Key ideas
Core concepts
- Variables can be statistically related without one causing the other; confounding third variables often explain apparent causal relationships.
- Establishing a causal relationship requires experimental design with proper controls.
Correlation Does Not Equal Causation is the principle that statistical association between variables doesn’t prove one causes the other (causal relationships require controlled experimental evidence).
Connected to
What Research Can You Trust | Replication Crisis | Jean Piaget | Learning Disabilities
Core principle
Statistical association between variables does not prove that one causes the other to change (Mill, 1843). This error is common in interpreting research and everyday observations.
To establish a causal relationship between variables, a control group is required (Campbell & Stanley, 1963). Without experimental control, correlation remains suggestive, not conclusive.
Ice cream and shark attacks
This absurd example illustrates how third variables create spurious correlations. As ice cream sales increase, shark attacks also increase (a clear correlation that seems meaningful until examined critically). One might conclude that ice cream sales cause shark attacks, but this interpretation is nonsensical. The actual explanation involves a third confounding factor: both ice cream sales and ocean swimming increase during summer months. The correlation between ice cream and shark attacks is real, but attributing causation misses the actual relationship.
Educational applications
Educational research frequently confuses correlation with causation, leading to misguided interventions based on misinterpreted associations.
Brain differences and learning
Brain scans show differences between dyslexic and typical readers (Gabrieli, 2009; Shaywitz et al., 2008), suggesting that brain differences cause reading difficulties. This is correlation, not causation. Brain differences could result from different teaching methods or reading experiences rather than causing them. The brain changes in response to what we learn, so brain differences might be a consequence rather than a cause of reading difficulties (Vellutino et al., 2004). Without experimental manipulation of brain structure or controlled comparison of identical brains with different reading experiences, the causal direction remains unclear.
Piaget’s stages
Piaget observed that children of certain ages show specific cognitive patterns (Piaget, 1952), leading to the conclusion that age causes cognitive abilities. However, prior knowledge and experience, rather than age, likely determine performance (Chi et al., 1981). Children of the same age with different educational experiences show vastly different cognitive abilities, which undermines age-based stage theories. Age correlates with cognitive abilities because it generally correlates with accumulated experience, but chronological age itself does not cause development (the learning and experience accumulated over time do).
Why correlation doesn’t imply causation
There are two potential problems with drawing causal inferences from correlational evidence (Stanovich & Stanovich, 2003).
The third-variable problem
The third-variable problem occurs when the correlation between two variables does not indicate a direct causal path between them but arises because both variables are related to a third variable that has not been measured. The ice cream and shark attacks example illustrates this: the third variable (summer weather) causes both increased ice cream sales and increased ocean swimming, creating a spurious correlation between ice cream and shark attacks.
In educational research, third variables frequently confound interpretations. Students who receive music lessons often achieve higher academic scores. However, music lessons correlate with family socioeconomic status, which independently predicts academic achievement. The correlation between music lessons and academic achievement may result entirely from the third variable of socioeconomic status rather than any causal effect of music instruction.
The directionality problem
Even if two variables have a direct causal relationship, the direction of that relationship is not indicated by the mere presence of the correlation. A correlation between variables A and B could arise because changes in A are causing changes in B or because changes in B are causing changes in A. The correlation alone does not allow us to decide between these possibilities.
Reading skill and vocabulary knowledge are correlated. This could arise because reading causes vocabulary growth (exposure to written language teaches new words), or because vocabulary knowledge causes reading improvement (knowing more words makes reading easier), or both. Correlational evidence alone cannot distinguish between these alternatives.
The logic of the experimental method
The heart of the experimental method lies in manipulation and control (Stanovich & Stanovich, 2003). In a correlational study, the investigator simply observes whether the natural fluctuation in two variables displays a relationship. In a true experiment, the investigator manipulates the variable thought to be the cause (the independent variable) and looks for an effect on the variable thought to be the effect (the dependent variable) whilst holding all other variables constant by control and randomisation.
This method removes the third-variable problem because, in the natural world, many different things are related. The experimental method pries apart these naturally occurring relationships. It isolates one particular variable (the hypothesised cause) by manipulating it and holding everything else constant through control.
Random assignment
When manipulation is combined with random assignment (in which subjects do not determine which experimental condition they will be in but are randomly assigned to one of the experimental groups), scientists can rule out alternative explanations of data patterns. By using manipulation, experimental control, and random assignment, investigators construct stronger comparisons so that the outcome eliminates alternative theories and explanations.
Random assignment ensures that groups are equivalent before the experimental manipulation. Any pre-existing differences between individuals are distributed equally across experimental and control groups. If groups differ after the manipulation, the difference can be attributed to the experimental treatment rather than pre-existing characteristics of the participants.
Control groups
Control groups provide the comparison necessary to isolate the effect of the independent variable. Without a control group, we cannot determine whether changes in the dependent variable result from the experimental manipulation or from other factors such as the passage of time, practice effects, or external events that affect all participants.
If students receive a new reading programme and their reading scores improve, we cannot conclude the programme caused the improvement without a control group. Students might have improved due to maturation, other aspects of schooling, or external factors. A control group that receives standard instruction lets us compare the improvement in the experimental group against the natural rate of improvement, isolating the programme’s effect.
Research implications
The distinction between correlation and causation affects how we evaluate research claims and educational practices (Pearl, 2009). Randomised controlled trials are the gold standard for establishing causation because they control for confounding variables (Shadish et al., 2002). By randomly assigning participants to treatment and control conditions, researchers can isolate the causal effect of a specific intervention.
Observational studies, by contrast, can reveal correlations but cannot prove causal relationships. These studies track existing patterns without manipulating variables, leaving open the possibility of confounding factors that explain observed associations. Critical evaluation of research requires asking “What else could explain this relationship?” before accepting causal claims.
References
Campbell, D. T., & Stanley, J. C. (1963). Experimental and quasi-experimental designs for research. Rand McNally.
Chi, M. T. H., Feltovich, P. J., & Glaser, R. (1981). Categorization and representation of physics problems by experts and novices. Cognitive Science, 5(2), 121-152. https://doi.org/10.1207/s15516709cog0502_2
Gabrieli, J. D. E. (2009). Dyslexia: A new synergy between education and cognitive neuroscience. Science, 325(5938), 280-283. https://doi.org/10.1126/science.1171999
Mill, J. S. (1843). A system of logic, ratiocinative and inductive. John W. Parker.
Pearl, J. (2009). Causality: Models, reasoning, and inference (2nd ed.). Cambridge University Press.
Piaget, J. (1952). The origins of intelligence in children. International Universities Press.
Shadish, W. R., Cook, T. D., & Campbell, D. T. (2002). Experimental and quasi-experimental designs for generalized causal inference. Houghton Mifflin.
Shaywitz, S. E., Morris, R., & Shaywitz, B. A. (2008). The education of dyslexic children from childhood to young adulthood. Annual Review of Psychology, 59, 451-475. https://doi.org/10.1146/annurev.psych.59.103006.093633
Stanovich, P. J., & Stanovich, K. E. (2003). Using research and reason in education: How teachers can use scientifically based research to make curricular and instructional decisions. National Institute for Literacy.
Vellutino, F. R., Fletcher, J. M., Snowling, M. J., & Scanlon, D. M. (2004). Specific reading disability (dyslexia): What have we learned in the past four decades? Journal of Child Psychology and Psychiatry, 45(1), 2-40. https://doi.org/10.1046/j.0021-9630.2003.00305.x