# IRS Discriminant Function **Entity class:** Tax-return examination scoring system ## Definition and history The IRS Discriminant Function, or DIF, began scoring returns for examination selection in 1969 after evaluation of a modified linear discriminant approach. [[articles/Intelligence Inside|Intelligence Inside]] treats it as an early supervised model allocating national institutional attention. A score ranks returns by expected significant tax change if examined; human classifiers then assess selected cases. The system's purpose and documented method are tax examination selection, rather than a general judgment of a taxpayer's character. ## Mechanism and limits Returns selected for examination generate outcome labels, creating a bias if training draws only from the selected cases. Statistically designed examination samples from TCMP and its successor NRP provide a second evidence channel for estimating population-level behavior and recalibrating the model. This makes sampling design part of the control loop, not a peripheral statistical footnote. ## Relationships **Operator:** [[wiki/Internal Revenue Service|Internal Revenue Service]]. **Training evidence:** [[wiki/Taxpayer Compliance Measurement Program|Taxpayer Compliance Measurement Program]] and [[wiki/National Research Program|National Research Program]]. **Problem:** [[wiki/Selective Labels|Selective Labels]]. **Article route:** [[articles/Intelligence Inside|Intelligence Inside]]. ## Sources [IRS Research Bulletin, DIF historical account](https://www.stayexempt.irs.gov/pub/irs-prior/p1500--2022.pdf); [[articles/Intelligence Inside|Intelligence Inside]].