Refrain from adopting the combination of citation and journal metrics to grade publications, as used in the Italian national research assessment exercise (VQR 2011-2014)

TL;DR

This paper critiques the Italian VQR 2011-2014 research assessment method, showing that combining citation counts with journal metrics does not improve prediction of long-term impact compared to simple citations.

Contribution

The study provides evidence that the combined citation and journal metric approach is less effective than using simple citation counts for research evaluation.

Findings

Combined metrics have lower predictive power than simple citations.

Effectiveness of metrics varies across disciplines.

Longer citation windows improve prediction accuracy.

Abstract

The prediction of the long-term impact of a scientific article is challenging task, addressed by the bibliometrician through resorting to a proxy whose reliability increases with the breadth of the citation window. In the national research assessment exercises using metrics the citation window is necessarily short, but in some cases is sufficient to advise the use of simple citations. For the Italian VQR 2011-2014, the choice was instead made to adopt a linear weighted combination of citations and journal metric percentiles, with weights differentiated by discipline and year. Given the strategic importance of the exercise, whose results inform the allocation of a significant share of resources for the national academic system, we examined whether the predictive power of the proposed indicator is stronger than the simple citation count. The results show the opposite, for all discipline in the sciences and a citation window above two years.