Scientific collaboration networks have been studied more often within social network analysis, as they are an example of a constantly changing network (Barabási et al. 2002). These studies have shown that scientific collaboration networks can often be described as highly clustered (Newman 2001), with scientists collaborating more often when they have a third common scientist as compared to two random scientists collaborating. Barabasi et al. (2002)also conclude that these scientific collaboration networks are often marked by preferential attachment, meaning in this case that scientists with many collaboration ties get even more ties. Furthermore, homophily is also present within networks of scientific collaboration, showing that scientists have a preference of working together with scientists of the same institution or with similar research interests (Wang and Zhu 2014).
These network structures that are typical for scientific collaboration networks are thus important to consider, and provide the opportunity to test the effects of several characteristics of scientists while also taking into account the consequences these network effects bring about. The effect of high clustering that Newman (2001) described, is also called “transitivity”. This describes the tendency among individuals to form a tie with another individual who already has a tie with someone known (e.g. becoming friends with friends of friends) (Block 2015). Scientists thus prefer to collaborate with scientists that have collaborated with one of my collaborators.
While the literature on (network effects within) scientific collaboration networks may be well-established (Barabási et al. 2002; Block 2015; Newman 2001; Wang and Zhu 2014 ), less is known about other factors that could have an effect of selection and influence in networks. While transitivity, homophily and preferential attachment provide clear expectations as to which scientist collaborates with whom and why, other elements could also determine collaborations between scientists. In this study, the influence of Twitter activity and its relation to popularity of scientific publications is investigated for co-publication networks of scientists.
Hall (2014) developed the Kardashian Index to quantify whether social media use of scientists is becoming more important than scientific output. This k-index is a ratio between the number of Twitter followers of a scientist and the number of citations for scientific articles. According to Hall (2014), scientists with a k-index higher than 5, are strongly overvalued and have built their public profile based on social media while not having highly cited articles.
The k-index is more often used within the literature. For instance, Kolahi and colleagues (2019) found that scientific dental articles that covered important topics, were often not among the highly tweeted articles. On the other hand, they also found that the highly tweeted articles had 12.96 cites on average, possibly demonstrating a correlation between number of tweets and citations. While Kohali and colleagues (2019) focused mostly on articles as subject of their research, Khan and colleagues (2020) investigated the k-index of cardiologists among a random sample drawn from cardiologists of the top 100 hospitals and found that the majority of the cardiologists had a rather low k-index, and that only some ranked above the number of 5.
Most of the literature using the k-index is descriptive in nature and does not look at any influences on and consequences of the k-index. Eacott (2020) did investigate the k-index and the Twitter tagging relations of scientists and concluded that scientists who get mentioned a lot are not necessarily of scientific value. What remains unclear, is whether this k-index also has an influence on social networks of scientists. As scientists that are popular on social media now get invited to speak at events because of their popularity (Hall 2014) and not because of highly cited publications, this may signal that scientists attach value to other scientists that have a strong social media outreach. In this context, the citations of a scientists may be of less importance, making scientists with a high k-index attractive for other scientists to collaborate. Therefore, I derive the first hypothesis: Scientists prefer to co-publish with other scientists that have a high k-index.
On the other hand, scientists are in general critical individuals. This may result in unpopularity of scientists with a high k-index, as they could be known for investing more time in social media than in contributing to science (Califf 2020). This leads to the second hypothesis: Scientists prefer to co-publish with other scientists that have a low k-index.
Furthermore, as explained above, the homophily effect is important within social networks (Wang and Zhu 2014). Possibly, not only the research topics of other scientists are taken into consideration when selecting others to co-publish with, but also a similar stance regarding the ratio between social media and scientific contributions. Thus, the scientist compares their own k-index with that of another scientist. In this regard, it can be expected that scientists select other scientists with a similar k-index for co-publication. Therefore, the last hypothesis can be derived as follows: Scientists prefer to co-publish with other scientists that have a similar k-index.