IntroductionLinda Nochlin’s question, "Why have there been no great women artists?"1, formulated 50 years ago, today is informed by widespread awareness of gender disparities across all disciplines, from science to industry. It is also supported by evidence of women’s under-representation in the art world, captured by multiple metrics, from disparities in the number of women and men artists2,3,4,5, to the lack of gender balance in museums and galleries4,6,7,8,9, and the lack of women artists in auction sales10,11,12,13. The root of these gender differences remains a topic of debate, with hypotheses ranging from genetic predisposition to discrimination10,14,15. In this article, we focus on the potential relationship of institutional practices on gender inequality in the art world. Regardless of whether institutional practices are grounded in potential differences in the artwork itself, or reflect discrimination of the artists, ultimately institutional practices combine with reputation shape access to representation, resources and rewards in art16,17,18. These mechanisms are exacerbated by lock-in effects that limit the access of some artists to future opportunities, while offering winners-take-all outcomes for others16. The interlocking effects suggest that to truly understand gender inequality in art, we need to understand the ecosystem that surrounds the production and the consumption of art, from the complete exhibition careers of all artists to the exhibition record of all institutions, together with the auction record of the secondary market, as they together shape the structural and institutional patterns that are related to gender bias. This research is methodologically aided by recent advances in quantifying the evolution, roots, and impact of gender disparities in science and other areas19,20,21,22,23,24,25,26,27,28,29,30. As scientific discourse is increasingly rooted in data, its production and impact is cataloged in a fashion that makes it amenable for quantitative analysis. In contrast, systematic and reliable datasets that allow us to quantify artistic careers have been difficult to develop and explore.Here, we remedy this situation by exploring a dataset with a broad coverage, capturing the exhibition careers of 496,354 artists in 23,570 institutions, representing both museums and galleries16,31. For the purposes of this study, we restricted our focus to contemporary, active artists who began exhibiting on and after 1990, hence their exhibition history can be accurately reconstructed, resulting in 65,768 artists participating in 378,517 exhibitions at 20,389 institutions worldwide (Fig. 1a, b, c). This focus is motivated by our goal of identifying relationships between institutional gender representation and active artist careers. To quantify gender-related patterns, we combine expert curated gender information with a gender identification service based on artists’ names genderize.io(21,32), allowing us to assign most artists a binary gender (See “Methods” section). To understand gender inequality of each art institution, we inspect each institution under two criteria of gender equity: gender-neutrality, when artists have gender-independent access to exhibition, and gender-balance, which aims for gender parity in representation. We find that 58% of art institutions are gender-neutral but only 24% are gender-balanced, and the fraction of man-overrepresented institutions increases with institutional prestige. To connect institutional gender inequality with an artist’s career, we define the co-exhibition gender of an artist to capture the gender inequality of the institutions that an artist exhibits under gender-neutral criteria. We find that artists with high prestige are more likely to be labeled as co-exhibition man, and co-exhibition gender tends to remain stable over an artist’s career. Finally, we build a logistic regression model to predict an artist’s access to the auction market, finding that co-exhibition gender has a stronger association with access to the auction market than artist’s gender.Fig. 1: Gender bias in art.Full size imagea The gender-identified artist population and exhibitions studied in this paper. See “Methods” section for the data cleaning and filtering process that resulted in the explored artist pool. b Number of artists over time, showing separately the number of men and women artists. The shadowed area corresponds to artists and exhibitions selected for our study. c Number of exhibitions over time, shown separately for men and women artists. The shadowed area corresponds to artists and exhibitions selected for our study.ResultsGender inequality in artist population and access to exhibitionTo begin, we identify gender differences in the number of artists and exhibitions. This initial overview provides context for the more nuanced analysis which follows and empowers us to capture institutional standpoint of gender imbalances in the art world. Our sample of 65,768 artists contains 41,738 men and 24,030 women, which corresponds to 1.74 men artists for every women artist (Fig. 1a, Table 1). This ratio captures the inherent population disparity, an elementary and widely explored evidence of gender imbalance2,4,8. However, population disparity reflects only one dimension of gender inequality. Another important metric is the exhibition disparity, which focuses on an artist’s access to exhibition opportunities. We found that men artists were exhibited 662,571 times, in contrast to women artists who were exhibited only 355,506 times, resulting in an exhibition gender ratio of 1.86 (Fig. 1a, Table 1). This exhibition disparity is higher than the population disparity, indicating not only that men artists outnumber women artists, but that men artists get more access to exhibition opportunities than expected based on their presence in the artist population.Table 1 Metrics of gender bias in artFull size tableTaken together, these results support previous observations relying on fragmented or local data, used by art theorists, activists, and artists to urge institutions to acknowledge and address gender imbalance4,6,7,8,11. These data further document a persistent gender imbalance both in terms of the number of exhibiting artists, and in terms of access to exhibition and auction venues.Gender-balance and gender-neutrality in artMuseums and galleries today are under tremendous professional, community, and societal pressure to address the widespread gender imbalance present in the art world. Operationally, however, this conversation becomes muddled, since there are many different criteria for assessing gender equality in art. Here, we quantitatively explore our progress under two criteria: (1) gender-neutral hypothesis, i.e., the expectation that all artists, independent of gender, should have the same equitable access to exhibition opportunities (Table 2); and (2) gender-balance, i.e., the important but different goal of achieving a 50/50 gender ratio in both population and representation in art (Table 2). Critically, gender-balance and gender-neutrality are two different objectives, which require different metrics and different kinds of interventions. Gender-balance, which aims for gender parity (50/50) in the artist population and representation, is a primary goal towards which the art community should and does strive. Achieving it requires a strategy with a longer time horizon, including investment in art education, balanced or affirmative enrollment in art programs, and most importantly, career-nurturing programs that help reduce the dropout of women artists, the root of gender-imbalance in many professions23. In contrast, gender-neutrality represents a more immediate goal of offering equal opportunities to all currently exhibiting artists, independent of their gender. Hence, gender-neutral representation could be achieved over shorter timescales, by engaging with the current gatekeepers of access, like curators, collectors, and directors at museums and galleries, offering them the metrics and structural insights to empower them to offer balanced representation.Table 2 Definition of gender-neutrality and gender-balance hypothesesFull size tableUltimately, the two goals are intertwined: to achieve a sustainable 50/50 gender ratio in exhibitions and auctions (gender-balance), it’s not sufficient to focus only on the artist population, but we must correct the institutional forces that affect gender-neutrality and inhibit equal exhibition and auction access to women and men. While in recent years, enrollment in many art schools has achieved gender parity13,33, the representation of those graduates is not 50/50, indicating a lack of institutional gender-neutrality in the art world. Yet, a gender-neutral objective alone has its limitations: it freezes the current status quo, and while it may improve access to exhibition opportunities for the currently practising women artists, it is only a step towards, but not a guaranteed path to gender-balance. In the following, our goal is to independently explore gender equality in art in terms of both gender-balance and gender-neutrality, asking how well the institutions that support the art world do in terms of these two criteria.Institutional representation inequalityInequalities in the institutional representation of artists are revealed by the extent to which the gender ratio of an institution’s list of exhibitions deviates from the baseline composition of the artist population. As only 36.54% of contemporary exhibiting artists are women (Fig. 1a, Table 1), the active artist population is highly gender-imbalanced. A gender-neutral institution, which offers equal access to all exhibiting artists, must therefore devote 36.54% of their exhibitions to women artists. In contrast, a gender-balanced institution strives for a 50/50 representation, independent of the gender composition of the artist population. It is important to clarify that the definitions of gender-neutrality and gender-balance are based on statistical observations of outcomes, and are not meant to infer the factors used by institutions in their decision-making practice when selecting artists for exhibition.Both gender-balance and gender-neutrality can be mathematically conceptualized with a two-tailed binomial hypothesis test, measuring the deviation between the observed proportion pi of exhibitions by women artists in institution i and the null hypothesis of p0. In both cases, an institution is assumed to be independently drawing their exhibitions from a universal pool of artists with replacement. Gender-balance implies p0 = 0.5 (50% of the exhibitions are of women artists), while the gender-neutral goal corresponds to p0 = 0.365 (corresponding to the current fraction of 36.54% contemporary women artists). Using the recorded number of exhibitions ni as the sample size for each institution, the binomial test’s Bayes Factor (BF) is used to categorize each institution i into one of four possible categories (see “Methods” for details). We define a gender as overrepresented in an institution when the observed proportion of exhibitions featuring artists of that gender deviates from the reference proportion p0 in a given direction.Institutions can therefore be classified into four categories. Man-overrepresented institutions (pi p0) are those where it exceeds p0. Gender-neutral or Gender-balanced institutions are those for which the observed proportion of women artists is close to the reference proportion (pi ≈ p0). Finally, uncategorized institutions are those for which the Bayes factors do not provide sufficient evidence to support any of the above classifications. We find that under the gender-balanced criteria, 12,489 institutions are uncategorized while under gender-neutral criteria, 11,493 institutions are uncategorized. Unless noted otherwise, we ignore these uncategorized institutions in our analysis.Due to the heavy-tailed nature of the exhibition distribution16, the uncategorized institutions account for only 27.64% and 19.85% of all exhibitions under the gender-balanced and gender-neutral criteria respectively, and therefore represent a minor contribution to most artists’ careers.We find that 57.80% of the art institutions are gender-neutral, but only 24.46% of the art institutions are gender-balanced (Fig. 2a). Under the gender-neutral criteria, 23.58% institutions are man-overrepresented and 17.62% institutions are woman-overrepresented (Fig. 2a). In contrast, under the gender-balanced criteria, 72.19% institutions are man-overrepresented and only 3.35% institutions are woman-overrepresented (Fig. 2a).Fig. 2: Gender-balanced vs gender-neutral institutional patterns.Full size imagea Number of man-overrepresented, woman-overrepresented and gender-neutral or gender-balanced institutions under the gender-neutral and the gender-balanced criteria. b Portion of man-overrepresented, woman-overrepresented, gender-neutral institutions of different institution prestige under the gender-neutral criteria. c Portion of man-overrepresented, woman-overrepresented, gender-balanced institutions of different institution prestige under the gender-balanced criteria. d Number of men and women exhibitions at 100 prominent institutions. The example compares an institution’s status to the decision area under both gender-neutral and gender-balanced criteria. For example, MoMA (The Museum of Modern Art) is a woman-overrepresented institution under the gender-neutral criteria but man-overrepresented under the gender-balanced criteria.Under the gender-neutral criteria (Fig. 2a), we find that a higher fraction of museums than galleries are gender-neutral (69.10% of museums vs 51.60% of galleries). Moreover, galleries show a stronger positive bias towards men artists (1541 man-overrepresented vs 929 woman-overrepresented), while museums are split between man-overrepresented (401) and woman-overrepresented (463). Under the gender-balanced criteria, we again observe that a higher fraction of museums are gender-balanced compared to galleries (27.77% vs. 22.76%), with strong statistical support for this difference (Bayes Factor 9.95 × 109). In contrast, we observe that a higher fraction of galleries than museums embrace women artists when evaluated using the gender-neutral criteria (3.76% vs. 2.55%). This points to a small but noteworthy subset of galleries that actively support the careers of women artists. However, the corresponding Bayes Factor of approximately 2.82 provides only weak to moderate evidence for this difference.Figure 2d shows the number of men and women exhibitions for 100 prominent institutions worldwide, plotted alongside the decision boundary for the gender-neutral and gender-balanced criteria based on the binomial test. The figure indicates, for example, that MoMA (The Museum of Modern Art) is gender-neutral (BF: 0.108) but man-overrepresented (BF: 4.229 × 107) under the gender-balanced hypothesis; Whitney Museum of American Art is woman-overrepresented (BF: 4.518) under the gender-neutral criteria but man-overrepresented (BF: 10.697) under gender-balanced criteria; and the Albertina is woman-overrepresented under both gender-neutral (BF: 45828.026) and gender-balanced (BF: 4.400) criteria. A similar classification for countries/regions (Fig. 3b, c) reveals that most countries/regions are man-overrepresented under the gender-balanced criteria. However, under the gender-neutral criteria (Fig. 3a, c), we can identify woman-overrepresented exhibition patterns in Austria (BF: 1073), Sweden (BF: 3.660 × 1064), Finland (BF: 6.183 × 1062), etc., and we find that United States (BF: 4.344 × 10−12), Germany (BF: 2.353 × 10−17), United Kingdom (BF: 0.027) etc. are characterized by gender-neutral exhibition patterns.Fig. 3: Institutional gender representation inequality across the world.Full size imageWorld map showing for each country its gender bias in art under the a gender-neutral and under the b gender-balanced criteria. c Number of men and women exhibitions of countries/regions. The gray shading shows the gender-neutral decision area and green shading shows gender-balanced decision area. For example, the figure reveals that Finland is a woman-overrepresented country under the gender-neutral criteria but man-overrepresented under the gender-balanced criteria.We next consider how institutional representation inequality varies with the institutional prestige. In the visual arts, reputation and networks of influence play an important role in determining access to resources and rewards. Here, we adopt a network-based measure of institutional prestige derived from the artist co-exhibition network, a directed weighted network connecting two institutions if an artist exhibits in the source institution before exhibiting in the target institution16. Specifically, the eigenvector centrality of an institution in the co-exhibition network correlates strongly with both expert assessment of institutional prominence and the average auction price of the exhibited artists, empowering us to assign a prestige score to all art institutions in our dataset16. To reveal the relationship between institutional prestige and gender representation inequality, we group institutions into three prestige categories: low prestige (lower than 40th percentile of all institutions), mid prestige (between 40th and 70th percentile) and high prestige (higher than 70th percentile). We find that under the gender-neutral hypothesis, the fraction of man-overrepresented institutions increases with the institutional prestige, while the fraction of woman-overrepresented institutions decreases (Fig. 2b). The effect is particularly strong under the gender-balanced hypothesis: while only 67.06% of the low prestige institutions are man-overrepresented, 80.20% of the high prestige institution are in this category (Fig. 2c). We also find that the percentage of woman-overrepresented institutions is tiny (between 1.41%-5.30%) in all three prestige categories (Fig. 2c). These results are robust to the specific number and placement of the prestige bins, and are qualitatively replicated on the subset of institutions with expert prestige scores (Supplementary Information S2.5, Supplementary Information Fig. 6).To offer an overview of the distribution of institutional inequality across the whole institutional space, in Fig. 4a we show the art institution network16, whose nodes are the museums and galleries, connected to each other if they exhibit the same artists in a statistically significant number. The network is laid out using a force-directed algorithm, that tends to place close to each other directly linked institutions, and helps uncover densely connected communities. We colored each node to capture the respective institution’s gender inequality status: man-overrepresented (blue), woman-overrepresented (red), and gender-neutral (gray). The emergence of clusters dense in blue or red nodes is evidence of inequality-based assortativity, indicating that institutions with comparable gender representation tend to be connected to each other, often exhibiting the same artists.Fig. 4: Institutional co-exhibition network.Full size imagea Each node is an art institution colored by the institutional gender representation under the gender-neutral criteria: man-overrepresented, woman-overrepresented or gender-neutral. Two institutions are connected by a directed, weighted link reflecting the number of artist-exhibitions in which the artist first exhibited in the source institution before later exhibiting in the target institution. The figure illustrates the inequality-based assortativity, namely, institutions with the same inequality status are more likely to connect to each other and belong to the same cluster. To better illustrate this effect, in (b) we show the sub-graph of several European countries: Germany, Austria, France, Spain and Portugal, allowing us to see the regional differences: Germany and Austria have more woman-overrepresented institutions while France, Spain and Portugal have more man-overrepresented institutions. We can also observe the presence of clusters based on gender representation within a region: woman-overrepresented institutions in Austria are close to woman-overrepresented institutions in Germany, while the few man-overrepresented institutions in Austria are closer to the other countries. For more details about network assortativity see Supplementary Information S1.2.To better illustrate this, we also show a case study on the sub-graphs of several European countries: Germany, Austria, France, Spain, and Portugal (Fig. 4b). The sub-graph illustrates the regional differences: Germany and Austria have more woman-overrepresented institutions while France, Spain, and Portugal have more man-overrepresented institutions. It also captures clusters based on gender representation within a region: for example, we can see that woman-overrepresented institutions in Austria are close to woman-overrepresented institutions in Germany, while the few man-overrepresented institutions in Austria are more close to man-overrepresented institutions in other countries. To quantify the magnitude and significance of this gender-representation assortativity, we calculate the percentage of weighted outgoing links to institutions with each gender representation under the gender-neutral criteria. We find that man-overrepresented institutions connect to other man-overrepresented institutions more than expected by institutions randomly connected to each other, 43.7% compared to the random baseline of 23.58%. Similarly, woman-overrepresented institutions connect to other woman-overrepresented institutions more than would be expected by institutions randomly connected to each other, 25.1% compared to the random baseline of 17.62% (Supplementary Information S1.2). The institutional variability in assortativity can be captured using a measure of multi-scale assortativity34 (Supplementary Information S1.2), which further illustrates the extent to which gender-representation shapes the connectivity of the co-exhibition network.In summary, we find that institutions vary widely in terms of their ability to establish a gender-neutral or gender-balanced exhibition schedule. Among categorisable institutions, 57.80% are gender-neutral, embracing a list of exhibitions that mirror the gender mix of the artist population. The institutions fare worse in terms of gender-balance, the majority of institutions displaying man-overrepresented exhibition patterns. Overall, museums are more likely to be both gender-neutral and gender-balanced than galleries. At the same time, we do observe the emergence of a small number of galleries that strongly embrace and promote women artists. Finally, we find that the proportion of institutions characterized by gender-balance and gender-neutrality varies with the prestige of the institutions: the higher the prestige, the institutions’ exhibition patterns are more biased towards men artists.Co-exhibition gender of artistsWe next explore the relationship between institutional gender representation inequality and artists’ careers. For this, we assign to each artist with more than 10 exhibitions a co-exhibition gender under the gender-neutral criteria for institutional gender representation (see Supplementary Information S2.2 for a similar analysis under gender-balanced criteria), to reflect the dominant co-exhibition community an artist exhibited in: an artist is labeled as co-exhibition woman (co-exhibition man/co-exhibition neutral) if the artist’s exhibition history contains an over-representation of woman-overrepresented (man-overrepresented/gender-neutral) institutions compared to what we would have expected if the exhibition institutions were selected at random (formal definition in “Methods” section). For 47.8% of men artists and 49.3% of women artists, the co-exhibition gender agrees with the artist’s gender. At the same time, 19.6% of women artists are labeled as co-exhibition man, 19.7% of men artists are labeled as co-exhibition woman, and about 32% of all artists from both genders are labeled as co-exhibition neutral (Fig. 5a). If we consider the relationship with an artist’s individual prestige (calculated as the average prestige of the institutions that exhibited the artist), both man and woman artists with high prestige are less likely to be labeled as co-exhibition woman (Fig. 5b).Fig. 5: Co-exhibition gender of artists under gender-neutral criteria.Full size imagea Fraction of co-exhibition man, co-exhibition woman, and co-exhibition neutral for men and women artists based on the gender-neutral criteria. b Fraction of co-exhibition gender for men and women artists based on the gender-neutral criteria, with different career prestige. c Transition probability of early co-exhibition gender to late co-exhibition gender based on the gender-neutral criteria. d Transition probability of early co-exhibition gender to late co-exhibition gender based on the gender-neutral criteria for artists with different career prestige.How stable is an artist’s co-exhibition gender designation, i.e., does it change throughout an artist’s career? For this, we compare the co-exhibition gender of an artist’s first 5 exhibitions to the co-exhibition gender of the artist’s last 5 exhibitions. Figure 5c shows the fraction of artists labeled using their early-career exhibitions (rows) and then using their late-career exhibitions (columns), both under the gender-neutral criteria (same analysis on gender-balanced criteria see Supplementary Information S2.2). We find that the highest density lies on the matrix diagonal, indicating that for all three categories (co-exhibition man, co-exhibition woman, and co-exhibition neutral) there is co-exhibition gender lock-in. In other words, the early co-exhibition gender label of an artist tends to match the late co-exhibition gender label. Overall, co-exhibition man artists have the highest lock-in intensity: 73% of initially co-exhibition men artists remain co-exhibition men, while only 54% of co-exhibition women artists and co-exhibition neutral artists maintain their initial label. The strength of this co-exhibition gender lock-in decreases slightly with an artists’ career prestige (Fig. 5d), yet even at high prestige only 11% of co-exhibition man artists transition to co-exhibition woman. Furthermore, the percentage of co-exhibition men artists that change their co-exhibition gender is less than co-exhibition women and co-exhibition neutral artists, indicating that the path towards high prestige institutions goes through man-overrepresented institutions.Gender disparity in auctionsSales, both in the primary (galleries) and the secondary (auctions) market, and the resulting valuation of an artist’s work, is a frequently used measure of artistic success. While gender disparity has been extensively documented in auction sales10,12,15,35, it is unclear if its degree reflects the gender disparities already present in the artist population and exhibition patterns, or represents additional dimensions of gender disparity. To address this question, we linked the exhibition record of each artist to his/her auction records (if any), allowing us to explore the degree to which gender disparities in the artist population and exhibition patterns translate to gender disparity in auction sales.We find that 10,179 men and 3761 women artists recorded at least one sale at auction (Fig. 6a, Table 1), revealing an auction population disparity of 2.71, larger than the population disparity (1.74) and the exhibition disparity (1.86). The gender gap is even more pronounced when we inspect the number of auctioned items (records): of the 125,682 auction records, 81.74% are for art created by men artists (102,729), reflecting an auction record disparity of 4.48 (Fig. 6a, Table 1). Looking at total auction sales in volume (calculated using normalized price), we found an auction sales disparity of 7.54 (Fig. 6a, Table 1).Fig. 6: Gender bias in auction.Full size imagea Gender bias in terms of auction population, number of auction records and total auction sales. b Access rate vs career length, indicating that the longer an artist’s career, higher the chances that of the artist enters auctions. c Access rate vs exhibition count per year, indicating that the higher the exhibition count per year, higher the chances that of the artist enters auctions.Combining exhibition and auction records allows us to explore the rate at which artists transition from exhibitions to auctions. We find that 21.83% of the artists that exhibit have at least one recorded auction sale, an aggregate rate that favors men artists: 24.39% of the exhibiting men artists transitioned to auction, in contrast with only 15.65% women artists. The number of auctions and the auction prices are also biased towards men artists (Table 1): on average men artists have 1.66 times more auctions than women artists (10.1 vs 6.1) and command a 1.75 times higher auction price (0.7 vs 0.4, normalized). Taken together, these numbers indicate that the auction process does not simply mirror the observed gender imbalance in terms of the artist population or exhibition patterns, but exacerbates that, representing a new dimension of gender disparity.The relationship between institutional representation inequality and auctionsMost importantly, the auction market dynamics allow us to quantify the relative importance of the different variables that correlate with the access to the auction market, helping us to gain a deeper understanding of gender and co-exhibition representation inequality. To achieve this, we built a series of logistic regression models (see Table 3) to predict the probability P(ai = 1) that artist i transitions to the auction market, first given only his/her average exhibition count per year, and career length in Model 1. Then, in Model 2, we add the artists’ gender, and in Model 3 the co-exhibition gender under gender-neutral criteria. For the gender and institutional gender variables, we employed a dummy-encoding with a baseline value of man and man-overrepresented respectively. Finally, in Model 4, we combine the artists’ gender and co-exhibition gender through an interaction term. Details of the logistic regression model can be found in Methods.Table 3 Logistic Regression auction access prediction models {showing Number of Observations, Degree of Freedom, Bayesian Information Criterion (BIC)}, coefficients (coef.), odds ratios (O.R.), standard errors (S.E.), statistical significance (P val., nominal, two-sided) and 95% confidence interval (Conf. Int.)Full size tableRather than predictive ability, our focus here is on the relative importance of each variable measured by the odds ratio.The model suggests a notable association between the number of exhibitions per year and career length with the likelihood of entering the auction market, showing that artists that are frequently exhibited and/or have a long career are more likely to enter the auction market, which is in line with the trends seen in Fig. 6b, c.The addition of artist gender into Model 2 reduces the Bayesian Information Criterion (BIC), a general measure of goodness of fit based on the log-likelihood discounted by the number of parameters used in the model, indicating a model improvement. The artist’s gender has a modest effect on auction access and reveals the different roles gender plays: the woman’s odds ratio is 0.634 (coefficient: −0.456, p value: p