The outline I’ve given above will work for almost any data. (Technically the within-subject deviations have to be normally distributed but this is will be the case for most commonly encountered measurement data). Given that I’m a gait analyst however there are a couple of points that are worth adding. Most of these stem from the fact that most gait analysis data is supplied as a time series of data points across the gait cycle and that we often capture data from several cycles as part of any gait analysis session.
I’m often asked how the SEM should be calculated given all these factors:
- Should I chose a representative cycle or the mean of several cycles?
- How do I adapt this if my focus is on the peak power generation during the gait cycle rather than the value at a particular time?
- Should I calculate the SEM across the gait cycle or at one specific instant?
The simple answer is that you should process the data for a repeatability study in exactly the same way as you are going to process the data in your definitive study. Thus if when you analyse your data you take the mean of several cycles whenever you make measurements on an individual then you should do this when calculating the SEM. If your primary outcome measure is the peak of power generation wherever it occurs in the gait cycle, or is the impulse of a motion over stance then you should do this calculation first for the repeatability study and calculate the SEM of the measurement you are interested in.
There are a growing number of repeatability studies for biomechanical models which produce time varying outputs across the gait cycle. These tend to be providing data for generic use rather than in a specific context. It is my suggestion in this case that the SEM is calculated separately at each interval of the time normalised gait cycle (generally 50 or 100 data points). To do this it is often easier to re-format the data so that all the data from a particular time point can be represented on a single line. Data from the next time point can be input onto the next line and the analysis over multiple time points can be replicated very quickly. This can be further enhanced by putting data from other gait variables in the same columns one below the other in which case the whole analysis can be performed as the analysis of a number of long columns.
The SEMs can be plotted across the gait cycle. For most kinematic variables the SEM is fairly constant and the average SEM across the gait cycle can be used as a broad indicator of the variability. For some other forms of data (particularly EMG) the SEM can vary quite substantially across the gait cycle and a single summary value may not be as useful.
Next page: A more complex (and realistic example)