Nontraditional Scoring of C-tests

TL;DR

This paper proposes a novel method for scoring C-tests by clustering items based on their inter-item distances and assigning weights to address local independence violations, aligning C-tests with classical test structures.

Contribution

It introduces a clustering-based scoring approach for C-tests that maintains local independence, improving their validity as standardized assessments.

Findings

Clustering items improves test validity.

Weighted scoring aligns C-tests with classical test structures.

Addresses local independence violation in C-tests.

Abstract

In C-tests the hypothesis of items local independence is violated, which doesn't permit to consider them as real tests. It is suggested to determine the distances between separate C-test items (blanks) and to combine items into clusters. Weights, inversely proportional to the number of items in corresponding clusters, are assigned to items. As a result, the C-test structure becomes similar to the structure of classical tests, without violation of local independence hypothesis.