ͯÑÕÊÓƵ

How much can we really trust a measurement? And how much data do researchers need to collect in order to draw reliable conclusions? Mohammad Reza Seydi at Umeå University has developed and tested new statistical methods to provide better answers to those questions.

When researchers compare how people move, it is not always enough to know whether there is a difference between two groups. It may also be important to identify when during the movement that difference occurs.

Seydi proposes a new way of calculating how many measurements are required to detect the specific differences a researcher is interested in. The calculation can, for example, be targeted at the part of a movement that is most important to investigate. This means that researchers can tailor the amount of data needed to what they actually want to detect already when planning a study.

When a measurement becomes an entire curve

To understand why this is needed, consider what happens when a researcher measures a movement. When a knee bends during a jump, the angle changes from one moment to the next. The result is therefore not a single measurement value, but an entire curve.

“Imagine measuring how your knee bends during a jump, or how the angle of a sprinter’s hip moves back and forth with each step. The measurement is not a single number, but an entire curve, or a continuous function, that changes from moment to moment throughout the movement,” says Mohammad Reza Seydi.

This type of measurement is known as functional data and is used in fields such as biomechanics and medicine. While well-established statistical methods have long existed for traditional data, methods for functional data are not yet as extensively developed and tested.

Seydi’s research also shows that the ability to detect differences in such data is influenced by how researchers define what should count as a statistically detectable difference. This forms the basis for the new way of calculating how much data a study requires.

How reliable is the measurement?

Another part of the research concerns how researchers can determine whether repeated measurements are reliable. Seydi has developed and tested methods for assessing measurement reliability and quantifying the uncertainty associated with those assessments.

Using simulations, he has also developed recommendations for how researchers can calculate the amount of data required when the object being measured is a curve or a function.

The methods can be applied in biomechanics and medicine, but also in other research fields where what is being measured changes over time. They are designed to be useful throughout the research process, from study planning to analysis and reporting of results.

“Taken together, these findings provide researchers, not only in biomechanics but in fact in any field working with curve data or functional data, with validated tools for statistical analysis, from the planning of a study to the reporting of its results,” says Mohammad Reza Seydi.