Normality for biomechanical data

Hi everyone!
I'd like to check my data for normality but I'm not sure how. I'm comparing different motion capturing systems with each other in terms of their accuracy. Therefore, I record the gait of a subject. This gait data is then normalized between 1-100%. Thus, there is a value for each percentage of the gait movement.
Because I want to perform an ANOVA, as far as I know, I need to check the normality of residuals.

In other tutorials of the internet, there is usually 1 value that is related to 1 person. In my case, I have 1 person that has 100 values (each % of the gait). How exactly do I check the normality? In SPSS do I just put in all 100 values and assign them to system 1 and then the next 100 value for system 2 and so on?

Any help is greatly aprecciated. Thank you!


Well-Known Member
While it isn't fully clear to me, perhaps an anova may not be what you want. For an anova, the responses need to be measuring the same thing, under different conditions. It sounds like your data for 1 person is a collection of measurements of different things. (I may well be wrong about this, and if so, please set me right.) kat
Hey, maybe I'll put in other words. I have 100 different values for a subject, corresponding to 1% of each phase of gait. I want to know if these values are normally distributed. So I believe that I would have to compare each percentage of the gait among each subject. Would I then just create 100 variables (corresponding to each percentage of gait) and enter each value of each participant, then run the normality analysis? Let's say I have 9 participants and each have 100 values. For normality purposes, I'd have to check every point of the 100 values between subjects?


Well-Known Member
It looks as if you have 100 response variables (the Points) and you plan to do 100 repeated measures anovas, each one to see if there is a difference between the systems at that point. If this is the case, normality should not be a problem.
But you have a bigger problem. If you do 100 anovas, with each using 0.05 as your cutoff for significance, then you will get about 5 points showing a significant difference, even with no real difference between the systems. The classic way to approach this problem is to do a multivariate anova MANOVA and use all 100 dependent variables at the same time for single p value.
Even this doesn't really answer your problem which is comparing the accuracy (post #1). Anova type analyses check the means of the two systems, but a difference in mean level might be simply fixed by calibration. Perhaps you could be comparing the spread? I don't know what is important in such systems.
I imagine that you need to decide what sort of thing makes one system more accurate than another before you start analyzing the data. kat