Abstract
To provide valuable insights for shaping future funding policies, in this study, we offer a comprehensive panorama of the research funding across 171 SCI disciplines in the decade 2011–2020, based on more than 13 million scientific literature records from the Web of Science. The relationship between funding and research impact is also explored. To this end, we employ two indicators: the universality and multiplicity of funding, to indicate the funding level and six indicators to gauge the impact advantages of funding. Our findings reveal an upward trend in both the universality (increasing from 66.30% to 74.26%) and multiplicity (increasing from 2.82 to 3.26) of funding over the decade concerned. The allocation of funding varies across disciplines, with life sciences and earth sciences receiving the highest percentage of funding (78.31%) and medicine having the highest multiplicity of funding (3.07). Engineering and computer science have seen relatively rapid growth in terms of universality and multiplicity of funding. Funded articles have a greater impact than unfunded ones. And this impact strengthens as the number of funding grants increases. Through regression analysis, the citation advantage of funding is also proven at the article level, although the usage advantage is not significant.
PEER REVIEW
1. INTRODUCTION
Nowadays, research funding is ubiquitous, particularly in SCI disciplines. Funding plays a crucial role in advancing technology and fostering innovative research (Lok, 2010; Wang, Liu et al., 2012; Wu, Yuan et al., 2018; Zhou, Cai, & Lyu, 2020; Zhou & Tian, 2014). Scientists are motivated to conduct scientific research by both personal interests and the strategic needs of the nation or society (Mutz, Bornmann, & Daniel, 2016; Ramos-Vielba, Thomas, & Aagaard, 2022; Salter & Martin, 2001; Wang, Wang et al., 2020). The latter, in particular, necessitates financial support and has emerged as the prevailing model for facilitating scientific research (Ramos-Vielba et al., 2022). By providing funding, scientists can establish and expand their social networks, enhance their social capital, and gain more opportunities to accumulate knowledge, skills, and experience, thereby enhancing their human capital (Bozeman & Corley, 2004; Jacob & Lefgren, 2011b). The enrichment of these forms of capital, in turn, contributes to the progression of researchers’ careers and the impact of scientific research. In summary, funding has become a prerequisite for scientists, especially junior scientists, to gain entry into the academic community (Ma, Mondragón, & Latora, 2015).
Existing research about funding has often been limited in its scope by focusing on a narrow selection of specific disciplines as samples, and failing to provide a comprehensive panorama of the funding landscape across all disciplines. Furthermore, there has been limited attention given to the intensity of funding. For instance, it is worthwhile to investigate whether articles receiving multiple funding grants are more likely to attain greater impact than those supported by a single funding source. Finally, discussions regarding the connection between funding and impact have predominantly relied on citation counts or citation-based metrics, with minimal attention given to the potential benefits of funding in terms of usage counts.
To fill these research gaps, this study introduces two indicators (i.e., universality and multiplicity) to evaluate the level of funding, and presents an extensive analysis of the funding landscape and trends across 171 SCI disciplines (included in the Science Citation Index within the Web of Science) over the 2011–2020 decade. A comprehensive overview of the impact advantages of research funding is also provided by introducing six additional metrics. We aim to offer researchers and policymakers novel insights into funding dynamics, deepen their understanding of various funding-related facets, and provide guidance for the formulation of future funding policies. Specifically, this study sets out to address four research questions as follows:
RQ1: Which SCI disciplines have the highest percentage of articles with research funding involved? And which disciplines have the highest average number of funding grants?
RQ2: How did funding trends in different SCI disciplines change in the decade under consideration? In comparison, which SCI disciplines have seen a ranking increase in the universality and multiplicity of funding, and which disciplines have witnessed a ranking decline?
RQ3: What is the impact advantage of research funding in different SCI disciplines? Do articles funded by multiple grants outperform those funded by a single one?
RQ4: After controlling for variables such as international collaboration, open access, and team size, does research funding still help boost the citation and usage counts of articles in different SCI disciplines? Do multiple grants have a great impact?
Answering the first two research questions has the following value. First, it can optimize scientific research investment decisions. Understanding which SCI disciplines receive the highest proportion of research funding and the average number of fundings can reveal the distribution pattern of funding, guiding policymakers and funding agencies to make more informed decisions in resource allocation. Second, it can promote interdisciplinary cooperation. By identifying disciplines where funding is highly concentrated, scientists can explore new opportunities for interdisciplinary collaboration, facilitating the transfer of knowledge and technology between disciplines. Lastly, it helps in predicting research hotspots. Tracking the changes in funding trends across different SCI disciplines can reveal the long-term dynamics of research investment, assisting the academic community and funding agencies in predicting future hot areas and research directions.
Furthermore, although numerous studies have already demonstrated the positive impact of funding on scientific performance, the answers to the latter two questions of this study can still be considered a beneficial supplement to this research direction. First, unlike previous studies, which often focused on specific disciplines, this research breaks through this limitation by providing a comprehensive, interdisciplinary analysis of funding. Second, this study delves into the issue of funding intensity, which helps to enhance the understanding of the relationship between the multiplicity of funding and scientific performance, a dimension rarely touched upon in existing literature. Finally, by introducing a multidimensional research performance evaluation metric, including usage counts, this study offers a more comprehensive framework for assessing the impact of funding. Despite the benefits mentioned, it is worth reiterating that this study was still conducted on the foundation of many previous relevant and valuable research efforts.
2. RELATED WORK
2.1. Research Funding in Different Disciplines
The allocation of scientific funding varies across disciplines. Xu, Tan, and Zhao (2015) proposed that the proportion of funding allocated to the natural sciences is higher than that in the social sciences. Zhao, Yu et al. (2016) confirmed that the proportion of funding in the field of economics is significantly lower than the average funding level in the social sciences. Álvarez-Bornstein, Díaz-Faes, and Bordons (2019) explored the funding patterns in biomedical research and discovered that basic medical research (e.g., “Virology”) receives more funding than clinical medicine (e.g., “Cardiac and Cardiovascular Systems”). In terms of funding amounts, Vanderelst and Speybroeck (2013) discovered that cancer research tended to be overfunded, whereas diseases of poverty, such as tropical diseases, which often occur in developing countries with smaller economies, received comparatively less funding. In comparing funding allocations for major disease burdens, Zhang, Zhao et al. (2020) found that China prioritized basic medical research, whereas the United Kingdom emphasized disease-oriented research, associated risk factors, and public health. Additionally, some scholars argue that the funding structure should be aligned with the stage of development of their country (Vinkler, 2008; Zhao, Tan et al., 2018), with less affluent countries allocating more funding toward education and engineering sciences (Rai & Lal, 2000).
2.2. Performance Evaluation of Research Funding
Performance evaluation is increasingly important for funding agencies and science policymakers as they seek to ensure that their investments generate commensurate returns (Huang & Huang, 2018; Osório & Bornmann, 2022; Pinar, 2020; Ubfal & Maffioli, 2011; Tian, Cai et al., 2024). Currently, the performance evaluation of funding primarily focuses on two aspects: productivity and impact. The former examines whether individuals, institutions, or countries exhibit increased paper production as a result of the funding (AlShareef, Alrammah et al., 2023; Auranen & Nieminen, 2010; Carayol & Matt, 2006; Sandström, 2009; Wang & Shapira, 2011; Wang et al., 2020). The latter examines whether funded articles demonstrate higher quality compared to nonfunded articles (Alvarez-Bornstein & Bordons, 2021; Costas & van Leeuwen, 2012; Rigby, 2013; Roshani, Bagherylooieh et al., 2021; Yan, Wu, & Song, 2018).
The advantages of funding in terms of productivity and impact have been widely demonstrated (Campbell, Picard-Aitken et al., 2010; Gök, Rigby, & Shapira, 2016; Trochim, Marcus et al., 2008; Zhao et al., 2018). For example, Jacob and Lefgren (2011a, 2011b) discovered that receiving an NIH postdoctoral fellowship could boost productivity by 20%, increasing the number of publications from 4.6 to approximately 5.2 within five years. Furthermore, the likelihood of publishing five or more articles was found to increase by 24% for fellowship recipients. AlShareef et al. (2023) discovered that funding could enhance productivity by 30%, as evidenced by a case study of Saudi Arabia. Regarding impact, Alvarez-Bornstein and Bordons (2021) demonstrated that funded research tends to be published in prestigious journals and receives more citations, based on a five-year analysis of scientific publications by Spanish researchers across seven disciplines.
However, the receipt of funding does not always guarantee advantages (Mariethoz, Herman, & Dreiss, 2021), as the research environment and the continuity of funding also play crucial roles in bolstering productivity (Álvarez-Bornstein & Montesi, 2020). Additionally, some studies suggested that the productivity advantage of funding may be attributed to selection bias (Ebadi & Schiffauerova, 2014). In other words, funded scientists have undergone a rigorous peer-reviewed selection process, which ensures that the chosen scientists are highly distinguished. This selection process amplifies the productivity advantage of funding. This explanation similarly applies to the impact advantage of funding.
3. DATA AND MEASUREMENTS OF RESEARCH FUNDING
3.1. Data Set
This study utilized research data collected from the Web of Science (WoS) database. In the WoS, funding information first appeared in the Science Citation Index (SCI) collection in 2008, and then was extended to the Social Sciences Citation Index (SSCI) in 2015 and the Arts and Humanities Citation Index (A&HCI) in 2017. It provides data support for researchers studying funding-related topics, thus broadening the scope of funding research.
This comprehensive database systematically categorizes each scholarly article based on its affiliated journal, leading to an allocation into one of 254 well-defined disciplines, including 178 in the SCI, 58 in the SSCI, and 18 in the A&HCI. As funding is much more universal in the natural sciences than in the social sciences and humanities (Xu et al., 2015), we focused exclusively on the SCI collections in this study. To ensure a more focused analysis, we excluded seven disciplines that exhibit a relatively stronger inclination toward the humanities and social sciences, including “Agricultural Economics & Policy,” “Behavioral Sciences,” “History & Philosophy of Science,” “Medical Ethics,” “Psychiatry,” “Psychology,” and “Substance Abuse.” The final data set comprised 13,287,984 articles (document type is limited to “article”) published between 2011 and 2020 in all 171 SCI disciplines. Correspondingly, the Journal Citation Reports (JCR) from 2011 to 2020 were obtained to determine journal partitioning information for analysis.
3.2. Indicators to Compare Research Funding in Different SCI Disciplines
Among the WoS data fields, the “FU” field contains information regarding funded projects, allowing us to determine whether an article has received funding and the number of grants it has been awarded. Based on this, we introduce two metrics to elucidate the funding landscape of each discipline: the universality of funding (UF) and the multiplicity of funding (MF). UF indicates the proportion of funded articles within a given discipline. In some disciplines, funding is indispensable and ubiquitous, while in others, articles with funding are relatively scarce. MF measures the number of funded projects in an article. MF allows us to assess the extent to which a study is considered significant, warranting substantial financial support for its execution. Articles that attract multiple grants suggest a high level of importance and garner a greater degree of financial backing. The calculation methods for these two indicators are as follows:
Universality of Funding (UF): The percentage of funded articles within a specific discipline.
Multiplicity of Funding (MF): The average number of funded projects per funded article within a specific discipline.
3.3. Impact Advantages of Research Funding
We used six indicators to measure the impact advantages of research funding based on the prestige of the publishing journal, the citations received by the article, and the corresponding usage counts.
Share of articles in Q1 journals (%Q1): Percentage of articles published in the top 25% of journals based on the journal impact factors within each discipline. This can be calculated by grouping articles by discipline, considering their publication year, and referencing the JCR for the corresponding year.
Share of articles in first decile journals (%D): Percentage of articles published in the top 10% of journals based on the journal impact factors within each discipline. This indicator can be calculated in the same way as %Q1.
Category Normalized Citation Impact (CNCI): Normalized citations of an article. The CNCI value of an article is obtained by dividing its actual citations by the average citations of all articles within the same discipline, publication year, and document type. The CNCI metric effectively mitigates the influence of variations in publication year, disciplinary field, and document type, enabling a comparison of the article’s performance against the global average. A CNCI value greater than one signifies that the article's academic performance surpasses the global benchmark; conversely, a CNCI value lower than one indicates that the article's academic performance falls below the global benchmark. The CNCI value for a discipline is obtained by averaging the CNCI values of all articles belonging to that discipline.
Category Normalized Usage Impact (CNUI): Normalized usage counts of an article. This indicator is calculated in the same way as CNCI. It is noteworthy that the usage counts refer to the total usage counts a paper has received since February 2013, which can be obtained from the “U2” field (for more detailed information, refer to https://images.webofknowledge.com/images/help/WOS/hp_usage_score.html).
Share of highly cited articles (%HCP): Percentage of articles within the top 10% of CNCI values in each discipline.
Share of highly usage articles (%HUP): Percentage of articles within the top 10% of CNUI values in each discipline.
Furthermore, we assess the impact advantage of research funding on the six aforementioned indicators. Taking the %Q1 indicator as an example, we initially categorized articles within a discipline into two groups: those funded by grants and those not funded. Then, by calculating the ratio of the %Q1 values for these two article subsets, we obtain research funding’s impact advantage on this specific indicator within the discipline, denoted as IA%Q1 (e.g., research funding’s impact advantage measured by %Q1). In a parallel manner, we employed the same methodology for the %D, CNCI, CNUI, %HCP, and %HUP indicators, labeling them as IA%D, IACNCI, IACNUI, IA%HCP, and IA%HUP, respectively. It is worth noting that if an IA value is greater than one, funded articles have a higher impact compared to nonfunded articles.
4. RESULTS AND DISCUSSION
4.1. Universality and Multiplicity of Research Funding in Different SCI Disciplines
Figure 1 displays the universality of funding for 171 SCI disciplines. Based on the classification system of disciplines developed by the CWTS at Leiden University (van Eck & Waltman, 2010), we further categorized these 171 disciplines into four groups: Computer sciences (20 purple nodes), Life and earth sciences (35 yellow nodes), Medicine and pharmacology (69 green nodes), and Engineering (47 blue nodes). In the figure, the node size represents the UF value, with larger nodes corresponding to higher UF values.
Overall, the subject area with the highest UF is Life and earth sciences (78.31%), followed by Engineering (74.27%). The UF values of Computer sciences (65.83%) and Medicine and pharmacology (63.62%) are very close. Specifically, the most funding-reliant discipline is “Evolutionary Biology” (89.60%), followed by “Cell Biology” (88.39%) and “Virology” (87.99%). Among the top 10 disciplines receiving the most funding, eight belong to Medicine and pharmacology. “Nanoscience & Nanotechnology” and “Astronomy & Astrophysics” in the field of engineering also rank among the top 10 in terms of UF. In contrast, “Emergency Medicine” and “Surgery” exhibit the lowest funding levels, with UF values below 30%, specifically 29.75% and 29.80%, respectively. Interestingly, all 10 disciplines with the lowest UF are related to the medical field, reflecting the funding disparity within this specific domain. Table S1 in the Supplementary material lists the top and bottom 10 disciplines based on UF.
Figure 2 shows the MF in each discipline. Similarly, the MF value determines the node size, with larger nodes corresponding to higher MF values. It can be seen that joint funding seems more prevalent in the medical sciences, with an MF value of 3.07. The MF values for the remaining three subject areas show minimal variation and are concentrated around 2.90. Specifically, the discipline with the highest MF is “Physics Particles & Fields” (5.98), followed by “Astronomy & Astrophysics” (5.85) and “Physics Nuclear” (5.03). These three disciplines are all physics-related and usually involve large-scale scientific engineering research. The disciplines with the lowest MF values are “Nursing,” “Education, Scientific Disciplines,” and “Medicine, Legal,” all exhibiting MF values below 2, specifically 1.84, 1.92, and 1.96, respectively. Table S2 in the Supplementary material lists the top and bottom 10 disciplines based on MF.
4.2. Trends of the Research Funding in Different SCI Disciplines
Over the decade investigated, scientific funding became increasingly important and reliable for supporting research endeavors. Figure 3A illustrates the growth in the presence and count of funding for each SCI paper during this period. In terms of UF, approximately 66.30% of all articles received funding from one or more grants in 2011. This proportion rose to 74.26% in 2020, representing an increase of about eight percentage points. Similarly, when considering MF, the average number of grants per funded article saw an upward trend. In 2011, funded articles had an average of 2.82 grants, which increased to 3.26 grants per article by 2020, representing an approximate 16% increase over the span of a decade. These findings demonstrate the growing prevalence and importance of scientific funding in research year by year.
Furthermore, we investigated the trends in distinct MF values over the decade under consideration. As depicted in Figure 3B, we divided MF into four groups: articles funded by 1, 2, 3, and 4 or more (defined as superfunded) funding sources. It can be observed that funding from more than three projects has become increasingly prevalent. Specifically, the proportion of superfunded articles rose from 24.93% in 2011 to 32% in 2020, representing an approximate 7% increase. We propose that two factors contribute to this phenomenon. On the one hand, the intensifying competitive research environment may drive researchers to secure multiple funding sources to ensure continuity and progress in their studies. On the other hand, the escalating complexity of research may necessitate expanding team sizes, leading to a greater number of funding sources for these studies. Correspondingly, the proportion of articles with 1–3 funding sources has experienced varying degrees of decline.
However, not all disciplines experienced the same increase. Governments tend to allocate more funding towards emerging research fields, prioritizing their investment in these fields. To identify the disciplines that have been more likely to attract funding favor over the decade under consideration, we investigated the changes in the rankings of UF and MF for all 171 SCI disciplines during the 10-year period.
Figure 4 illustrates the changes in UF rankings for each discipline. Overall, the two discipline clusters of Life and earth sciences and Computer sciences showed relatively stable rankings. However, significant variations were observed in Medicine and Engineering. For Medicine and pharmacology, most nodes were distributed below the diagonal line, and the overall ranking of the discipline cluster dropped by 12.57 places. In contrast, most of the nodes in Engineering are distributed above the diagonal line, and the overall ranking of the discipline cluster rose by 18.96 places. In terms of specific disciplines, “Instruments & Instrumentation” experienced the greatest increase in ranking, rising from 126 to 55, a remarkable gain of 71 places. In contrast, “Ornithology” suffered the greatest decline, falling from 28 to 101, with a decrease of 73 places. Most of the disciplines experiencing ranking increases belong to engineering. Nuclear Science, Petroleum Energy, Aviation Materials, and Robotics were among the disciplines that witnessed the greatest increase in funding over the 10-year period. Table S3 in the Supplementary material lists the top 10 disciplines with the largest increases and decreases in UF rankings.
Figure 5 illustrates the changes in MF rankings for each discipline. At the discipline cluster level, MF exhibits consistency with UF in terms of changes. Engineering and Computer sciences experienced an increase of 6.47 places and 5.95 places in MF, respectively. Conversely, Medicine and pharmacology dropped by 4.87 places. In terms of specific disciplines, “Computer Science, Cybernetics” and “Materials Science, Textiles” showed the most significant increase in MF rankings. The former rose from 86 to 31, while the latter rose from 152 to 97, both of them increasing by 55 places. In contrast, “Medicine, General & Internal” faced the most substantial decline, falling from 27 to 103, a decrease of 76 places. Table S4 in the Supplementary material lists the top 10 disciplines with the largest increases and decreases in MF rankings.
Engineering and computer science disciplines have experienced a funding preference over the decade under consideration. On the one hand, this can be attributed to the already high funding rates in certain medical fields, leading to a relatively slower growth rate. On the other hand, the development of interdisciplinary collaboration has played a significant role. The increasing complexity of modern scientific problems has driven experts from various disciplines to collaborate. As a crucial component of such interdisciplinary research projects, the engineering domain has experienced a surge in funding needs. Additionally, the prominence of artificial intelligence governance and digital transformation as primary funding priorities for governments worldwide has contributed to the flourishing funding landscape in the field of computer science.
4.3. Impact Advantages of Research Funding in Different SCI Disciplines
4.3.1. Descriptive analysis
As illustrated in Figure 6, the values of IA%Q1, IA%D, IACNCI, IACNUI, IA%HCP, and IA%HUP across various SCI disciplinary are all greater than 1, indicating that funded articles exhibit higher scientific and societal impact compared to nonfunded articles. Among these metrics, IA%Q1, IA%D, IACNCI, and IA%HCP values are relatively higher compared to IACNUI and IA%HUP, suggesting that the scientific impact advantage of funded articles is stronger than their societal impact advantage. Moreover, the IA%D metric stands out with an average value of 2.46, meaning that funded articles are 2.46 times more likely to be published in journals with an impact factor within the top 10% compared to nonfunded articles. Within the medical field, the IA%D value peaks at 3.12, while other fields have IA%D values ranging from 1.69 to 2.16. Similarly, the IA%HCP metric demonstrates a notable impact advantage, averaging at 1.87. This denotes that funded articles have a 1.87-fold higher probability of citations ranking within the global top 10% compared to nonfunded articles.
The impact advantages associated with different numbers of funding projects are depicted in Figure 7. As the number of funding grants received by an article increases, its impact advantage also increases. Similarly, the indicators that best reflect the impact advantage of multiple funding are IA%D and IA%HCP. Specifically, the IA%D value ranges between 1.3 and 1.9 for articles receiving single funding, while the IA%D value ranges between 2.1 and 4.7 for those that are superfunded. Consequently, the likelihood of superfunded articles being published in journals with a top 10% impact factor is 1.5 to 2.5 times higher than that of articles receiving single funding. Similarly, the probability of highly cited articles receiving superfunding ranking in the top 10% globally is approximately twice that of articles receiving single funding. It is important to note, however, that at the paper level alone, due to the law of diminishing marginal effects, articles with more funding grants do not always guarantee a greater impact advantage. For instance, except in the medical field, there was no statistically significant difference in the impact advantage between articles receiving three funding grants and those receiving more than three grants in other subject areas, based on IA%Q1 and IA%D metrics.
4.3.2. Regression analysis
Dependent variables . | Variables . | Medicine and pharmacology . | Engineering . | Life and earth sciences . | Computer sciences . |
---|---|---|---|---|---|
Dependent variables: CNCI | Funding | 0.270*** | 0.350*** | 0.202*** | 0.185*** |
(0.002) | (0.002) | (0.003) | (0.005) | ||
Control variables | Yes | Yes | Yes | Yes | |
Control discipline | Yes | Yes | Yes | Yes | |
Control year | Yes | Yes | Yes | Yes | |
Constant | 0.311*** | 0.712*** | 0.528*** | 0.717*** | |
(0.008) | (0.005) | (0.006) | (0.014) | ||
Dependent variables: CNUI | Funding | 0.341*** | 0.372*** | 0.236*** | 0.165*** |
(0.007) | (0.005) | (0.077) | (0.020) | ||
Control variables | Yes | Yes | Yes | Yes | |
Control discipline | Yes | Yes | Yes | Yes | |
Control year | Yes | Yes | Yes | Yes | |
Constant | 0.765*** | 0.760*** | 0.749*** | 0.761*** | |
(0.024) | (0.014) | (0.029) | (0.055) |
Dependent variables . | Variables . | Medicine and pharmacology . | Engineering . | Life and earth sciences . | Computer sciences . |
---|---|---|---|---|---|
Dependent variables: CNCI | Funding | 0.270*** | 0.350*** | 0.202*** | 0.185*** |
(0.002) | (0.002) | (0.003) | (0.005) | ||
Control variables | Yes | Yes | Yes | Yes | |
Control discipline | Yes | Yes | Yes | Yes | |
Control year | Yes | Yes | Yes | Yes | |
Constant | 0.311*** | 0.712*** | 0.528*** | 0.717*** | |
(0.008) | (0.005) | (0.006) | (0.014) | ||
Dependent variables: CNUI | Funding | 0.341*** | 0.372*** | 0.236*** | 0.165*** |
(0.007) | (0.005) | (0.077) | (0.020) | ||
Control variables | Yes | Yes | Yes | Yes | |
Control discipline | Yes | Yes | Yes | Yes | |
Control year | Yes | Yes | Yes | Yes | |
Constant | 0.765*** | 0.760*** | 0.749*** | 0.761*** | |
(0.024) | (0.014) | (0.029) | (0.055) |
Note. Robust standard errors in parentheses. *p < 0.05, **p < 0.01, ***p < 0.001. The controlled variables include whether the article engaged in international collaborations, employed open access, and the size of the research team.
Dependent variables . | Variables . | Medicine and pharmacology . | Engineering . | Life and earth sciences . | Computer sciences . |
---|---|---|---|---|---|
Dependent variables: CNCI | Funding | ||||
Num = 2 | 0.061*** | 0.112*** | 0.075*** | 0.070*** | |
(0.004) | (0.002) | (0.003) | (0.009) | ||
Num = 3 | 0.122*** | 0.228*** | 0.142*** | 0.131*** | |
(0.004) | (0.002) | (0.004) | (0.011) | ||
Num ≥ 4 | 0.394*** | 0.546*** | 0.282*** | 0.383*** | |
(0.004) | (0.002) | (0.003) | (0.009) | ||
Control variables | Yes | Yes | Yes | Yes | |
Control discipline | Yes | Yes | Yes | Yes | |
Control year | Yes | Yes | Yes | Yes | |
Constant | 0.454*** | 0.915*** | 0.649*** | 0.820*** | |
(0.009) | (0.006) | (0.007) | (0.021) | ||
Dependent variables: CNUI | Funding | ||||
Num = 2 | 0.108*** | 0.112*** | 0.131*** | 0.092*** | |
(0.011) | (0.007) | (0.016) | (0.036) | ||
Num = 3 | 0.196*** | 0.233*** | 0.198*** | 0.224*** | |
(0.013) | (0.007) | (0.018) | (0.042) | ||
Num ≥ 4 | 0.310*** | 0.474*** | 0.354*** | 0.278*** | |
(0.010) | (0.006) | (0.016) | (0.036) | ||
Control variables | Yes | Yes | Yes | Yes | |
Control discipline | Yes | Yes | Yes | Yes | |
Control year | Yes | Yes | Yes | Yes | |
Constant | 1.043*** | 0.998*** | 0.895*** | 0.800*** | |
(0.028) | (0.017) | (0.033) | (0.082) |
Dependent variables . | Variables . | Medicine and pharmacology . | Engineering . | Life and earth sciences . | Computer sciences . |
---|---|---|---|---|---|
Dependent variables: CNCI | Funding | ||||
Num = 2 | 0.061*** | 0.112*** | 0.075*** | 0.070*** | |
(0.004) | (0.002) | (0.003) | (0.009) | ||
Num = 3 | 0.122*** | 0.228*** | 0.142*** | 0.131*** | |
(0.004) | (0.002) | (0.004) | (0.011) | ||
Num ≥ 4 | 0.394*** | 0.546*** | 0.282*** | 0.383*** | |
(0.004) | (0.002) | (0.003) | (0.009) | ||
Control variables | Yes | Yes | Yes | Yes | |
Control discipline | Yes | Yes | Yes | Yes | |
Control year | Yes | Yes | Yes | Yes | |
Constant | 0.454*** | 0.915*** | 0.649*** | 0.820*** | |
(0.009) | (0.006) | (0.007) | (0.021) | ||
Dependent variables: CNUI | Funding | ||||
Num = 2 | 0.108*** | 0.112*** | 0.131*** | 0.092*** | |
(0.011) | (0.007) | (0.016) | (0.036) | ||
Num = 3 | 0.196*** | 0.233*** | 0.198*** | 0.224*** | |
(0.013) | (0.007) | (0.018) | (0.042) | ||
Num ≥ 4 | 0.310*** | 0.474*** | 0.354*** | 0.278*** | |
(0.010) | (0.006) | (0.016) | (0.036) | ||
Control variables | Yes | Yes | Yes | Yes | |
Control discipline | Yes | Yes | Yes | Yes | |
Control year | Yes | Yes | Yes | Yes | |
Constant | 1.043*** | 0.998*** | 0.895*** | 0.800*** | |
(0.028) | (0.017) | (0.033) | (0.082) |
Note. Robust standard errors in parentheses. *p < 0.05, **p < 0.01, ***p < 0.001. The controlled variables include whether the article engaged in international collaborations, employed open access, and the size of the research team.
In the formulas mentioned, Y represents CNCI or CNUI, which are the dependent variables. Funding (indicating whether the article received funding, with values of 0 or 1) and MFunding are the core explanatory variables (indicating the amount of funding received based on being funded, with values ranging from 1 to 4, where 1 serves as the reference category). Control refers to control variables, including whether the article involved international collaboration (values of 0 or 1), whether it was open access (values of 0 or 1), team size (a continuous variable with no transformations applied), the discipline to which it belongs (dummy variables), and the year of publication (dummy variables).
We first did a descriptive analysis of the association between the control variables (international collaboration, open access, and team size) that were added in the regression analysis and the research funding (see Figures 8, 9, and 10). Information regarding whether an article is classified as open access can be obtained from the “OA” field in the WoS database. Details about an article’s collaboration can be extracted from the “C1” field containing author address information. Information on the team size can be obtained from the “AF” field.
Figure 8 shows the association between funding and collaboration. The funding-related increase in international collaboration ranges from approximately 8% to 14%, with the highest increase observed in the medical domain and the lowest in the engineering domain. Compared to a single grant, the presence of multiple grants substantially elevates the proportion of international collaboration. To be more specific, in all subject areas, the proportion of superfunded articles (with more than four grants) exhibits an international collaboration rate approximately 16% higher than those with a single grant.
Figure 9 illustrates the association between funding and open access. Compared to articles without any funding, those supported by grants exhibit a higher proportion of open access. The more grants involved, the higher the proportion of open access. This association is particularly evident in the medical field. Specifically, in the medical domain, the proportion of open access articles among funded research is approximately 30% higher than that among nonfunded articles, while in other fields the difference is only about 10%. Moreover, in the medical field, the open access proportion of superfunded articles is approximately 20% higher than that of articles with a single grant. In comparison, in other subject areas, the increase in open access proportion for articles with multiple grants ranges from 4% to 7%.
Figure 10 displays the association between funding and team size. Compared to nonfunded articles, those that received funding had a larger team size. The more grants involved, the larger the team size. Specifically, the funded articles have an average team size of approximately two more members than the unfunded articles. Furthermore, in the medical field, superfunded articles have a team size that is eight (12 versus four) members larger than those with a single grant.
Regarding the regression results, Table 1 shows that funding brings citation and usage advantages. Compared to articles that did not receive funding, funded articles tend to receive higher citation and usage counts. This promotional effect is most evident in the engineering field, followed by the medical field. Table 2 reveals that, compared to articles with single funding, those with superfunding often receive more citations and usage counts. Specifically, for each additional funding grant, the regression coefficient approximately doubles.
4.3.3. Robustness test
We enhanced the robustness of the regression results by introducing extra control variables, namely, journal reputation and author reputation. The journal ranking in JCR is used to signify journal reputation, where a higher journal ranking indicates a better reputation. Author reputation is measured using author productivity, which is represented by the number of publications by the author. The greater the number of publications, the higher the author’s reputation.
Due to the lack of author disambiguation in the WoS database, it is challenging to effectively calculate author reputations. To address this issue to some extent, we adopted a novel database, the OpenAlex database (formerly known as MAG). The OpenAlex database includes author disambiguation and directly provides author reputation data (Liu, Jones et al., 2023). Based on this, we conducted a matching process between the WoS database and the OpenAlex database by article DOIs. This allowed us to obtain the publication counts for each author in every article. Because we utilized the 2022 version of the OpenAlex database, the author reputation reflects the total number of publications for each author up to the year 2022. Meanwhile, to ensure a sufficient time window for articles to accumulate an adequate number of citations, we specifically chose data from the year 2019 for conducting robustness tests (the retrieval of WoS data was conducted in February 2023).
Furthermore, we used the reputation of the corresponding author for each article as the measure of author reputation for that specific article. For articles with multiple corresponding authors, we selected the highest reputation value among them. It is worth noting that out of the 1,492,816 articles retrieved from WoS in 2019, there were 262,615 articles that did not successfully match with the OpenAlex database. As a result, we retained 1,230,201 articles for the robustness test, including 187,479 articles with multiple corresponding authors.
The results of the robustness test are shown in Tables 3 and 4. From Table 3, it can be seen that, even after controlling for author reputation and journal reputation, the impact of funding on citation counts remains significantly positive, except in the field of computer science. However, the influence of funding on usage counts is no longer significant. Furthermore, we further verified the positive impact of funding on citation counts in the fields of “Medicine and pharmacology,” “Engineering,” and “Life and earth sciences” by selecting a subdiscipline in each area and using the classical matching method, propensity score matching (PSM). Details can be found in Tables S5 and S6 in the Supplementary material. Table 4 reveals that articles receiving multiple grants are more likely to receive higher citation counts compared to those receiving single grants, but this phenomenon is less pronounced in terms of usage counts.
Dependent variables . | Variables . | Medicine and pharmacology . | Engineering . | Life and earth sciences . | Computer sciences . |
---|---|---|---|---|---|
Dependent variables: CNCI | Funding | 0.037*** | 0.038*** | 0.025*** | 0.005 |
(0.011) | (0.004) | (0.006) | (0.010) | ||
Control variables | Yes | Yes | Yes | Yes | |
Author reputation | Yes | Yes | Yes | Yes | |
Journal reputation | Yes | Yes | Yes | Yes | |
Control discipline | Yes | Yes | Yes | Yes | |
Constant | −0.316*** | −0.725*** | −0.358*** | −0.391*** | |
(0.016) | (0.010) | (0.011) | (0.021) | ||
Dependent variables: CNUI | Funding | 0.220*** | −0.034 | 0.005 | 0.009 |
(0.048) | (0.040) | (0.067) | (0.065) | ||
Control variables | Yes | Yes | Yes | Yes | |
Author reputation | Yes | Yes | Yes | Yes | |
Journal reputation | Yes | Yes | Yes | Yes | |
Control discipline | Yes | Yes | Yes | Yes | |
Constant | 1.583*** | 1.129*** | 1.485*** | 0.986*** | |
(0.118) | (0.086) | (0.105) | (0.277) |
Dependent variables . | Variables . | Medicine and pharmacology . | Engineering . | Life and earth sciences . | Computer sciences . |
---|---|---|---|---|---|
Dependent variables: CNCI | Funding | 0.037*** | 0.038*** | 0.025*** | 0.005 |
(0.011) | (0.004) | (0.006) | (0.010) | ||
Control variables | Yes | Yes | Yes | Yes | |
Author reputation | Yes | Yes | Yes | Yes | |
Journal reputation | Yes | Yes | Yes | Yes | |
Control discipline | Yes | Yes | Yes | Yes | |
Constant | −0.316*** | −0.725*** | −0.358*** | −0.391*** | |
(0.016) | (0.010) | (0.011) | (0.021) | ||
Dependent variables: CNUI | Funding | 0.220*** | −0.034 | 0.005 | 0.009 |
(0.048) | (0.040) | (0.067) | (0.065) | ||
Control variables | Yes | Yes | Yes | Yes | |
Author reputation | Yes | Yes | Yes | Yes | |
Journal reputation | Yes | Yes | Yes | Yes | |
Control discipline | Yes | Yes | Yes | Yes | |
Constant | 1.583*** | 1.129*** | 1.485*** | 0.986*** | |
(0.118) | (0.086) | (0.105) | (0.277) |
Note. Robust standard errors in parentheses. *p < 0.05, **p < 0.01, ***p < 0.001. The controlled variables include whether the article engaged in international collaborations, employed open access, and the size of the research team.
Dependent variables . | Variables . | Medicine and pharmacology . | Engineering . | Life and earth sciences . | Computer sciences . |
---|---|---|---|---|---|
Dependent variables: CNCI | Funding | ||||
Num = 2 | 0.032*** | 0.015*** | 0.019*** | −0.011 | |
(0.006) | (0.004) | (0.006) | (0.013) | ||
Num = 3 | 0.099*** | 0.075*** | 0.051*** | 0.034* | |
(0.007) | (0.005) | (0.007) | (0.017) | ||
Num ≥ 4 | 0.354*** | 0.316*** | 0.171*** | 0.185*** | |
(0.007) | (0.005) | (0.007) | (0.016) | ||
Control variables | Yes | Yes | Yes | Yes | |
Author reputation | Yes | Yes | Yes | Yes | |
Journal reputation | Yes | Yes | Yes | Yes | |
Control discipline | Yes | Yes | Yes | Yes | |
Constant | −0.444*** | −0.863*** | −0.419*** | −0.496*** | |
(0.020) | (0.012) | (0.012) | (0.027) | ||
Dependent variables: CNUI | Funding | ||||
Num = 2 | 0.080 | 0.012 | 0.118 | −0.006 | |
(0.068) | (0.031) | (0.062) | (0.119) | ||
Num = 3 | 0.286*** | 0.079 | 0.077 | −0.054 | |
(0.072) | (0.041) | (0.063) | (0.117) | ||
Num ≥ 4 | 0.395*** | 0.122*** | 0.226** | 0.054 | |
(0.062) | (0.034) | (0.078) | (0.125) | ||
Control variables | Yes | Yes | Yes | Yes | |
Author reputation | Yes | Yes | Yes | Yes | |
Journal reputation | Yes | Yes | Yes | Yes | |
Control discipline | Yes | Yes | Yes | Yes | |
Constant | 1.917*** | 1.150*** | 1.429*** | 1.373** | |
(0.155) | (0.103) | (0.129) | (0.484) |
Dependent variables . | Variables . | Medicine and pharmacology . | Engineering . | Life and earth sciences . | Computer sciences . |
---|---|---|---|---|---|
Dependent variables: CNCI | Funding | ||||
Num = 2 | 0.032*** | 0.015*** | 0.019*** | −0.011 | |
(0.006) | (0.004) | (0.006) | (0.013) | ||
Num = 3 | 0.099*** | 0.075*** | 0.051*** | 0.034* | |
(0.007) | (0.005) | (0.007) | (0.017) | ||
Num ≥ 4 | 0.354*** | 0.316*** | 0.171*** | 0.185*** | |
(0.007) | (0.005) | (0.007) | (0.016) | ||
Control variables | Yes | Yes | Yes | Yes | |
Author reputation | Yes | Yes | Yes | Yes | |
Journal reputation | Yes | Yes | Yes | Yes | |
Control discipline | Yes | Yes | Yes | Yes | |
Constant | −0.444*** | −0.863*** | −0.419*** | −0.496*** | |
(0.020) | (0.012) | (0.012) | (0.027) | ||
Dependent variables: CNUI | Funding | ||||
Num = 2 | 0.080 | 0.012 | 0.118 | −0.006 | |
(0.068) | (0.031) | (0.062) | (0.119) | ||
Num = 3 | 0.286*** | 0.079 | 0.077 | −0.054 | |
(0.072) | (0.041) | (0.063) | (0.117) | ||
Num ≥ 4 | 0.395*** | 0.122*** | 0.226** | 0.054 | |
(0.062) | (0.034) | (0.078) | (0.125) | ||
Control variables | Yes | Yes | Yes | Yes | |
Author reputation | Yes | Yes | Yes | Yes | |
Journal reputation | Yes | Yes | Yes | Yes | |
Control discipline | Yes | Yes | Yes | Yes | |
Constant | 1.917*** | 1.150*** | 1.429*** | 1.373** | |
(0.155) | (0.103) | (0.129) | (0.484) |
Note. Robust standard errors in parentheses. *p < 0.05, **p < 0.01, ***p < 0.001. The controlled variables include whether the article engaged in international collaborations, employed open access, and the size of the research team.
Overall, the results of the robustness tests also demonstrate the positive impact of funding on citation counts. Articles receiving multiple grants are more likely to achieve higher citation counts compared to those with single grants. However, funding does not have a significant influence on usage counts.
5. CONCLUSIONS
In this study, we utilized a large-scale data set to outline the funding situation for 171 SCI disciplines, using a comprehensive disciplinary map as a framework. Specifically, we proposed two indicators, including the universality of funding (UF) and multiplicity of funding (MF), to examine the characteristics of scientific funding in different SCI disciplines over the period 2011–2020. We also investigated the association between these two funding indicators and research impact. To this end, we employed six impact indicators to study the impact advantages of funding. Through analyzing the aforementioned indicators, this study provides valuable insights for funders and researchers, facilitating a comprehensive understanding of funding characteristics and differences across various SCI disciplines as well as the advantages it generates. This can be helpful in formulating better scientific research policies and funding strategies.
The role of funding in research has become increasingly prominent. In the decade under consideration, the percentage of articles receiving funding rose from 66.30% to 74.26%, accompanied by an increase in the average number of grants per article from 2.82 to 3.26. Although a noteworthy improvement has been observed in both the universality and multiplicity of funding, the pace of growth appears to have slowed down. In addition, the proportion of articles that receive single funding is decreasing year by year, while the proportion of superfunded articles is increasing, with an annual increase of about 0.7 percentage points.
There are differences in funding among different disciplines. First, the UF in life science and earth science is high, reaching 78.31%, while the medical field has the highest MF at 3.07. Second, there is an imbalance within disciplines, with the UF of “Evolutionary Biology” being over three times higher than that of “Emergency Medicine.” Third, “Physics, Particles & Fields” and “Astronomy & Astrophysics” have the highest MF, with a value above 5. In addition, we have observed the liquidity of funding among different disciplines in the decade under consideration. As evidenced by the two levels of UF and MF, funding in medicine flowed to engineering and computer science.
Funded articles have a greater impact compared to unfunded articles. This trend becomes more pronounced with an increasing number of funding grants. Specifically, superfunded articles have a twofold higher likelihood of being published in the top 10% of high-impact journals and becoming the top 10% of highly cited articles than those receiving a single grant. While it is acknowledged that superfunding can lead to a certain level of research resource monopolization (Yan et al., 2018), it is still necessary and justified for certain research endeavors to receive superfunding to tackle challenges based on the appropriate allocation of resources. The citation advantage of research funding was further confirmed at the article level through regression analysis, although the usage advantage is not significant.
In addition, it is essential to explore how we can expand the evaluation horizon to better assess the performance of funding. This expansion should go beyond the sole consideration of scientific benefits and encompass a broader range of social and economic benefits (Lane & Bertuzzi, 2011; Lane, Owen-Smith et al., 2015; Weinberg, Owen-Smith et al., 2014). In particular, the evaluation of social benefits can include factors such as the project's contribution to social equity, educational accessibility, and cultural dissemination. The evaluation of economic benefits can cover the project's contribution to employment opportunities, industrial development, innovation capacity, etc. To achieve this goal, it is necessary to establish a comprehensive evaluation framework that incorporates scientific, economic, and social indicators. This framework will not only provide a solid basis for decision-makers but also help funding recipients better engage in dialog and cooperation with social stakeholders and realize the organic integration of scientific research and social development.
For future research, this article suggests three potential avenues of investigation. First, the indicators proposed herein can be used to comprehensively analyze the funding picture in the fields of social sciences (included in the SSCI) and humanities (included in the A&HCI). Second, the moderation effect can be used to explore the underlying mechanisms of author reputation or journal reputation on the relationship between research funding and citation counts of articles. Finally, it is possible to incorporate more control variables for a more rigorous regression analysis and even a causal analysis of the relationship between research funding and citation counts.
ACKNOWLEDGMENTS
The authors are grateful to the anonymous reviewers for their helpful comments and suggestions.
AUTHOR CONTRIBUTIONS
Wencan Tian: Conceptualization, Data curation, Formal analysis, Methodology, Writing—original draft, Writing—review & editing. Ruonan Cai: Data curation, Methodology, Software, Visualization, Writing—review & editing. Zhichao Fang: Data curation, Formal analysis, Writing—review & editing. Qianqian Xie: Conceptualization, Formal analysis, Investigation, Validation. Zhigang Hu: Funding acquisition, Investigation, Supervision, Writing—review & editing. Xianwen Wang: Conceptualization, Funding acquisition, Supervision, Writing—review & editing.
COMPETING INTERESTS
The authors have no competing interests.
FUNDING INFORMATION
This study is supported by the Major Project of National Social Science Foundation of China (22&ZD194), the National Natural Science Foundation of China (71974030, 71974029, 72304274), the LiaoNing Revitalization Talents Program (XLYC2007149), and the Scientific Research Funding of Renmin University of China (23XNF037). Wencan Tian is financially supported by the China Scholarship Council (202106060134).
DATA AVAILABILITY
Data can be accessed from: https://github.com/Tianwencan/funding.
REFERENCES
Author notes
Handling Editor: Vincent Larivière