Knowledge MaxxingMeasurement bias
Measurement bias occurs when a feature is measured less accurately or consistently for some groups than others, distorting what the model actually learns about them.
Prerequisite: Label bias
What matters
Because the model treats measured values as ground truth, an unevenly accurate measurement process can teach a model wrong patterns specifically for the poorly measured groups.
Example
A health app's step counter that's less accurate for certain phone models used more by one demographic can introduce measurement bias into any model built on that data.
Common misconception
Measurement bias is not the same as sampling bias; it's specifically about the accuracy of how a value is measured, not which groups are represented in the data at all.
Recall question
How does measurement bias differ from sampling bias?