Information value (IV)
Information value (IV) measures how strongly a characteristic, such as an income band, separates goods from bads across its bins. It is calculated by multiplying each bin’s difference between its share of goods and share of bads by that bin’s weight of evidence, then summing the results.
Example
For illustration, suppose an income characteristic has two bins. In one bin, goods account for 60% of all goods and bads for 30% of all bads, with a weight of evidence of 0.69; its IV contribution is (0.60 − 0.30) × 0.69 = 0.207. If the other bin’s shares are 40% and 70%, with a weight of evidence of −0.56, its contribution is (0.40 − 0.70) × −0.56 = 0.168, making the total IV 0.375.
How each bin contributes to information value
Work out the contribution for one bin at a time. For illustration, if a bin contains 25% of all goods and 15% of all bads, the difference is 0.10; with a weight of evidence of 0.40, the contribution is 0.04.
Repeat the calculation for each bin, using its own shares and weight of evidence, then add the contributions. The signs matter: a negative share difference multiplied by a negative weight of evidence gives a positive contribution.
Reading information value when choosing characteristics
A common reading treats IV below 0.02 as not predictive, 0.02 to 0.1 as weak, 0.1 to 0.3 as medium, and above 0.3 as strong. A value above 0.5 is worth checking for leakage. These are common interpretations, not regulatory thresholds.
More bins can leave very small groups, and very small bins can inflate IV. Bins are usually kept to at least about 5% of the sample, so an unusually strong value should be considered alongside how the characteristic was binned.
Questions we get asked
How is information value different from weight of evidence?
Weight of evidence is calculated for each bin, while information value combines the bins into one measure for the characteristic. IV uses each bin’s difference in goods and bads shares together with its weight of evidence.
Can information value tell whether one applicant is good or bad?
No. It describes how a characteristic separates goods from bads across bins in a sample, not the outcome for one applicant. It is a measure used when choosing scorecard variables.
Can two characteristics with the same IV separate goods and bads differently?
Yes. IV sums the contributions from each bin, so different bin-level patterns can produce the same total. Comparing the bin contributions shows how each characteristic’s total is formed.
Decisimo’s free information value calculator calculates weight of evidence and information value by bin from goods and bads counts, or bins an uploaded file.
