MULTIVARIATE REGRESSION MODELS FOR ESTIMATING JOURNAL USEFULNESS IN PHYSICS

Abstract

Multiple regression models can be used to rank physics journals in approximately the same order as the journals are perceived useful by actual users. Four such regression models are reported here, each having a multiple R value of ·74 or greater. Perceived usefulness, the dependent variable used in constructing the models, was obtained from a survey of 167 physicists in the US and Canada. The independent, or predictor variables include easily obtainable bibliometric statistics such as number of source items published, immediacy index, ratio of citations received to citations made, total citations received, impact factor and others. Regression models that combine certain of these statistics can predict user valuation of the journals better than any single bibliometric predictor alone can do. Their advantage for serials management is in ease of estimating usefulness as judged by users, a much more difficult statistic to obtain. Where these models may not apply, it is relatively simple to construct similar models based upon surveys of other user groups. It appears likely that good models of this type can also be developed for many other disciplines.