Predicted correlation. Issue 1 (2nd January 2022)

• Record Type: Journal Article
• Authors: Forthmann, Boris; Szardenings, Carsten
• Is Part Of: Collnet journal of scientometrics and information management. Volume 16:Issue 1(2022)
• Page Start: 7
• Page End: 18
• Publication Date: 2022-01-02
• Subjects: Pearson correlation -- Predicted correlation -- Simulation -- Reanalysis -- Independence; Science -- Periodicals; Science indicators -- Periodicals; Information science -- Periodicals; Social sciences -- Periodicals; 505
• DOI: 10.1080/09737766.2021.1989988
• Languages: English
• ISSNs: 0973-7766

Abstract

Abstract : Correlations are ubiquitous in scientometric research. The present work illustrates a formula to quantify the predicted correlation between a composite indicator and a primary indicator (i.e., the composite indicator can be expressed as a weighted sum of the primary indicator), for example. Total citations received and number of self-citations or total publications and number of first-authorship publications, for example, represent such variable pairs. However, predicted correlation has a far wider range of potential applications in scientometrics. It is demonstrated that the predicted correlation provides a useful reference that allows a more conclusive interpretation of the data. Ignoring the outlined approach can result in overlooking of robust correlational patterns in the data. This is illustrated by a small simulation and two illustrations based on re-analyses of previous work. The approach can be used in new studies to understand the complete correlational pattern. In addition, the outlined approach can be used to revisit past findings reported in journal articles.