Earning Equations: Unraveling Gender Disparities in Income through Regression Analysis

 

Mathematik  |  Informatik

 

Nina Bugajska, 2006 | Zürich , ZH

 

This study investigates gender disparities in income by applying advanced statistical modeling to a large-scale dataset of over 1 million anonymized records from the Swiss Federal Statistical Office (BFS). The central research question explores how multiple linear regression and Generalized Additive Models (GAMs) can be used to elucidate the varying influences of demographic and professional factors on income, particularly between genders.

Introduction

The study aims to answer: How does multiple linear regression analysis elucidate the varying influences of demographic and professional factors on income between genders?

Methods

The analysis uses a dataset of over one million observations from the 2020 Swiss Wage Structure Survey (LSE), containing rich demographic and professional information. After cleaning and processing the data in R (version 4.4.3), the study initially applied a multiple linear regression model, then refined it through the use of interaction terms and GAMs to account for non-linear relationships and group-specific effects.
Variables included gender, age, education, years of service, professional position, employment level, professional requirements, and occupational sector. All continuous predictors were centered, and the dependent variable (income) was log-transformed to address right-skewness. To improve model validity, 2,119 influential observations were removed based on standardized residuals, leverage, and Cook’s Distance.

Results

The final GAM interaction model explained 88% of income variance and confirmed that gender significantly moderates the effects of all key predictors. Smooth terms for age, service years, and employment level revealed non-linear relationships that differ by gender:
Age: Men’s income increased steadily until age 50, while women’s plateaued earlier.
Years of Service: Income gains per year of tenure were stronger for men than women.
Employment Level: Full-time work benefited both genders, but men’s income rose more sharply.

Interaction terms for education, professional role, job complexity, and sector showed further discrepancies:
Education: Women faced steeper penalties for lower education and required higher qualifications to match men’s earnings.
Professional Position: Although penalties for lower roles were smaller for women, this likely reflects baseline disparities rather than relative advantage.
Job Complexity: Women benefited slightly more from complex roles, possibly due to higher performance expectations or standardized compensation.
Occupational Sector: Gender income gaps persisted within sectors, with women often earning less even in the same industry.

Overall, women earned approximately 15% less than men on average, even when controlling for all observed factors.

Discussion

The analysis confirms the alternative hypothesis: gender significantly influences how demographic and professional characteristics affect income. The model not only reveals persistent income gaps but demonstrates that the effects of age, education, experience, and sector are not uniform across genders. These disparities likely reflect both structural labor market biases and unobserved factors such as negotiation behavior or career interruptions.
Methodologically, the use of GAMs allowed for a multifaceted modeling of non-linear, gender-specific relationships. While the model performs well, limitations include the exclusion of unobservable variables (e.g., parental leave, full employment history), cross-sectional data constraints, and reliance on automated exclusion of outliers due to scale.

Conclusions

This study offers a data-driven analysis of gendered income disparities in Switzerland. It highlights the need for policy interventions targeting gender inequality not only through wage transparency but also through structural changes in career progression, education valuation, and occupational access. Future research should incorporate time series data, additional social variables, and potentially machine learning techniques to enhance predictive accuracy and policy relevance.

 

 

Würdigung durch den Experten

Dr. Marcel Dettling

In der Arbeit von Nina Bugajska wird mit dem Thema Lohngleichheit zwischen den Geschlechtern ein gesellschaftlicher Brennpunkt aufgegriffen. Sie analysiert die schweizweiten Daten der Lohnstrukturerhebung des Bundesamtes für Statistik aus dem Jahr 2020. Durch die gekonnte Anwendung von in Eigenregie erarbeiteten, fortgeschrittenen Methoden der statistischen Regression folgert Frau Bugajska korrekt, dass nach wie vor erhebliche Unterschiede in der Entlöhnung zwischen den Geschlechtern bestehen. Darüber hinaus liefert eine vertiefte Analyse der Resultate vertiefte Einblicke über deren Ursprung.

Prädikat:

sehr gut

 

 

 

Realgymnasium Rämibühl, Zürich
Lehrer: Valentin Künzle