Hello, my first post. I have long enjoyed reading your posts here! I work in orthopedics at a university hospital in Sweden.

My question is about Statistical Parameter Mapping, a Statistical method used more and more when analyzing for example gait, where you measure some angles in the foot while walking.

“spm1d is a package for one-dimensional Statistical Parametric Mapping (SPM). spm1d uses Random Field Theory expectations regarding smooth, one-dimensional (random) Gaussian fields to make statistical inferences regarding a set of 1D measurements.”

So if you have two groups and want to compare the whole movement there is a value for each second and the question is if there is a difference in the whole movement, not just at some point during the movement as is frequently done.

Todd Pataky has created a module that runs in MatLab that researchers has been using in research and it is picking up momentum and is now in many published articles.

My question is if this is a valid approach from a statistical point of view?
If anyone is familiar with the method?

What does this solve that the older methods of longitudinal modeling with smooth nonlinear time trends (using spline functions) doesn’t solve? With longitudinal models once can estimate the difference in trajectories and test whether two trajectories differ in slope or shape or just differ in vertical shift.

For those unfamiliar, this seems to be an adaptation of Keith Worsley’s approach, statistical parametric mapping or SPM, for mapping behavioral experiments onto locations in the brain.

In this context, random field theory was developed to enable inference on highly correlated coefficients (more precisely on their inferential statistics) obtained from solving multivariate regression (multiple predictors and multiple outcomes)

At first glance, I don’t follow why one would need 1D version. It seems to me that much simpler methods from time-series analyses that employ good change detection statistics would be adequate.

You should be aware that SPM makes strong assumptions about spatial regularity. When these assumptions are not true, it makes the resulting p-values invalid. I’m grossly simplifying the nuances here but there was is some influential body of work on this a while back.

Thank you!
I myself do not know the answer to your question. But i have not seen any of that approach you mention in any research for gait analysis for example, where you have a value measured every half second during a gait cycle.

Before the method of Statistical Parameter Mapping there was only some arbitrary point or region during the gait cycle that was analyzed.

But most often the clinicians wants to know if the whole gait cycle differs between some patologies for example.

Thank you for this discussion, it will certainly aid me in the talks I have with researchers at our hospital.