Item Response Theory for trait assessment in randomized item pool for computer based test

The rapid adoption of computer-based examinations (CBE) has raised concerns regarding fairness and reliability in assessing student performance. This study examines randomized test item fairness to test takers’ traits in CBE through the application of Item Response Theory (IRT), with a focus on the two-parameter logistic (2PL) model. Data were collected from 500 level 100 undergraduate students who responded to ten Algebra test items. Raw scores were dichotomized into binary outcomes and analyzed to estimate each item’s difficulty level ( b ) , item’s discrimination power ( a ) , student ability (latent trait) ( θ ) and overall test performance. The estimated item discrimination parameter ranged from 0.87 to 2.14, while the item difficulty parameter estimate ranged from 0.26 to 1.91. These estimates indicated moderate to strong discriminatory power across most test items and its effectiveness between low and high ability examinees, as well as a balanced mix of less difficult, moderately difficult, and highly difficult items, respectively. The test-level analyses using the Test Characteristic Curve (TCC) and Test Information Curve (TIC) showed that the test was most reliable for candidates with average ability levels, clustered at latent traits of 0 ≤ θ ≤ 2 . The item-fit metric evaluation utilizing the S- X 2 statistic ranging from 1.182 to 5.304, with corresponding p-values between 0.505 and 0.978 above the chosen significance level ( α = 0 . 05 ), and RMSEA values close to 0.000 highlighted eight out of the ten items within the acceptable model fit criteria. This implied that although the majority of examinees receive fair and reliable measurements from the randomized test, some items may need to be revised in order to increase measurement accuracy and equity. The study shows how IRT may be used practically for trait assessment and test item fairness evaluation in computer-based exams.