"Donald Burrill" <[EMAIL PROTECTED]> wrote in message [EMAIL PROTECTED]">news:[EMAIL PROTECTED]... > This sounds as though you have a N-by-T data matrix: N cases or > observations (as rows in the matrix), and T variables (as columns). > If you had something else in mind, my subsequent remarks may verge on > nonsense. Using "T" leads one to wonder whether the data in fact form a > time series? Yes it is a time series, in fact they are measured impulse responses! And yes the data form (or could form) a N-by-T data matrix, as you mentioned. > > Each fn(t) can be considered to have one part, fsame(t), that is the > > same throughout the dataset and another that is varying, > > fn_different(t). > > In other words, for fn(t), t = 1,...,T, fn(t) = constant for all N rows, > t = 1,...,(say) k; and fn(t) varies from row to row, t = k+1,...,T ? > Do you know in advance which t are associated with constant fn(t) and > which t are associated with varyhing fn(t)? Or is that part of what is > to be inferred from the data? Let me try to be more specific: The idea of the specific measuring method that I'm using, is that when averaging the N fn(t), ie. 1/N*sum(fn(t)), fn_different(t) averages out for large N, but not totally for N <"large", and I'm left with app. fsame(t). This demands that fn_different(t) is statistically independent/uncorrelated with each other. If this was true, for large N this would give me fn_different(t)=1/N*sum(fn(t))-fn(t). I know that this is not 100% true, but I would like to determine to what degree they are stat.ind. / uncorr. and establish knowledge on how each fn_different(t) depends on each other. I do not know in advance which t that are associated with constant fn(t) and which t that are associated with the varying part of fn(t). > > I'm interested in performing some sort of correlation/statistical > > analysis of the data, that can tell me how the part of the data in > > fn(t), that are varying (ie. fn_different(t)) are dependent of > > each other, ie. are the parts statistically independent or not, and if > > not with which distribution do they depend of each other or ? > > This sounds as though you want to know what correlational structure > exists among the (T-k) variables fn(t), t = k+1,...T. What kinds of > models of relationships among variables are you interested in > considering? What particular models are important to you? It is important to me to test the hypothesis of the method, i.e. are the fn_different(t) statistically really independent of each other ? Regarding the models I don't have a clue! I considered using the above statement fn_different(t)=1/N*sum(fn(t))-fn(t) and calculate the crosscorrelation for lag 0, between them, but this only tells me to what degree they are uncorrelated, not which kind of statistical dependence there are between them. If it means anything I'm using Matlab. kind regards Johnny ================================================================= Instructions for joining and leaving this list and remarks about the problem of INAPPROPRIATE MESSAGES are available at http://jse.stat.ncsu.edu/ =================================================================
