Who Comes First and Who Gets Cited? A 25-Year Multi-Model Analysis of First-Author Gender Effects in Web of Science Economics

Abstract

The aim of this research is to provide a 25-year multi-model analysis of gender dynamics in economics articles that include at least one Romanian-affiliated author, published in Web of Science journals between 2000 and 2025 (2025 records current as of 15 May 2025). Drawing on 4030 papers, we map the bibliometric gender gap by examining first-author status, collaboration patterns, research topics and citation impact. The results show that the female-to-male first-author ratio for Romanian-affiliated publications is close to parity, in sharp contrast to the pronounced under-representation of women among foreign-affiliated first authors. Combining negative binomial, journal fixed-effects Poisson, quantile regressions with a text-based topic analysis, we find no systematic or robust gender penalty in citations once structural and topical factors are controlled for. The initial gender gap largely reflects men’s over-representation in higher-impact journals rather than an intrinsic bias against women’s work. Team size consistently emerges as the strongest predictor of citations, and, by extension, scientific visibility. Our findings offer valuable insights into gender dynamics in a semi-peripheral scientific system, highlighting the nuanced interplay between institutional context, research practices, legislation and academic recognition.

1. Introduction

Over the past three decades, collaborative authorship of scientific papers by research teams has become standard practice across academia [1,2], reshaping virtually every discipline, from mathematics, physics, chemistry and engineering to the social sciences and humanities. This development has boosted research productivity, fostered interdisciplinary dialogue and methodological rigour and enhanced the visibility and replicability of scientific results. The shift from single-authored works to those produced by research teams is equally evident in economics. Ref. [3] reported a persistent rise in the mean number of authors per economics article, from 1.56 in 1991 to 2.23 in 2013. Bibliometric evidence consistently shows expanding team sizes, increasing international collaboration and a broader adoption of interdisciplinary approaches. Today, most papers published in elite economics journals are collaborative endeavours involving authors affiliated with institutions in at least two countries [3,4,5].

The proportion of women in scientific research has risen markedly [6]; nevertheless, they remain under-represented, especially at the highest levels, across several dimensions: productivity, collaboration patterns, first-author position and citation impact. Gender disparities in scientific productivity and first-author status were noted as early as the 1980s. Ref. [7], for example, analysed a sample of 526 scholars who earned their doctorates between 1969 and 1970 and found that, on average, in the first twelve post-doctoral years, men published 11.2 papers, compared with 6.4 for women, yielding a female-to-male ratio of 0.57; this difference is statistically significant. The result is corroborated by [8], who emphasised that the productivity gap among biochemists appears very early in their careers and persists over the long term. Ref. [9] documented a similar effect across disciplines, showing that female researchers publish fewer articles than their male counterparts, not because of a direct gender effect, but because they typically have fewer resources, weaker institutional positions and smaller collaboration networks to sustain their productivity.

Ref. [10] examined the academic labour market and showed that collaboration networks and institutional factors shape the likelihood of publishing as first author; although these mechanisms still favour men, gender gaps in access to collaborative networks evolve over time as women become better represented in academia and research. A gender disparity in first-author positions in top journals, again to the disadvantage of women, is likewise documented by [11] for geoscience. In economics, the number of female first authors in elite journals, and their position within collaboration networks, is markedly lower and more clustered than that of men [10,12,13].

Citations, often used as a proxy for scientific impact and visibility, should ideally reflect a paper’s intrinsic quality and originality, yet recent bibliometric research reveals asymmetric gender effects. Ref. [14] showed that publications with a woman as first or last author attract fewer citations than those with men in these positions. At the national level, the evidence is mixed. Ref. [15] found no significant gender differences in citation or self-citation rates among Danish researchers, except in medical fields, while [16] reported no marked citation bias against women in countries with mature science systems (Spain, the United Kingdom and the United States) but pronounced disadvantages in Turkey and India; he also stresses that conclusions depend on the normalisation method used to compare disciplines. In gender-mixed teams, both male and female researchers garner more citations when they have a female co-author than when the co-author is male [13]. Ref. [17] further showed that gender-mixed co-authorship does not confer the same career benefits on women as on men. Women receive lower rewards for collaborating with men, whereas men are evaluated similarly whether they publish alone or with others. This so-called female co-author penalty [17,18,19,20] undermines women’s visibility and the recognition of their scientific contribution. The international academic community is therefore debating whether journals should explicitly state each author’s individual contribution to curb gender bias.

Ref. [21] showed that gender-mixed teams accrue 10–20% more citations than single-gender teams. Team heterogeneity thus acts both as a driver of visibility and as a potential locus of asymmetric reward allocation, and collaboration practices vary substantially across specialties and disciplines. Moreover, [22] demonstrated that gender imbalance shapes not only publication volumes but also the topology of topic networks. Similar analyses by [23,24] further clarify the mechanisms that generate and perpetuate gender bias in scientific research.

Bibliometric analysis has concentrated on mature science systems in the United States and Western Europe, or on broad international samples, whereas semi-peripheral research systems, such as Romania’s, remain under-explored, despite transformations that may interact uniquely with gender dynamics. Since Romania’s accession to the European Union in 2007, Romanian economic research has internationalised rapidly, offering an ideal setting in which to test whether patterns observed in core countries also apply in a semi-peripheral context.

Addressing this gap, the present study aims to examine: (1) whether gender disparities exist in the first-author position for economics articles published in Web of Science journals between 2000 and 2025 that include at least one author affiliated with a Romanian institution, with nuanced analyses and sub-grouping by the local or international affiliation of the first author; (2) how first-author gender affects article visibility, measured by Web of Science citations, comparing papers led by women and men in the overall corpus and in five additional subsets detailed below; (3) the influence of legislative changes during this period on the gender dynamics of publications; and (4) the extent to which article topic moderates these gender differences, that is, whether particular subjects favour female first authors or, conversely, penalise them in terms of citations. Using a multi-model strategy, with negative binomial models, journal fixed-effects Poisson models, quantile regressions, Structural Topic Model and extensive robustness checks, we investigate whether and why the gender penalty persists in both first-author roles and citation counts and examine how the Romanian institutional context and article topic mitigate or exacerbate these inequalities.

To our knowledge, no empirical study has yet examined Romanian peer-reviewed economic publications over a 25-year horizon. Our contribution therefore lies both in systematically charting the bibliometric gender gap across a quarter-century of Romanian economic research and in clarifying the role of gender in researchers’ scientific visibility.

2. Materials and Methods

2.1. Data Gathering and Variables

The statistical analysis rests on a corpus of articles indexed in the Web of Science Core Collection (WoS) [25], Economics category, published between 2000 and 2025, that include at least one author affiliated with a Romanian institution. The query was “WC = (Economics) AND CU = (Romania) AND PY = (2000–2025) AND DT = (Article),” executed on 15 May 2025. The WoS export returned 6725 records, downloaded in seven Excel files (the WoS limit is 1000 per file) and subsequently imported, concatenated and processed in SPSS 26.0 and R 4.4.3.

To obtain a homogeneous sample, the raw set underwent rigorous filtering. We retained only records marked with the Publication Type = J (the WoS code for journal articles, 6510 in total, 96.8% of the raw set), excluded duplicate items (WOS:000422179700019 and WOS:000422170800044) and articles that, although retrieved under “CU = Romania” had no author affiliated in Romania (WOS:000456093800005, WOS:000306250600005, WOS:000422170800049, WOS:000758146600003, WOS:001252822700006 and WOS:000915982800001). After cleaning, the analysis set comprised 6502 articles.

The initial query also retrieved 2458 articles classified as Agricultural Economics and Policy and 14 articles classified as Agricultural Economics and Policy, Food Science and Technology. These items lacked the Economics label and were assigned the research areas Agriculture or Agriculture, Food Science and Technology. Because the present study focuses on WoS Economics articles, we retained only papers carrying the Economics label, either alone or together with other economics labels, and excluded the 2472 agriculture items. The resulting dataset, containing 4030 observations, includes extended bibliometric metadata (authors’ full names, article title, journal name, language, keywords, Keywords Plus, abstract, affiliations and addresses, WoS category, citation count, open-access status and so forth).

Several additional analytical variables were derived. From Author Full Names we extracted the given name of the first author, and using the paid version of GenderAPI, we assigned each name a gender label (female or male). Given names assigned with a probability below 80% were checked manually. We then reviewed and corrected missing or inaccurate information in the first author’s affiliation. A binary variable, Ro_author (True or False), flags whether the first author is affiliated with a Romanian institution; the True category includes both Romanian scholars as well as a small number of foreign doctoral candidates enrolled at Romanian universities and foreign professors affiliated with Romanian universities or research institutes, while the False category covers foreign scholars and Romanians affiliated abroad. One article had an empty Affiliations_first field, although the author’s name is Romanian; manual inspection in WoS revealed a Romanian address, so the record was retained.

Based on the 4030 Economics-labelled articles, we constructed, for robustness, another five subsets: Economics pure (1719 articles containing only the Economics label), Ro_author total (3475 articles), Ro_author Economics pure (1414 articles), Foreign_author total (555 articles) and Foreign_author Economics pure (305 articles). Using WoS Categories, we defined the MultiDisc variable to capture multidisciplinarity coded 0 for monodisciplinary articles (Economics only, 1719 articles) and 1 for multidisciplinary articles (at least one additional label beyond Economics, 2311 articles). We also recorded Num_authors, the total number of authors per paper, used as lnAuthors (natural logarithm of Num_authors) in the econometric models. We constructed three dummies for open-access designation: OA_GG = 1 if the label contains gold or green, including any hybrid combination (1575 articles); OA_Unknown = 1 if the label is exactly “Unknown” (2300 articles); closed/hybrid = 1 when the label contains neither gold nor green and is not “Unknown,” that is, cases labelled bronze or solely hybrid (155 articles). In regressions, closed/hybrid serves as the reference category.

In

Table 1. Analysed datasets and variables.

Abbreviation	Definition	N
Total	Final WoS Economics-labelled dataset	4030
Economics pure	Coded only under WoS “Economics”	1719
Ro_author total	First author affiliated in Romania	3475
Ro_author Economics pure	First author affiliated in Romania and Economics pure	1414
Foreign_author total	First author affiliated outside Romania	555
Foreign_author Economics pure	First author affiliated abroad and Economics pure	305
MultiDisc = 1	≥1 additional WoS category beyond “Economics”	2311
OA_GG = 1	Open access (gold or green, incl. hybrids)	1575
OA_Unknown = 1	OA status = “Unknown” in WoS	2300
Closed/Hybrid = 1	Neither gold nor green and not “Unknown”	155

Source: Author’s processing.

, we present, in a structured manner, the analysed datasets and the variables used in the econometric analysis, along with their definitions and the number of articles (N) for each.

We underline that the unit of analysis is the individual article, each observation representing a study published in a journal indexed by WoS in the Economics category.

2.2. Econometric Framework

As a methodological approach, we begin by applying a ${\chi }^2$ goodness-of-fit test with one degree of freedom and reporting Cramer’s V effect size to assess whether the gender distribution of first authors departs from theoretical parity (50-50). In the next step, we model the number of WoS citations, denoted by $Y_{\mathit{i}}$ as an overdispersed count using a log-link negative binomial regression in the NB2 parametrisation. The conditional mean is ${\mu }_i=\exp \left({\beta }_0+{\beta }_1{Female}_i+{\beta }_2{MultiDisc}_i+{\beta }_3{lnAuthors}_i+{\beta }_{4}O{A_{GG}}_i+{\beta }_{5}O{A_{Unknown}}_i\right)$, and the variance is $Var\left(Y_i|X_i\right)={\mu }_i+{{\mu }_i}^2/\theta$, with $\theta >0$. The baseline specification, estimated for the complete corpus and for each of the five subsets is as follows:

$$\mathrm{ln\; E}\left[Y_i|X_i\right]={\beta }_0+{\beta }_1{Female}_i+{\beta }_2{MultiDisc}_i+{\beta }_3{lnAuthors}_i+{\beta }_{4}O{A_{GG}}_i+{\beta }_{5}O{A_{Unknown}}_i$$

(1)

where $Y_i~NB\left({\mu }_i, \theta \right).$ This specification explicitly accommodates overdispersion and yields robust estimates of the effects of the explanatory variables considered, first-author gender, the article’s multidisciplinary scope, team size (natural log of authors) and open access status, on the frequency of WoS citations. Regression coefficients, interpreted on the incidence-rate ratio (IRR) scale, quantify proportional changes in the expected number of citations associated with each covariate. Overdispersion is diagnosed using the Pearson ${\chi }^2$ divided by the degrees of freedom and, complementarily, the Cameron–Trivedi score test. Model adequacy and parsimony are evaluated using Akaike Information Criterion (AIC) and likelihood-ratio tests.

To assess a possible excess of zeros in the citation counts, we estimate a zero-inflated negative binomial (ZINB) specification as a robustness check and compare the performance of NB against ZINB using information criteria. In addition to reporting IRRs, we compute Average Marginal Effects (AME) with HC0 robust standard errors, implemented in R 4.4.3 with the MASS, marginaleffects, dplyr, and estimatr packages. As a robustness check for cross-journal heterogeneity, we also estimate a Poisson model with journal fixed effects. To account for shifting policy regimes that may shape publication and citation behaviour, we re-estimate the NB model across three legislative windows (2000–2010; 2011–2016; 2017–2025) and run additional checks with year fixed effects.

Finally, to mitigate bias induced by thematic variation, we apply a Structural Topic Model (STM) to article titles and abstracts after preprocessing, such as lowercasing, tokenisation, stemming and removal of English stop-words. We determine k using the searchK() procedure in R 4.4.3 and include the resulting document-level topic prevalences ${\theta }_{ik}\in \left[\mathrm{0,1}\right]$ that sum to 1, as k − 1 explanatory variables in the negative binomial model. In addition to the main estimates, we also estimate within-theme models (per topic and per keyword-defined group) and, as an alternative topical control, add dummies for the thirty most frequent Keywords Plus terms. We chose to use Keywords Plus instead of author-supplied keywords because the latter are subjective and can sometimes be incomplete or inaccurate. By contrast, Keywords Plus are automatically generated by the WoS platform and provide an objective perspective on the topic, indicating how the scholarly community perceives and frames the article.

All analyses are implemented in R 4.4.3 and, where indicated, cross-checked in SPSS 26.0. We note that WoS provides the aggregate “Citations_WoS” (self-citations not identified separately), thus counts are used as reported.

3. Results

As a first step in the empirical analysis, we applied a ${\chi }^2$ goodness-of-fit test (df = 1), under the null hypothesis that the proportions of women and men are equal. The results, namely the ${\chi }^2$ and Cramer’s V effect size, are presented in

Table 2. Results of ${\chi }^2$ Test.

Sample	Women	Men	N	${\mathit{\chi }}^2$(1)	p-Value	Cramer’s V
Total	1957	2073	4030	3.339	0.068	0.029
Economics pure	841	878	1719	0.796	0.372	0.022
Ro_author total	1782	1693	3475	2.279	0.131	0.026
Ro_author Economics pure	748	666	1414	4.755	0.029	0.058
Foreign_author total	175	380	555	75.721	<0.001	0.369
Foreign_author Economics pure	93	212	305	46.430	<0.001	0.390

Source: Author’s calculations using SPSS 26.0 and R 4.4.3.

.

The ${\chi }^2\left(1\right)$ on the representation of women as first authors indicated a nuanced pattern that varies across sub-groups. For the full dataset, the female-to-male proportion, 48.6% versus 51.4%, was only marginally significant at p = 0.068 (significant at the 10% level but not at the 5% level), and the effect size is very small, V = 0.0287. In the Economics pure subset, the distribution remains virtually equal, ${\chi }^2\left(1\right)=0.796, p=0.372, V=0.022$. The picture changes, however, once affiliation is considered. Among articles whose first author is affiliated with a Romanian institution, there is no statistically significant overall difference, yet the data show a slight over-representation of women. More strikingly, in the specific case of Economics pure articles with a Romanian-affiliated first author, the difference is statistically significant in favour of women, ${\chi }^2\left(1\right)=4.755, p=0.029$, although the effect is small, V = 0.058. The sharpest contrast is found in articles whose first author is affiliated outside Romania: women account for only 31.53% in the full corpus and 30.49% in Economics pure. These gender gaps are highly significant and associated with moderate effect sizes, V = 0.369 and V = 0.390, respectively. In conclusion, the gender imbalance observed in the overall sample is largely explained by articles whose first authors are based abroad, whereas Romanian-affiliated economics publications remain close to gender parity, deviating only marginally.

In the next stage, we analysed the factors that influence the number of Web of Science citations, focusing on the first author’s gender. Given the common overdispersion of citation counts, we estimated a negative binomial (NB2) model as the baseline specification and reported the estimated dispersion parameter (θ) as direct evidence of overdispersion. For completeness, we also fitted a Poisson model and used it solely as a diagnostic benchmark (outputs available upon request). Ex post diagnostics confirmed marked overdispersion; therefore, we focused on NB2 results. The incidence rate ratios (IRR) presented in

Table 3. Summary results of the negative binomial model.

Variables	Total	Economics Pure	Ro_Author Total	Ro_Author Economics Pure	Foreign_ Author Total	Foreign_ Author Economics Pure
Female	0.901 [0.825–0.984] **	0.865 [0.752–0.996] **	1.005 [0.917–1.102]	0.917 [0.787–1.068]	0.869 [0.669–1.139]	1.037 [0.721–1.514]
MultiDisc	0.980 [0.895–1.072]	-	1.079 [0.980–1.188]	-	1.035 [0.810- 1.323]	-
lnAuthors	1.633 [1.512–1.764] **	1.891 [1.678–2.131] **	1.300 [1.195–1.414] **	1.572 [1.379–1.792] **	2.069 [1.561–2.763] **	1.446 [0.876–2.381]
OA_GG	1.102 [0.865–1.388]	0.892 [0.649–1.200]	1.080 [0.825–1.392]	0.865 [0.607–1.201]	1.394 [0.805–2.293]	1.149 [0.557–2.175]
OA_Unknown	1.095 [0.860–1.375]	0.942 [0.685–1.268]	1.007 [0.770–1.296]	0.870 [0.609–1.209]	1.610 [0.935–2.631]	1.230 [0.604–2.284]
Dispersion (θ)	0.542	0.507	0.581	0.537	0.501	0.479

Source: Author’s calculations in R 4.4.3 (MASS package); results cross-checked in IBM SPSS Statistics 26.0; ** p < 0.05; “-” variable constant within the subset (not included in the model). Dispersion parameter θ is reported, with values around 0.48–0.58, indicating pronounced overdispersion.

were computed as $\mathrm{e}\mathrm{x}\mathrm{p}\left({\beta }_j\right)$.

The negative binomial estimates indicate that, in the full corpus, economics papers whose first author is a woman receive, on average, 9.9% fewer WoS citations than those whose first author is a man, a difference that is statistically significant. The gap widens in the Economics pure subset, reaching −13.5%. By contrast, for articles whose first author is affiliated with a Romanian institution, no statistically significant gender difference emerges; the same holds for Romanian-affiliated authors publishing in Economics pure journals. Hence, the citation penalty for female first authors is concentrated in the overall (and especially the Economics pure) international literature, while it disappears in output led from within Romania, suggesting that citation dynamics for women vary with national context and sub-field characteristics.

Multidisciplinarity has no significant effect on citation counts in any subset. Team size, however, is the strongest predictor: a one-unit increase in lnAuthors is associated with a 63.3% rise in citation rate, a large and highly significant effect. The association is even stronger in Economics pure articles (+89.1%). For locally affiliated first authors, the effect is positive and significant, though smaller (+30%), whereas for first authors based abroad, it is largest: a one-unit rise in lnAuthors corresponds to a +106.9% increase, again highly significant. In the first author affiliated abroad Economics pure subset, the coefficient remains positive but is not statistically significant.

Open-access route, gold or green versus unknown versus closed/hybrid, does not significantly influence citation counts in any subset (confidence intervals include 1, and coefficients fail to reach the 5% level). Overall, the gender citation penalty is confined to internationally produced economic papers (particularly Economics pure), whereas editorial practice in Romania appears neutral with respect to the first author’s gender. The benefits of larger collaborations are strong and gender-neutral, and multidisciplinarity does not confer additional visibility in the analysed data.

Residual diagnostics support the negative binomial specification. Pearson and deviance residuals are centred around zero, with dispersion ratios of 2.06 for Pearson ${\chi }^2$/df and 1.12 for deviance/df. As we can notice from Figure 1, the Tukey hanging rootogram fluctuates around zero, with no systematic pattern either at zero or in the upper tail.

To assess whether the number of uncited articles affects the estimates, we re-estimated the models using the zero-inflated negative binomial (ZINB) specification. The regression coefficients and their statistical significance remained virtually identical to those obtained from the negative binomial model, and the improvement in model fit was modest (ΔAIC $\approx 2$ across the six datasets). Thus, we present the main results from the negative binomial model, in line with the principle of parsimony. The complete ZINB estimation outputs for all six datasets are available upon request.

To test whether the impact of gender on citation counts varies according to an article’s multidisciplinary character, we used the baseline negative binomial model, adding the interaction term Female × MultiDisc and retaining the same control variables. The likelihood-ratio test showed no significant improvement in model fit (${\chi }^2\left(1\right)$ = 1.00; p = 0.32 for the full set, ${\chi }^2\left(1\right)$ = 2.49; p = 0.11 for articles whose first author is affiliated abroad, and ${\chi }^2\left(1\right)$ = 3.24; p = 0.07 for articles whose first author is locally affiliated). In subsets that contain no multidisciplinary papers with a female first author, the interaction cannot be estimated. Consequently, we kept the specification without interactions and conclude that any citation advantage or disadvantage associated with multidisciplinarity does not differ statistically by the gender of the first author. The complete estimation outputs for all six datasets are available upon request.

After estimating the negative binomial models for the total corpus and the five subsets, we computed average marginal effects with the MASS, marginaleffects, dplyr and estimatr packages, using HC0 robust standard errors. The average marginal effect (AME) of the Female variable represents the estimated difference in the expected citation count between papers whose first author is a woman and those whose first author is a man, holding all other control variables constant. The AME estimates and their 95% confidence intervals are presented in

Table 4. AME results, values expressed in WoS citations.

Sample	AME	SE	CI_Lower	CI_Upper	z	p-Value
Total	−2.479	2.262	−6.913	1.955	−1.096	0.273
Economics pure	−0.810	0.690	−2.162	0.543	−1.173	0.241
Ro_author total	0.175	0.402	−0.612	0.963	0.437	0.662
Ro_author Economics pure	−0.140	0.571	−1.260	0.979	−0.246	0.806
Foreign_author total	−1.654	2.591	−6.732	3.424	−0.638	0.523
Foreign_author Economics pure	1.518	2.773	−3.917	6.952	0.547	0.584

Source: Author’s calculations using R 4.4.3; MASS package, marginaleffects, dplyr and estimatr.

.

Re-estimating the negative binomial models and computing AMEs with robust standard errors shows that Female has no statistically significant effect on WoS citation counts in any of the six samples. For the full corpus, the estimated difference is –2.479 citations (p = 0.273), suggesting a downward tendency but lacking statistical significance. Across the thematic and affiliation-based subsets, Economics pure, local affiliated authors or abroad affiliated authors, the AMEs range from –2.479 to +1.518, and all their 95% confidence intervals include zero. Consequently, there is no robust evidence of a gender gap in citations, either for the overall body of articles or for first authors with local affiliation.

To capture the heterogeneity of the citation distribution, which features a long tail and a dense mass of poorly cited papers, we estimated quantile regressions at four representative percentiles, τ = 0.25, τ = 0.50 (the median), τ = 0.75 and τ = 0.90, using ln(1 + Citations_WoS) as the dependent variable. This monotonic transformation preserves the ranking of papers, retains uncited articles and reduces the amplitude of extreme values, yielding more stable and interpretable quartiles. Unlike mean-based models such as negative binomial or OLS, quantile regression yields separate coefficients for each segment of the distribution, addressing the following question: what is the effect for an article at quantile τ? Observations are ordered directly by their actual citation counts, avoiding distortions from journal-level bibliometric indicators. This approach allows us to test whether first-author gender, interdisciplinarity or team size affect impact uniformly or only among highly cited papers. The quantile regression results are summarised in

Table 5. Quantile regression results.

Variables	τ = 0.25	τ = 0.50	τ = 0.75	τ = 0.90
Total
Female	−0.000 [−0.027; 0.027]	0.001 [−0.098; 0.100]	0.000 [−0.104; 0.104]	−0.051 [−0.186; 0.084]
MultiDisc	0.000 [−0.288; 0.288]	0.001 [−0.117; 0.118]	−0.026 [−0.137; 0.085]	−0.101 [−0.258; 0.057]
lnAuthors	0.000 [−0.124; 0.124]	0.263 [0.172; 0.353] **	0.393 [0.272; 0.514] **	0.493 [0.374; 0.613] **
OA_GG	0.000 [−0.654; 0.654]	−0.001 [−0.275; 0.273]	−0.180 [−0.465; 0.104]	0.153 [−0.138; 0.445]
OA_Unknown	0.000 [−0.643; 0.643]	0.000 [−0.267; 0.267]	−0.180 [−0.461; 0.101]	0.049 [−0.231; 0.330]
Economics pure
Female	0.000 [−0.307; 0.307]	0.000 [−0.187; 0.187]	−0.059 [−0.239; 0.120]	−0.171 [−0.397; 0.055]
MultiDisc	-	-	-	-
lnAuthors	0.431 [0.050; 0.811] **	0.485 [0.253; 0.718] **	0.589 [0.414; 0.764] **	0.562 [0.379; 0.745] **
OA_GG	−0.000 [−0.545; 0.545]	−0.223 [−0.550; 0.104]	−0.309 [−0.599; −0.018] **	−0.182 [−0.640; 0.276]
OA_Unknown	−0.096 [−0.699; 0.506]	−0.248 [−0.577; 0.081]	−0.249 [−0.548; 0.049]	−0.171 [−0.622; 0.281]
Ro_author total
Female	0.000 [−0.056; 0.056]	0.020 [−0.092; 0.133]	0.019 [−0.091; 0.130]	0.080 [−0.062; 0.221]
MultiDisc	0.000 [−0.617; 0.617]	0.199 [0.005; 0.392] **	0.019 [−0.100; 0.139]	0.001 [−0.151; 0.153]
lnAuthors	−0.000 [−0.124; 0.124]	0.100 [−0.103; 0.302]	0.278 [0.165; 0.391] **	0.355 [0.228; 0.481] **
OA_GG	0.000 [−0.561; 0.561]	−0.158 [−0.447; 0.130]	−0.232 [−0.516; 0.052]	−0.033 [−0.229; 0.163]
OA_Unknown	0.000 [−0.544; 0.544]	−0.109 [−0.396; 0.177]	−0.212 [−0.496; 0.071]	−0.087 [−0.293; 0.119]
Ro_author Economics pure
Female	0.000 [−0.433; 0.433]	0.000 [−0.208; 0.208]	−0.008 [−0.201; 0.185]	−0.000 [−0.258; 0.258]
MultiDisc	-	-	-	-
lnAuthors	0.431 [0.036; 0.825] **	0.314 [0.057; 0.571] **	0.435 [0.245; 0.624] **	0.428 [0.246; 0.609] **
OA_GG	−0.220 [−0.909; 0.469]	−0.288 [−0.741; 0.166]	−0.478 [−0.817; −0.139] **	−0.118 [−0.398; 0.162]
OA_Unknown	−0.220 [−0.942; 0.502]	−0.288 [−0.732; 0.157]	−0.435 [−0.771; −0.099] **	−0.069 [−0.390; 0.252]
Foreign_author total
Female	0.405 [−0.069; 0.880]	0.037 [−0.284; 0.358]	−0.109 [−0.458; 0.241]	−0.038 [−0.430; 0.355]
MultiDisc	0.000 [−0.365; 0.365]	0.096 [−0.229; 0.421]	0.051 [−0.225; 0.328]	−0.171 [−0.575; 0.232]
lnAuthors	0.000 [−0.476; 0.476]	0.472 [0.068; 0.877] **	0.623 [0.264; 0.983] **	0.569 [0.233; 0.905] **
OA_GG	0.693 [0.013; 1.373]	0.376 [−0.413; 1.165]	0.320 [−0.494; 1.134]	0.261 [−0.499; 1.021]
OA_Unknown	0.693 [0.046; 1.341]	0.231 [−0.506; 0.968]	0.307 [−0.495; 1.109]	0.279 [−0.532; 1.089]
Foreign_author Economics pure
Female	0.405 [−0.207; 1.018]	0.141 [−0.312; 0.594]	0.045 [−0.357; 0.447]	0.093 [−0.515; 0.701]
MultiDisc	-	-	-	-
lnAuthors	0.000 [−0.472; 0.472]	0.122 [−0.663; 0.906]	0.126 [−0.481; 0.733]	0.188 [−0.501; 0.876]
OA_GG	0.288 [−0.766; 1.341]	0.386 [−0.351; 1.122]	0.166 [−0.937; 1.270]	−0.044 [−0.837; 0.749]
OA_Unknown	0.000 [−1.013; 1.013]	0.041 [−0.680; 0.762]	−0.038 [−1.140; 1.065]	0.243 [−0.636; 1.121]

Source: Author’s calculations in R 4.4.3; results cross-checked in IBM SPSS Statistics 26.0; “**” p < 0.05.

.

The results of quantile regressions presented in Table 5 confirm that WoS citation impact is shaped almost entirely by team size, with effects that strengthen toward the upper tail. Regarding first-author gender, we find no robust evidence of a female penalty at any of the analysed quantiles, either for the full corpus or across the other five subsets. The coefficient of lnAuthors is zero at τ= 0.25 and becomes clearly positive at the median and rises further at τ= 0.75–0.90 in the total dataset. Interpreted as the effect of doubling the number of co-authors, this corresponds to 26.3% more citations at the median and 39.3–49.3% more in the upper quartiles. The MultiDisc effect is not statistically significant in the total dataset. Patterns by subset are consistent with this picture. The Economics pure set shows a strong, monotonic lnAuthors effect across all quantiles. OA_GG is negative and statistically significant at the 0.75 quantile, while other OA effects are small and non-significant. In the Ro_author total set, multidisciplinarity appears to help only in the middle of the distribution, not in the tails. In the Foreign_author total, the effect of team size becomes significant in the upper half of the distribution (τ = 0.50–0.90), proving that even when the first author is affiliated abroad, a larger co-author team pushes outcomes upward at the median and upper quantiles. In the Foreign_author Economics pure, none of the covariates reaches conventional significance, reflecting limited within-quantile variation.

Overall, the quantile-regression results show that team size is the principal determinant of citations across the impact distribution, whereas first-author gender and multidisciplinarity do not exhibit robust or consistent effects.

To investigate whether national promotion policies shaped the gender distribution of publications and their scientific visibility, we split the entire 2000 to 2025 period into three time windows aligned with the major changes to the promotion criteria for professors and associate professors in economic sciences: first, 2000 to 2010, the pre-standardisation period, when the only legal framework was Ref. [26]; second, 2011 to 2016, starting with Ref. [27], which introduced national minimum standards for academic appointments and titles. Ref. [28] provided the first stable, detailed set of threshold criteria for each scientific field, including economics, explicitly requiring a minimum number of Web of Science (WoS) articles and citations and publication in journals with a specified impact factor or recognised indexing. These rules substantially increased the pressure on academic staff to publish in prestigious WoS international journals and accumulate citations, bringing Romanian criteria in line with international academic standards. The third period, 2017 to 2025, is when Ref. [29] strengthened the promotion standards (for example, mandating a minimum number of WoS articles with an Article Influence Score above 0.15 and specific citation counts). The standards stayed in effect until Ref. [30], and the new standards apply starting with the 2026–2027 academic year. The heightened demands intensified the drive to publish in high-quality journals and to generate impact, encouraging larger teams and more international collaboration.

This segmentation allows us to test directly whether the introduction of bibliometric standards in 2011 and their subsequent strengthening in 2016 amplified or reduced gender differences in scientific visibility. Such standards can influence publishing behaviour and journal placement, thereby indirectly affecting the distribution of citations across genders. For this analysis, we employed the same negative binomial model as in the previous sections, estimating it separately for each of the three policy-defined periods: 2000 to 2010, 2011 to 2016 and 2017 to 2025. The results are presented in

Table 6. Incidence-rate ratio (IRR) for the explanatory variables, across the three legislative windows.

Sample	2000–2010	2011–2016	2017–2025
Female
Total	1.258 [1.017–1.557] **	0.761 [0.655–0.884] **	0.943 [0.834–1.066]
Economics pure	0.967 [0.647–1.461]	0.802 [0.610–1.057]	0.896 [0.749–1.072]
Ro_author total	1.326 [1.067–1.650] **	0.847 [0.727–0.987] **	1.046 [0.915–1.193]
Ro_author Economics pure	1.089 [0.713–1.681]	0.867 [0.654–1.151]	0.927 [0.758–1.132]
Foreign_author total	0.951 [0.349–2.916]	0.692 [0.391–1.293]	0.966 [0.717–1.312]
Foreign_author Economics pure	0.214 [0.046–0.987] **	0.902 [0.390–2.393]	1.106 [0.735–1.694]
MultiDisc
Total	0.746 [0.592–0.935] **	1.207 [1.024–1.421] **	0.986 [0.871–1.116]
Economics pure	-	-	-
Ro_author total	0.749 [0.588–0.949] **	1.414 [1.187–1.681] **	1.059 [0.924–1.213]
Ro_author Economics pure	-	-	-
Foreign_author total	0.958 [0.444–2.078]	0.963 [0.579–1.610]	1.032 [0.775–1.377]
Foreign_author Economics pure	-	-	-
lnAuthors
Total	1.529 [1.262–1.854] **	1.642 [1.425–1.892] **	1.668 [1.493–1.865] **
Economics pure	1.226 [0.859–1.756]	2.268 [1.747–2.950] **	2.007 [1.719–2.341] **
Ro_author total	1.422 [1.162–1.741] **	1.296 [1.110–1.512] **	1.330 [1.179–1.501] **
Ro_author Economics pure	1.051 [0.710–1.562]	1.698 [1.295–2.236] **	1.701 [1.435–2.016] **
Foreign_author total	2.289 [0.908–6.250]	2.838 [1.559–5.509] **	2.133 [1.496–3.073] **
Foreign_author Economics pure	5.561 [1.240–25.798] **	4.759 [1.115–22.762] **	1.581 [0.832–2.985]
OA_GG
Total	0.983 [0.150–4.156]	1.613 [0.954–2.604]	1.097 [0.820–1.442]
Economics pure	0.226 [0.003–2.084]	0.805 [0.336–1.665]	0.934 [0.654–1.300]
Ro_author total	1.073 [0.023–9.740]	1.494 [0.888–2.410]	1.054 [0.754–1.439]
Ro_author Economics pure	0.822 [0.300–3.079]	0.700 [0.293–1.446]	0.908 [0.607–1.316]
Foreign_author total	1.074 [0.109–9.013]	1.542 [0.031–11.581]	1.443 [0.801–2.450]
Foreign_author Economics pure	0.075 [0.003–1.846]	1.805 [0.018–17.795]	1.139 [0.527–2.232]
OA_Unknown
Total	0.333 [0.057–1.060]	0.837 [0.506–1.314]	1.354 [1.006–1.793] **
Economics pure	0.198 [0.003–1.288]	0.609 [0.256–1.241]	1.074 [0.746–1.512]
Ro_author total	0.863 [0.020–5.944]	0.830 [0.506–1.296]	1.091 [0.775–1.503]
Ro_author Economics pure	-	0.553 [0.234–1.122]	0.925 [0.610–1.363]
Foreign_author total	0.308 [0.047–1.428]	0.847 [0.017–6.000]	2.087 [1.155–3.558] **
Foreign_author Economics pure	0.036 [0.002–0.390] **	1.121 [0.011–10.201]	1.415 [0.659–2.744]

Source: Author’s calculations in R 4.4.3; results cross-checked in IBM SPSS Statistics 26.0; “**” p < 0.05.

.

Table 6 shows that although the gender gap fluctuated with each set of bibliometric standards, team size consistently remained the primary driver of citation visibility, whereas multidisciplinarity and open access exerted only temporary, limited effects. In the pre-standardisation period (2000–2010), the global Female coefficient was positive, implying 25.8% more citations for papers whose first author was a woman, statistically significant save for a pronounced disadvantage among Economics pure articles by foreign affiliated authors (IRR = 0.214, p < 0.05). During the same interval, multidisciplinary papers were penalised (IRR = 0.746, p < 0.05), and gold/green or unknown open access had no significant impact. After the WoS thresholds were introduced (2011–2016), a significant overall penalty emerged for Female (IRR = 0.761, p < 0.05), yet the gap vanished in all subsets led by local or abroad affiliated Economics pure authors; this window also brought a visibility bonus for multidisciplinarity (IRR = 1.207, p < 0.05) and a one-unit rise in lnAuthors lifted citations by 64.2% (IRR = 1.642, p < 0.05). Gold/green open access remained non-significant, and OA_Unknown still had no effect. After the criteria were tightened (2017–2025) all Female IRRs reverted to roughly 1 and lost significance, multidisciplinary ceased to offer a consistent advantage and open access stayed inconclusive apart from a modest positive effect for OA_Unknown globally (IRR = 1.354, p < 0.05). Team size, by contrast, continued to dominate in every period as a one-unit increase in lnAuthors produced between +52.9% and +66.8% more citations (all p < 0.05) in the full corpus and most subsets. In conclusion, the gender gap surfaced only sporadically, chiefly in the international arena and immediately after 2011, while multidisciplinarity and OA failed to yield a robust visibility premium; team size, however, exerted a constant and substantial influence throughout 2000–2025.

To probe the robustness of our findings, we re-estimated a Poisson model with journal fixed effects, using the fepois function from the fixest package in R 4.4.3. In our analysis, the fixed effects are constructed from the Web of Science Source Title (journal) field. The MultiDisc variable is omitted, as the fixed effects implicitly control for journal type. The obtained results show that the coefficient on Female is not statistically significant across most datasets, apart from a marginal disadvantage, significant at p < 0.10, for first authors affiliated abroad. This result suggests that the initial gender gaps mainly reflected the over-representation of men in higher-impact journals, rather than any intrinsic quality differences. Team size remains a robust predictor: doubling the number of authors raises citations by about 17.8% in the full corpus and 11.5% among locally affiliated first authors. Gold/green open access boosts visibility by 89.6% in the full corpus and 91.3% in the Ro_author total subset, with the OA_Unknown label likewise positive and significant in those samples. The results presented in

Table 7. Poisson model with journal fixed effects, estimated using the fepois function (fixest package).

Sample	N	IRR Female	Δ %Citations (Doubling Authors)	IRR OA_GG	IRR OA_Unknown
Total	4030	0.916	+17.8%	1.896 **	1.849 **
Economics pure	1719	0.908	+18.2%	1.531	1.525
Ro_author total	3475	1.009	+11.5%	1.913 **	1.814 **
Ro_author Economics pure	1414	1.015	+12.7%	1.453	1.471
Foreign_author total	555	0.794 *	+17.5%	1.720	1.715
Foreign_author Economics pure	305	0.752	+7.0%	1.209	1.479

Source: Author’s calculations using R 4.4.3; ** p < 0.05; * p < 0.10; elasticity with respect to a doubling of team size is $(e^{\beta ln2}-1) \times 100; \beta$= Poisson coefficient of lnAuthors.

therefore reinforce the conclusion that there is no systematic citation penalty for women. A small, marginally significant exception remains for first authors based abroad.

Residual diagnostics indicate substantial overdispersion in the Poisson model with journal fixed effects. The results show the dispersion ratios Pearson ${\chi }^2/\mathrm{d}\mathrm{f}$= 12.23 and deviance/df = 9.18. Moreover, 25 journal fixed effects (31 observations) were dropped because all outcomes were zero in those groups. Residual analysis indicates that while the Poisson model with journal fixed effects is useful for absorbing journal heterogeneity, it does not model the variance of citation counts well. Therefore, we rely on the negative binomial models for inference and retain Poisson with journal fixed effects only as a robustness check.

To check whether the estimated penalty simply reflects the fact that women publish more often in recent years, years that have had less time to accumulate citations, we re-estimated every negative binomial model with year fixed effects (2000–2025). As we can notice from

Table 8. Effect of the Female variable on citations after controlling for year dummies, negative binomial models, IRR and 95% CI.

Sample	IRR	95% CI	p-Value
Female
Total	0.928	0.854–1.009	0.08 *
Economics pure	0.918	0.804–1.049	0.21
Ro_author total	1.053	0.966–1.148	0.24
Ro_author Economics pure	0.965	0.837–1.113	0.63
Foreign_author total	0.992	0.782–1.258	0.94
Foreign_author Economics pure	1.214	0.871–1.693	0.25
MultiDisc
Total	0.884	0.809–0.965	0.01 **
Economics pure	-	-	-
Ro_author total	1.035	0.942–1.136	0.48
Ro_author Economics pure	-	-	-
Foreign_author total	0.807	0.643–1.013	0.06 *
Foreign_author Economics pure	-	-	-
lnAuthors
Total	1.833	1.690–1.998	<0.001 **
Economics pure	2.268	2.011–2.559	<0.001 **
Ro_author total	1.451	1.333–1.580	<0.001 **
Ro_author Economics pure	1.805	1.589–2.050	<0.001 **
Foreign_author total	2.119	1.621–2.770	<0.001 **
Foreign_author Economics pure	1.513	0.941–2.434	0.09 *
OA_GG
Total	0.739	0.586–0.932	0.01 **
Economics pure	0.679	0.504–0.916	0.01 **
Ro_author total	0.673	0.521–0.870	<0.001 **
Ro_author Economics pure	0.581	0.417–0.809	<0.001 **
Foreign_author total	1.065	0.657–1.725	0.80
Foreign_author Economics pure	0.999	0.539–1.851	1.00
OA_Unknown
Total	0.755	0.598–0.952	0.02 **
Economics pure	0.775	0.559–1.021	0.07 *
Ro_author total	0.591	0.458–0.764	<0.001 **
Ro_author Economics pure	0.573	0.409–0.802	<0.001 **
Foreign_author total	1.213	0.750–1.960	0.43
Foreign_author Economics pure	1.076	0.583–1.988	0.81

Source: Author’s calculations in R 4.4.3; results cross-checked in IBM SPSS Statistics 26.0; ** p < 0.05; * p < 0.10.

, once publication year and the other controls are absorbed, the Female coefficient remains insignificant in almost all datasets, showing only marginal significance in the full corpus (p = 0.08). By contrast, the lnAuthors retains a strong positive effect everywhere (except Foreign_author Economics pure, where it is marginally significant). The multidisciplinarity indicator generally reduces citations: in the full set IRR = 0.884 (95% CI 0.809–0.965; p = 0.01), in Ro_author total IRR = 1.035 (p = 0.48, not significant) and in Foreign_author total IRR = 0.807 (p = 0.06, marginal). Articles tagged open access-gold/green (OA_GG) attract fewer citations in most sub-groups; the effect is not significant for abroad-affiliated authors. Papers with OA_Unknown status likewise show a significant negative effect in the total corpus, in Ro_author total and in Ro_author Economics pure, but not among authors based abroad. The results show that once year dummies and covariates are included, Female is consistently non-significant and lnAuthors reliably lifts citation counts, while multidisciplinary articles and those labelled OA_GG or OA_Unknown tend to receive fewer citations, particularly among Romania-affiliated authors.

The Web of Science export provides only the aggregate indicator Citations_WoS, without separating author self-citations, so all our analyses rely on raw counts. Earlier studies suggest that self-citations typically account for less than 10% of citations in economics; nevertheless, we cannot rule out a slight upward or downward bias in the estimated gender gap, which remains a limitation of the study.

To examine whether gender differences are topic-specific, we decomposed the Keywords Plus field (lowercasing and discarding blanks), retained only terms appearing in at least ten papers and calculated for each the share of articles with a female versus a male first author. The difference between the two shares signals gender over-representation. This procedure highlights topics favoured by women and those dominated by men and later allows us to test whether certain Keywords Plus bring a citation bonus independent of gender, team size and journal. Of the entire corpus (4030 papers), 2614 papers (64.8%) list at least one Keywords Plus entry (the remainder are “Unknown”); within this subset, 1262 have a female first author, and 1352 have a male first author.

Table 9. Keywords Plus with the highest female over-representation (≥60%).

Nr.crt	Keywords Plus	Articles	Average Citations	% Female	% Male	Δ% (F–M)
1.	urbanization	13	11.6	84.6	15.4	+69.2
2.	future	17	12.4	82.4	17.6	+64.7
3.	age	11	7.1	81.8	18.2	+63.6
4.	error-correction	10	7.8	80.0	20.0	+60.0
5.	government	10	8.8	80.0	20.0	+60.0
6.	happiness	10	11.7	80.0	20.0	+60.0
7.	outcomes	10	22.8	80.0	20.0	+60.0
8.	transformation	10	7.6	80.0	20.0	+60.0

Source: Author’s calculations using R 4.4.3. Only Keywords Plus terms that occur in at least ten articles were included. Δ% indicates the percentage-point difference between the share of papers with a female first author and the share with a male first author.

lists the Keywords Plus terms with the strongest female over-representation (≥60%). The pattern is only moderately polarised: for example, 84.6% of papers tagged “urbanization” have a female first author, whereas “credit” is 92.3% male led. Among the most heavily “feminised” topics, “outcomes”, “future” and “urbanization” also attract relatively high mean citation counts. Overall, just eight terms exceed the 60-point threshold in favour of women, covering 88 papers and representing 2.2% of the full corpus, so thematic specialisation cannot fully account for gender differences in citations.

To check whether gender differences in citations could be driven by topic specialisation, we re-estimated the negative binomial model, adding dummies for the thirty most frequent keywords. In this specification, each paper is compared with other articles published in the same journal and year, with the same team size, open-access status and explicit topic. The Female coefficient is not significant (IRR = 0.947, p = 0.221), confirming that subject distribution does not account for the initial gender differences. The model highlights several topics with a significant citation surplus, for example, “countries”, “management”, “unit root”, “social responsibility”, “governance”, “emissions”, “consumption”, “models” and “impact”. No topic is associated with a significant citation penalty. The results are reported in

Table 10. Negative binomial model with dummies for the thirty most frequent Keywords Plus.

Variables	β	SE	z	p	IRR
Intercept	1.305	0.122	10.678	0.001	3.686 **
Female	−0.054	0.044	−1.224	0.221	0.947
lnAuthors	0.425	0.040	10.590	0.001	1.529 **
OA_GG	0.054	0.117	0.462	0.644	1.056
OA_Unknown	0.073	0.116	0.627	0.531	1.075
Keywords Plus associated with a citation bonus (β > 0 and p < 0.05)
countries	0.648	0.147	4.412	0.001	1.912 **
management	0.545	0.118	4.634	0.001	1.724 **
unit root	0.503	0.217	2.319	0.020	1.654 **
social responsibility	0.497	0.242	2.059	0.040	1.644 **
governance	0.461	0.176	2.615	0.009	1.585 **
emissions	0.450	0.185	2.435	0.015	1.568 **
consumption	0.442	0.154	2.871	0.004	1.556 **
models	0.347	0.143	2.433	0.015	1.416 **
impact	0.285	0.087	3.276	0.001	1.330 **
Keywords Plus associated with a citation penalty (β < 0 and p < 0.05)
No keyword is significantly associated with a citation penalty

Source: Author’s calculations using R 4.4.3; ** p < 0.05; β-coefficients are on the log-count scale; an IRR > 1 indicates a citation surplus, whereas an IRR < 1 indicates a citation penalty.

.

To determine whether women and men specialize in different scientific topics and whether such specialization might mask or amplify gender-based citation gaps, we ap- plied the Structural Topic Model (STM) on the full corpus of 4030 Web of Science articles published between 2000 and 2025, combining titles and abstracts. All texts were lowercased, tokenized and stemmed and had English stop-words removed. The optimal k = 8 was selected using the searchK() procedure, balancing held-out likelihood, semantic coherence and exclusive topics. Estimation was carried out in R 4.4.3, with stm(v. 1.3.7), using spectral initialisation and fixed random seed (2025) for reproducibility.

The eight topics span a broad spectrum of economics and business research (Figure 2). The most prevalent themes are Topic 6, Research/Business/Management, and Topic 3, Macro and Countries/EU, each covering slightly more than 17% of the corpus. Topic 7, Models/Methods/Estimation, also features prominently. Mid-sized topics include Topic 5, Markets/Banking/Prices, and Topic 8, Firms/Performance/Finance. The least common themes are Topic 1, Work/COVID and HR, and Topic 2, Tech/Digital/Innovation, each at roughly 8–9%.

Having established the thematic structure of the corpus, we next examine whether topic prevalence differs by first-author gender and whether such differences help explain citation patterns. The marginal effects of the Female explanatory variable point to modest but clear gender-based thematic specialization within the full corpus (Figure 3). Female first authors are significantly more likely to publish on Topic 1, Work/COVID and HR, Topic 2, Tech/Digital/Innovation, Topic 3, Macro and Countries/EU, and Topic 6, Research/Business/Management. In contrast, Topic 5, Markets/Banking/Prices, and Topic 7, Models/Methods/Estimation, are significantly more prevalent among male first authors. However, the effect sizes are small in absolute terms. Topic 8, Firms/Performance/Finance, shows no significant gender difference.

When the eight topic probabilities, (document-topic shares, θ), are added to the negative binomial citation model (

Table 11. Negative binomial regression of Web of Science citations with STM topics.

Variable	IRR	95% CI	p-Value
Intercept	9.021 **	[4.96, 16.42]	<0.001
Female	1.002	[0.92, 1.09]	0.956
lnAuthors	1.692 **	[1.56, 1.83]	<0.001
Spline Year (1)	1.291	[0.95, 1.76]	0.106
Spline Year (2)	0.024 **	[0.01, 0.07]	<0.001
Spline Year (3)	0.091 **	[0.07, 0.11]	<0.001
Topic 1—Work/COVID and HR	0.979	[0.63, 1.52]	0.926
Topic 2—Tech/Digital/Innovation	3.421 **	[2.24, 5.22]	<0.001
Topic 3—Macro & Countries/EU	0.864	[0.64, 1.17]	0.350
Topic 4—Energy/Sustainability/Policy	2.116 **	[1.47, 3.04]	<0.001
Topic 5—Markets/Banking/Prices	1.682 **	[1.20, 2.37]	0.003
Topic 6—Research/Business/Management	0.465 **	[0.33, 0.66]	<0.001
Topic 7—Models/Methods/Estimation	0.483 **	[0.34, 0.68]	<0.001

Source: Author’s calculations in R 4.4.3.—“**” p < 0.05; Dispersion θ = 0.637 (S.E. = 0.016; 95% C.I. [0.606, 0.669]), indicating substantial over-dispersion relative to a Poisson model; Topic 8 is omitted from the regression to serve as the reference category.

), with Topic 8 omitted, as is the reference category, the coefficient for Female is small and remains insignificant (IRR = 1.002; p = 0.956), indicating that thematic specialization does not mask a gender-based visibility gap. Citation rates rise with team size (IRR = 1.692; p < 0.001) and vary sharply across topics. Several topics are associated with higher citation rates compared to the reference category: Topic 2, Tech/Digital/Innovation, IRR = 3.421, p < 0.001; Topic 4, Energy/Sustainability/Policy, IRR = 2.116, p < 0.001; and Topic 5, Markets/Banking/Prices, IRR = 1.682, p = 0.003. In contrast, Topic 6, Research/Business/Management, IRR = 0.465, p < 0.001, and Topic 7, Models/Methods/Estimation, IRR = 0.483, p < 0.001, are associated with lower citation rates. The differences between Topic 1, Work/COVID and HR, Topic 3, Macro and Countries/EU, and the reference category are not statistically significant.

We replicated the STM and negative binomial procedure for the other five subsets (Economics pure, Ro_author total, Ro_author Economics pure, Foreign_author total and Foreign_author Economics pure). For each dataset, the coefficient on Female remains statistically insignificant once both topic composition and paper age are controlled for. Hence, the large raw citation differentials are determined by topic choice, not by the gender of the first author. Full STM outputs and regression tables for these five supplementary subsets are available upon request.

4. Discussion

The aim of this study was to examine gender disparities in first-author status, topic choice and Web of Science citation impact for 4030 economics articles published between 2000 and 2025 (records for 2025 current as of 15 May 2025) that included at least one author affiliated with a Romanian institution.

Our contribution was to provide the first systematic 25-year map of bibliometric gender gap in Romanian economics and to investigate how gender related to first-authorship, research topics and scientific visibility. We used a multi-model framework that triangulated evidence from a goodness-of-fit ${\chi }^2$ test and negative binomial, Poisson with journal fixed effects and quantile regressions, complemented by keyword-based topic analyses, a Structural Topic Model and the integration of topic probabilities into the citation regressions, ensuring that any gender effect was evaluated net of thematic composition and publication age.

The ${\chi }^2$ test showed a marginally significant female versus male distribution in the full corpus (48.6% versus 51.4%; ${\chi }^2$(1) = 3.339, p = 0.068, V = 0.0287), an effect that vanished in the Economics pure subset. Women were over-represented as first authors in Romanian-affiliated Economics pure papers, but with a small effect size. By contrast, foreign-affiliated papers exhibited a pronounced female under-representation (31.53% overall; 30.49% in Economics pure), with moderate-to-large effect sizes. Therefore, the modest imbalance observed in the full corpus is driven almost entirely by papers led from abroad, whereas Romanian-affiliated outputs remain close to parity.

The baseline negative binomial model indicated that, after controlling for multidisciplinarity, team size and OA status, papers with a female first author received on average 9.9% fewer citations, rising to 13.5% in Economics pure. Yet the Female coefficient became insignificant in all affiliation-defined subsets, indicating that the citation gap was driven mainly by the international, Economics pure segment. Multidisciplinarity showed no robust advantage, whereas team size was the strongest predictor; a one-unit increase in the lnAuthors implies 63.3% more citations in the full set and up to +106.9% for foreign-led articles. The open access route showed no consistent effect across specifications.

Quantile regressions (0.25, 0.50, 0.75, 0.90) confirmed the dominance of team size across the citation distribution and clarified where it matters most. Female remained statistically insignificant everywhere, while the impact of team size increases toward the upper tail of the citation distribution. Splitting the data into three policy windows (2000–2010, 2011–2016 and 2017–2025) revealed that a significant female penalty emerged immediately after the 2011 WoS thresholds but disappeared in 2017–2025. Multidisciplinarity delivered a visibility bonus only in 2011–2016, and open access never produced a stable effect.

Robustness checks, Poisson models with journal fixed effects and negative binomial models with year dummies support the core conclusion: once journal prestige and year effects are absorbed, there is no systematic, robust female penalty. The initial gap reflects men’s over-representation in higher-impact journals rather than an intrinsic bias against women’s work; only a marginally significant disadvantage persists for foreign-affiliated first authors.

Topic analysis showed moderate polarisation. A few Keywords Plus (e.g., urbanization, future, outcomes) were strongly “female,” yet adding keyword dummies left the Female coefficient non-significant. Several topics (countries, management, impact etc.) carried independent citation bonuses, regardless of author gender.

The STM strengthened these results by making thematic composition explicit. Across the full corpus, eight latent topics were recovered, ranging from Work/COVID and HR to Models/Methods/Estimation. Women were significantly more active in four themes (Work/COVID and HR, Tech/Digital/Innovation, Macro and Countries/EU, Research/Business/Management), whereas men dominated two others (Markets/Banking/Prices and Models/Methods/Estimation). However, once topic probabilities of each paper were added to the negative binomial regression, the Female coefficient lost significance, while topic effects were large and heterogeneous. Replicating the STM and NB models for the other five subsets (Economics pure, Ro_author total, Ro_author Economics pure, Foreign_author total and Foreign_author Economics pure) emphasized the same pattern: topic choice and team size drive citation inequality, not author gender.

A limitation of our study is the lack of a self-citation indicator in the WoS Export. Although prior studies suggest that self-citations account for less than 10% in economics, it could slightly shift the estimated gaps. In addition, the WoS database was interrogated on 15 May 2025, while the calendar year was still in progress. Thus, coverage and citation counts for 2025 papers are provisional and may evolve as indexing and citations accrue.

The results of our study underline that once structural and topical factors are controlled, gender no longer explains visibility differences in Romanian economics research. Nevertheless, there remains a wide scope for investigation into how institutional context, promotion policies, and cultural norms influence the evolution and perception of female first authors in economic sciences within semi-peripheral research systems.