Statistical analysis
Due to the appearance of novel social platforms where TE research is shared, we have also performed a statistical analysis including three different tests: Spearman correlation test (Asuero, Sayago, & Gonzalez, 2006), factorial analysis (Taherdoost, Sahibuddin, & Jalaliyoon, 2014) and linear regression model (M. Thelwall & Wilson, 2014). In this sense, the collection of cites using traditional metrics requires several years and the availability of data provided by Almetrics.com before 2015 is not extensive due to the platform was founded in 2012. Thus, the correlation and factorial studies were performed for 2015 metadata. This strategy has been employed in other scientific disciplines (Mike Thelwall & Nevill, 2018). Furthermore, all citation and mention counts were transformed with the formula Ln(1+x) before processing to reduce skewing (Mike Thelwall & Nevill, 2018).
First, the Kolmogorov-Smirnov test was calculated to verify the dataset do not follow a normal distribution for the next 16 variables that overall characterize TE online dimension: 1) WoS citations 2) News, 3) Blogs, 4) Policy, 5) Twitter, 6) Patents, 7) Peer review, 8) Weibo, 9) Facebook, 10) Wikipedia, 11) Google, 12) Reddit, 13) F1000, 14) Q&A, 15) Video and 16) Mendeley readers. Furthermore, the Spearman correlation was obtained for the variables previously described, and the statistical significance was defined as a probability of less than 0.01.
Second, once the correlation data were obtained, we performed a factorial analysis. In this sense, factorial analysis is a statistical model that allows us to identify the common variables or factorsthat could explain the previously observed correlation. Bartlett Sphericity test (BS) and Kaiser-Meyer-Olkin (KMO) were previously performed to assess the suitability of factorial analysis (Chan & Idris, 2017).
Finally, the linear regression model was employed to achieve a more accurate expression of the relation between TE academic model and the new social paradigm of it. In this sense, we have developed a mathematical model exposing the influence of new altmetric scores in TE classical measure of impact, that is, the number of cites received by TE original articles. Thus, the linear regression equation constructed for TE behaviour contents a standard measure of impact as WoS citations and a group of alternative metrics previously identified in the factorial analysis. The free software JASP (University of Amsterdam, Amsterdam, The Netherlands) was employed to perform all the statistical analyses (Love et al., 2019).