Measuring Discrimination

1
Measuring Discrimination

• Statistical Analysis of Observational Data
  Oaxaca-Blinder Decompositions

2
Outline

• Experiments, audits and discriminatory behavior
  in the neoclassical model. (What would Gary
  Becker say?)
• Oaxaca-Blinder statistical decomposition
• Problems - omitted variable bias, sample
  selection bias
• Proof of Discrimination and the Law

3
Statistical Analysis

• Race-specific intercepts
• WabHRwhere H is productivity and R is equal 0
  if white and equal to 1 if nonwhite.
• Graph
• Example
• Race-specific wage equations
• WwawbwHWnwanwbnwH

4
Race Specific Wage Equations
White
Ww
Nonwhite
Wbnd
Wb
Hw
Hnw
5
(No Transcript)
6
Omitted Variable Bias

• If we omit a variable that affects wage and is
  correlated with Race, we can either underestimate
  or overestimate racial disparity.
• In general, productivity can not be directly
  observed either by researcher or by employing
  firm.
• AFQT as proxy for productivity?

7
Source Neal and Johnson, 1996
8
Sample Selection Bias
9
Recent Discrimination Cases

• http//www.eeonews.com/news/race/index.html
• http//www.usdoj.gov/crt/emp/papers.html
• http//www.eeoc.gov/litigation/settlements/index.h
  tml