3.2. Research Method
The research methodology employed is explanatory, using the technique of multiple regression analysis based on the study of the relationship between variables measured on a quantitative scale.
On 18 April 2020, this research was initiated by conducting a search of the main collection of the Web of Science, limiting SSCI between the years 1900 and 2020 to “all types of documents” in the theme “LinkedIn,” resulting in 616 documents (of which 132 are Open Access). Refining the search by document type to “items only” yielded 411 items, and then refining it again by business* manage* company* gave a result of 211 items. A total of 588 authors or coauthors were counted. It should be noted that the maximum number of authors per article found was 23. Of the 588 authors, we finally focused on the search for data from 521, as some of them participated in more than one article [
80].
Another search was also performed by adding the term LinkedIn to Twitter, which extended the results to 1111 articles. We decided to focus on LinkedIn only due to the professional nature of the network, although in the empirical part of our work, we have included the presence of researchers on Twitter as an additional tool that provides us with more data on the professional profile of the academic. LinkedIn, as they define themselves, have as their vision “to create economic opportunities for every member of the global labor market through the continuous development of the world’s first economic chart” and have as their mission “to connect professionals from all over the world to help them be more productive and achieve all their career goals” [
81], while Twitter is a generalist communication tool, defining themselves as “advocates for free expression and protecting the health of the public conversation around the world” [
82].
The basic research objective, as stated above, is to see the degree of connection between academia and professional visibility, understood today as the appearance and level of activity in the main professional networks on the Internet: LinkedIn and Twitter. For academic activity, we used two productivity measurement indices: WOS h-index and Google Scholar h-index. For the activity in OSNs, we focused on LinkedIn and Twitter; in addition, we collected the activity in ASN ResearchGate and presence in Publons. As intermediate instruments to find academics, when it is difficult to locate them through the previously discussed networks, we will use Facebook, YouTube or Instagram.
The 211 articles resulting from our selection written by 521 research authors on the subject of “business” have given rise to two types of analysis items, the article item and the author item. From the item article, we extracted the following data: Keywords, Theories used in the study, Type of analysis (Quantitative/qualitative, Theoretical/empirical), Conclusions, Indexation in Journal Citation Reports (Quartile), Category, and Observations.
From the item author, we extracted the following data: Name, Gender, ORCID, Contact Academic, University, Country, h-index WOS, h-index Google Scholar, Profile in Publons, ResearchGate presence/number of items, ResearchGate Score, RG Photo, RG Number of Followers, RG Number of Followed People, LinkedIn Presence, LinkedIn Number of Contacts, LinkedIn Photo, Twitter professional profile, Tw Number of Tweets, Tw Number of followers, Tw Number of followed people, and Observations.
In our work, we used two data sets. First was data obtained from 211 articles filtered from the Web of Science (SSCI) with the criteria that have been detailed in the methodology, analysing the information from 7 records extracted from each of these articles. This resulted in a total of 1477 data points in this first set.
Second, each of the authors who signed the articles was analysed, with a total of 521 authors, of which a total of 22 records were collected from each of them, ranging from affiliation to different indicators of academic productivity, as well as other indicators of each of the social networks analysed in the articles, such as ResearchGate, LinkedIn, and Twitter. This results in a total of 11,462 data points in this second set. All of this information has brought the size of the total data set of our research to 12,939 data points. Therefore, the sample size in our study, based on previous research, is considered appropriate [
83].
In summary, the methodology is compiled in
Table 1:
3.3. Model Variables
The variables used in the proposed models (
Table 2), both dependent (h-index of Web of Science and h-index of Google Scholar) and independent (the rest of them), are presented and defined below:
H-index of Web of Science (hindexWOS): the index that the Web of Science gives for each author.
H-index of Google Scholar (hindexGoogleScholar): the index that Google Scholar offers for each author.
JCR: indexing of the Journal in the Journal Citation Reports list on the Web of Science and, if applicable, the quartile in which it is located. It takes the values (0, 1, 2, 3, 4), 0 in the case of non-indexation or 1 to 4 depending on the quartile in which the journal is positioned at the time of the search [
84].
ORCID: Open Research and Contributor ID is a unique identifier whose main purpose is to provide researchers with a persistent and unambiguous author code that clearly distinguishes their scientific production and avoids confusion linked to scientific authorship and the existence of similar names [
85]. It takes the values (0, 1), 0 if no ORCID and 1 if yes.
Academic contact (AC): e-mail of the university or affiliation centre that the author puts either in the article or in social or professional networks to be contacted. It takes the values (0, 1) in the case of inexistence or existence.
Publons (PUB): presence on your website [
86]. On this platform, part of Web of Science, the author follows the impact of his publications, as well as his reviews of other works. It takes the values (0, 1) in the case of inexistence and existence of a profile.
ResearchGate (RG): purely academic social network. They define themselves as “the professional network for scientists and researchers. Over 17 million members from all over the world use it to share, discover, and discuss research. We’re guided by our mission to connect the world of science and make research open to all.”
In our study, RG will be represented by a composite index. To calculate it, a factorial analysis has been done as a method of validation of the same [
87] with the idea of looking for the minimum number of dimensions capable of explaining the maximum of information contained in the data. The variables taken into account to carry out the factorial analysis were:
RGPresence: takes the value 0 when there is no presence in this network. It takes a numerical value equal to or greater than 1 when the author has put contributions in this network depending on the number of items he has updated.
RGScore (RGS): takes the value 0 when it has no mark in this network, and a positive value otherwise.
RGPhoto (RGPh): takes the values (0, 1) depending on the existence or absence of a personal photo.
RGFollowers (RGFWers): quantitative value in the interval of 0 to the maximum number of followers in the network.
RGFollowing (RGFWing): quantitative value in the interval of 0 to the maximum number of followed academics.
To determine the degree of presence of the authors in social networks, we used various indicators derived from their appearance on LinkedIn and Twitter. A social network presence metric that jointly integrates both networks has not been found in the academic literature. All of the authors use as metrics those thrown by each social network [
88]. In the case of LinkedIn, presence is measured through the existence of an account, the number of contacts, and the existence of a photo. Regarding Twitter, the data we extracted as indicators of presence in this network are the existence of a professional Twitter profile and the numbers of Tweets, Followers, and People Followed. Karampela et al. [
89] use as a scale of measurement of social presence the variables of Ou et al. [
90] adapted to their study (There is a sense of human contact on supplier X’s social media. There is a sense of personalness on supplier X’s social media. There is human warmth on supplier X’s social media).
In our case, we created a composite indicator taking into account the five variables referred to ResearchGate, which is adjusted, although the variables explain only 59.24% of the index created. The most irregular items are photo included, with a coefficient extracted from the factor analysis of 0.418 and number of people followed, with a coefficient extracted from the factor analysis of 0.370, so these variables are not advisable to measure the constructed index. The remaining variables (RGPres, RGS, and RGFWers) have an acceptable coefficient because they are closer to one. We perform a second factorial analysis for this index by eliminating these two items. The result is a much tighter model with a Bartlett’s significance of 0.000 and a total variance explained of 80%, where the variables that compose its RGPres, RGS, and RGFWers have a coefficient extracted from the second factorial analysis between 0.760 and 0.817.
LinkedIn (LK): They define themselves as the world’s largest professional network with over 645+ million users in over 200 countries and territories whose vision it is to create economic opportunities for every member of the global labour market through the continued development of the world’s first economic chart and whose mission it is to connect professionals around the world to help them be more productive and achieve all their career goals [
81].
LinkedIn is a composite indicator obtained through factorial analysis. The variables that have been taken into account in its creation are:
LKPresence (LKPres): takes the values (0, 1) based on if the profile exists in this network.
LKNumberofContacts (LKNc): if yes, number of contacts in the network.
LKPhoto: LKPh: takes the values (0, 1) for the existence or absence of a photo in this network.
The variable LKNc has a very low value, very close to 0 (coefficient extracted from the factorial analysis of 0.014), so it must be extracted from the composition of the index. We carry out a second factorial analysis for this index, eliminating this item. The result is a very tight model with a Bartlett’s significance of 0.000 and a total explained variance of 85%, where both variables that compose it, LKPres and LKPh, have a coefficient extracted from the second factorial analysis of 0.850.
Twitter (TW): professional network specialized in communication. On their website, they state as their “philanthropic mission to spread and expand the power of Twitter and the talent of our employees through direct civic interaction, staff volunteerism, charitable contributions, in-kind donations, and using the Twitter service in a positive way” [
82].
Twitter is a composite indicator also obtained through factorial analysis. The variables that have been taken into account for its formation are:
TWPresence (TWPres): takes the values (0, 1) depending on the presence of the researcher in this network.
TWFWers: quantitative value indicating the number of followers in the network.
TWFWing: quantitative value indicating the number of people followed by the researcher in the network.
The result is a poorly adjusted model with a very low value of the item TWPres (coefficient extracted from the factor analysis of 0.072), so it must be removed from the composition of the indicator. When the factorial analysis is performed again, the result is that the TWFWers and TWFWing variables that now make up the indicator explain 70% of it, being well adjusted with a Bartlett significance of 0.000 and a coefficient extracted from the second factorial analysis of 0.997.
To test our hypothesis, we proposed two models, one with each of the dependent variables, the h-index of Google Scholar (h_indexGS) and the h-index of Web of Science, as follows:
It must be taken into account when evaluating the results of both models that Google Scholar has more flexibility and is more generic in nature because it is freely accessible and each author designs his or her own academic profile, while Web of Science has restricted access and follows strict control of references to indexed documents and the author does not have access to his or her profile (it is the system itself that includes them), resulting in a certain index [
91].
It should be noted that the variables RG, LK, and TW have been included in the model as composite indices resulting from the grouping of each of the individual reference variables.