Imagine two hospitals treating the same disease. Hospital A has a better survival rate than Hospital B for mild cases, and a better survival rate for severe cases too. Yet somehow, when you combine all the patients together, Hospital B comes out ahead overall. This is not a trick or a typo. It is a real and well-documented statistical phenomenon called Simpson’s paradox, and it shows up far more often than most people realize.
What Is Simpson’s Paradox?
Simpson’s paradox occurs when a trend that appears in several different groups of data disappears, or even reverses, when those groups are combined into one. It is named after statistician Edward Simpson, who described the effect in a 1951 paper, although earlier statisticians had noticed similar patterns before him. The paradox is not a flaw in mathematics; the numbers are entirely correct at every step. Instead, it reveals how a hidden variable can completely change the story the data seems to tell, depending on how the data is grouped.
A Classic Example: Hospital Survival Rates
To see exactly how the paradox works, here is a simplified but faithful version of the classic pattern, broken down by case severity and then combined:
| Group | Hospital A | Hospital B |
|---|---|---|
| Mild cases: survivors / total | 81 / 87 (93%) | 234 / 270 (87%) |
| Severe cases: survivors / total | 192 / 263 (73%) | 55 / 80 (69%) |
| Combined total: survivors / total | 273 / 350 (78%) | 289 / 350 (83%) |
Notice what happened. Hospital A has a higher survival rate for mild cases and a higher survival rate for severe cases, yet Hospital B has the higher overall combined rate. The explanation is that Hospital A treats a much larger share of severe, harder-to-save cases, which pulls its combined average down, even though it is actually performing better within each category.
Why Does This Happen?
Simpson’s paradox arises when there is a lurking variable, sometimes called a confounding variable, that is unevenly distributed across the groups being compared. In the hospital example, the severity of the case is the lurking variable. Because Hospital A sees proportionally more severe cases than Hospital B, simply adding up totals without accounting for case severity produces a misleading combined figure. The paradox is really a warning about the danger of aggregating data across groups that are not actually comparable.
Other Real-World Examples
Simpson’s paradox is not just a hospital curiosity. It has appeared in many well-documented real situations:
- University admissions: A university can appear to favor one gender overall in admission rates, while actually admitting a higher percentage of both genders within nearly every individual department, simply because one gender applied more heavily to more competitive departments.
- Batting averages in sports: A baseball player can have a higher batting average than a teammate in each individual season, yet a lower combined average across both seasons, if the number of at-bats differs significantly between the seasons.
- Public health trends: A treatment can appear less effective overall while being more effective within every age group, if the age groups receiving the treatment are distributed unevenly.
How to Spot and Avoid Being Misled
Because Simpson’s paradox can make a true underlying trend look exactly backward, it pays to build a habit of questioning aggregated statistics. Here are practical steps to protect yourself from being fooled:
- Ask how the groups were combined. Whenever you see an overall average or rate, ask whether it is hiding meaningful subgroups underneath.
- Look for uneven group sizes. Simpson’s paradox is far more likely to appear when the groups being compared are not proportionally similar in size.
- Break the data down by category. Whenever possible, examine the trend separately within each relevant subgroup before trusting a combined number.
- Identify potential lurking variables. Think about what factors, like case severity, department choice, or experience level, might be unevenly distributed across the groups.
- Be skeptical of single summary statistics. A single combined percentage rarely tells the whole story on its own.
Simpson’s paradox is a powerful reminder that correlation, causation, and aggregation are three very different things, and confusing them can lead to conclusions that are the exact opposite of the truth.
Why This Matters Beyond Statistics Class
Simpson’s paradox is not just an academic curiosity for a textbook. It has real consequences in medicine, policy, business, and everyday decision-making. A manager comparing overall sales performance across teams, a journalist reporting on national trends, or a voter evaluating a policy’s effectiveness can all be misled by aggregated numbers that hide an important lurking variable. Recognizing the possibility of this paradox encourages a healthier, more careful relationship with data: always ask what is underneath the summary number before drawing a conclusion.
The Bigger Lesson
At its core, Simpson’s paradox teaches that context changes meaning. The same set of raw numbers can support two opposite conclusions depending on how they are grouped and analyzed. This is not a reason to distrust statistics altogether, but rather a reason to engage with data more carefully, asking not just “what does the number say” but “what is the number actually measuring, and what might be hidden inside it.”
If exploring how statistics can both reveal and disguise the truth fascinates you, Cursa has free courses on statistics and data analysis that walk through paradoxes like this one in much greater depth.
