It Is Surprisingly Difficult to Measure Income Segregation

Area of Science:

• Sociology
• Demography
• Urban Studies

Background:

• Existing U.S. Census and American Community Survey (ACS) income segregation estimates are known to have upward finite sampling bias.
• Two additional, larger biases—measurement error-induced attenuation and temporal pooling—complicate trend analysis.

Purpose of the Study:

• To identify and quantify multiple sources of bias in income segregation estimates.
• To develop a method for simultaneously correcting finite sampling, attenuation, and temporal pooling biases.
• To produce corrected estimates of U.S. income segregation from 1990 to 2019.

Main Methods:

• Formalized three sources of bias: finite sampling, measurement error-induced attenuation, and temporal pooling.
• Developed a novel bias-correction method using decennial census and ACS data (1990-2019).
• Applied the method to public data to generate bias-corrected income segregation estimates.

Main Results:

• Bias-corrected income segregation is approximately 50% higher than previously estimated.
• The increase in segregation between 2000 and 2005-2009 was substantially underestimated.
• Income segregation has shown a decline since the 2005-2009 period.

Conclusions:

• Accurate measurement of income segregation requires correcting for multiple biases.
• Reliability of self-reported income and year-to-year neighborhood income volatility are crucial for bias correction.