Introduction

The gender wage gap has received significant attention in the economic literature, and while it has narrowed, it persists (Kunze, 2018). In a survey of gender wage gap estimates for developed and transition countries, Kunze (2018) shows that the gender wage gap ranges from 5% to almost 30%, averaging 14%. For a large range of countries, the literature shows that differences in education and work experience explain an important part of the gender wage gap but less than half of it, suggesting discrimination against women (Kunze, 2018). Blau & Kahn (2017) show that in the U.S., the unexplained portion of the gender wage gap (associated with discrimination) increased from 71% in 1980 to 85% in 2010. In a study for the U.S., Goldin (2014) shows that as college education has increased, the explained portion of the gender wage gap has decreased, and the unexplained portion has increased. Recent evidence confirms that traditional economic factors fail to explain the gender wage gap (Cortes et al.,2023; Gharehgozli & Atal, 2020).

The literature has shown that while the gender wage gap has decreased in recent decades, greater equality is mainly observed among low- and middle-earning workers, with a significant gap remaining among high-wage workers and highly educated individuals (Binder et al., 2014; Quadlin et al., 2023; England et al., 2020; Yavorsky et al., 2019). While we can observe important progress in female labor force participation, educational attainment, and participation in occupations typically held by men, this has not resolved lower wages among more educated women (Petrongolo & Ronchi, 2020; Goldin, 2006). For example, there is evidence of heterogeneous gender wage gaps between STEM majors (science, technology, engineering, and math) and other educated individuals (Michelmore & Sassler, 2016).

There is also evidence that confirms that shocks impact men differently from women. For example, recent research has indicated that after the COVID-19 pandemic, female labor outcomes were damaged in the U.K., and the gender wage gap widened in Canada, which especially affected full-time female workers (Singh et al., 2022; Blundell et al., 2020). Alon et al. (2020) reported that the COVID-19 recession increased the gender wage gap by five percentage points. Evidence from the U.S., Germany, and Singapore shows that after the COVID-19 pandemic, unemployment and reduced working hours were more prevalent for women than men (Reichelt et al., 2021). Importantly, in the post-COVID-19 economy, research has shown that the gender wage gap is more significant in occupations with a high prevalence of working from home (Bonacini et al., 2024).

In this paper, we study the case of Chile. We estimate the gender wage gap for Chilean workers with higher education degrees by field of study. While Chile has experienced substantial increases in higher education and female enrollment, there are still substantial enrollment gaps in favor of men in areas such as engineering and technology and in favor of women in health and medicine (Dieder, 2021a; Bordon et al., 2020). Evidence indicates that expanding higher education has not impacted the gender wage gap between white-collar workers or improved the glass ceiling effect (i.e., a gender wage gap that increases among high-wage workers or highly educated individuals) (Dieder, 2021b). In addition, Chile has lower female labor participation rates than other Latin American countries despite being one of the most developed countries in the region, with mixed female incorporation into the labor market, gender segregation by industry, and lower-quality jobs for women (Cruz & Rau, 2022; Verick, 2014).

Recent evidence shows that in Chile, there is an estimated gender wage gap in the range of 20-28% that can be explained by market frictions, monopsony power, sorting within the firm, and professional education credentials, among other factors (Dieder, 2021a; Cruz & Rau, 2022; Parada-Contzen, 2022; Sanchez et al., 2022). Using data from 1990 to 2017, Kristjanpoller et al. (2023) show that the gender wage gap in Chile increased over this period, illustrating that education is an important variable in determining this gap.

For the estimation, we consider both college and technical college degrees. We compare two periods: before and after the COVID-19 pandemic. Chile faced the longest school lockdowns among OCDE countries (Hofflinger et al., 2024), and evidence indicates that economic recovery after COVID-19 impacted the labor market structure. While output levels recovered by 2022, employment levels were still 10% below their level in January 2020 (Egana-del-Sol et al., 2022). Note that the changes that COVID-19 introduced in the labor market are not heterogeneous across genders and fields. Scarce evidence for Latin America shows that women have higher average automation, although the risk is heterogeneous across fields (Egana-del-Sol et al., 2022). There is also evidence that young workers in Chile, particularly women, are less exposed to artificial intelligence than their counterparts in developed countries (Egana-del-Sol y Bravo-Ortega, 2025).

We use two cross-sections (2017 and 2022) of the Chilean National Socioeconomic Household Characterization Survey for our analysis. We estimate a Mincerian model with Oaxaca-Blinder decomposition (Oaxaca, 1973; Blinder, 1973) for ten fields of study: (1) Education; (2) Arts & Humanities; (3) Social Sciences & Communications; (4) Business & Law; (5) Natural Sciences, Math, & Statistics; (6) Information & Communication Technology; (7) Engineering, Industry, & Construction; (8) Agriculture, Forestry, Fishing, & Veterinary; (9) Health; and (10) Services. The Oaxaca-Blinder decomposition allows us to evaluate whether wage differentials among genders come from observed or unobserved characteristics to the econometrician. Our paper contributes to the literature on the glass ceiling with evidence from a developing economy, an effect that persists in developed countries (Sandner & Yukselen, 2024).

Our results indicate that the aggregated gender wage gap for individuals with higher education (i.e., across all fields of study) ranges between 17-25% before COVID-19 and 15-20% after COVID-19, with statistically significant changes. While we show that there was a reduction in the gender wage gap between 2017 and 2022, our results indicate that the portion of wages that observed characteristics cannot explain has increased among educated workers. Similar effects are found in other countries. For example, Blau & Kahn (2017) found an important decrease in the gender wage gap in the U.S. in 3 decades, but with increases in the unexplained portion. Piazzalunga & Di Tommaso (2019) find a similar increase in discrimination in Italy after the 2008 crisis. Xiao (2021) finds that gender quota policies in Finland may improve women’s representation at top jobs but increase wage discrimination.

The results across fields are heterogeneous. For example, we find statistically significant decreases in the gender wage gap in fields such as Education, Health, and Services after COVID-19. Education is an exceptional case where the gender wage gap only comes from observed individual and job characteristics in 2022. In contrast to the abovementioned cases, we detect substantial increases in the gender wage gap among STEM majors after COVID-19. Most of the contribution to the gender wage gap across STEM degrees comes from unobserved heterogeneity. Among STEM degrees, the notable exception is for graduates from Information and Communication Technology, with a significant decrease in the gender wage gap. In Information and Communication Technology, all differences in wages among genders come only from observed characteristics. Lastly, a distinctive result comes from graduates in Engineering. In Engineering, we find that observed individual and job characteristics are in favor of women by reducing the gender wage gap.

This paper’s results are relevant for designing policies that seek to reduce the gender wage gap and discrimination against women. Our findings suggest that these policies should be sector-specific, as graduates face field-specific labor market characteristics. For example, more attention should be given to sectors where workers hold STEM degrees than to sectors where workers hold education-related degrees.

The rest of the paper proceeds as follows. In the next section, we present the relevant background literature. In Section 3, we present the model and data, while in Section 4, we present the results. Finally, Section 5 concludes.

Literature review

A generally accepted result in the literature indicates that women’s educational progress has contributed to closing the gender wage gap (Blau & Kahn, 2017). However, these developments have not been homogenous for all workers. For example, there is a widening gender wage gap among college graduates (Golding et al., 2014). Based on 30 years of data (1980-2010), Rotman & Mandel (2023) find that returns to education and work experience in the U.S. are larger for men than for women, and this difference in returns has increased over time.

Based on evidence that the gender wage gap has not improved among educated workers, recent literature has focused on the particularities of the educational elite and STEM workers (Science, Technology, Engineering, and Mathematics). Evidence from the U.S. shows that the gender wage gap remains large for the educational elite and that acquiring advanced degrees does not help close the gap (Torche, 2018). Torche (2018) also shows that in the long run, educated women converge to lower earnings than men across generations. High-wage gender inequality also relates to gender gaps in college premiums, where returns to education for the top 10 percent of earners have grown faster for men than for women (Mandel & Rotman, 2021). The evidence indicates that the gap varies across fields of study, indicating that fields with the highest returns at the top of the income distribution are male dominated (Quadlin et al., 2023).

Cross-country analyses indicate that there is a critical glass-ceiling effect in European countries, with gender wage gaps that vary across different regions of Europe (Arulampalam et al., 2007; Cukrowska-Torzewska & Lovasz, 2020). Evidence from Australia indicates that men are overrepresented in the highest contributions/income quartiles for all cohorts and that there is a high-income disparity for younger female workers (Feng et al., 2019). Among educated workers, gender segregation across college majors has contributed to the gender wage gap (Blau & Kahn, 2017).

There are several studies on the gender wage gap among STEM majors in the U.S. Previous evidence shows positive returns in STEM careers, where women are typically underrepresented (Brenøe & Zölitz, 2020; Peri et al., 2015; Arcidiacono, 2004). Research focusing exclusively on the gender wage gap among STEM workers has shown that in fields with higher female participation, the gender wage gap is explained mainly through observed characteristics. In contrast, in fields with a lower concentration of women, the gender wage gap has unobserved components (Michelmore & Sassler, 2016).

While one hypothesis is that the gender wage gap in STEM fields may be due to abilities and preferences, the evidence indicates that the gender wage gap is explained by the underrepresentation of women in less math-intensive STEM majors and that graduates from those majors are more likely to take jobs in non-STEM occupations (Jiang, 2021). Evidence in the post-COVID-19 era indicates that there has been an increase in wages among STEM workers in Canada (Singh et al., 2022).

A trend in the literature suggests that educational mismatch might contribute to the gender wage gap. The literature started with a model that considers both individuals’ educational attainments and the requirements of their jobs (Duncan & Hoffman, 1981). Educational mismatch significantly affects wage inequality (Tang & Wang, 2021). Evidence from Spain shows that educational mismatch penalizes women’s return to a greater extent than men’s, where women get lower return to the years of overeducation (Salinas-Jiménez et al., 2013). Evidence from Belgium shows a similar direction of effect (i.e., wage penalties for over-educated women are larger than for over-educated men) (Jacobs et al., 2022).

For Chile, there is a wide range of estimates of the gender wage gap. Using the Equal Pay for Equal Work Lab implemented in 2009 with double fixed effects (at the individual and firm level), Cruz & Rau (2022) find a gender wage gap averaging 24.5%. When decomposing the sample by educational groups, they find that college graduates face a higher wage gap (32.5%), with firms contributing 30.9% to the gap. By analyzing firm power, Sanchez et al. (2022) find a gender wage gap of 22%, explained by differences in labor supply elasticities at the firm level. They suggest that gender wage gaps are related to structural factors that generate gender sorting at the firm. Gender sorting implies that men and women are allocated differently into different jobs or firms (Bamieh & Ziegler, 2024). García-Echelar et al. (2024) investigate gender wage gaps in the maritime industrial sector in Chile. Regarding glass-ceilings estimates, recent evidence for Chile indicates that the gender wage gap increases as experience increases (García-Echelar et al., 2024). They find mixed results regarding the gender wage gap among different educational levels, with the gap decreasing as education increases but peaking for individuals with post-secondary education.

Generally, we expect the gender wage gap to be higher in developing and developed countries. There is, however, high variability in estimates. Kunze (2018) reports gender wage gaps in developed and transition countries. The largest gap is in Japan (28.7%), followed by Israel, Switzerland, and Turkey (around 20-21%). Not only economic conditions impact the gender wage gap but also institutional factors, preferences, and social norms, among others. Evidence from Europe indicates that glass ceiling effects are more important in central and eastern European countries versus developed European countries, which suggests that the glass ceiling decreases with development (Ciminelli et al., 2021).

Perticará & Tejada (2022) study the sources of gender wage gaps in Latin America. They find the highest wage gap in Chile, with a raw difference of 28% between men and women. It follows Ecuador (19%), Paraguay (13%), Peru (12%), and Colombia (11%). These results suggest that more developed countries in Latin America do not seem to necessarily have better equality estimates, as in the case of Chile. In Perticará & Tejada (2022), Argentina and Bolivia show the highest level of equality (4% and 6% respectively). Gomes Mantovani et al. (2020) estimate the gender wage gap in Brazil for three types of workers (managers, technicians, and operational workers). They find the gender wage gap to be in similar ranges to Chile, with the largest for managers reaching 30%. Technicians and operational workers face gender wage gaps of 27% and 21%, respectively. The same authors find a gender wage gap of 25-27% for the states of Paraná and Bahia. They find that the gender wage gap is larger for managers and high-school technicians than for operational workers, suggesting substantial glass ceiling effects (Gomes Mantovani et al., 2021). Similarly, Alves et al. (2019) find that in Brazil, for women with high qualifications, there is gender discrimination, which increases with education.

Evidence regarding the impact of economic shocks on the gender wage gap shows that the 2007-2009 recession in the U.S. reduced the gender wage gap by 2 percentage points (Alon et al., 2020). In Germany, evidence indicates that after job displacement, the gender wage gap increases, and the effect is persistent five years after the event (Illing et al., 2021). Specific evidence for Chile indicates that during the contraction phases (2011-2015) of the mining industry, men’s wages were more affected than women’s wages were, resulting in a reduction in the gender wage gap (Chávez & Rodríguez-Puello, 2022). This is because of the male-dominant nature of the industry, meaning that a contraction of labor demand affects the demand for male labor and their wages, reducing the gender wage gap. In contrast, the opposite is true in expansion phases.

Economic shocks have different impacts on the gender wage gap across the wage distribution. Evidence from crisis periods indicates that, for example, in Italy, the gender wage gap increased by more than 5 pp after the 2008 crisis, driven by fiscal measures for public workers (Piazzalunga & Di Tommaso, 2019). However, this effect was not observed in richer European Union countries. In middle-income countries such as Georgia (sometimes classified as upper-middle income), evidence indicates that after the 2008 crisis, the shape of the gender wage gap changed, with heterogeneous effects across high- and low-income workers, with the gap contracting at the top of the distribution (Khitarishvili, 2016).

The COVID-19 shock had important implications for gender equality during the breakdown and recovery periods. Evidence indicates that, differently from other recessions, lockdown measures impacted female employment (Alon et al., 2020). The literature argues that these differences come from breakdowns and differences in family responsibilities across genders, where schools and daycare facilities were closed. In addition, grandparents stopped providing childcare due to mortality risks (Alon et al., 2020). Evidence from Latin America showed that the impact of COVID-19 on the gender wage gap was more significant in Chile than in other countries such as Brazil or the Dominican Republic (Viollaz et al., 2023).

In this context, Chile faced the most extended school closures among OECD countries, with 259 school days closed between 2020 and 2022 (Hofflinger et al., 2024). Lockdowns in Chile remained even after 60% of the population was fully vaccinated (BBC, 2021). Lockdown periods were differentiated by municipalities and geographical regions. The first lockdown was announced in March of 2020. Schools were forced to open only in March 2022, two years after the first lockdown. COVID-19 was one of the worst economic crises for the country, with unemployment rates that increased from about 7% to 13% (Economic Commission for Latin America and the Caribbean, 2020).

Methods and data

Regression model with homogenous effects across genders

We estimate a wage equation (in logs) as a function of socio-demographic and job characteristics. We estimate two types of models. The first model assumes that control variables have the same effect across genders, i.e., homogeneous effects in the wage equation. This model follows a linear Mincer regression of the following form, in equation (1):

$${\mathrm{ln}}\left({w}_{{ij}}\right)={\beta }_{0}+{\beta }_{1}{Femal}{e}_{{ij}}+{\ddot{\beta}}{X}_{{ij}}+{\varepsilon }_{{ij}}$$
(1)

where \({w}_{i}\) represents the hourly earnings of individual \(i\) in field \(j\), \({Female}\) is an indicator variable that takes a value of 1 if the worker is female and 0 otherwise, and \({X}_{i}\) is a vector of control variables that capture individual and job characteristics. We estimate the equation for \(j=1,\ldots ,\,10\) subsamples representing the fields of study. Unobserved heterogeneity \(\varepsilon\) is assumed to be normally distributed.

We include the following control variables: age, age squared, degree (college degree versus technical school degree), marital status (married, cohabitating, single), firm size (1-9, 10-49, 50-199, 200+ workers), employment status (contract or not), and geographical region (north, metropolitan region, south). Note that we do not include controls for occupation. Occupation classifications are highly correlated with the field of study, meaning we cannot control for occupation and study fields. For example, the traditional classification from the International Labor Organization defines occupation depending on the skill level and specialization required to perform the tasks of the occupations (ILO, 2024). This skill level includes formal education degrees. We do not control the economic sector or industry. This is because, given that we consider only educated individuals, we do not have enough data variation between the fields of study in majors and the work industry. Both aspects are left to future work.

Our vector of coefficients is \(\beta =\left({\beta }_{0},{\beta }_{1},\,\ddot{\beta }\right){\boldsymbol{.}}\) The coefficient of interest is \({\beta }_{1}\), which indicates the difference in wages for females with respect to males after controlling for individual and job characteristics. A statistically significant \({\beta }_{1}\) indicates that there is a gender wage gap. If \({\beta }_{1} < 0\), then women earn less than their gender counterparts after controlling for other characteristics (i.e., ceteris paribus). We statistically compare the coefficients across fields and waves to understand how the gender wage gap behaves.

Regression model with heterogeneous effects across genders

To determine whether the gender wage gap comes from observed or unobserved characteristics, we use the Oaxaca-Blinder decomposition method (Oaxaca, 1973; Blinder, 1973), where the estimation sample is divided into two groups (male and female, in this case). The wage gap is divided into an explained component based on observed characteristics and an unexplained residual component. Equation (2) describes the wage decomposition:

$${\log (w}_{m})-{\log (w}_{f})={\left({{\bar{X}}_{m}-\bar{X}}_{f}\right)\beta }_{f}+{\bar{X}}_{m}{(\beta }_{m}-{\beta }_{f})$$
(2)

The first term represents differences in characteristics (explained component), whereas the second term is the unexplained component due to differences in returns on observed characteristics. The second term is typically interpreted as proxy discrimination against women since individual or job characteristics do not explain it. Note that this term (i.e., the unexplained portion of the gender pay gap) does not exclusively capture discrimination but also the effect of other unobserved factors. For example, it may include unmeasured productivity, competitiveness, or risk aversion (Blau & Kahn, 2017). We use the same control variables as those used for the first model. The complete decomposition is presented in Appendix A.

The difference between the estimation strategies for Eqs. (1) and (2) is that for (1), we assume that the coefficients of the control variables are equal for men and women (Słoczyński, 2015). The Oaxaca-Blinder decomposition (Oaxaca, 1973; Blinder, 1973) is equivalent to a linear regression with a full set of interactions, meaning that it extends Model (1) (Jann, 2008). We use the model in Eq. (2) to decompose the contribution of observed and unobserved factors to the gender wage gap.

Data source and research sample

We use the 2017 and 2022 waves of the CASEN (acronym from its name in Spanish, Encuesta de Caracterización Socioeconómica Nacional). The CASEN is a national and regional representative cross-sectional survey with information at the individual and household levels. The Ministry of Social Development manages the survey, and it is publicly available (Ministerio de Desarrollo Social, 2023). The objectives of the survey are to design public policies to measure poverty, estimate income inequality among Chilean households, and evaluate the impact of social policies in Chile. The CASEN survey has been carried out periodically since 1990. The CASEN is implemented every 2-3 years, with 2022 being the last available survey.

In this paper, we excluded the 2020 cross-section because it was a reduced survey with a limited number of questions, which does not allow us to disaggregate the information by field of study. Thus, we consider the last two full complete surveys available. We interpret the waves as the before-COVID-19 scenario and the after-COVID-19 scenario.

Our study classification is based on the CASEN survey. We use the highest level of data disaggregation available, defining fields by the individual’s college or technical degree. The STEM fields are (5), (6), (7), and (8) of the following list:

  1. (1)

    Education

  2. (2)

    Arts & Humanities

  3. (3)

    Social Sciences & Communications

  4. (4)

    Business & Law

  5. (5)

    Natural Sciences, Math, & Statistics

  6. (6)

    Information & Communication Technology

  7. (7)

    Engineering, Industry, & Construction

  8. (8)

    Agriculture, Forestry, Fishing, & Veterinary

  9. (9)

    Health

  10. (10)

    Services

Because we separate the sample into ten subsamples by field of study, we do not include controls for occupation. The occupation classification is highly correlated with the field of study, which leads to multicollinearity. When restricting the sample of workers to those with higher education, we cannot add control variables for economic sectors due to insufficient data variation.

The research sample comprises individuals with a higher education degree (college or technical degree), who are working, and who have information on key characteristics. For the 2017 wave, we have 15,430 individuals and 22,691 for the 2022 round. Summary statistics for the research sample are presented in Table 1. Generally, the research samples are very similar across waves, where the average individual is a female worker who is 38-39 years old, single, and has a college degree.

Table 1 Summary statistics for research sample (2017 and 2022).
Full size table

Details regarding sample sizes by field of study are presented in Table 2. The larger sectors are Business and Law, Education, Engineering, Industry, and Construction. Their importance is relatively stable in 2022 with respect to 2017. We only observe statistically significant differences in industry size over time in Education (decrease of 3.5 pp) and Engineering, Industry, and Construction (increase of 3.4 pp). For the rest of the fields, there are no significant differences in their share of workers. Employment equilibrium for most fields of study remains constant in 2022 with respect to 2017. Note that for some fields of study, sample sizes are relatively small. For these sectors, we recommend being cautious with the interpretation of results.

Table 2 Characterization of fields of study for the research samples.
Full size table

In Table 3, we explore the gender composition of workers with higher education across fields. The fields with less female participation are Engineering, Industry and Construction, Information and Communication Technology, Agriculture, Forestry, Fishing, and Veterinary. Women constitute the majority in fields such as Health, Education, Arts & Humanities, Social Sciences & Communications, and Business & Law. We observed statistically significant increases in the share of women in the following fields: Education (an increase of 2.9 percentage points), Business and Law (3.8 pp), Engineering, Industry, and Construction (3.7 pp), Agriculture, Forestry, Fishing and Veterinary (5.6 pp), and Services (3.7 pp). There are no significant differences in the gender composition for the remaining fields.

Table 3 Gender composition across fields of study.
Full size table

The fields with a notoriously important share of women are Education and Health, whereas fields with a specific low presence of women are associated with Engineering, Information, and Agriculture. There were 26.7% and 29.7% of workers in STEM fields in 2017 and 2022, respectively, where the share of women over total workers in STEM fields was 20.66% and 22.64% in 2017 and 2022, respectively. The sample shares align with those in the literature (Cruz & Rau, 2022; OECD, 2017). This is an important characteristic to consider, as gender differences in presence in different fields might be related to gender wage gaps. Note that STEM majors are typically associated with higher pay.

Hourly wages are expressed in dollars for 2022, considering reported monthly wages and weekly hours worked. No treatment for outliers was required. For categorical variables, we construct the classification of variables considering the survey alternatives for answering each question. All variables are self-reported. The survey management constructed the field of studies, considering individuals’ responses regarding educational degrees.

Results

Regression results for the homogenous treatment

We start by presenting the results of the homogenous model so that we can compare them with other results in the literature. The estimated gender wage gap across fields is presented in Table 4. All the models are globally significant at the 99% confidence level. In most models (i.e., aggregating all fields and by fields), the gender coefficient is statistically significant. All coefficients (at the aggregated level and by field) are significantly different in 2022 from 2017. The changes occur in both directions (i.e., increasing and decreasing the gender wage gap), depending on the field of study. At the aggregated level, we find that the gender wage gap among individuals with higher education significantly decreased in 2022 with respect to 2017 by 2 pp.

Table 4 Estimated gender wage gap coefficient by field of study and wave.
Full size table

With respect to STEM fields, generally, we find increases in the gender wage gap after the COVID-19 pandemic for all STEM sectors except for Information & Communication Technology. The results indicate that fields that show a substantial decrease in the gender wage gap are Health (9.1 pp), Information & Communication Technology (8.3 pp), Education (4.7 pp), and Services (3.7 pp). Conversely, we find an extensive and statistically significant increase in the gender wage gap in Natural Sciences, Math, & Statistics (18.9 pp), and Social Sciences & Communications (18.8 pp). There are smaller but significant increases in the gender wage gap in Engineering, Industry, & Construction (3.9 pp), and Agriculture, Forestry, Fishing, & Veterinary (2.3 pp).

Complete estimation results are presented in Appendix B. Regarding other wage determinants, we find results that are consistent with literature. We find that on average, wages increase with firm size and firms in the main metropolitan region in Chile pay better off. On average, individuals with college degrees have substantially higher wages that individuals with technical college degrees. Self-employed individuals received on average lower wages than dependent workers. Note that returns on observables are lower in 2022 than in 2017. Because some sample sizes are small some results should be taken with caution. For example, Natural Sciences in 2017 have less than 200 observations. Some other fields are in the range of 200 to 400 observations.

Regression results for the heterogeneous treatment and Oaxaca-Blinder decomposition

We present the estimated gender wage gap when allowing for full interaction across controls and the decomposition of explained and unexplained factors contributing to the gap. When allowing full interaction across control variables, we find, on average, a larger estimate of the gender wage gap across fields. The main advantage of considering this decomposition is understanding how unexplained wage variations vary across different fields of study. A summary of the results for the gender wage gap and the Oaxaca-Blinder decomposition are presented in Table 5. Complete regression results are presented in Appendix B.

Table 5 Oaxaca-Blinder decomposition for the gender wage gap by field of study.
Full size table

Specifically, for all fields, the baseline gender wage gap in 2017 was 24.7%. After COVID-19, the gender wage gap decreased by 4.7 pp. Importantly, 72% of the gender wage gap in 2022 was due to unexplained factors, the share of which increased with respect to 2017. Both components are positive (favoring men over women) and statistically significant. The importance of the unexplained variation in wages among genders increased after the COVID-19 pandemic, suggesting a deterioration in the portion of wages that is explained by observed characteristics.

The estimates range when allowing heterogeneous treatments are close to those in the literature. Our estimates indicate that the average wage gap in 2017 was 24.7% and 20% in 2022. Didier (2021a) estimates a gender wage gap of approximately 28% for Chile using the 2013 and 2017 CASEN waves. He finds evidence of glass ceiling effects among professionals, where the quality of the school decreases the gender wage gap. In a different household survey from 2002 to 2009, Parada-Contzen (2022) found a significant gender wage gap of 19.6%; using monthly administrative records from 2010 to 2019, Sanchez et al. (2022) estimated a gender wage gap of 22%.

The pattern of change in 2022 with respect to 2017 varies across fields of study. For all other fields, the wage penalty for women ranges between 10.2% (Engineering) to 24.6% (Business and Law). After COVID-19, we found improvements in the gender wage gap for five fields: Education, Business and Law, Information and Communication, Health, and Services. A notorious decrease in the gender wage gap after the pandemic was observed in fields associated with health (27.7 pp). This result is consistent with the increased demand for healthcare workers during the pandemic. We also observe significant improvements in Business and Law (17.3 pp). This sector, however, presents a larger gender wage gap in 2022.

Notably, with improvements in the wage gap in Education and Information and Communication Technology, no gap is attributed to unobserved factors. For the remaining fields that show a reduction in the gender wage gap, the unexplained component is statistically significant and represents 95% (Health), 64% (Business and Law), and 42% (Services) of the gap. Note that two sectors with a notorious share of female workers are Education and Health. In both cases, we find that after COVID-19, women are better off because the gender wage gap has decreased. However, the field that shows better results is Education, where all the gaps are due to observed characteristics.

In contrast, although women represent more than 70% of the sample in Health majors, the gender wage gap is explained by unobserved characteristics. In the information sector, we also find that no gap is unexplained; we draw this conclusion with more caution, as it is a small field, and we are working with smaller sample sizes. Importantly, however, this sector may have faced growth during the COVID-19 pandemic, as it is related to the increase in the use of technology as mobility restrictions were imposed.

When focusing on STEM fields such as Engineering, Industry, and Construction, we find an increase in the gender wage gap in 2022 compared to 2017, when no significant gender wage gap was detected. Engineering, however, is the only sector where individual and job characteristics are observed to have an impact in favor of women by reducing the gender wage gap by 25% of the 10 log points. This result relates to the discussion in Cruz & Rau (2022) regarding the impact of better-paying fields and the gender wage gap.

Our findings are consistent with the literature, which argues that specific labor market segments have different market equilibriums (Webber, 2016; Sanchez et al., 2022; Sandner & Yukselen, 2024). Our estimates show that gender wage gap levels and their evolution are very different according to the field of study for educated individuals, suggesting that the specific characteristics of each labor market affect gender equality in earnings. There is a recent trend in the literature of studies on the gender wage gap and glass ceiling effects by field of study, with mixed evidence. Our results are consistent, for example, with Webber (2016), who shows that in the U.S., the healthcare sectors are less competitive for women. For Germany, there is evidence that in all fields of study of university graduates, except for medical doctors, there is a gender wage gap that starts at the beginning of their career, but in fields such as math, the gender wage gap has different behavior (Sandner & Yukselen, 2024)

Conclusion

This paper estimates and compares the gender wage gap for individuals with college or technical college degrees in Chile before and after the COVID-19 pandemic. We estimate the gender wage gap by field of study depending on the declared major. Our results contribute to the literature that explores the glass ceiling effect and returns on education among professionals. While the literature on the wage gap among educated workers is extensive, most evidence concentrates on developed countries. This paper provides evidence for a developing country where higher education has expanded. However, fewer individuals have college and technical school degrees than less educated workers.

Our results indicate that at the aggregated level (i.e., all majors), the gender wage gap decreased by 2-5 pp after the COVID-19 pandemic, ranging between 15-20% in 2022. This result suggests an improvement in the glass ceiling phenomenon documented in the literature. However, there is a 6 pp increase in the share of the gap that does not depend on individual and job characteristics. In 2022, 28% of the gender wage gap comes from observables, but the rest is unexplained. While we show a reduction in the gender wage gap between 2022 and 2017, the importance of the unexplained variation increased after the COVID-19 pandemic, suggesting worsening wage differentials that cannot be explained due to observed characteristics.

We find heterogeneous effects across fields of study. For STEM majors, there is an increase in the gender wage gap, ranging from 10-24% in favor of men. Among the STEM majors, the fields with the lowest gender wage gap are Engineering, Industry, and Construction. Notably, in Engineering, we find that observed individual and job characteristics impact the favor of women by reducing the gender wage gap. Engineering is the only field where this direction of effect is detected.

Two fields also presenting unique results were Education and Information and Communication Technology, where the gap decreased after COVID-19. Both sectors are important examples of policy design since they do not present unexplained contributions to the gender gap, meaning that all the variation comes from control variables. These fields could be studied more deeply to use them as examples for reducing discrimination against women in the labor market. While over 70% of workers in the education sector are women, similar shares can be observed in other sectors (e.g., health), but a significant unexplained wage gap persists.

Because specific labor markets may be heterogeneous across fields, future research should explore the possibility of model selection into employment. We expect that if labor possibilities and unemployment levels vary across fields, then the gender wage gap may also differ. Our paper only considers the field of study. Our data does not allow us to explore whether individuals are employed in their field of study or whether they are over or under-employed. Future work could explore this area to test whether a potential educational mismatch affects gender wage gaps as well as other sources of selection into employment.

In this paper, several considerations should be noted. First, we do not control for the specific occupation of the worker, which may affect wages. Since we are working only with a sample of educated individuals, we do not have enough data variation given the occupation classification in the dataset. The same happens with the industry where the individual works. Second, we do not have enough variation to consider differences across specific majors within a field. For example, in the health field, we do not observe whether the worker is a nurse, a medical doctor with a rare specialty, or a general physician. Third, some fields have small sample sizes, which may affect the results.

Additionally, because sample sizes are small, we cannot explore the gender wage gap through the wage distribution. Future research could explore the impact of specialization and occupation levels and the work industry to contribute to designs that seek to reduce glass ceilings. With larger sample sizes, consideration over the distribution should also be considered.