Arranging data for test of statistical significance from multiple subjects

Hi,

I have 20 different feature matrix as (NxM) from 20 different subjects. Where N (e.g. 1... 10 different features) is fixed but M varies (i.e. 1...., observations) for each subject.

How should I arrange my data such that I will know the statistical significance of each feature by including all the subject's data?

Which kind of tests are suitable when M is varying for each subject? Should I use ANOVA, MANOVA, multicompare etc. or any other?

Any help in this regard will be highly appreciated.

Thanks!

Answer 1

[Tom Lane]

It sounds like you have a univariate rather than multivariate problem. So anova1 rather than manova1.

You have two choices. First, put your data in the N-by-M matrix and pad with NaNs so make up the M total for each subject. Or put your data in a vector V and create a separate vector S giving the subject number for each corresponding element of V. The anova1 function will accept either form.

[LRBLFAST]

Thanks Mr. Tom Lane for your kind reply, But, isn't repeated measures ANOVA http://www.mathworks.in/matlabcentral/fileexchange/30737-n-way-repeated-measures-anova will be suitable in this case? If yes then how to arrange the data again as they've not explained explicitly about the M which varies (unbalanced).

[Tom Lane]

Maybe repeated measures is appropriate. I didn't understand that from the original description. There's currently no repeated measures functionality in the Statistics Toolbox. You may be able to handle this using anovan and specifying subject as random. Not sure.

[LRBLFAST]

Sorry for late replying, My data description is as follows: I've 10 subjects data. From each subject I've extracted 20 features (N) which were sampled over 10 seconds window,but due to variation in records for each subject the no. of samples for a feature is different (i.e. M). For e.g. the size of the feature matrix for Sub 1 = 20x217 whereas for Sub 2 = 20x245 and so on (i.e. N=fixed, M=varying). In total I have 10 such matrices, so putting them as vector (V) will be too lengthy(second choice suggested by you). The link provided in my earlier comment describes N-way repeated measures ANOVA. Will that be valid? If yes kindly suggest how to arrange?

[Tom Lane]

Sorry, I'm not familiar with that File Exchange program. Based on a quick look it seems to require balanced data. You might try writing to the author on the File Exchange page. He seems to respond to questions.

[LRBLFAST]

Thanks very much for your time Mr. Tom Lane.