liveBook · Manning

Expected value of sample information

As explained in Timeless Algorithms, The Foundational Ideas

The expected value of sample information is the expected improvement in decision outcome from imperfect information before its acquisition cost is deducted.

A test can change beliefs without changing what action is best. Expected value of sample information (EVSI) measures the expected improvement in utility from imperfect evidence before its acquisition cost is deducted. It accounts for possible results, their probabilities, the resulting posterior beliefs, and the actions chosen in response. section 7.1.3

EVPI ≥EVSI≥0

EVSI is nonnegative before costs and cannot exceed the value of perfect information. section 7.1.3

Net value of sample information=EVSI-Cost of information

Subtract the information-acquisition cost to determine whether gathering evidence has positive net value. section 7.1.3

A second diagnostic test

The positive and negative results produce posterior disease probabilities of 0.484 and 0.0055, both below the 0.5238 treatment threshold. Since neither result changes the preferred action, EVSI is zero; a five-unit test cost makes its net value −5. section 7.1.4

How Timeless Algorithms, The Foundational Ideas covers it

How deep it goes
Calculates information value from possible results, posterior beliefs, and resulting actions, then separates gross value from acquisition cost.
Examples
Uses a second diagnostic test whose results shift disease probabilities but leave the preferred treatment action unchanged.
What it assumes
Assumes probabilities, posterior beliefs, and utilities can be specified for possible evidence and actions.
What you can do after
Evaluate whether imperfect information improves expected utility enough to justify its acquisition cost.
How it connects
Relates EVSI to expected utility, expected value of perfect information, Bayesian updating, and decision trees.

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.