Multiple regression and Oaxaca-Blinder decomposition for pay gap analysis
A statistical method combining multiple regression to control for legitimate pay determinants and the Oaxaca-Blinder decomposition to quantify the portion of a pay gap attributable to differences in characteristics versus discrimination or unexplained factors.
This skill provides empirical, legally-defensible evidence to diagnose systemic pay inequity, enabling targeted remediation and mitigating regulatory and reputational risk. It directly impacts talent retention, brand equity, and compliance by transforming subjective perception into quantifiable, actionable data.
1Careers
1Categories
8.7 Avg Demand
20%
Avg AI Risk
How to Learn Multiple regression and Oaxaca-Blinder decomposition for pay gap analysis
1. **Foundational Statistics:** Solidify understanding of linear regression, coefficient interpretation, and dummy variables. 2. **Core Economic Theory:** Study the human capital model (education, experience as legitimate pay drivers) and the concept of the 'unexplained' residual. 3. **Basic Software:** Learn to run a simple linear regression in a tool like Stata, R, or Python (statsmodels).
1. **Model Specification:** Master selecting and operationalizing control variables (e.g., job level, tenure, performance rating) to avoid omitted variable bias. 2. **Decomposition Implementation:** Execute the standard Oaxaca-Blinder decomposition in code, correctly interpreting the 'endowments' vs. 'coefficients' components. 3. **Common Pitfalls:** Recognize and address issues like multicollinearity, heteroscedasticity, and the limitation of using job titles as a coarse proxy for work content.
1. **Strategic Analysis Design:** Structure analyses for intersectionality (e.g., gender-by-race), multiple job families, or global subsidiary comparisons while maintaining statistical power. 2. **Methodological Extensions:** Apply and interpret advanced techniques like RIF (Recentered Influence Function) regressions for distributional analysis or the Cotton-Neumark decomposition to account for potential discrimination in the male-dominated wage structure. 3. **Executive Communication:** Translate complex decomposition results into a concise, compelling narrative for CHROs and Legal, focusing on actionable drivers and remediation pathways.
Practice Projects
Beginner
Project
Internal Salary Data Audit & Simple Regression
Scenario
You have a CSV file of 500 employees from a single department with columns: Annual Salary, Years of Experience, Highest Education Level (coded), Gender (0/1). The goal is to see if a raw gender pay gap exists and if it changes after controlling for experience and education.
How to Execute
1. **Data Prep:** Clean the data, handle missing values, and code categorical variables (e.g., education into dummy variables). 2. **Run Two Models:** First, regress Salary on Gender alone. Second, regress Salary on Gender, Experience, and Education dummies. 3. **Interpret Results:** Compare the Gender coefficient (and its significance) between the two models. The reduction in the coefficient magnitude after adding controls suggests part of the raw gap is explained by human capital differences. 4. **Document:** Create a short report stating the raw gap, the controlled gap, and the limitations (e.g., lack of job-level controls).
Intermediate
Case Study/Exercise
Full Oaxaca-Blinder Decomposition of a Gender Pay Gap
Scenario
A mid-sized tech firm's People Analytics team has data for 2000 software engineers. Variables include: Salary, Gender, Years of Experience, Performance Rating (last 2 years), Job Level (L1-L4), and Tenure at Company. The VP of HR wants a breakdown of the 7% average salary gap between male and female engineers.
How to Execute
1. **Specify the Model:** Define the pay equation: ln(Salary) = β0 + β1*Experience + β2*Performance + β3*Tenure + β4-6*JobLevelDummies + ε. 2. **Run Separate Regressions:** Estimate this model separately for the male and female groups. 3. **Execute Decomposition:** Use the Oaxaca-Blinder formula: Gap = [X_male - X_female] * β_female (explained/differences in characteristics) + X_male * [β_male - β_female] (unexplained/potential discrimination). 4. **Analyze & Report:** Quantify how much of the 7% gap is due to women, on average, having lower job levels, less experience, etc., versus the portion that remains 'unexplained' given identical observed characteristics. Highlight the largest 'unexplained' coefficient drivers (e.g., the return on tenure).
Advanced
Project
Multi-Country, Intersectional Pay Equity Analysis with Strategic Recommendations
Scenario
A multinational corporation faces pressure from investors and regulators on DEI metrics. The task is to conduct a comprehensive pay equity analysis across its 3 largest markets (US, UK, Germany) for gender and race/ethnicity intersections, and to propose a data-driven remediation plan with a 3-year budget projection.
How to Execute
1. **Complex Data Integration:** Harmonize disparate HRIS and payroll data across countries, defining consistent occupational codes and compensation components. 2. **Intersectional Decomposition:** For each country, run decompositions for Gender, Race (where legally permissible to collect), and Gender*Race groups. Use RIF regressions to analyze gaps not just in means but across the salary distribution (e.g., at the 25th, 50th, 90th percentiles). 3. **Root Cause Hypothesis:** Integrate decomposition results with qualitative data (e.g., promotion rates by group) to formulate hypotheses for large unexplained gaps (e.g., biased performance calibration). 4. **Develop Remediation Model:** Create a predictive model to estimate the cost of closing the unexplained gap via targeted salary adjustments. Build a phased 3-year plan prioritizing roles with the largest unexplained gaps and highest business impact. 5. **Board Presentation:** Prepare a slide deck that moves from technical findings to strategic business risks and a costed action plan.
Tools & Frameworks
Statistical Software & Packages
Stata (`oaxaca` command)R (`oaxaca` package or `blinder_oaxaca` from `fixest`)Python (`statsmodels` for regression, then custom Oaxaca code or the `linearmodels` package)
Stata's `oaxaca` is the industry standard for its ease and robustness. R and Python offer more flexibility for advanced modeling (like RIF regressions) and integration into automated data pipelines. The choice depends on team expertise and the need for reproducibility at scale.
Mental Models & Methodological Frameworks
The Human Capital ModelThe Oaxaca-Blinder Decomposition FormulaThe Three-Term (Cotton-Neumark) Decomposition
The Human Capital Model (Becker) provides the theoretical basis for 'legitimate' pay determinants. The OB decomposition is the core execution framework. The Cotton-Neumark extension is a critical mental model for discussions on whether the male or female wage structure is the 'non-discriminatory' reference, impacting how the unexplained gap is sized.
Careers That Require Multiple regression and Oaxaca-Blinder decomposition for pay gap analysis