The Roles of Female Involvement and Risk Aversion in Open Access Publishing Patterns in Vietnamese Social Sciences and Humanities

Given the obstacles mentioned above, research has found several strategies to help female scientists to thrive in academia. Fathima et al. (2020) find the three most common ones used by female scientists across 55 countries: self-confidence, dedication, and hard work. Also, as publications and citations are vital for academic promotion, funding proposals, and various aspects of scholars’ lives, understanding how female scientists get their works published is crucial to develop and implement the right policies for women in science. On e of the potential methods to improve female researchers’ visibility and impact is open access (OA) publishing.

Despite its short history in science, open access has received support from major international institutions to improve accessibility to scientific knowledge, researchers, and the public. OA publishing is advocated for its ability to disseminate scientific knowledge and improve research transparency (Chan, Kirsop, & Subbiah, 2006; Shuva & Taisir, 2016). Many powerful institutions join to promote open access, including cOAliationS (Plan S), Bill and Melinda Gates Foundation, Wellcome Trust, and so on. Academic publishers seem not to miss the trend and even accelerate it. Recently, there was notable news that the publisher of Nature, one of the most prestigious academic publishers, has agreed to an unprecedented open access deal that permits researchers to publish in the journal and other Nature-branded titles under OA terms (Van Noorden, 2020). John Wiley & Sons, Inc. has acquired Hindawi, an OA publisher, with a total expenditure of $298 million, to show its commitment to the open access movement (Hoboken, 2021).

One benefit of OA publishing is that it makes the published research highly accessible among academia and the public. By removing paywalls, OA publishing is far more within reach than non-OA publishing (e.g. subscription-based publishing) (Gadd & Troll Covey, 2019). Piwowar et al. (2018) provide evidence about the visibility of OA publications. They find 47% of publications read by Unpaywall users are OA, mostly Bronze OA, despite OA publications are estimated to only account for 28% of scholarly literature. Moreover, as OA publications are free to spread, scientists will gain more recognition for their intellectual work (Davis et al., 2008). OA publications are also more likely to have a citation advantage than non-OA publications (Norris, Oppenheim, & Rowland, 2008; Sotudeh, Ghasempour, & Yaghtin, 2015), though this assumption remains controversial. Gaulé and Maystre (2011) find a significantly higher number of citations in OA publications than non-OA ones in the Proceedings of the National Academy of Sciences from three prolific journals (Nature, Cell, and Science). Having yet found a significant causal relationship between citation and open access, they attribute the self-selection of higher quality articles into OA, thus helping OA works gain more citations.

Apart from OA publishing in an official journal, some authors are determined to self-archive their works on academic social networks, such as ResearchGate and Academia.edu. Self-archiving on these networks might seem similar to green OA. Nonetheless, such practices are controversial since some content is illegally posted (Jamali, 2017). Moreover, ResearchGate and Academia.edu are independent for-profit companies, so their ethics and sustainability are still contentious (Fortney & Gonder, 2015). Two scholarly publishing giants, Elsevier and American Chemical Society (ACS), even filed a lawsuit in Germany against ResearchGate for violating the paywalled publications’ copyright. In 2013, Elsevier also sent Academia.edu 2800 takedown notices but did not take the company to court (Chawla, 2017).

Because of the advantages of OA publishing, scientists from minority groups or with less reputation might perceive OA publishing as an effective strategy to gain more recognition in the scientific community in which readers are more favorable with well-known researchers’ publications. Given the pressure that females face in academia, OA publishing, which offers greater perceived visibility and impact, could be a critical factor when female scientists consider a place to publish. For investigating this assumption, we proposed the following research question:

RQ1: Were publications with female researchers’ participation more likely to be published under OA terms than those without female researchers’ participation?

However, even though OA publishing might help female authors to reach a wider readership, it is also costly. In a study analyzing the experience with OA publishing of UK-based researchers, male scientists are found to be more experienced than female ones (Zhu, 2017). The cost of publishing OA might make it expensive for female researchers, who often have fewer resources than male researchers. Due to modest salary and funding (Manh, 2015; Shuva & Taisir, 2016), the article processing charge (APC) is a huge barrier that constrains researchers to publish OA. Q. H. Vuong et al (2021) find that gender inequality might have created more barriers towards OA publishing among Vietnamese female scientists. As collaboration is usually viewed as a useful cost-sharing model in Vietnam (T.-T. Vuong et al., 2020) and publications with all-female researchers or solo female researchers are least likely to be published under OA terms (H. T. T. Nguyen et al., 2021), we suspect that female researchers tend to collaborate with male colleagues to lessen OA's cost. Therefore, the second research question was proposed:

RQ2: Were female-involved publications with a higher percentage of female researchers less likely to be published under OA terms?

Although OA publishing provides various benefits to multiple parties, the incomplete OA evaluation, funding, and regulation models create opportunities for many inappropriate practices. Researchers in emerging countries tend to capitalize on low-quality predatory or junk OA journals for increasing their scientific output (Bayry, 2013). These inappropriate practices lead to negative views towards researchers publishing OA (Hien, 2020; Quyen, 2020), which, in turn, adversely affects their scientific career and increases the perceived risk of publishing OA.

Doing research is a highly competitive job; thus, not only is scientific productivity crucial, but the scientific quality is also emphasized for acquiring funding and maintaining public trust (Edwards & Roy, 2016). A researcher's ability to publish in journals with high impact factors is a common criterion to evaluate their competitiveness (Chapman et al., 2019). Despite the longstanding criticism against the use of journal impact factor (JIF) as the quality judgment of scientific papers or scientists (Alberts, 2013), this index is still frequently used in the evaluation process for grants, promotion, and even academic success (van Dijk, Manor, & Carey, 2014). In their paper, van Dijk et al. (2014) find that the impact factor of the journals in which the articles were published is a more significant predictor of an author's academic success than the number of citations divided by the impact factor.

Suppose a researcher(s) simultaneously acknowledges the advantages of the OA publishing model and the risk of public criticism on the quality of OA publishing. In that case, he/she might try to find an OA publishing approach that can prove the publications’ quality to minimize the risk. This tendency might be more likely to happen among risk-averse people because their perceived demand for mitigating risk increases. Given that females are more risk-averse than males (Charness & Gneezy, 2012; Eckel & Grossman, 2008), we expect female researchers to consider journals with high JIF as a critical criterion publishing OA for reducing the risk of negative criticism from colleagues and the public. We proposed the following research questions to clarify our assumptions:

RQ3: Were OA publications more likely to be published in lower JIF than non-OA publications?

The above research questions are examined by analyzing 3,122 Vietnamese researchers’ humanities and social sciences publications and employing the Bayesian analysis method aided by the Markov Chain Monte Carlo (MCMC) technique.

This paper extracted data from the Social Sciences & Humanities Peer Awards (SSHPA) database (https://sshpa.com/). The database is the result of a national project that aims to record the international scientific output of Vietnamese scientists in SS&H since 2008. It is important to note that this database covers only publications by Vietnamese researchers or about Vietnam. A publication must meet at least one of the following criteria to be recorded in the SSHPA database: (i) the author has an affiliation in Vietnam; (ii) Its topic concerns Vietnam. Further information related to the data collection, verification, and recording procedure is explained by Q.-H. Vuong et al. (2018). The dataset of 3,122 publications employed in the current study was extracted on October 24, 2020, with information related to authors (gender and nationality), title, DOI, journal, year of publication, and publisher and stored in an excel worksheet.

Since the SSHPA does not include JIF and OA status information, JIF was extracted from the 2019 Journal Citation Reports provided by Clarivate Analytics. As for OA status, Unpaywall's Simple Query Tool was employed (link: https://unpaywall.org/products/simple-query-toolle-query-tool). Some publications had no DOI and could not be searched using Unpaywall's Simple Query Tool, so manual checks were conducted by examining the journal's websites and eligible repositories that assign OA status. The procedure strictly conformed with Unpaywall's instructions and the definitions of Piwowar et al. (2018), as follow:

Gold Access: A publication that is published in a fully open access source.

Data collectors conducted a cross-check after the data regarding OA status were recorded to ensure the data quality and accuracy.

Bayesian analysis was employed in this research due to several reasons. First of all, the Bayesian approach, which treats all properties probabilistically (including uncertainty or unknown parameters) (Contreras, Brown, & Ruest, 2018), might provide a better estimation for an exploratory study than using a frequentist approach. Moreover, future research could update the current study's findings when additional data are available (Wagenmakers et al., 2018) because the simulated posterior distribution of Bayesian analysis is computed based on the prior and the likelihood simultaneously. Bayesian analysis provides considerable flexibility in fitting models with various measurement scales by incorporating the Markov Chain Monte Carlo (MCMC) simulation technique (Dunson, 2001). This advantage enables the current study to examine models with high complexity constructed using the Mindsponge mechanism as a theoretical foundation. To elaborate, those models all incorporate non-linear relationships (e.g. interaction variables).

In this study, two models are constructed using four variables to examine the four hypotheses proposed above. It should be noted that we did not include green OA status for the OpenAccess variable because the authors’ self-archiving on open repository might be free of charge. Descriptions of four variables are shown in Table 1. Model 1 was simulated to investigate Research Questions 1 and 2, whereas Model 2 was fitted to examine Research Questions 3 and 4. In Model 2, we did not include FemaleRatio but only Female and its interaction with OpenAccess because it would cause a cofounding effect between Female and FemaleRatio.

Both models were constructed following the law of parsimony, which asserts the avoidance of complex models without necessity. We simulated them using the bayesvl R package and following the protocol suggested by Q.-H. Vuong et al. (2020) due to its ease of use and ability to generate valuable graphics (Q.-H. Vuong et al., 2020). As the current study is explorative in nature, we set all coefficients’ priors as uninformative. All models are simulated using 5,000 iterations, 2,000 warm-up iterations, four cores, and four Markov chains.

Interpreting Bayesian inference's results is dissimilar from those using conventional frequentist methods. Whereas in conventional methods, researchers make a dichotomous judgment on the associations between predictor and outcome variables based on p-value, those associations analyzed using Bayesian inference are visually judged based on the parameters’ probability distributions. The probability distribution is a normal distribution with the mean value indicating the value that has the highest possibility to happen.

Nevertheless, before assessing the parameters’ probability distributions, it is essential to assess whether the simulated results meet the Markov chain central limit theorem, which assumes that the simulated samples must be independent or not autocorrelated. In other words, the Markov chains have to be convergent. Effective sample size (n_eff) and Gelman values (Rhat) are two fundamental statistics for convergence assessment. The two Markov chains’ convergence signals are the n_eff larger than 1,000 and the Rhat value equal 1. The convergence can be visually evaluated using the trace plot. If the Markov chains shown in the trace plot fluctuate around a central equilibrium, the Markov chain central limit theorem is held.

From 2008 to 2019, we observed 3,122 publications recorded in the SSHPA database, of which 1,202 documents were OA. Among 1,202 publications, more than 60% had the involvement of female researchers (768 documents). Figure 1 illustrates the annual OA publications involving female researchers during the 2008–2019 period. It can be seen that the annual number of OA papers rose rapidly from 2012 (26 documents) to 2019 (210 documents). In the last decade, the proliferation of OA publications was mostly driven by Gold Access publications, but not much change happened to other OA types, especially Hybrid and Bronze Access. Since 2015 the number of Gold Access papers has changed radically and passed the number of Green Access papers. The highest annual number of Gold Access publications was recorded in 2019, which accounted for more than 86% of the total OA publications in 2019. These statistics hint at the fact that OA publishing, especially the Gold Access mode, was recently becoming more prevalent among female researchers.

Figure 1

Among the top ten journals with the highest JIF, female-involved publications tend to be published in journals with higher JIF than publications with no female involvement (see Table 2). During the 2008–2019 period, three journals with the highest JIF in which Vietnamese social scientists published OA were Nature, Nature Human Behaviour, and PLOS Medicine. Two out of them published research that had the participation of female researcher(s). Besides, the JIF of most journals in which publications were published without female researchers’ involvement was lower than those with female researchers’ participation. Scientific Data was the only journal within the top ten published two publications without female researcher's involvement. In contrast, there were three journals in the top ten publishing two or more publications that had female researchers’ contribution: Bulletin of the World Health Organization (7 publications), PLOS Medicine (2 publications), and Journal of International Business Studies (2 publications). Even though the descriptive statistics hint that female researchers tend to publish under OA terms in high impact factor journals, this assumption still needs to be tested using inferential statistics.

Figure 2

In this sub-section, we present the simulated results of Models 1 and 2 and provide interpretation for their meaning.

3.2.1

Model 1: Gender effects on OA publishing

The first model aims to test the effects of female involvement and its interaction with the female ratio against OA publishing probability. The model's simulated results are expected to answer Research Questions 1 and 2. Before interpreting the simulated posterior results, it is vital to check the model's Markov chains’ convergence. Visually, the trace plots show the stationarity and adequate mixing of chains, confirming a good convergence of coefficients’ Markov chains (see Figure 2). In addition to the trace plots, coefficients’ effective sample size (n_eff) and the Gelman shrink factor (Rhat) are other diagnostics of the Markov chains’ convergence. The effective sample size was all over 1,000, and the Gelman shrink factor of the current model's coefficients equals 1 (see Table 3), suggesting a healthy convergence of the Markov chains.

Table 3

Simulated posterior coefficients of Model 1.

	Mean	SD	n_eff	Rhat
Constant	−0.98	0.07	5,250	1
Female	1.10	0.13	4,320	1
Female*FemaleRatio	−1.52	0.19	5,180	1

Table 3 shows that the presence of a female researcher in a publication was positively associated with the decision to publish open access (μ_{Female_OpenAccess} = 1.10 and σ_{Female_OpenAccess} = 0.13). Specifically, the Female's posterior lies entirely on the x-axis's positive side (see the cyan histogram in Figure 3), indicating a robust positive relationship between the female researcher's involvement and OA publishing probability. Meanwhile, the interaction between Female and FemaleRatio posed an opposite effect on the OA publishing probability. The interaction's posterior, with μ_{Female*FemaleRatio_OpenAccess} = −1.52 and σ_{Female*FemaleRatio_OpenAccess} = 0.19, suggests that the presence of a higher female ratio in a publication negates the impact of female researcher's participation on OA publishing probability. The posterior distribution is wholly located on the x-axis's negative side, indicating a reliable negative association between the interaction and the OA publishing likelihood.

Figure 3

3.2.2

Model 2: Gender and OA effects on JIF

Were OA publications more likely to be published in lower JIF than non-OA publications? Were OA publications that have female involvement more likely to be published in higher impact factor journals? Model 2's simulated results might answer these two questions. The second model examines the effects of open access status, female researcher's involvement, and their interaction on the journal's impact factor in which the publication was published.

Similar to Model 1, diagnostic statistics were checked to ensure the good convergence of Markov chains. All coefficients’ effective sample size (n_eff) surpasses 1,000, and the Gelman shrink factor (Rhat) equals 1 (see Table 4), signaling robust correlations between the independent variables and the dependent variable. The stationarity and adequate mixing of chains displayed in the trace plots also confirm the Markov chains’ convergence (see Figure 4).

Figure 4

Table 4

Simulated posterior coefficients of Model 2.

	Mean	SD	n_eff	Rhat
a_JIF	1.43	0.06	5,165	1
Constant	−0.78	0.11	5,227	1
Female	0.02	0.08	5,024	1
OpenAccess*Female	0.98	0.14	5,170	1

We found a negative association between the OA status and JIF, providing a “yes” answer to Research Question 3 (μ_{OpenAccess_JIF2019} = −0.78 and σ_{OpenAccess_JIF2019} = 0.11). At the same time, the negative association between JIF2019 and the interaction of OpenAccess and Female was also observed, which underscores the impact of female researcher's participation on increasing the impact factor of journals in which the publications were published OA (μ_{OpenAccess*Female_JIF2019} = 0.98 and σ_{OpenAccess*Female_JIF2019} = 0.14). Figure 5 illustrates the simulated posterior distribution of OpenAccess (x-axis) and its interaction with Female (y-axis) on a two-dimensional graph. The OpenAccess's simulated posterior lies entirely on the axis's negative side, whereas OpenAccess*Female's simulated posterior is distributed entirely on its positive side. These distributions highlight the robustness of OpenAccess's and its interaction's impacts on the JIF2019.

Figure 5

Notably, the female researcher's participation was found to have no clear effect on the JIF2019. Its standard deviation was much higher than its mean, implying the high variation among simulated posteriors. Together with the above associations, this finding suggests that female-involved publications were more likely to be published under OA terms in high impact factor journals than other scenarios, such as no female involvement – OA, female involvement – no OA, or no female involvement – no OA.