Oaxaca-Blinder Decomposition

The Oaxaca-Blinder decomposition splits the difference in mean outcomes between two groups into a part explained by different characteristics and an unexplained part due to different regression coefficients.

Oaxaca-Blinder 분해는 두 집단의 평균 차이를 특성(설명변수) 차이로 설명되는 부분과 회귀계수 차이에서 오는 설명되지 않는 부분으로 나눕니다.

How much of the wage gap between two groups is explained by differences in education and experience?

두 집단 사이 임금 격차 가운데 학력·경력 차이로 설명되는 몫은 얼마인가요?

Separate linear regressions are fitted in each group, and the mean gap is written as differences in average predictors times reference coefficients plus differences in coefficients weighted by group means. The method reports the explained and unexplained totals with standard errors and each variable's contribution.

When to use Oaxaca-Blinder Decomposition

• You need to quantify how much of a group gap in a continuous outcome is due to observable characteristics.
• Studying wage, health or test-score gaps between two groups.
• You want per-variable contributions to the explained part.

When to use something else

• The outcome is binary or a count; use nonlinear decompositions such as Fairlie.
• You want the gap across the whole distribution rather than at the mean; use RIF or quantile decompositions.

Assumptions and requirements

• A linear model is adequate within each group.
• The same predictors are measured in both groups with overlapping values.
• The unexplained part includes all unmeasured factors and is not proof of discrimination.

What the analysis reports

• Mean gap with explained and unexplained parts and delta-method or bootstrap SEs
• Threefold endowments, coefficients and interaction parts
• Per-variable contributions
• Percentage of the gap explained

How to report it

The mean difference in [outcome] between [group A] and [group B] was [value]; differences in characteristics explained [value] (SE = [value], [value]% of the gap) and the unexplained part was [value] (SE = [value], p = [value]).

Related methods

• Multiple Linear Regression
• Difference-in-Differences
• Propensity Score Matching

Run Oaxaca-Blinder Decomposition on Quorant