Expand your stats expertise with SPM

What is SPM?

SPM (Statistical Parametric Mapping) is a statistical tool that can be used with biomechanical data. It allows for the application of traditional hypothesis testing across a waveform of registered data. For example, let’s say you are working to understand the effect of Botulinum Toxin (Botox) treatment on gait in children diagnosed with spastic cerebral palsy. You complete a gait analysis before and after the Botox injection for the children enrolled in your study. You are curious if there were any differences in lower limb movement between the two time points, an paired samples t-test. However, what joint angle value best represents what happens during the gait cycle? Average? Max? Minimum? Instead of having to choose one value to represent the entire gait cycle movement pattern, you can perform a statistical comparison of the entire kinematic curve with SPM. SPM will allow you to identify where in the time series data there is a significant difference between the two time points (if a difference exists). Nieuwenhuys et al. (2016) actually did this study. They reported differences in ankle dorsiflexion throughout the gait cycle and increased knee extension during terminal stance for the post-Botox treatment time point, concluding that Botox treatment is a valuable option to improve gait for children with spastic cerebral palsy. – A great example of SPM use with biomechanical time series data.

How much of the data that you collect are normalized time-series?  Maybe SPM is a new way for you to draw inferences about your data. We think SPM is a great tool to have as a part of your stats toolbox. Actually, we are so interested in SPM we asked the experts of SPM use for biomechanical analyses to teach a two day workshop for us in July (2019).  And those of you also interested in getting started with SPM can join us! To get a better feel for the value of SPM and the use of stats in general, we had the SPM course instructors write a guest blog post:

Not a fan of statistics? One way or ANOVA we will expand your expertise with SPM.

By Todd Pataky (Kyoto University), Mark Robinson (Liverpool John Moores University) & Jos Vanrenterghem (KU Leuven)

Who has not been an undergraduate student who was forced to go through the perils of extremely abstract statistics lectures in their first years at uni? Perhaps you may have enjoyed learning how to best design an experiment, but surely at that time you were also wondering why on earth every science student has to go through these horrifying experiences of learning about randomness, normal distributions, and the huge variety of statistical tests that each come with their idiosyncrasies. If at that stage you were still paying attention to your stats teacher, then surely you got lost once the practical classes required a magician’s hand to shake the appropriate results out of a complicated stats package printout, right?!

Obviously, as a biomechanist it is easier to hate statistics than to love it. In fact, our definition of fun is getting a better understanding of how humans and/or animals move, how we can help those that are struggling to move (read: injury prevention), or how we can make them move better than the crowd (read: performance enhancement). But why is it then that some biomechanists eventually do get to love statistics, fully understand its value, and strategically incorporate it into their work? Well, if you ask those few, then you will likely get to hear one of the following explanations.

Statistics is the mathematical heart of good science.

Science has been around for a while, and scientists have throughout history had to fight very hard to be taken seriously. Ultimately, technological development and innovations in general depend on solid science, with an ability to identify observations that somehow stand out from the crowd. Fortunately, what we do in scientific experiments can be described mathematically and represented in terms of probability distributions. And if we can model probability appropriately, then that is what gives us the most reliable way to find needles in haystacks. In other words, better statistics and statistical tools lead to more trustworthy science.

Statistics evolves alongside our evolving scientific needs.

It does not take much effort to convince any biomechanist that research topics that were of great interest in our field during the 70’s and 80’s are no longer of interest to the field now. This evolution has come about through rapid technological developments, evolving societal needs, altered funding streams, and sometimes even through fairly random shifts in interests because of historical events or discoveries. These evolutions tend to add complexity to modern day experiments, and with this a constant need to update our statistical tools. For this, statisticians are working tirelessly to push the boundaries of statistics by developing and testing mathematical descriptions of ever more complex research experiments.

✅ P(data|H0)

Statistics are objective.

Unlike humans neither statistics nor science can lie. However, we must always be aware of what we are asking statistics and science, because if we don’t understand the question we will never understand their answers. For example, if a biomechanist needs to observe movement over time to understand how this movement may lead to problems for that individual, then we should at all times carefully chose the observation(s) that is required for it, that is, prior to our experiment. It needs no saying that reducing our observation(s) for practical conveniences, or shifting our focus during our experiment, will make choosing an appropriate statistical model impossible. True, exploratory experiments also exist, in which one may improve the focus of observations, but that is why there are exploratory tools, such as Functional Data Analysis, Principal Component Analysis, Neural Networks, etc. If, however, one has chosen to test a hypothesis that involves an observation over time, then the statistical model should appropriately deal with, for example, the fact that this observation is not a discrete value. The use of SPM can accomplish this. The beauty of a suitable statistical model is then that we can keep our statistical reporting objective!

Modern statistics loves data complexity.

As mentioned earlier, our experiments tend to gain in complexity, mostly because we need to take into account a host of precursor knowledge from prior experiments. Improved technology makes observing less invasive and quicker to analyse, so our biomechanical datasets are also often large and spatiotemporally complex. On the one hand we don’t want to run overly simplistic experiments, as this will often prevent interpretations with relevance to the societal problem, but on the other hand it can be daunting to attempt analyses of these complex datasets.  Fortunately, modern statistical tools, like SPM, can help us to resist overly simplistic interpretations.

Are you still reading but haven’t been convinced to enter a love affair with statistics? Perhaps it is then advised not to start a pub conversation with a statistician, because they will start throwing random arguments (got the pun?) in your direction for hours to follow, none of which will likely do the job of convincing you anyway. However, not loving statistics does not mean that one cannot become friends with statistics. Particularly for biomechanists, there have in the past decade been extensive efforts to upgrade our commonly used statistical tools so that we can deal with complicated experiments and observations. The authors of this blog tend to claim that the introduction of Statistical Parametric Mapping for one-dimensional data (SPM1D) is one essential upgrade that was duly needed in our field. Actually, our experience tells us that trying out SPM1D can, if not more, at least improve one’s friendship status with statistics!

Ready to learn more?

Join the SPM team in Calgary July 29 and 30 2019 for a two day course on getting started with SPM. Early registration closes 5/31/2019.


Nieuwenhuys, A., Papageorgiou, E., Pataky, T., De Laet, T., Molenaers, G., & Desloovere, K. (2016). Literature review and comparison of two statistical methods to evaluate the effect of Botulinum Toxin treatment on gait in children with Cerebral Palsy. PLOS One, 11(3), e0152697.