What's the difference between practical and theoretical measures of data utility?

Question

Accepted Answer

Practical utility measures performance on a specific use case (and must be tested experimentally); theoretical utility is a generic, math-based measure of information loss.

Practical Utility
Theoretical Utility

Nature
Use-case dependent
Mathematical, generic

Measures
How well anonymized data performs for its purpose
Information loss, independent of use

Examples
ML model accuracy, query-answer accuracy, business-decision quality
Normalized Certainty Penalty (NCP), Discernibility Metric, average equivalence-class size

Evaluation
Must be tested experimentally — no universal formula
Computed directly; enables comparing configurations without knowing the end use

You typically use theoretical metrics for a quick, use-case-agnostic baseline when comparing anonymization settings, and practical metrics (run the real downstream task) for the final verdict, because no formula predicts every application's needs.
Tip: Theoretical utility tells you how much information you lost; practical utility tells you whether you lost the information that mattered.
Go deeper:

Data anonymization (Wikipedia) — information-loss vs use-case utility framing.

	Practical Utility	Theoretical Utility
Nature	Use-case dependent	Mathematical, generic
Measures	How well anonymized data performs for its purpose	Information loss, independent of use
Examples	ML model accuracy, query-answer accuracy, business-decision quality	Normalized Certainty Penalty (NCP), Discernibility Metric, average equivalence-class size
Evaluation	Must be tested experimentally — no universal formula	Computed directly; enables comparing configurations without knowing the end use