A data collection procedure is described, which can be performed automatically for each subject of a study as the participant produces the data being collected. In one case, the procedure transforms the data matrix X (of the participants' data) to AXB, where matrix A is a row operator that transforms data records (cases) in X and matrix B is a column operator that transforms data attributes (variables) in X, and the keys to generate these random operators are held separately by different parties. In another case, each participant's data is decomposed into a sum of k vectors before being collected and variously transformed by a plurality of masking service providers.