The technique of partial aggregation is explored as a means of preserving the confidentiality of data while enabling research scholars to utilize the information for analytic purposes. For this purpose, two criteria are developed for evaluating the analytic consequences of partial aggregation: One measure indicates the degree of divergence or non-conformity between estimates produced by unaggregated data and partially aggregated data; and the other measure pertains to efficiency loss and expresses the fraction of the useful information in the unaggregated data which remains after the data have been grouped or partially aggregated. These measures are then applied in an experimental test using data from the Call Reports and the Income and Dividend Statements of nearly 5400 member banks of the Federal Reserve System. This experiment consists of evaluating the effect on twenty different regression models of three different levels of aggregation and seven different rules for arraying the data prior to aggregation.