pathway connecting them. The challenge of modern genetics is figuring out which scenario you're looking at when you find a statistical link in your data.

SECTION 4

How Scientists Measure Correlation in Genetics

In genetic studies, researchers use math to measure how strongly two things are linked. The most common tool is the correlation coefficient, symbolized by the letter r. This number tells you how closely two variables move together.

CORRELATION COEFFICIENT RANGE

−1 ≤ r ≤ +1

r = +1: Perfect positive correlation (as one goes up, the other always goes up). r = 0: No correlation (no pattern at all). r = −1: Perfect negative correlation (as one goes up, the other always goes down).

In genetics, researchers also use a statistic called a p-value to determine whether a correlation is likely real or just due to chance. A p-value tells you the probability that you'd see a result this strong even if there were no real connection between the gene and the trait.

STATISTICAL SIGNIFICANCE THRESHOLD

p < 0.05 (standard) or p < 5 × 10⁻⁸ (GWAS)

A p-value below 0.05 means there's less than a 5% chance the result is a coincidence. In GWAS studies, which test millions of gene variants at once, scientists use an extremely strict threshold of 5 × 10⁻⁸ to avoid false positives.

ODDS RATIO (OR)

OR = (a × d) ÷ (b × c)

In a 2 × 2 table comparing people with/without a gene variant and with/without a disease: a = have variant & disease, b = have variant & no disease, c = no variant & disease, d = no variant & no disease. An OR > 1 means the variant is associated with higher disease risk.

⚠

Important Reminder

Even a very strong correlation (r close to +1 or −1) and a very small p-value do not prove causation! These numbers only tell you that the pattern is unlikely to be random. You still need experimental evidence or a known biological mechanism to claim one thing causes the other.

SECTION 5

Types of Genetic Studies: From Correlation to Causation

Not all genetic studies are created equal. Some can only find correlations, while others can provide evidence of causation. Understanding the different study types helps you judge how much weight to give a finding.

This ladder of evidence shows how different types of genetic studies provide increasingly strong evidence. Observational studies (left) can only find correlations. Moving right, each type adds more experimental control, building toward causal proof.

The key lesson from this ladder is that no single study proves causation. Scientists build a case by combining multiple types of evidence. A GWAS might identify a suspicious gene variant. Then family studies check whether the variant tracks with the disease across generations. Then a lab experiment using CRISPR (a gene-editing tool) might disable that gene in mice to see if the disease appears. When all these lines of evidence agree, scientists grow confident that the relationship is truly causal.

ℹ

Real-World Example

The gene BRCA1 was first identified through family studies of women with breast cancer. GWAS confirmed the correlation. Then laboratory experiments showed exactly how mutations in BRCA1 break the cell's DNA-repair system, allowing cancer to develop. This combination of evidence is why we confidently say BRCA1 mutations cause increased cancer risk — not just correlate with it.

SECTION 6

Worked Example: Evaluating a Genetic Study

Let's walk through a realistic scenario. Imagine you're a scientist who just read a study with the following finding: "People who carry the variant rs12345 in the FTO gene are, on average, 3 kg heavier than people without the variant (p = 0.001, OR = 1.4)." Does this gene variant cause weight gain?

Is the FTO Gene Variant a Cause of Weight Gain?

1
Step 1 — Identify What Kind of Study This Is

The study reports an odds ratio and a p-value from comparing two groups of people. This is an observational study — it looked at data from a population but did not manipulate any genes. Observational studies sit at the bottom of our evidence ladder.

This study can show correlation but not causation on its own.
2
Step 2 — Check the Statistics

The p-value is 0.001, which is less than the standard threshold of 0.05. This means there is only a 0.1% chance the result is due to random chance. The odds ratio (OR) of 1.4 means people with the variant are 1.4 times more likely to be in the heavier group. The correlation appears statistically significant.

p = 0.001 < 0.05 → statistically significant. OR = 1.4 → moderate association.
3
Step 3 — Look for Confounding Variables

Could something else explain this link? Diet, exercise, socioeconomic status, and ethnic background can all affect body weight. If the variant is more common in a population that also has different dietary habits, then diet — not the gene — could be the real driver. A well-designed study should control for these factors (account for them mathematically).

Must rule out confounders like diet, ancestry, and lifestyle.
4
Step 4 — Search for a Biological Mechanism

Does the FTO gene do something biologically related to weight? Yes! Research shows the FTO protein is expressed in the brain and helps regulate appetite and energy balance. The rs12345 variant changes how much FTO protein is produced, which affects hunger signals. This gives us a plausible mechanism — a logical chain from gene to trait.

Biological mechanism identified: FTO affects appetite regulation in the brain.
5
Step 5 — Look for Converging Evidence

Has this result been replicated in other populations? Yes — multiple GWAS across different countries found the same association. Have lab experiments confirmed it? Yes — when researchers knocked out the FTO gene in mice, the mice became leaner. With observational data, biological mechanism, and experimental confirmation all pointing the same direction, the evidence shifts from mere correlation toward strong support for causation.

Conclusion: Multiple lines of evidence support that the FTO variant causally influences body weight.

SECTION 7

Correlation vs. Causation: Side-by-Side Comparison

Let's put correlation and causation side by side so you can clearly see how they differ. Understanding these distinctions will help you evaluate any genetic study you encounter.

Feature	Correlation	Causation
Definition	Two things happen together in a pattern	One thing directly makes the other happen
Direction	Does not tell you which causes which	Specifies A → B (direction is known)
Evidence needed	Statistical test showing a significant pattern (e.g., p < 0.05)	Controlled experiments, biological mechanism, and replication
Study type	Observational (GWAS, surveys)	Experimental (gene knockouts, CRISPR)
Confounders	Can be explained by hidden third variables	Controlled for through experimental design
Genetics example	"Gene variant X is found more often in people with asthma"	"Mutation in CFTR gene produces a faulty protein → thick mucus → cystic fibrosis"
Can prove alone?	No — correlation alone never proves a cause	Yes — when mechanism + experiment + replication all agree

Side-by-side comparison of correlation and causation in genetic research

✦

KEY TAKEAWAY

Here's a helpful way to remember it: correlation is like seeing footprints at a crime scene — it tells you someone was there, but not who did it or why. Causation is like having a video of the crime — you can actually see what happened, step by step. In genetics, GWAS studies give us the footprints (correlations), and lab experiments like CRISPR give us the video (causal proof).

SECTION 8

Connection to Advanced Genetics & Genomics

As you continue studying genetics, you'll encounter more advanced methods that blur the line between observational and experimental studies. One important technique is called Mendelian Randomization. This clever approach uses the fact that your gene variants are randomly assigned at conception (like a natural experiment) to test causal claims without actually doing a lab experiment.

Feature	Standard GWAS (This Lesson)	Mendelian Randomization (Advanced)
Goal	Find correlations between gene variants and traits	Use gene variants as natural experiments to test causation
How it works	Compares genetic markers across large populations	Uses genetic variants as "instruments" to mimic random assignment
Can establish causation?	No — only correlation	Potentially yes — if assumptions are met
Handles confounders?	Must control for them statistically	Naturally avoids many confounders because genes are randomly inherited
Example question	"Is vitamin D level correlated with depression?"	"Do genes that cause higher vitamin D levels also lead to lower depression rates?"

How Mendelian Randomization extends the concepts from this lesson

Another advanced topic you'll encounter is polygenic risk scores. These scores combine the effects of hundreds or thousands of small genetic correlations to predict your overall risk for a disease. While each individual correlation might be weak, together they can be quite powerful. However, because they're built from correlations, they come with all the limitations we've discussed — confounders, population differences, and the inability to prove that any single variant is truly causal.

💡

Looking Ahead

Technologies like CRISPR base editing and single-cell sequencing are making it faster and cheaper to test whether specific gene variants actually cause disease. In the coming years, many correlations found by GWAS will be tested and either confirmed as causal or revealed to be coincidences. Your ability to tell the difference will be more important than ever.

SECTION 9

Practice Problems

Question 1: CONCEPTUAL

A student reads a headline that says: "Scientists discover the gene for intelligence." The study behind the headline found that a certain gene variant appeared more often in people who scored higher on IQ tests (p = 0.02). Does this study prove that the gene causes intelligence? Explain why or why not.

Show answer

No, this study does not prove causation. A p-value of 0.02 means the correlation is statistically significant (unlikely to be due to chance alone), but correlation does not equal causation. The gene variant might be more common in certain populations that also have better access to education (a confounding variable). Additionally, intelligence is influenced by hundreds of genes and many environmental factors. A single observational study can only show a pattern — it cannot prove that this specific gene directly produces higher intelligence. To move toward causation, scientists would need to identify a biological mechanism, replicate the result across different populations, and ideally perform functional experiments.

Question 2: BASIC CALCULATION

In a study of 1,000 people, researchers build a 2 × 2 table. Among people with the gene variant: 120 have the disease (a) and 280 do not (b). Among people without the variant: 80 have the disease (c) and 520 do not (d). Calculate the odds ratio (OR) and determine whether the gene variant is positively or negatively associated with the disease.

Show answer

Using the odds ratio formula: OR = (a × d) ÷ (b × c) = (120 × 520) ÷ (280 × 80) = 62,400 ÷ 22,400 ≈ 2.79. Because the odds ratio is greater than 1, the gene variant is positively associated with the disease — people who carry the variant are about 2.79 times more likely to have the disease compared to people without it. However, this is still only a correlation. The OR tells us the association is fairly strong, but we would need additional evidence (biological mechanism, experimental tests, replication in other populations) before concluding that the variant actually causes the disease.

Question 3: INTERMEDIATE

A GWAS study conducted only in people of European ancestry finds a strong correlation between gene variant rs99999 and Type 2 diabetes (p = 3 × 10⁻¹⁰). However, when the same study is repeated in people of East Asian ancestry, the correlation disappears (p = 0.45). What are two possible explanations for this discrepancy, and what does it suggest about the original finding?

Show answer

Two possible explanations are: (1) Population stratification as a confounding variable — the variant rs99999 might be common in a subgroup of European-ancestry people who also have higher diabetes rates due to shared environmental or lifestyle factors, rather than because of the gene itself. When the confounding population structure changes (East Asian ancestry), the false signal disappears. (2) Linkage disequilibrium differences — rs99999 might not cause diabetes itself but might sit near a true causal gene on the chromosome in European populations. In East Asian populations, different patterns of genetic recombination may have broken this linkage, so rs99999 no longer tags the causal variant. This discrepancy suggests that the original finding might reflect a correlation driven by confounders or linkage rather than true causation, and further investigation (like functional experiments) is needed.

Question 4: APPLIED

A pharmaceutical company wants to develop a drug that targets Protein X. Their reasoning: a GWAS found that people with a variant in Gene X (which produces Protein X) have lower rates of heart disease. Before investing billions of dollars, what specific types of evidence should the company gather to make sure the relationship is causal and not just a correlation?

Show answer

The company should gather evidence at multiple levels. First, they should confirm the GWAS result by replicating it in independent populations with different ancestries to rule out population stratification. Second, they should perform functional laboratory experiments: use CRISPR to modify Gene X in animal models and observe whether this changes heart disease outcomes. Third, they should establish a biological mechanism — does Protein X do something relevant to heart health, like reducing cholesterol or preventing inflammation? Fourth, they could use Mendelian Randomization to test whether people who naturally produce more or less Protein X (due to their genetic variants) have correspondingly different heart disease rates, controlling for confounders. Only with converging evidence from observational studies, mechanism studies, animal experiments, and Mendelian Randomization should the company feel confident that targeting Protein X will actually prevent heart disease.

Question 5: CRITICAL THINKING

Consider this claim: "Because your DNA sequence is fixed at birth and cannot be changed by diseases, any correlation found between a gene variant and a disease must be causal — the gene must cause the disease, since the disease can't change the gene." Evaluate this argument. Is the logic correct? Identify at least two flaws in this reasoning.

Show answer

While it's true that your DNA sequence is generally fixed at birth, this argument contains at least two major flaws. First, it ignores confounding variables entirely. A gene variant and a disease can be correlated without either one causing the other if a third factor (like shared ancestry, socioeconomic status, or environmental exposure) influences both the frequency of the variant in a population and the rate of the disease. The direction-of-causation argument only eliminates reverse causation (disease → gene), not confounding. Second, the argument ignores linkage disequilibrium. A gene variant might correlate with a disease not because it causes the disease, but because it sits close to another variant on the same chromosome that is the true cause. Third, while DNA sequence is fixed, gene expression is not — epigenetic changes and environmental interactions can make the same gene variant behave differently in different people, complicating the picture further. So even though reverse causation is largely ruled out in genetics, that alone is not enough to claim causation.

SUMMARY

Lesson Summary

Summary

Correlation means two variables change together in a pattern, while causation means one directly produces a change in the other. In genetic studies, GWAS and other observational studies can identify correlations between gene variants and traits, but they cannot prove causation on their own. Hidden factors called confounding variables (like shared ancestry or environmental differences) and linkage disequilibrium (where nearby genes are inherited together) can create false associations that look meaningful but aren't. Statistical tools like the p-value and odds ratio measure the strength of a correlation but cannot, by themselves, establish a cause.

To move from correlation toward causation, scientists climb a ladder of evidence: from observational studies, to family and twin studies, to functional experiments like gene knockouts and CRISPR editing, and finally to techniques like Mendelian Randomization that use natural genetic variation as a built-in experiment. True causal proof in genetics requires converging evidence — multiple independent lines of study all pointing in the same direction. Remember: correlation is the starting point, not the finish line.