An Item Response Theory Analysis of the Korean Version of the CRAFFT Scale for Alcohol Use Among Adolescents in Korea

Abstract

This study aimed to validate the psychometric properties of the CRAFFT (Car, Relax, Alone, Forget, Family/Friends, Trouble) by using item response theory (IRT) and further examine gender differences in item-level responses. Methods: This study used the 13 th (2017) Korea Youth Risk Behavior Survey data conducted by the Korean Centers for Disease and Prevention and analyzed data of 8,568 students who reported drinking alcohol in the previous 30 days. IRT assumptions including unidimensionality, local independence, and monotonicity were tested. A one-factor two-parameter IRT model was fitted for item parameterization (M 2 ¼ 76.92, 20 df, p < .001, root mean square error of approximation ¼ 0.02). Graphics for item characteristic curves and item and test information curves were provided. Differential item functioning (DIF) analysis was evaluated to measure item equivalence between boys and girls. Results: Unidimensionality and local independence were satisfied, and the dominant factor eigenvalue and variance were 2.48 and 41.3% in boys and 2.08 and 34.7% in girls, respectively, and residual correlations in paired items were less than .2. The discrimination parameters were estimated as 1.32e3.59 for boys and 1.30e3.64 for girls. The severity parameters estimated verity as À0.23 to 1.67 for boys and À0.32 to 2.17 for girls. DIF was detected for four of six items, "Car," "Relax," "Forget," and "Family/ friends." However, its impact on the total score was negligible. Conclusion: The CRAFFT is a valid and reliable screening tool to identify alcohol use problems for both boys and girls, exhibiting good discrimination, good coverage of severity, and negligible DIF.

Quality of multiple-choice questions in medical internship qualification examination determined by item response theory at Debre Tabor University, Ethiopia

“…Evidence shows the validity of assessment results is affected by the content tested, quality of test items, qualification of item writers, number of test items, presence of item writing flaws, and psychometric characteristics of items [ 5 , 9 – 11 , 14 , 20 , 30 , 31 ]. Item difficulty, measured by the percentage of examinees that correctly answered the item, runs from 0 to 1; easy items have a higher difficulty index [ 32 ]. Most studies classify item difficulty as too easy (≥ 0.8), moderately easy (0.7–0.8), desirable (0.3–0.7), and difficult (< 0.3) [ 22 , 33 – 37 ].…”

Evaluation of Studies on the Measurement Properties of Self-Reported Instruments

“…Three studies applied IRT/Rasch analysis to structural validity. However, for one study [27] there were no values reported for the assumption tests for IRT. For another of the studies [29], model fit values (infit and outfit mean squares) were reported for the assumption test of unidimensionality, while no IRT-related values were reported for the third study [31].…”

“…In this case, the application of multiple-group CFA is recommended for investigating structural invariance between the cultural groups rather than separately conducting CFA in each population. Another study [27] used DIF to investigate item invariance by gender based on IRT.…”

The Classical Test or Item Response Measurement Theory: The Status of the Framework at the Examination Council of Lesotho

“…The responses may, therefore, be influenced by factors other than what the instrument was designed to measure. Given an individual's score on the latent trait, the observed items should be independent of each other (Debelak & Koller, 2020;Song et al, 2019). Independent means are statistically independent.…”

“…A test information function may be used to balance multiple alternate test forms for the same exam. TIF values should be the same across all alternate forms (Song et al, 2019).…”