A Multiple Linear Regression Analysis to Measure the Journal Contribution to the Social Attention of Research

Abstract

This paper proposes a three-year average of social attention as a more reliable measure of the social impact of journals since the social attention of research can vary widely among scientific articles, even within the same journal. The proposed measure is used to evaluate a journal’s contribution to social attention in comparison to other bibliometric indicators. This study uses Dimensions as a data source and examines research articles from 76 disciplinary libraries and information science journals through multiple linear regression analysis. This study identifies socially influential journals whose contribution to social attention is twice that of scholarly impact, as measured by citations. In addition, this study finds that the number of authors and open access have a moderate effect on social attention, while the journal impact factor has a negative effect and funding has a small effect.

1. Introduction

Several factors influence the relationship between the public attention a paper receives and its scholarly impact. The academic impact is primarily influenced by the perceived quality of the research, the reputation of the authors, their institutions, and the journals in which they publish. Social attention, on the other hand, is influenced by a broader range of factors, including the topic of the publication, the demographics of the authors who are active on social media, and the current trends and interests of the general public. For example, topics that are controversial or trending tend to generate a lot of social attention, regardless of the scholarly impact of the publication.

Since the term “altmetrics” was first coined in 2010 [1], both theoretical and practical research has been conducted in this area [2]. In addition, governments are now encouraging researchers to engage in activities that have a social impact, such as those that bring economic, cultural, and health benefits [3].

While altmetric data can increase citation rates by accelerating the accumulation of citations after publication [4], they show only a moderate correlation with Mendeley readership [5] and a weak or negligible correlation with other altmetric indicators and citations [6,7]. Consequently, altmetrics may capture different types of impact beyond citation impact [8].

There are many and varied factors that influence the public exposure of research [2]. These include collaboration, research funding, mode of access to the publication, citation, and journal impact factor, which will be discussed below.

Collaboration is becoming increasingly important in scientific research as it allows the combination of knowledge and skills to generate new ideas and research avenues [9]. Co-authorship analysis of research papers is a valid method for studying collaboration [10]. While researchers have increased their production of research articles in recent decades, the number of co-authors has also increased, resulting in a steady publication rate [11,12]. Collaborative research has been associated with higher citation rates and more impactful science [13]. Scientists who collaborate more often tend to have higher h-indices [14]. In addition, both citation and social attention increase with co-authorship, although the influence becomes less significant as the number of collaborators increases [15].

Funding is another important input into the research process. The authors found that 43% of the publications acknowledge funding, with considerable variation between countries [16]. They also found that publications that acknowledge funding are more highly cited. However, citations are only one side of a multidimensional concept, such as research impact, and alternatives have been considered to complement the impact of research. Other authors conclude that there is a positive correlation between funding and usage metrics, but with differences between disciplines [17].

Another factor to consider when analyzing research performance is the type of publication access. The impact advantage of open access is probably because greater access allows more people to read articles that they would not otherwise read. However, the true causality is difficult to establish because there are many possible confounding factors [18,19].

In this context, in this present paper, I quantify the contribution of the journal to societal attention to research. I compare this contribution with other bibliometric and impact indicators discussed earlier, such as collaboration, research funding, type of access, citation, and journal impact factor. To this end, I propose a measure of social influence for journals. This indicator is a three-year average of the social attention given to articles published in the journal. The data source is Dimensions, and the units of study are research articles in library and information science. The methodology used is multiple linear regression analysis.

2. Social Attention to Research and Traditional Metrics

In the field of research evaluation, scholars have studied the impact of research papers. However, traditional metrics such as citation analysis, impact factors, and h-index tend to focus only on the academic use of research papers and ignore their social impact on the Internet [20]. Web 2.0 has transformed social interaction into a web-based platform that allows two-way communication and real-time interaction, creating an environment in which altmetrics has emerged as a new metric to measure the impact of a research paper. The term “altmetrics” was first introduced in 2010 by Priem, who also published a manifesto on the subject [1]. However, the correlation between citation counts and alternative metrics is complicated because neither are direct indicators of research quality, making it nearly impossible to achieve a perfect correlation unless they are unbiased [21].

Some research studies have focused on exploring the correlation between traditional citation metrics and alternative metrics [21,22]. Such studies are important for understanding how research performance is evaluated, particularly with respect to measuring impact through citation counts and altmetric attention scores (AAS). While citation counts have been the primary means of assessing research performance, the importance of AAS is increasing in today’s social media-driven world. This is because citation counts have limitations, such as delays in adding a publication to citation databases and potential biases due to self-citation.

A study compared citation data from 3 databases (WoS, Scopus, and Google Scholar) for 85 LIS journals and found that Google Scholar citation data had a strong correlation with altmetric attention, while the other 2 databases showed only a moderate correlation [23]. However, for the nine journals that were consistently present in all three databases, there was a positive but not significant correlation between citation score and altmetric attention. Although there was no correlation between citation count and altmetric score, a study found a moderate correlation between journal impact factor and citation count, a weak correlation between journal tweets and impact factor, and a strong correlation between journal tweets and altmetric score [24].

3. Materials and Methods

This study employed a rigorous and systematic process for collecting and analyzing bibliometric data using the Dimensions database to ensure that the results were reliable, valid, and informative for the field of library and information science.

• Unit of Analysis: The unit of analysis for this study is the “research article” in the field of library and information science.
• Data Source: The data source for this study is the Dimensions database, which provides social attention data at the article level. This database was selected because of its comprehensive coverage of scholarly publications in a variety of disciplines.
• Journal Selection: The JCR Journal Impact Factor in the Web of Science database was used to select journals in the library and information science category. Of the 86 journals identified in the 2020 edition, 10 journals were excluded because they were not indexed in the Dimensions database. This step ensured that only high-quality, peer-reviewed journals were included in the analysis. The final dataset included 76 library and information science journals. These journals were selected based on their relevance and impact on the field.
• Timeframe: Research articles indexed in the Dimensions database between 2012 and 2021 were included in the analysis. This timeframe ensured that the analysis covered recent publications, while also allowing for the collection of sufficient data.
• Search Criteria: In the Dimensions database, the following search criteria were used for each journal X. Source Title was set to X, Publication Type was set to Article, and Publication Year was set to the range 2012 to 2021.
• Final Dataset: All retrieved records were then exported to a file in CSV (comma-separated values) format. The export file contained data on each research article’s bibliographic information, social attention metrics, and other relevant variables.
• Data Collection: Data was collected on 6 June 2022. A total of 49,202 research articles were analyzed in this study.

Note that Altmetric is the source of altmetric data in Dimensions, and it is one of the earliest and most popular altmetric aggregator platforms. Digital Science launched this platform in 2011, and it tracks and aggregates mention and views of scholarly articles from various social media channels, news outlets, blogs, and other platforms. It also calculates a weighted score, called the ‘altmetric attention score’, in which each mentioned category contributes differently to the final score [25].

The altmetric attention score measures the amount of social attention an article receives from sources such as mainstream and social media, public policy documents, and Wikipedia. It assesses the online presence of the article and evaluates the discussions surrounding the research. To avoid confusion, this paper uses the term “social attention score” or simply “social attention” is used to refer to this metric.

This paper proposes a journal-level measure of social attention to research. This measure is defined as the average social attention of articles over a three-year window. Note that the Dimensions database does not provide journal-level impact indicators. Therefore, I included another measure of journal impact in the dataset. I used the Journal Impact Factor provided by the JCR Web of Science database for 2020, the year available at the time of data collection.

The methodology consists of a multiple linear regression analysis. Thus, the dependent variable is the social attention of the article, and the independent variables are the proposed measure of the social attention of the journal, the number of authors, the type of access to the publication {open access = 1, closed = 0}, the funding of the research {funded = 1, unfunded = 0}, the citations of the article, and the impact factor of the journal.

4. Results

The article-level dataset is described in

Table 1. Descriptive statistics of the dataset at the article level. Category: Library and Information Science. Data source: Dimensions.

Years Since Pub.*	Year of Pub.	Num. Art.	Num. Authors (Mean)	OA Art. (%)	Funded Art. (%)	Citations (Mean)	Art. Social Attention
							Mean Score	Marg. Var.
1	2021	6156	3.15	39.57%	23.93%	2.79	4.41
2	2020	5687	3.10	44.43%	26.01%	9.14	5.39	0.97
3	2019	4880	2.88	43.03%	23.55%	11.87	5.31	−0.08
4	2018	4811	2.73	43.17%	22.49%	14.94	5.57	0.26
5	2017	4897	2.63	45.01%	21.14%	16.91	5.17	−0.40
6	2016	4592	2.72	42.29%	18.47%	20.61	5.51	0.34
7	2015	4521	2.64	39.02%	20.97%	23.14	4.92	−0.58
8	2014	4551	2.67	37.84%	19.12%	22.73	3.89	−1.03
9	2013	4613	2.45	34.92%	17.86%	24.36	3.33	−0.57
10	2012	4494	2.37	34.40%	15.20%	25.77	2.57	−0.75
All		49,202	2.76	40.50%	21.12%	16.52	4.63

* Average years from publication to the time of data collection in the first half of 2022.

. The information in this table is presented according to the time elapsed since publication (in average years from publication to the time of data collection in the first half of 2022). It can be observed that the maximum social attention of scientific research in library and information science is reached on average 4 years after its publication, with an average score of 5.57. However, there are no significant differences after the second year. The highest values, more than five points, are observed between the second and sixth year after publication. Nevertheless, the marginal variation between years is only relevant in the second year, with an increase of 0.97 points compared to the first year.

Table 1 also shows how the average number of authors per article in library and information science has gradually increased over the past decade, from an average of 2.37 authors per article in 2012 to an average of 3.15 authors per article in 2021. This 33% increase in co-authorship in a decade is remarkable. The increase in co-authorship may partially explain the 37% increase in research article production over the decade in the Library and Information Science category, from just under 4500 articles in 2012 to more than 6100 articles in 2021.

In the dataset, 40% of the articles are open access, and 21% of publications indicate in the acknowledgments section that the authors have received some form of funding, with a sustained increase in most of the years analyzed. In terms of citations, the increase observed in Table 1 was to be expected, from 2.8 cites per article in the first year after the publication to an average of 25.8 cites at the end of the decade. Significant marginal increases are observed up to the seventh year after publication, highlighting the increase of 6.35 citations that occurs in the second year.

4.1. Journal Social Attention: Definition and Consistency of the Indicator

When aggregating the data at the journal level, I observed a large interannual variability in the average social attention per article when the time window was reduced to a single year. That is, for each journal, the average social attention of the articles in a single year differs significantly from that of the articles in the previous and subsequent years of the series. This large variability in the average social attention of each journal over time means that the one-year average is not a consistent measure of social attention for journals. This weakness observed for social media mentions also occurs for other citation-based indicators, such as the impact factor, with short time windows.

One reason for this high interannual variability is the low correlation between the individual scores of articles and the average scores of journals when the time window in which citations or mentions are collected is short. Thus, a small proportion of articles from each journal receive a large proportion of scientific citations and social mentions. In order to increase the consistency of a measure by partially reducing the interannual variability, it is often chosen in bibliometrics to increase the size of the window for counting observations (citations or mentions). In the case of the impact factor, the various databases thus provide indicators for two, three, four, and even five years.

In this study, I chose a three-year window as a compromise between the advantages and disadvantages of considering large time windows. That is, four-year and five-year windows require a long waiting period before social attention can be measured for a journal, while a two-year window still produces a high interannual variability in the dataset.

Therefore, I propose the following definition for the social attention measure at the journal level. The journal social attention in year y counts the social attention received in years y–2, y–1, and y for research articles published in those years (y–2, y–1, and y) and divides it by the number of research articles published in those years (y–2, y–1, and y). For example, the journal social attention in 2021 counts the social attention received in 2019–2021 for research articles published in 2019–2021 and divides it by the number of research articles published in 2019–2021.

The journal-level dataset is described in

Table 2. Description of the dataset at the journal level in the subject category Library and Information Science (JCR). Data sources: Web of Science and Dimensions.

	Journal	Num. Art. 2020	JIF 2020	JIF Percentile 2020	JIF Quartile 2020	5-Year JIF 2020	Journal Social Attention 2021
1	ASLIB J INFORM MANAG	54	1.903	44.12	Q3	2.343	1.83
2	CAN J INFORM LIB SCI	5	0.000	0.59	Q4	0.420	0.78
3	COLL RES LIBR	52	2.381	52.35	Q2	2.204	5.55
4	DATA TECHNOL APPL	51	1.667	39.41	Q3	1.667	0.37
5	ELECTRON LIBR	56	1.453	34.71	Q3	1.540	0.36
6	ETHICS INF TECHNOL	64	4.449	74.71	Q2	3.925	10.70
7	EUR J INFORM SYST	64	4.344	71.18	Q2	7.130	6.43
8	GOV INFORM Q	71	6.695	91.18	Q1	8.293	7.66
9	HEALTH INFO LIBR J	45	2.154	47.65	Q3	2.187	6.25
10	INFORM SOC-ESTUD	32	0.311	8.82	Q4	0.313	0.39
11	INFORM MANAGE-AMSTER	91	7.555	94.71	Q1	9.183	2.48
12	INFORM ORGAN-UK	15	6.300	90	Q1	5.866	6.24
13	INFORM DEV	74	2.049	46.47	Q3	2.205	0.99
14	INFORM PROCESS MANAG	237	6.222	88.82	Q1	5.789	2.06
15	INFORM RES	75	0.780	20.59	Q4	1.197	1.42
16	INFORM SOC	25	4.571	77.06	Q1	3.936	6.94
17	INFORM SYST J	49	7.453	93.53	Q1	8.814	7.05
18	INFORM SYST RES	69	5.207	82.94	Q1	6.888	7.54
19	INFORM TECHNOL MANAG	15	1.533	38.24	Q3	2.627	0.19
20	INFORM TECHNOL PEOPL	121	3.879	67.65	Q2	4.238	1.65
21	INFORM TECHNOL LIBR	27	1.160	27.65	Q3	1.351	8.90
22	INFORM TECHNOL DEV	66	4.250	70	Q2	4.221	6.06
23	INT J COMP-SUPP COLL	18	5.108	80.59	Q1	4.966	4.09
24	INT J GEOGR INF SCI	162	4.186	68.82	Q2	4.645	3.66
25	INT J INFORM MANAGE	203	14.098	99.41	Q1	13.074	4.99
26	INVESTIG BIBLIOTECOL	39	0.475	13.53	Q4	0.535	0.27
27	J ACAD LIBR	104	1.533	38.24	Q3	2.023	5.03
28	J COMPUT-MEDIAT COMM	25	5.410	85.29	Q1	9.953	25.01
29	J DOC	98	1.819	40.59	Q3	1.988	3.43
30	J ENTERP INF MANAG	128	5.396	84.12	Q1	5.839	0.26
31	J GLOB INF MANAG	38	1.373	33.53	Q3	1.550	0.75
32	J GLOB INF TECH MAN	14	3.519	66.47	Q2	2.631	0.94
33	J HEALTH COMMUN	94	2.781	59.41	Q2	3.468	8.44
34	J INF SCI	120	3.282	65.29	Q2	2.904	2.38
35	J INF TECHNOL-UK	21	5.824	86.47	Q1	9.439	2.28
36	J INFORMETR	77	5.107	79.41	Q1	5.421	7.12
37	J KNOWL MANAG	162	8.182	97.06	Q1	8.720	0.60
38	J LIBR INF SCI	123	1.992	45.29	Q3	2.009	2.47
39	J MANAGE INFORM SYST	40	7.838	95.88	Q1	8.335	3.53
40	J ORGAN END USER COM	23	4.349	72.35	Q2	2.808	0.08
41	J SCHOLARLY PUBL	19	1.512	35.88	Q3	1.245	3.77
42	J STRATEGIC INF SYST	16	11.022	98.24	Q1	11.832	1.37
43	J AM MED INFORM ASSN	209	4.497	75.88	Q1	5.178	14.38
44	J ASSOC INF SCI TECH	150	2.687	54.71	Q2	3.854	8.09
45	J ASSOC INF SYST	47	5.149	81.76	Q1	6.780	2.06
46	J AUST LIB INF ASSOC	28	0.725	19.41	Q4	0.851	1.64
47	J MED LIBR ASSOC	60	3.180	61.76	Q2	3.874	4.57
48	KNOWL MAN RES PRACT	106	2.744	58.24	Q2	3.027	1.24
49	KNOWL ORGAN	34	1.000	25.29	Q4	0.979	0.18
50	LEARN PUBL	53	2.506	53.53	Q2	2.659	20.19
51	LIBR INFORM SCI RES	32	2.730	57.06	Q2	2.778	5.15
52	LIBR HI TECH	83	2.357	51.18	Q2	2.065	1.32
53	LIBR QUART	26	1.895	42.94	Q3	2.277	1.27
54	LIBR RESOUR TECH SER	3	0.424	12.35	Q4	0.541	0.45
55	LIBR TRENDS	38	1.311	31.18	Q3	1.354	5.70
56	LIBRI	25	0.521	14.71	Q4	0.706	0.75
57	MALAYS J LIBR INF SC	20	1.250	28.82	Q3	1.320	0.00
58	MIS QUART	58	7.198	92.35	Q1	12.803	0.84
59	MIS Q EXEC	17	4.371	73.53	Q2	7.563	6.12
60	ONLINE INFORM REV	82	2.325	50	Q3	2.883	3.92
61	PORTAL-LIBR ACAD	36	1.067	26.47	Q3	1.285	1.93
62	PROF INFORM	169	2.253	48.82	Q3	2.285	6.99
63	QUAL HEALTH RES	192	3.277	64.12	Q2	5.038	7.03
64	REF USER SERV Q	27	0.650	17.06	Q4	0.581	0.22
65	REF SERV REV	38	0.831	22.94	Q4	1.221	1.74
66	RES EVALUAT	20	2.706	55.88	Q2	3.434	12.70
67	RESTAURATOR	13	0.296	7.65	Q4	0.427	0.11
68	REV ESP DOC CIENT	29	1.276	30	Q3	1.259	2.71
69	SCIENTOMETRICS	454	3.238	62.94	Q2	3.702	7.97
70	SERIALS REV	36	0.324	10	Q4	0.425	1.53
71	SOC SCI COMPUT REV	93	4.578	78.24	Q1	5.194	11.99
72	SOC SCI INFORM	28	0.714	18.24	Q4	0.966	4.48
73	TELECOMMUN POLICY	84	3.036	60.59	Q2	3.500	9.72
74	TELEMAT INFORM	91	6.182	87.65	Q1	6.769	3.61
75	TRANSINFORMACAO	23	0.648	15.88	Q4	0.561	5.20
76	Z BIBL BIBL	19	0.125	1.76	Q4	0.071	0.31

Note. The shade of green corresponds to the magnitude of each number relative to the other values in its column. As such, the larger the number, the deeper the green hue used to represent it.

. This table also includes the measure of the journal social attention. Due to space limitations, I only show the information corresponding to the year 2020 for the production and impact indicators (the last year available at the time of data collection) and the year 2021 for the journal social attention (time window 2019–2021). The graphical description of the data is shown in Figure 1 and Figure 2.

A box-and-whisker plot by quartile for the 5-year journal impact factor (2020 edition) and the journal social attention (2021 edition) is shown in Figure 1. The differences between the groups are all significant for the journal social attention (p < 0.01), except between Q1 and Q2. Note that the journal social attention decreases in groups Q2 to Q4 as the journals reduce their impact factors. This trend is observed in both the mean and the median, and even in the remaining quartiles of the distribution represented by the boxes and whiskers in the figure. However, this is not the case when moving from Q1 to Q2.

As can also be seen in Figure 1, the distributions in the group of journals with a higher impact factor (Q1) are highly skewed, especially with respect to social attention. Note that the mean, represented by the cross, is much higher than the median, represented by the central line in the box.

Figure 2 shows the enormous variability in social attention, regardless of the journal’s impact factor. Note that in the group of medium-impact journals (Q2 and Q3), there are articles that receive a lot of social attention.

4.2. Multiple Linear Regression Analysis

I would like to know if and how the social attention of articles can be predicted from the social attention of journals and several bibliometric characteristics. The description of the variables can be found in

Table 3. Variables and description in the Multiple Linear Regression model.

Name	Measure	Variable Description
Article Social Attention	Altmetric attention score for research articles published in 2019	Natural number N = {0, 1, 2, ...}
Journal Social Attention	Journal social attention 2021: average altmetric attention score for research articles published in 2019–2021	Positive real number R+0 = [0, +∞)
Num. Authors	Number of authors for research articles published in 2019	Positive natural number N+ = {1, 2, 3, ...}
OA Article	Type of access to the research articles published in 2019	Dichotomous {open access = 1, closed = 0}
Funded Article	Research funding declared in the acknowledgments section for the research articles published in 2019	Dichotomous {funded = 1, unfunded = 0}
Article Citations	Times cited in Dimensions for research articles published in 2019	Natural number N = {0, 1, 2, ...}
Journal Impact Factor	5-year journal impact factor in the 2020 issue of JCR	Positive real number R+0 = [0, +∞)

.

The dependent variable is the article social attention for publications in the year 2019, in short, “article social attention”. The independent variables are the journal social attention (2021), the number of authors for publications in the year 2019, the type of access to these publications {open access = 1, closed = 0}, the funding of the research {funded = 1, unfunded = 0}, the article citations, and the journal impact factor (2020 edition).

I checked for missing values in the descriptive statistics of all variables (see

Table 4. Means, SDs, and Pearson correlations between the dependent and all independent variables.

Variable	Mean	SD	1	2	3	4	5	6	7
1. Article Social Attention	5.31	22.42		0.22 **	0.09 **	0.10 **	0.04 **	0.10 **	0.03 *
2. Journal Social Attention	5.08	4.29			0.20 **	0.19 **	0.20 **	0.03 *	0.15 **
3. Num. Authors	2.88	2.10				0.07 **	0.33 **	0.11 **	0.19 **
4. OA Article (1, Closed = 0)	0.43	0.50					0.04 **	0.00	−0.11 **
5. Funded Article (1, Unfunded = 0)	0.24	0.42						0.11 **	0.17 **
6. Article Citations	11.87	26.62							0.39 **
7. Journal Impact Factor	4.24	3.03

* p < 0.05. ** p < 0.01.

for the mean and standard deviation). Note that I have N = 4880 independent observations in the dataset. The distributions in the histograms are likely for all variables and there are no missing values. I have also checked for curvilinear relationships or anything unusual in the plot of the dependent variable against each independent variable.

Note that each independent variable has a significant linear relationship with the article’s social attention (see Table 4). Therefore, the multiple linear regression model could estimate the article’s social attention from all independent variables simultaneously. I checked the correlations between the variables (Table 4). The absolute correlations are low (none of the correlations exceeds 0.39), and multicollinearity problems are discarded for the actual regression analysis.

In general, the observed correlations are low. The highest correlations are between article citations and journal impact factor (0.39) and between the number of authors and funding (0.33). All other correlations are below 0.22. The only negative correlation is observed between the type of access and the impact factor.

The regression model according to the b-coefficients in

Table 5. Regression coefficients predicting Article Social Attention. Standard multiple linear regression analysis.

Variable	B (Coeff.)	95% CI	β (Standardized Coeff.)	t	p (Sig.)
Constant	0	-	0	-	-
Journal Social Attention	0.741	[0.595, 0.887]	0.207	9.956	0.000
Num. Authors	0.412	[0.215, 0.609]	0.051	4.093	0.000
OA Article (1, Closed = 0)	1.755	[1.040, 2.471]	0.048	4.811	0.000
Funded Article (1, Not = 0)	1.552	[0.358, 2.746]	0.022	2.548	0.011
Article Citation	0.090	[0.079, 0.101]	0.104	16.373	0.000
Journal Impact Factor	−0.784	[−0.969, −0.600]	−0.044	−8.336	0.000

Note. Adjusted R-square R²_adj = 0.197 (N = 4880, p = 10⁻⁴). CI = confidence interval for B.

is as follows:

Article_Social_Attention_i = 0.74 ⋅ Journal_Social_Attention_i + 0.41 ⋅ Num_Authors_i
+ 1.75 ⋅ OA_Article_i + 1.55 ⋅ Funded_Article_i
+ 0.09 ⋅ Article_Citations_i − 0.78 ⋅ JIF_i

(1)

where Article_Social_Attention_i denotes the predicted social attention for article i, i = 1, 2, ..., 4880.

R-squared is the proportion of the variance in the dependent variable accounted for by the model. I have reported the adjusted R-squared in Table 5. In this model, R²_adj = 0.197. This is considered acceptable by social science standards. Furthermore, since the p-value found in the ANOVA is p = 10⁻³, I conclude that the entire regression model has a non-zero correlation.

Note that each b-coefficient in Equation (1) indicates the average increase in social attention associated with a one-unit increase in a predictor, all else equal. Thus, a 1-point increase in social attention to the journal is associated with a 0.74 increase in social attention to the article.

Similarly, an additional co-author increases the social attention of a study by an average of 0.41 points. Furthermore, 1 additional citation increases the social attention of an article by an average of 0.09 points, or, alternatively, every 10 citations increase the social attention of a study by 0.9 points, all else being equal. Similarly, a 1-point increase in the journal’s impact factor is associated with a 0.78 decrease in the social attention to the article.

For the dichotomous variables, a 1 unit increase in open access is associated with an average 1.75 point increase in the social attention to the article. Note that open access is coded in the dataset as 0 (closed access) and 1 (open access). Therefore, the only possible 1 unit increase for this variable is from closed (0) to open (1). Therefore, I can conclude that the average social attention for open articles is 1.75 points higher than for closed articles (all other things being equal). Similarly, the average social attention for funded articles is 1.55 points higher than for unfunded articles, all else being equal.

The statistical significance column (Sig. in Table 5) shows the 2-tailed p-value for each b-coefficient. Note that all of the b-coefficients in the model are statistically significant (p < 0.05), and most of them are highly statistically significant with a p-value of 10⁻³. However, the b-coefficients do not indicate the relative strength of the predictors. This is because the independent variables have different scales. The standardized regression coefficients or beta coefficients, denoted as β in Table 5, are obtained by standardizing all the regression variables before calculating the coefficients and are, therefore, comparable within and across regression models.

Thus, the 2 strongest predictors in the coefficients are the social attention of the journal (β = 0.207) and the citations received by the article (β = 0.104). This means that the journal is the factor that contributes the most to the social attention of the research, about twice as much as the citations received. In addition, the number of authors in the research (β = 0.051) contributes about half as much as citations and slightly more than open access to the publication (β = 0.048). Journal impact factor (β = −0.044) contributes as much as open access but in the opposite direction. Finally, research funding (β = 0.022) contributes half as much as the impact factor.

Regarding the multiple regression assumptions, each observation corresponds to a different article. Therefore, I can consider them independent observations. The regression residuals are approximately normally distributed in the histogram. I also checked the assumptions of homoscedasticity and linearity by plotting the residuals against the predicted values. This scatterplot shows no systematic pattern, so I can conclude that both assumptions hold.

5. Discussion

Social attention to research is crucial for understanding the impact and dissemination of scientific research beyond traditional citation-based metrics and has practical implications for academic publishing, funding decisions, and science communication.

First, it provides a measure of the impact of scientific articles beyond traditional citation-based metrics, such as the number of times an article is shared, downloaded, or discussed on social media platforms. This can help researchers, publishers, and funding agencies better understand the impact and reach of their research. Second, social attention to research can provide insights into how scientific information is disseminated and consumed by different audiences, which can inform public engagement and science communication strategies. Third, social attention research can highlight emerging trends and issues in science and technology that can guide future research agendas and funding decisions.

The results suggest that public attention to research occurs mainly in the first year after publication and to a lesser extent in the second year. However, a more detailed analysis of the dataset shows that the largest increase in social attention is observed in the first months after its publication.

Some considerations can be made about the negative signs observed in the interannual marginal variation (decrease in the average social attention compared to the previous year). This decrease could be due to several factors. First, the increasing use of social networks and the growing number of platforms from which social attention is collected. Second, the social attention of research is measured with regularly updated data on the social presence on the Internet (from June 2022 in the dataset). Since some mentions in social media may be ephemeral and disappear after a while (unlike citations in the databases that index the documents), the negative signs in the marginal variation could also be due to this circumstance. Finally, the observations (research articles) differ between years, so this result is, therefore, plausible.

The average number of authors per article in library and information science has gradually increased over the last decade, from an average of 2.37 authors per article in 2012 to an average of 3.15 authors in 2021. This 33% increase in co-authorship in a decade is relevant in terms of social attention, as discussion on the web is often driven by the authors of the research. More authors, therefore, mean more presence on social networks.

In the dataset, 40% of library and information science publications are open access. In addition, 21% of the publications indicate in the acknowledgments section that the authors have received some form of funding, with a sustained increase over the years. Note that this percentage of funded articles in LIS is half the average for all research fields [12].

The social attention of journals decreases in quartiles Q2 to Q4 as journals reduce their impact factors. However, there are no significant differences between the two highest quartiles (Q1 and Q2). This means that the journals that are most cited by researchers are not necessarily the ones that receive the most social attention. Note that this may be due to the subject category analyzed. For example, in the Library and Information Science category, there are also prestigious journals in the second quartile. This is the case, for example, for the journal Scientometrics. Journals with low obsolescence are penalized by the impact factor compared to other journals with higher obsolescence, which accumulate most of their citations in the first years after publication [26].

I found low correlations between the variables. The highest correlations are between citation count and journal impact factor (0.39) and between the number of authors and funding (0.33). All other correlations are lower than 0.22. The only negative correlation is observed between the type of access and journal impact factor. This is because, in the library and information science category, open access publishing is not yet widespread among the journals with the highest impact factors. Surprisingly, however, there is no correlation between access type and citations. In other words, open-access articles do not receive more citations than closed articles. The reason for this is the same as that mentioned above. Open access in library and information science is not generalized in high-impact journals, which are those with the greatest visibility of research [27].

I observed that a 1-point increase in the social attention of the journal is associated with an average 0.74 increase in the social attention of the article, all else equal. Similarly, an additional co-author contributes an average 0.41 increase in the social attention of the research. Furthermore, every 10 citations increase the social attention of a paper by 0.9 points, all else being equal. I also concluded that the average social attention for open articles is 1.75 points higher than for closed articles, all else being equal. Similarly, the average social attention for funded articles is 1.55 points higher than for unfunded research.

The finding that a 1-point increase in the journal impact factor is associated with a 0.78 decrease in the social attention to the article suggests that there is a negative relationship between the 2 metrics. One possible explanation for this finding is that the number of citations, which is the basis of the journal impact factor, is an indicator of the influence of research in the academic world. The academic impact of a publication is determined by several factors, such as the reputation of the authors, the standing of the institutions with which they are affiliated, and the perceived importance and quality of the research. Therefore, journals with high impact factors tend to publish research that is more specialized and may be of interest primarily to researchers in a particular field, resulting in fewer social mentions.

However, social attention is influenced by a wider range of factors, including the topic of the publication and the current trends and interests of the general public. For example, controversial or fashionable topics tend to generate a lot of social attention, regardless of the journal impact factor of the publication. Therefore, papers in high-impact factor journals that do not address current social trends or controversial topics may not receive as much social attention as papers in low-impact factor journals that do address such topics.

Another factor that may contribute to the negative correlation between the journal impact factor and the social attention of a paper is the demographics of the authors who are active on social media. Younger researchers tend to be more active on social media than their more established counterparts, and they may be more likely to publish in low-impact journals due to their less extensive research experience. This could also contribute to the negative correlation between the two measures.

In summary, the negative association between the journal impact factor and the social attention of a paper can be explained by the different factors that influence the two metrics. While the journal impact factor is primarily influenced by academic factors such as the reputation of the authors and their institutions, the social attention of a paper is influenced by a wider range of factors, including the subject of the publication and the demographics of the authors.

The standardized regression coefficients indicate that the social attention of the journal and the citations received by the article are the two strongest predictors of the social attention of the article. The analysis shows that the journal is the most influential factor, contributing about twice as much as the citations received. The number of authors in the research contributes about half as much as the citations and slightly more if the publication is open access. The impact factor of the journal has a similar influence as open access but in the opposite direction. Finally, research funding contributes about half as much as the impact factor.

There are some points to note regarding hybrid indicators and the “altmetric attention score” used in this research. A hybrid indicator can combine different sources to create a single score [25]. However, hybrid indicators are not robust and, therefore, should not be used to evaluate researchers, especially for hiring or internal promotions. In this study, the indicator was used to evaluate the research process rather than the researchers themselves.

6. Conclusions

Understanding societal attention to research is important because it can help researchers identify emerging or pressing societal issues, prioritize research questions, and engage with stakeholders and the public. It can also inform efforts to communicate research findings to a broader audience, promote evidence-based policy, and increase public trust in science.

Although most of the social attention to research occurs in the first year, even in the first few months, a robust measure with low variability over time is preferable for identifying socially influential journals. This paper proposes a three-year average of social attention as a measure of social influence for journals. I used a multiple linear regression analysis to quantify the contribution of journals to the social attention of research in comparison to other bibliometric indicators. Thus, the data source was Dimensions, and the unit of study was the research article in disciplinary journals of Library and Information Science.

As a main result, the factors that best explain the social attention of the research are the social attention of the journal and the number of citations. There are socially influential journals, and their contribution to the social attention of the article multiplies by two the effect attributed to the academic impact measured by the number of citations. Furthermore, the number of authors and open access has a moderate effect on the social attention of research. Funding has a small effect, while the impact factor of the journal has a negative effect.

It should be emphasized that low R-squared values may indicate that the predictions of an article’s social attention are not very accurate. In addition, altmetric indicators have the advantage of measuring different types of impact beyond scholarly citations, and they have the potential to identify earlier evidence of impact, making them valuable for self-assessment. Furthermore, they are useful for investigating scholarships, as in this study. Nevertheless, it is important to use social attention with caution, as it can provide a limited and biased perspective on all forms of social impact.

This study analyzed a specific area of research over a specific period of time. However, in order to apply the findings to other areas, it is recommended that further studies be conducted with more diverse data. In terms of future research directions, incorporating author characteristics, such as research experience, h-index, affiliations, and social media presence, into the model could provide insights into the social impact of their research and its correlation with citations received. In addition, the inclusion of these and other variables may improve the model’s R-squared.