DataTerminology.comNEW
Also known as: Surrogate variable
A substitute variable used to approximate an unobservable or hard-to-measure quantity.
DESCRIPTION
A proxy variable serves as a stand-in for a latent or difficult-to-measure construct. Proxies enable practical analysis when direct measurements are unavailable, but can introduce bias if poorly correlated with the target. Careful validation, domain knowledge, and sensitivity analysis help assess proxy quality. In AI ethics, proxies can reflect sensitive attributes unintentionally.
ALTERNATIVE DEFINITIONS
Wikipedia
In statistics, a proxy or proxy variable is a variable that is not in itself directly relevant, but serves in place of an unobservable or immeasurable variable.
Emphasizes substituting for an unobservable variable.
OECD Glossary of Statistical Terms
Proxy variables are used as substitutes for variables that are either unobservable or difficult to measure.
Highlights practical use when measurement is infeasible.
NIST
A surrogate measurement is used to stand in for a direct measurement when the latter is impractical or impossible.
Uses surrogate measurement concept aligned with proxies.
RELATED TERMS
Latent Variables Confounding Instrumental Variable Measurement Error Feature Engineering Systematic Error Construct Validity Label Proxy Outcome Proxy
COMPETING TERMS
Direct Measurement Oracle
RELATIONS
- Narrower: Envoy Proxy Reverse Proxy
KNOWLEDGE AREAS
Data Science AI Data Management
TARGET AUDIENCE
Data Scientist AI Engineer Data Analyst Data Steward
EXTERNAL SOURCES
- Wikipedia - Proxy (statistics)
- OECD - Statistical terminology glossary
- NIST - Statistical methods handbook
- Fairness and Machine Learning - Proxy variables and bias discussion
TOP QUESTIONS
When should a Proxy variable be used?
Use proxies when direct measurement is infeasible or expensive, ensuring strong correlation and validity with the target construct.
Category: Application
What risks do Proxy variables introduce?
They can encode bias, reduce validity, and mislead decisions if poorly correlated or confounded.
Category: Risk
How do you validate a Proxy variable?
Assess convergent validity, correlation with ground truth, temporal stability, and robustness via sensitivity analyses.
Category: Requirements
What are common mistakes with Proxy variables?
Assuming causality from correlation, ignoring confounders, and failing to reassess proxies over time.
Category: Mistakes
How do proxies relate to AI fairness?
Proxies may unintentionally encode sensitive attributes, necessitating fairness testing, de-biasing, and careful feature selection.
Category: Opinion
What is an example of a Proxy variable?
Using zip code as a proxy for socioeconomic status can be practical but risky due to potential bias.
Category: Example
What are alternatives to Proxy variables?
Collect direct measurements, use instrumental variables for causal inference, or model latent variables explicitly.
Category: Alternatives
How to document Proxy variables?
Include rationale, expected correlation, limitations, and monitoring plans in data documentation and model cards.
Category: Management
What hot topics involve Proxy variables?
Detecting proxies for protected attributes, causal feature selection, and privacy-preserving measurement innovations.
Category: Topic