Correlation vs causation feels confusing in IB Maths statistics because the same data can provide strong evidence that two variables move together while providing little evidence about why they move together. A correlation coefficient describes an association, but causation is a claim that changing one variable produces a change in another. Those are different conclusions requiring different kinds of evidence.
For an IB exam, the safest principle is simple: correlation alone does not establish causation. Even if Pearson's correlation coefficient is extremely close to or , you should not conclude that one variable causes the other unless the study design and context provide appropriate causal evidence. This article explains the mathematical distinction, why students blur it, and how to answer interpretation questions precisely.
What is the difference between correlation and causation?
Correlation means that two variables are statistically associated. When one variable changes, the other tends to change in a recognizable direction or pattern.
Causation means that changing one variable directly affects the other variable, with credible alternative explanations controlled or ruled out. It is therefore a stronger claim about the process generating the data.
FeatureCorrelationCausationMain questionDo the variables vary together?Does changing one variable affect the other?Typical evidenceScatter diagram, correlation coefficient, regression modelCarefully designed experiment or a wider body of causal evidenceDirectionSymmetric: the correlation of with equals that of with Directional: causing is different from causing Third variablesMay be present and unmeasuredMust be controlled or convincingly addressedSafe IB wording“There is a strong positive linear correlation”“The evidence supports a causal effect of on ”
The mathematical symmetry is particularly important. If study time and examination score have correlation , reversing the variable labels does not change . However, the statement “study time affects examination performance” is not equivalent to “examination performance affects previous study time.” Correlation has no built-in causal direction.
Why does correlation vs causation feel confusing?
A strong pattern looks like an explanation
Humans naturally interpret patterns as stories. If students who study longer tend to obtain higher marks, it feels reasonable to say that studying caused the improvement.
That explanation may be partly correct, but the graph alone cannot establish it. Prior attainment, motivation, sleep, tutoring, course difficulty and the accuracy of self-reported study time could also affect the observed relationship. A convincing-looking line is still a description of the data, not proof of its generating mechanism.
The same variables can be correlated and causally connected
The statement “correlation does not imply causation” does not mean correlated variables can never have a causal relationship. It means correlation by itself is insufficient evidence.
For example, fertilizer quantity and plant growth could be positively correlated because fertilizer promotes growth. They could also be correlated because larger, healthier plants were deliberately given more fertilizer. Deciding between those explanations requires information about how the observations were produced.
Everyday language is less precise than statistical language
In ordinary speech, students often use “leads to,” “results in” and “affects” when they only mean “is associated with.” In statistics, those expressions make causal claims.
IB interpretation questions reward controlled language because conclusions must not exceed the evidence. Compare these statements:
-
Too strong: Students with more sleep achieve better marks because sleep increases performance.
-
Supported by correlation: In this sample, sleep duration and examination mark have a positive linear association.
-
Carefully qualified: Greater sleep may contribute to improved performance, but the observed correlation alone does not establish causation.
What does Pearson's correlation coefficient actually tell you?
In IB Maths statistics, Pearson's product-moment correlation coefficient, written as , measures the direction and strength of a linear association between two quantitative variables:
-
indicates a positive linear correlation.
-
indicates a negative linear correlation.
-
close to or indicates a strong linear association.
The word linear matters. A curved relationship can be strong and predictable even when Pearson's is close to zero. This is why you should inspect the scatter diagram rather than interpreting the coefficient in isolation.
There are also no universal IB cut-offs that make values such as automatically “strong” in every context. Descriptive labels depend on the subject, sample, scatter pattern and purpose of the analysis. If a question or resource supplies a classification scale, use it; otherwise describe the value sensibly and avoid presenting an informal boundary as an official rule.
The current IB Mathematics courses include work with bivariate data, correlation and statistical interpretation. RevisionDojo's IB Maths AI correlation notes provide the wider calculation and regression coverage, while the IB Maths AI Statistics and Probability hub and IB Maths AA statistics notes consolidate the topic-wide, exam-focused material. This article focuses specifically on the conceptual boundary between association and cause.
Why a high value of r does not prove causation
Suppose data from 20 students give between weekly study time and mathematics score. This supports the conclusion that the sample has a strong positive linear correlation. It does not prove that increasing any student's study time will cause a predictable increase in their score.
Several alternative explanations remain possible.
A lurking or confounding variable
A lurking variable is an unmeasured variable that could affect the observed relationship. A confounding variable is related to both the proposed explanatory variable and the response, making their separate effects difficult to distinguish.
For study time and marks, prior mathematical ability or motivation could influence both. Highly motivated students may study longer and use more effective revision methods, so part of the correlation attributed to time may actually reflect motivation or study quality.
Reverse causation
Students sometimes assume that must cause because appears on the horizontal axis. Axis placement does not establish a causal direction.
For example, confidence and practice time may be positively correlated. Practice could improve confidence, but confident students might also choose to practise more. Both directions could operate at once.
Coincidence and multiple comparisons
A sample can display a correlation by chance, especially when it is small or when many pairs of variables are tested. If a researcher compares enough unrelated measurements, some may appear strongly associated without a stable underlying relationship.
This does not make correlation useless. It means the pattern should be evaluated using sample size, context, statistical testing where appropriate, and independent evidence.
Selection and measurement problems
The observed sample may not represent the intended population. A study based only on volunteers from an advanced mathematics class, for example, may not support a conclusion about every IB student.
Measurements can also be inaccurate. Self-reported study hours may include distracted time, use inconsistent definitions, or be affected by students' memory. A precise calculator output cannot repair weak data collection.
IB Maths examples of correlation without established causation
Ice cream sales and sunburn cases
Suppose a city records a strong positive correlation between ice cream sales and reported sunburn cases. It would be unreasonable to conclude that buying ice cream causes sunburn.
Temperature or sunny weather is a plausible lurking variable. Warmer, sunnier conditions increase both ice cream purchases and exposure to sunlight.
Shoe size and reading score in young children
Larger shoe sizes may correlate positively with reading scores in a primary school sample. Shoes do not improve literacy.
Age affects both foot size and reading development. This example shows how a very real correlation can arise without a direct causal link between the measured variables.
Screen time and examination performance
A negative correlation may appear between daily recreational screen time and examination marks. It is tempting to conclude that screen use lowers performance, but the association could involve sleep, stress, parental supervision, time management or prior academic difficulties.
A responsible conclusion is that greater screen time is associated with lower marks in the sample. Further evidence would be required to determine whether reducing screen time itself would improve results.
When can causation be investigated more convincingly?
A randomized controlled experiment provides stronger causal evidence than an observational study. Researchers impose a treatment, randomly assign experimental units to groups, include an appropriate comparison group, and keep other conditions as similar as practicable.
Random assignment matters because it tends to balance both known and unknown confounding factors across treatment groups. If the groups then differ systematically in their outcomes, the treatment becomes a more credible explanation.
Observational studyRandomized experimentResearchers observe naturally occurring variablesResearchers impose a treatmentParticipants may select their own exposureTreatments are assigned randomlyConfounding is difficult to eliminateRandom assignment reduces systematic confoundingUsually supports association claimsCan support causal conclusions when well designed
Do not confuse random sampling with random assignment. Random sampling helps make a sample representative and supports generalization to a population. Random assignment supports causal inference by creating comparable treatment groups.
Experiments are not always ethical or practical. Researchers cannot randomly assign people to smoke for decades, for example. In such cases, causal conclusions may be developed from multiple observational studies, temporal order, plausible mechanisms, consistency, dose-response evidence and sophisticated control of confounding. For an ordinary IB correlation question, however, the expected conclusion is usually that observational correlation alone cannot prove cause and effect.
How to answer correlation and causation questions in an IB exam
Use a three-step structure:
-
Describe the association. State its direction, strength and linear form where appropriate.
-
Reject an unsupported causal conclusion. Explain that correlation alone does not establish causation.
-
Give a contextual reason. Identify a plausible lurking variable, reverse direction or study-design limitation.
For example:
There is a strong positive linear correlation between weekly exercise time and reported wellbeing in this sample. However, this does not establish that additional exercise causes improved wellbeing. A variable such as physical health, income or available leisure time could influence both measurements.
This is stronger than writing only “correlation does not imply causation.” The standard phrase states the principle, but the contextual explanation demonstrates that you understand why it applies.
Match your wording to the command term
-
Describe: Report the visible or calculated relationship without explaining its cause.
-
Interpret: Explain what the statistic means in the context of the variables.
-
Comment: Make a relevant judgment, often including limitations or reliability.
-
Suggest: Offer a plausible explanation, such as a lurking variable, without claiming certainty.
The RevisionDojo explanation of why interpretation matters in IB statistics develops this distinction further. You can then practise the language through the AI correlation Questionbank or the broader Statistics and Probability Questionbank.
Common mistakes that lose marks
-
Treating a large as causal proof. Strength of linear association is not strength of causal evidence.
-
Writing “no relationship” when . The data may have a strong nonlinear relationship.
-
Assuming the -variable causes the -variable. Axis placement and regression notation do not establish direction.
A strong correlation can also produce poor predictions when the relationship changes outside the observed range, contains influential outliers or is applied to a different population. The RevisionDojo guide to strong correlations and weak predictions explains why predictive usefulness and causality are separate issues.
A practical revision method
Create two columns headed What the data show and What the data do not prove. For every practice scatter plot, write one sentence in each column.
For example:
What the data showWhat the data do not proveTaller plants tended to have larger leaf areas.Increasing leaf area would cause a plant to become taller.Temperature and electricity use had a strong positive linear correlation.Higher temperature directly caused all of the additional electricity use.Revision time was negatively associated with errors.Adding one hour of revision would reduce every student's errors by the regression slope.
After learning the distinction, complete several short interpretation questions rather than repeating calculator procedures. Use RevisionDojo's correlation topic page for focused resources, then ask Jojo AI to evaluate whether your conclusion describes association or accidentally claims causation.
Conclusion
Correlation vs causation feels difficult because a strong pattern encourages a causal story, even though the mathematics only measures association. Pearson's describes the strength and direction of a linear relationship; it does not identify a mechanism, establish direction or remove alternative explanations.
In an IB exam, describe the observed relationship precisely, avoid unsupported causal verbs, and identify a plausible confounding factor or design limitation when appropriate. RevisionDojo's Study Notes and Questionbank can reinforce the wider statistics topic, while Jojo AI can help you refine short, context-specific interpretations.
