Predicting downloads of acadamic articles to inform online content management

We examine 1,510 journals from a major research university library, representing more than 40% of the university's annual financial cost for electronic resources at the time of the study. In this research, we utilize a web analytics approach for the creation of a linear regression model to predict usage among these journals. We categorize metrics into global (e.g., journal impact factor, Eigenfactor, etc.) that are journal focused and local (e.g., local downloads, local citation rate, etc.) classes that are institution focused. By means of 275 journals for a training set, our analysis shows that a combination of both global and local metrics creates the strongest model for predicting full-text downloads. These research results establish the value in better informed purchasing decisions by creating local metrics versus relying solely on global metrics for the evaluation of library content collections. The linear regression model has an accuracy of more than 80% in predicting downloads for greater than 80% of the 1,235 journals in our test set.