Developing a Computerized Adaptive Test to Assess Stress in Chinese College Students

Stress is among the most prevalent problems in life; thus, measurement of stress is of great importance for disease prevention and evaluation. This work aims to develop a computerized adaptive test (CAT) application to measure stress (CAT-S) based on item response theory (IRT). Two types of analyses were performed. The first analysis was to meet the psychometric requirements of the CAT-S. A Paper and Pencil (P&P) test involving 226 items was developed based on eight stress-related scales, and 972 Chinese college students completed the test. The first seven scales were used to build the item bank, and the last scale (i.e., the Perceived Stress Scale, PSS) was used to determine the convergent validity of the CAT-S. With some statistical considerations, such as item fit, discrimination, differential item functioning (DIF), and the assumption of unidimensionality, the final item bank comprised 93 items. The second analysis was to simulate the CAT adaptively using the existing item response. A Bayesian method called Expected a Posterior method (EAP) was used to estimate θ. For the item selection strategy, the greatest item information was considered at each step. The stopping rule was determined by the fixed length (10, 11, 12, …, 20, and 93) or the prespecified level of measurement precision (standard errors of 0.3, 0.4, 0.5, 0.6, 0.7, and 0.8). Finally, the criterion validity was tested by using PSS as a criterion and analyzing the effect of CAT-S diagnosis with a receiver operating curve (ROC). The results showed that (1) the final stress item bank had good quality based on the psychometric evaluation, (2) the CAT-scores were highly correlated with the scores of the final item bank, (3) the scores of the P&P form of PSS were correlated with those of the CAT-S (r > 0.5), (4) the value of the area under the ROC curve (AUC) was greater than 0.7 under each stopping rule, and (5) the CAT-S needed only a small number of items to obtain a highly precise measure of stress. Therefore, the CAT-S presented the theoretically expected advantages, which enabled a rapid, accurate, and efficient dynamic and intelligent measurement of stress.