Local Computation of Answers to Table Queries on Summary Databases

We address the problem of evaluating table queries from a summary database formed by a collection of pre-computed tables on certain measure variables. We assume that every table query asks for the distribution of a measure variable of interest, and that the summary database contains tables on the variable of interest as well as on other measure variables. If the requested distribution is none of the base tables and cannot be exactly derivable from none of them, then the answer to the query will be the result of an estimation procedure, which may bring up another measure variable that is correlated to the measure variable of interest. We give an estimation procedure that combines the "divide-and-conquer" principle with tree computations.