Which bias arises when data reflect existing societal inequities?

Which bias arises when data reflect existing societal inequities?

• A. Historical bias
• B. Sampling bias
• C. Representation bias
• D. Proxy label bias

Answer: (A)

Historical bias? reflect the social and institutional structures of the past, including discrimination and unequal opportunities. When data come from a time or system with entrenched inequities, the patterns learned by a model mirror those inequities, leading to biased outcomes that reproduce historical harms. For instance, loan approval data drawn from periods with redlining and biased lending practices can cause a lending model to favor certain groups over others, even when the algorithm is otherwise neutral. This differs from sampling bias, which stems from how data are collected and may skew who appears in the dataset; or representation bias, which concerns underrepresentation of certain groups in the data; or proxy label bias, which involves labels that encode sensitive information indirectly. Historical bias is about the imprint of past injustice on the data themselves.

Historical bias? reflect the social and institutional structures of the past, including discrimination and unequal opportunities. When data come from a time or system with entrenched inequities, the patterns learned by a model mirror those inequities, leading to biased outcomes that reproduce historical harms. For instance, loan approval data drawn from periods with redlining and biased lending practices can cause a lending model to favor certain groups over others, even when the algorithm is otherwise neutral.

This differs from sampling bias, which stems from how data are collected and may skew who appears in the dataset; or representation bias, which concerns underrepresentation of certain groups in the data; or proxy label bias, which involves labels that encode sensitive information indirectly. Historical bias is about the imprint of past injustice on the data themselves.