A tertile is a statistical term for dividing a dataset into three equal groups. If you ranked everyone in a study from lowest to highest on some measure, the tertiles would split them into a bottom third, a middle third, and a top third. You’ll most often encounter this term in medical and health research, where it’s used to compare outcomes across these three groups.
How Tertiles Work
To create tertiles, you first sort all the values in your dataset from smallest to largest. Then you find the two cut points that split the data into three groups of roughly equal size. The first tertile contains the lowest third of values, the second tertile the middle third, and the third tertile the highest third. With 900 people in a study, each tertile would contain about 300 people.
Technically, the word “tertile” refers to the two dividing points themselves (at the 33rd and 67th percentiles), but in everyday use, people almost always use it to mean the three groups those points create. When a study says “participants in the highest tertile,” it means the top third of the sample.
With discrete or whole-number data, the groups won’t always be perfectly equal. If several people share the same value right at a cut point, they all get placed in the same group, which can make one tertile slightly larger than the others. Statistical software handles this automatically, but it’s why you’ll sometimes see tertile groups with slightly uneven numbers.
How Tertiles Compare to Other Quantiles
Tertiles are one member of a larger family called quantiles, all of which do the same basic thing: split ranked data into equal-sized groups. The differences are just in how many groups you get.
- Tertiles: 3 groups (each containing 33.3% of the data)
- Quartiles: 4 groups (25% each)
- Quintiles: 5 groups (20% each)
- Deciles: 10 groups (10% each)
- Percentiles: 100 groups (1% each)
Researchers choose between these based on their sample size and what they’re trying to show. With a smaller sample, tertiles make sense because each group still has enough people to draw meaningful comparisons. With thousands of participants, quintiles or deciles can reveal more fine-grained patterns. Tertiles are the simplest split beyond just dividing data in half, which makes them a popular middle ground.
How Researchers Use Tertiles
Tertiles show up constantly in health and epidemiology research as a way to turn a continuous measurement into something easier to compare. Instead of trying to describe the effect of every incremental change in, say, dietary fat intake, researchers can group people into low, medium, and high intake and compare health outcomes across those three groups.
The standard approach is to label the tertiles T1 (lowest), T2 (middle), and T3 (highest), then use the lowest tertile as the reference group. This lets researchers say something like “people in the highest tertile of sodium intake had a 40% greater risk of high blood pressure compared to people in the lowest tertile.” That kind of statement is much easier to grasp than a per-unit risk estimate.
One study on the relationship between neighborhood supermarket access and body mass index, for example, divided participants into tertiles based on the number of supermarkets within 5 kilometers of their home: 0 to 9 supermarkets, 10 to 14, and 15 or more. People in the highest tertile had a BMI about 1.5 points lower than those in the lowest tertile. A study tracking socioeconomic changes from birth to adulthood used income tertiles at two time points (birth and age 19), classifying the lowest third as “poor” and the upper two thirds as “non-poor.” By crossing the two time points, researchers created nine possible trajectories of economic change.
Limitations Worth Knowing
Tertiles are useful, but they come with a real tradeoff: they erase the detail within each group. Everyone in the top third gets treated the same, whether they’re barely above the cut point or at the extreme high end. This can sometimes mask important differences or make relationships between variables look different than they actually are.
The cut points themselves are also somewhat arbitrary. If you randomly split the same population into two separate samples and calculated tertiles for each, the cut points could land in noticeably different places. Research on built environment and health outcomes has demonstrated this directly, showing that tertile boundaries can shift between random subsamples of the same data, potentially changing which group a person falls into and altering the study’s conclusions.
This is why some statisticians argue that keeping variables continuous, when possible, preserves more information than chopping them into groups. Still, tertiles remain widely used because they make results intuitive and easy to communicate, especially when the goal is comparing a “high” group to a “low” group in plain terms.