Multivariate Statistics with holy data

*To*: mathgroup at smc.vnet.net*Subject*: [mg25619] Multivariate Statistics with holy data*From*: Chris Johnson <cjohnson at shell.faradic.net>*Date*: Mon, 16 Oct 2000 03:04:41 -0400 (EDT)*Sender*: owner-wri-mathgroup at wolfram.com

This may be heretical, but I'm trying to calculate a covariance matrix on historic data where I'm missing some points in some series. This was being done in Excel, which varied the size of n for each pair of vectors when constructing the matrix. Mathematica seems to use the symbol Null for missing data, and creates a cov matrix symbolically with the n constant for each series. Question 1: Is there a way to make Mathematica ignore the missing data points automatically? Question 2: The resulting matrix will probably not be positive definite (which CovarianceMatrixMLE[] will reject). Are there any references out there to learn more about alternatives like the DispersionMatrix function? Thanks for any help/advice. Chris