Analysis on an online placement exam for college algebra

Abstract

An online placement exam was administered to 2800 entering freshmen, 700 of whom enrolled in College Algebra during the succeeding Fall semester. Problems on the placement exam were clustered using several different techniques including both expert analysis and item response theory. Student scores on these groupings of problems were then compared to their scores on the first two hour exams in the course (representing the first half of the material in the course) and also on ACT data. Based on this comparison, certain problems were selected as more or less informative for purposes of placement. A model was created using previously available ACT data along with the new placement data to predict initial student success in the course. This model explains 50% more of the variance in student scores than the previously available ACT data alone. Suggestions for improvements to the test and the placement methodology are made based on our analysis.