Over the past decades, women have achieved progress in labor force participation, breaking the glass ceiling, and narrowing the gender wage gap. And even though women constitute approximately 50 percent of the world population, they still represent only 21.3% of computer programmers, 26.9% of chief executives, 39.9% of financial analysts, 40.2% of physicians and surgeons. On the contrary, there are more than 80% of female physician assistants, nurses, elementary school teachers, and teacher assistants. In 2018, women CEOs of the Fortune 500 companies were outnumbered by men named James running such top companies.
There is an ongoing debate regarding the reasons behind the underrepresentation of women in different occupations, especially mathematically-intensive. Studies highlight that gender stereotyping in family and society influence educational and occupational choices. This, in fact, may lead to occupational segregation between men and women. However, the adverse effect of gender stereotyping can be broken through exposure to experts and professional role models of the same gender.
Besides, relevant role models may also influence the aspirations of women and their sense of competitiveness in the labor market of male-dominated occupations. And, therefore, facilitate gender equality in the labor market. Although economic research more and more draws attention to gender equality and stereotypes, the channels of this concept remain understudied.
The general idea
The study aims to address the topic of gender stereotypes from the perspective of media economics. News media provide role models and shape attitudes of people towards political and social issues, economic and everyday decisions. Anecdotal evidence suggests that despite the evolution of gender norms, women are still portrayed in news media in accordance with traditional stereotypes as home-makers and caregivers. Conversely, images of men appear more frequently in articles about success and business, authority and power. All of these factors may facilitate labor market discrimination through implicit bias and influence occupational choices of women.
The main research question is the following. Are women, contrary to men, are less likely to be portrayed as workers and professionals in articles published by the New York Times – the largest news media in the US?
Facial Recognition: measuring the nonverbal gender slant
With almost 90,000 articles published by the New York Times, I use a facial recognition algorithm to identify the gender of people depicted on the main images of each article. Figure presents the realization of this algorithm. It detected attributes of the famous French philosopher and feminist Simone de Beauvoir.
Let’s have a look at the shares of women in images by news desks. Figure indicates that women are underrepresented in news categories mostly related to professional topics and overrepresented in sections about fashion, entertainment, and home.
Are women less portrayed as professionals than men do? I use image attributes, dictionary-based method, and a simple econometric model to answer this question. Results indicate that the more professional article is, the smaller, on average, the share of women is portrayed on its main image.
These results indicate that women are underrepresented in news categories related to science and technology, economics and politics, business, and finance. If we are to accept the idea of news media influencing educational and occupational choices of women through role models, then role models based on gender stereotypes may discourage women from starting a career such as financial analysts and computer scientists.
Gendered language: text analysis
However, not only shares are important, the context in which women appear is also of a paramount importance. I address the question: what are the words with the strongest explanatory power for the nonverbal gender of each article? That is, which words are associated with gender of person on article images?
The main challenge of addressing this question is a high-dimensional word-level data with more than 50,000 unique words. I use the lasso logistic regression and find the following results.
Articles, where women are depicted in images, focus more on physical appearance, family, and emotions. On the contrary, articles having images of men, tend to discuss career perspectives, expertise, and professionalism.
Although the causal impact of gendered language on both occupational and educational choices is hard to identify, this result is in line with evidence on occupational segregation and labor market discrimination.
Finally, the study addresses the question of how gender stereotypes in news media evolved over time. I present the dynamics of women representation in different news categories.
Except for business and economics, all news categories tend to converge to the equal share of men and women. In addition, I estimate the effect of the empowering movement #MeToo. in October 2017. The main cause of the movement originates from gender stereotyping at a workplace, which makes the event particularly suitable for this study. The regression analysis reveals a slight improvement towards the more equal media world.
Methods used in this study advance the conventional methodological approach employed by researchers in media economics. Existing studies in media economics have primarily focused on textual analysis. However, the most popular media consumed today is nonverbal such as videos and images. Moreover, nonverbal information is considered to be more convincing and influential than verbal information. The methodological idea of this paper is to look at the relationship between nonverbal and textual data.
Since academic papers have commonly relied on hand collection and coding of nonverbal media content, researchers were restricted to small samples and a number of sources. I expand data limitations and use Machine Learning algorithm to automatically process a large amount of nonverbal data to identify the gender of people on 89,818 images. This method allows me to construct a 100-times larger dataset than in previous media studies. For the text analysis, I use a lasso logistic regression model. The model is suitable for dealing with high-dimensional data, digital texts, as it can include more than 1,000 predictors represented by words.
As the main results shed more light on the gender stereotypes in news media, I sincerely hope to advance the discussion of gender equality in media. If we are to design the world that is meant to work for everyone, we cannot afford to underrepresent women because, at best, we will be left with only half of the truth. Failing to include women in professional articles strengthens the unconscious bias, which, in turn, is responsible for much discrimination on the labor market (Bohnet, 2018).
Accordingly, there are to be many policy implications. By introducing regulations for more equal representation of women in media, there would be more relevant role models to be identified with. Seeing is believing, and for younger generations, such changes may influence both occupational and educational choices.
Автор: Мария Гребенщикова
 U.S. Bureau of Labor Statistics (2019). Women in the labor force: a databook : BLS Reports. Retrieved May 15: https://www.bls.gov/opub/reports/womens-databook/2019/home.htm.
 Kahn, S. and Ginther, D. (2017). Women and stem. Working Paper 23525, National Bureau of Economic Research.
 Goldin, C. (2014). A grand gender convergence: Its last chapter. American Economic Review, 104(4):1091–1119.