Short Forms and Computerized Adaptive Tests With Monosyllabic Words Can Efficiently Measure Speech Recognition

Construct validity of the Northwestern University Auditory Test No. 6 (NU-6) monosyllabic word lists or shortened versions have not been examined using the Rasch measurement theory. The study purposes were to test the fit of the Rasch measurement model to monosyllabic word lists and whether short forms and computerized adaptive testing can measure speech recognition.

A cross-sectional study design with 50 persons (average age = 71 years; 35 female, 15 male) with mild-to-moderate hearing loss was used to test the fit of NU-6 Word Lists 1 and 2 to the Rasch measurement model. Pearson's correlations quantified the accuracy of person measures from short forms or computer-adaptive testing simulations with measures from the full lists. Average standard error of person measures quantified measurement precision.

Word lists were unidimensional, had negligible misfit, and had high person reliability. Nineteen- and 11-word short forms were made per list. Person measures from 19-word short forms had a high linear association with person measures from the full word lists (List 1: r = .92, p < .0001, SE = 0.76; List 2: r = .91, p < .0001, SE = 0.74) compared to moderate association for the 11-word short form (List 1: r = .84, p < .0001, SE = 0.99; List 2: r = .81, p < .0001, SE = 0.97). Person measures from computerized adaptive testing simulation reached a correlation threshold, r > .90, after 15-20 words were administered for both lists. A precision-based stopping rule used an average of 18 (List 1) or 20 (List 2) words.

Short forms with 19 words and computerized adaptive testing may accurately and precisely measure speech recognition.