Exploring paths for the normalization of altmetrics : applying the Characteristic Scores and Scales

The probability of a paper being mentioned on Twitter, Facebook, in a blog post or the news or being saved to Mendeley depends on its discipline. For example, 31.7% of recent biomedical and health sciences journal articles get tweeted, while the Twitter coverage in mathematics and computer science is as low as 7.5% (Haustein, Costas, & Larivière, 2015). The coverage, density and distribution of various altmetrics differ substantially across disciplines (Costas, Zahedi, & Wouters, 2015). The need to take these field differences into account in the calculation of altmetric indicators represents a critical item in the agenda of altmetric research (Wouters & Costas, 2012). Recently there have been some suggestions of normalization approaches for Mendeley similar to the normalization of citations (Haunschild & Bornmann, 2016) as well as for Twitter (Bornmann & Haunschild, 2016). However, a broader discussion about the potential problems that the distribution of metrics across fields can pose for the extensive normalization of altmetric indicators is still missing. This paper addresses these issues and specifically answers the following research questions: