Inferential Analysis and Predictive Modeling of Wages in Brazil: Empirical Evidence from RAIS Administrative Records (2010–2023)
Using Brazilian RAIS administrative data from 2010 to 2023, this study combines inferential statistics and predictive modeling to reveal persistent structural wage inequalities, demonstrating that gender and racial disparities widen significantly at higher wage percentiles and are driven largely by unexplained differentials, while ensemble methods outperform traditional regression in predicting earnings.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
Understanding how people earn their living is one of the oldest questions in economics, but in a country as vast and varied as Brazil, the answer is rarely simple. For decades, researchers have tried to untangle why some workers earn significantly more than others, even when they seem to have the same skills or work in the same places. The central puzzle involves three main forces: the value of education, the influence of where and what someone does for a living, and the persistent gaps that remain based on gender and race. While it is widely known that women and Black workers often earn less, the critical question is whether this gap exists because they have less education or work in lower-paying jobs, or if the gap persists even when those factors are accounted for. This distinction matters deeply for public policy, because if the gap is caused by differences in qualifications, the solution is better schools; if the gap remains after accounting for qualifications, it suggests something deeper is at play, such as discrimination or barriers to advancement.
A team of researchers from the University of São Paulo has tackled this question using a massive, detailed record of the Brazilian formal labor market. They analyzed data from over 600,000 employment relationships spanning from 2010 to 2023, drawn from a government registry that functions almost like a census of formal workers. By combining traditional statistical methods with modern computer algorithms, the researchers mapped out how wages are determined and whether the rules of the game have changed over time. Their work reveals that the Brazilian labor market has undergone a significant structural shift in the last decade, but it also confirms that deep-seated inequalities remain stubbornly fixed. The study finds that while education and experience are powerful drivers of income, they do not tell the whole story. Instead, the data shows that the system systematically undervalues women and Black workers, a pattern that becomes even more severe the higher up the wage ladder one looks.
The researchers began by examining the raw numbers to see if the labor market of the 2020s looked different from the 2010s. They found a clear break between the two periods, suggesting that the economic landscape has been reshaped by recent crises, reforms, and changes in how data is collected. However, despite these shifts, the fundamental inequalities did not disappear. When the researchers controlled for where a person lives, what industry they work in, and how old they are, a significant wage gap remained for women. In fact, the data showed that women in Brazil often possess higher levels of education on average than men, yet they still earn less. This paradox led the team to dig deeper, using a method that separates the wage gap into two parts: the portion explained by observable differences, like schooling, and the portion that cannot be explained by these factors.
The results of this separation were striking. The researchers calculated that if women had the exact same observable characteristics as men, their wages would actually be predicted to be higher, not lower. Yet, in reality, men still earn more. The entire gap, and then some, is driven by the "unexplained" portion. This unexplained difference, which accounts for more than 15 percent of the gap, suggests that women are receiving lower returns for their skills and experience compared to men. The study also found that the model used to predict wages consistently underestimated the pay of women and Black workers, further indicating that the system is not just missing data, but is structurally biased against them.
Perhaps the most revealing part of the study came from looking at the wage distribution not as a single average, but as a spectrum from the lowest earners to the highest. The researchers discovered that inequality is not uniform; it intensifies as one moves up the ladder. This phenomenon, often called a "glass ceiling," means that the penalties for being a woman or a Black worker are much smaller at the bottom of the income scale but grow dramatically at the top. For instance, the wage gap between men and women is about 12 percent for those in the lowest tenth of earners, but it more than triples to nearly 47 percent for those in the top tenth. Similarly, the penalty for being Black grows from about 4 percent at the bottom to over 13 percent at the top. This pattern suggests that while entry-level jobs might be relatively accessible, the barriers to reaching the highest-paying leadership and specialized roles are formidable and disproportionately affect women and Black workers.
The study also examined who actually occupies these different levels of the wage spectrum. The data revealed a stark demographic shift: while Black and mixed-race workers make up the majority of the lowest-paid segment, their representation shrinks drastically as wages rise, while white workers become increasingly dominant at the top. This indicates that the "glass ceiling" is not just about how much people are paid for the same work, but also about who gets access to the high-paying jobs in the first place. The combination of these two forces—unequal access to top positions and unequal pay for those who do reach them—creates a compounding effect that keeps inequality high.
To ensure their findings were robust, the researchers tested their conclusions using advanced computer models known as machine learning, which are designed to find complex patterns in large datasets that traditional statistics might miss. These models, which learned from the data without being told specific rules, confirmed that education and the economic sector are the strongest predictors of wages. However, the machine learning models also showed that the relationship between these factors and pay is not simple or linear; it involves complex interactions that vary across different groups. While these computer models were better at predicting exact wage numbers than the traditional equations, the researchers emphasized that the simpler statistical models were essential for understanding the underlying causes of the inequality.
The authors conclude that the Brazilian labor market is characterized by a dual reality. On one hand, there has been progress and structural change over the last decade. On the other, the mechanisms of inequality have evolved rather than vanished. The persistence of a large unexplained wage gap for women, despite their educational advantages, points to structural discrimination that is not captured by simple measures of job title or location. The deepening of racial and gender penalties at the highest income levels suggests that the path to the top is blocked by barriers that go beyond individual merit. The study does not claim to have solved the problem or identified every single cause, but it provides a clear, data-driven map of where the inequalities lie. It shows that fixing wage gaps will require more than just expanding education; it will require addressing the specific barriers that prevent women and Black workers from accessing and thriving in the highest tiers of the formal economy.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.