A "Roziah" by any other name: a simple Bayesian method for determining ethnicity from names

Area of Science:

• Computational epidemiology
• Population health
• Biostatistics

Background:

• Accurate ethnicity identification is crucial for epidemiologic research.
• Missing ethnicity data presents a significant challenge in population health studies.
• Current methods often require extensive name-ethnicity databases.

Purpose of the Study:

• To introduce a novel, data-efficient Bayesian strategy for ethnicity classification.
• To evaluate the performance of a substring-based Bayesian approach for name-ethnicity association.
• To provide an alternative method for ethnicity identification in epidemiologic analyses.

Main Methods:

• Utilized a naïve Bayesian strategy based on 3-letter name substrings.
• Employed training (n=10,104) and testing (n=9,992) datasets from Malaysian ethnic groups (Malay, Indian, Chinese).
• Assessed classification performance using Cohen's kappa, sensitivity, and specificity.

Main Results:

• High classification performance observed with minimal difference between training (κ=0.93) and test (κ=0.94) datasets.
• Excellent sensitivity and specificity achieved for Malay, Indian, and Chinese ethnic groups on test data.
• Demonstrated robust performance, suggesting the method's reliability.

Conclusions:

• The naïve Bayesian substring method is a promising tool for ethnicity classification.
• This approach performs comparably to more complex methods and is suitable for smaller datasets.
• Further research into substring lengths and diverse ethnic groups is recommended.