liveBook · Manning

Net value of sample information

As explained in Timeless Algorithms, The Foundational Ideas

The net value of sample information is the expected value of sample information after subtracting the cost of acquiring the information.

Net value of sample information compares the expected utility of acquiring imperfect information and then acting with the expected utility of acting immediately, after accounting for acquisition cost. The expected value of sample information (EVSI) measures the improvement before cost; net value is EVSI minus cost. section 7.1.3 section 7.1.4

Net value of sample information=EVSI-Cost of information

Subtract the information’s acquisition cost from its expected decision benefit to obtain its net value. section 7.1.3

Information has positive decision value only when its possible results can improve the action chosen:

  • If every result leaves the preferred action unchanged, EVSI is zero, even if the information changes estimated probabilities. section 7.1.3 section 7.1.4
  • When acquisition cost exceeds EVSI, net value is negative and acting without the information yields greater expected utility. section 7.1.3

A second diagnostic test

A positive result gives a disease probability of 0.484, below the 0.5238 treatment threshold; a negative result gives 0.0055. Both results leave “do not treat” preferred, so EVSI is zero. With a test cost of five units, net value is −5: expected utility falls from 91 when acting immediately to 86 after testing. section 7.1.4

How Timeless Algorithms, The Foundational Ideas covers it

How deep it goes
Calculates EVSI and net value through result probabilities, posterior probabilities, action choices, and expected utilities in a diagnostic-test example.
Examples
Uses a second disease test whose results shift estimated risk but leave the treatment decision unchanged.
What it assumes
Assumes probabilities, posterior disease estimates, and action-specific expected utilities can be compared.
What you can do after
Evaluate whether an information-gathering option improves expected utility enough to cover its cost.
How it connects
Connects expected utility, expected value of sample information, posterior probabilities, 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.