Continuous data appear everywhere in sports science and sports medicine: workload measures, range of movement, muscle strength tests, risk scores, golf shot distances, etc. The variations in these all contain useful information.

But too often that information is lost when the data are categorised. We may only be told whether an athlete’s shoulder external rotation is ‘normal’ versus ‘abnormal’; if their leg strength asymmetry on the single leg vertical jump test is >10% or <10%; if a patient’s risk score places them into a ‘high’ or ‘low’ category; or whether a given correlation coefficient or effect size is ‘strong,’ ‘moderate’ or ‘weak.’

Two often unstated assumptions are required for these methods to be valid: (1) there are no meaningful variations within categories and (2) there are meaningful differences around category cut-points. There are a limited number of situations in which both of these assumptions are reasonable in clinical research.1 2 Despite this, the unnecessary and inappropriate categorisation of continuous variables is widespread in much of the medical literature, including sports science and sports medicine.2 3

In this paper, we discuss why researchers often choose to categorise continuous data, the assumptions required, and why it is frequently an inappropriate choice. We then provide a brief example from the sports medicine literature. Finally, we discuss methods to analyse continuous variables without categorisation. We hope to dissuade researchers from categorising continuous data in the most cases.

Researchers commonly encounter continuous data as independent variables or dependent variables of a statistical model. They also encounter other continuous quantities such as correlation coefficients and effect sizes (eg, Cohen’s d), which for simplicity we will also call ‘data’ (table 1).

Recommendations for handling various types of continuous data

[bjsports-2023-107599supp001.pdf]

The choice to categorise may be guided by the misperception that it is simpler to conduct …