Use of Previously Published Data in Statistical Estimation

Traditionally, researchers collect and analyze their own data, or use published results to perform meta-analysis. However, they rarely combine the experimental data with already published findings, which is a more efficient and cost effective approach for experimental planning and data analysis. In this work, we present two methods on the use of previously published data. One method targets variance minimization and another minimizes mean squared error (MSE). Bayesian approaches to prior information are not considered in this work. Variance minimization is designed to work in a class of unbiased estimators, where both (1) the estimators based on experimental data and (2) the estimators available as additional information (previously published results) are unbiased. MSE minimization relaxes the unbiasedness assumption on additional information and assumes that bias may be present. The use of these methods is illustrated for the analysis of association between gestational age at birth and third grade academic performance.