Examination on the Reliability of a Racquetball Skills Test Battery Using Univariate and Multivariate Models

The American Alliance for Health, Physical Education, Recreation, and Dance has consistently promoted the need for reliability and validity evidence of tests and test batteries. Wood and Safrit (1987) noted that estimates of battery reliability have typically been derived from estimates of individual test items and that multivariate models may be a better method of evaluating reliability of a multivariate construct (e.g., skills test). The purpose of this investigation was to compare the results of both univariate and multivariate models in the evaluation of battery reliability. Participants (N=131) were voluntary undergraduate students (mean age=23 years old) from university racquetball courses. A racquetball skills test battery (Lam & Zhang, 2002) was administered to participants on 2 occasions separated by 2-4 days. Test items included service placement to the left and right, forehand and backhand power drive, forehand and backhand power shot placement, ceiling shot, and wall rally. The mean of 20 trials for each item was used as the test item score. Wall rally was the lone exception, with the mean of 3 trials accounting for the score. One-way intraclass correlation coefficients (ICC) were used to determine test-retest reliability of individual items. Canonical correlation analysis and generalizability (G) theory were used to estimate test-retest reliability of the battery. Intraclass correlation coefficients ranged from .49-.90 across items, with 6 of the 8 coefficients below .80. Canonical correlation analysis revealed moderate shared variance between test and retest scores (R_C1=.914, R_{D Total}=.45, p<.05). The results were lower than the theoretical upper limit to battery reliability based on 1 administration of the battery (R_C1=.975, R_{D Total}=.82). Service placement was not included in the G-study because of its low validity. The G coefficients of the remaining items ranged from .61-.79. Based on the results of this study, the following conclusions seem warranted: 1) the racquetball skills test battery is moderately reliable; 2) service placement items and power placement items need adjustment to improve reliability (ICC <.70) but still contribute to the test battery; and 3) the univariate model and multivariate models complimented the other in the evaluation of reliability and revealed independent information.