Integrated Testlets and the Immediate Feedback Assessment Technique

TL;DR

This paper introduces integrated testlets using the immediate feedback assessment technique (IF-AT), enabling multiple-choice exams to mimic constructed-response questions with partial credit and interdependent items, enhancing assessment effectiveness.

Contribution

It presents a novel hybrid testing format combining MC and CR features through integrated testlets and IF-AT, improving discrimination and partial credit allocation.

Findings

High test reliability (α=0.82) for the exams.

Excellent discrimination with mean item-total correlation of 0.41.

Partial credit effectively discriminates student understanding.

Abstract

The increased use of multiple-choice (MC) questions in introductory-level physics final exams is largely hindered by reservations about its ability to test the broad cognitive domain that is routinely accessed with typical constructed-response (CR) questions. Thus, there is a need to explore ways in which MC questions can be utilized pedagogically more like CR questions while maintaining their attendant procedural advantages. we describe how an answer-until-correct MC response format allows for the construction of multiple-choice examinations designed to operate much as a hybrid between standard MC and CR testing. With this tool - the immediate feedback assessment technique (IF-AT) - students gain complete knowledge of the correct answer for each question during the examination, and can use such information for solving subsequent test items. This feature allows for the creation of a new type of context-dependent item sets; the "integrated testlet". In an integrated testlet certain items are purposefully inter-dependent and are thus presented in a particular order. Such integrated testlets represent a proxy of typical CR questions, but with a straightforward and uniform marking scheme that also allows for granting partial credit for proximal knowledge. We present a case study of an IF-AT-scored midterm and final examination for an introductory physics course, and discuss specific testlets with varying degrees of integration. In total, the items are found to allow for excellent discrimination, with a mean item-total correlation measure for the combined 45 items of the two examinations of $\bar{r}'=0.41\pm 0.13$ (mean $\pm$ standard deviation) and a final examination test reliability of $\alpha=0.82$ ($n=25$ items). Furthermore, partial credit is shown to be allocated in a discriminating and valid manner in these examinations.