Statistical Innovation to Promote Fairness in Modern Educational Assessment Tests

Educational assessment tests are used to measure the proficiencies of examinees worldwide, and the test results are often used for decision-making at the individual level. These tests have a massive impact on modern society, and cheating seriously challenges fair assessments.

The consequences of cheating can be serious, leading to for example unjust admission decisions to university programs.

The consequences can even be severe if for example the cheating is conducted on a medical examination by an individual becoming a surgeon.

A big test security threat is that of questions (referred to as items) getting leaked, meaning that items get exposed to some part of the population of examinees before the test is administered.

This project will develop statistical methods that can simultaneously detect cheaters and leaked items for both classic paper-and-pencil tests and state-of-the-art digital assessments.

Statistical decision rules that quantifies the uncertainty of classifying an examinee as a cheater and an item as leaked will also be proposed.

This project will furthermore propose the first statistical model that takes the presence of cheating and leaked items into account when adjusting scores in-between test administrations for the purpose of reporting test results that only are affected by the abilities of the examinees, a statistical process known as equating.

All methods that will be developed will be made available through the free software environment R.