liveBook · Manning

Value of information

As explained in Timeless Algorithms, The Foundational Ideas

The expected improvement in a decision outcome attributable to acquiring additional information before acting.

Before acting under uncertainty, a decision maker may order a test, conduct a survey, or run an experiment. The value of information compares the expected utility of learning first with the expected utility of acting immediately. section 7.1.3

Expected value of information is the difference between expected utility with information and expected utility without it, before acquisition costs. Information has decision value only if some possible result changes the preferred action and improves expected utility. Subtract the cost to obtain its net value. section 7.1.3

EVPI ≥EVSI≥0

The expected value of perfect information (EVPI) bounds the expected value of sample information (EVSI), which cannot be negative before costs. section 7.6

Net value of sample information=EVSI-Cost of information

For sample information, subtract its acquisition cost from EVSI to find net value. section 7.1.3

Considering another medical test

A patient’s decision maker can compare acting on current beliefs with obtaining an additional, imperfect test. The test’s value depends on whether its possible results change the treatment choice enough to justify its cost. section 10.2.6

The calculation depends on the probabilities, consequences, and values assigned to outcomes, as well as the available actions and the information’s cost and accuracy. Better reasoning can improve decision quality, but cannot remove uncertainty or guarantee a favorable result. section 10.2.6 section 7.6

How Timeless Algorithms, The Foundational Ideas covers it

How deep it goes
Defines expected value, distinguishes preposterior from terminal analysis, and relates EVSI and EVPI without detailing a numerical calculation.
Examples
Uses tests, surveys, and experiments as information choices, including an additional medical test before treatment.
What it assumes
Combines probabilities for uncertain states with utilities or losses for possible outcomes.
What you can do after
Evaluate whether additional evidence is worthwhile by comparing expected utility gains with acquisition costs.
How it connects
Connects Bayesian inference, expected utility, decision trees, and decision analysis.

Related concepts

This explainer was generated by Manning's knowledge graph from the text of Timeless Algorithms, The Foundational Ideas. The section links point to the parts of the book it draws on.