Source URL: View this document on the ASHA website
This article demonstrated that despite the 10 years that had passed between the publication of McCauley and Swisher (1984) and this article, the vast majority of commercially available norm-referenced tests did not provide psychometric measures deemed necessary in order to establish a test as valid.
Background: This article was a follow up to “Psychometric Review of Language and Articulation Tests for Preschool Children” by McCauley and Swisher (1984), which established 10 psychometric measures necessary for norm-referenced tests. These 10 measures included descriptions of the normative sample, sample size, item analysis, means and standard deviations, concurrent validity, predictive validity, test-retest reliability, inter-examiner reliability, description of test procedures, and description of test qualifications.
Methodology and Participants: Twenty-one tests of language skills that included norms for children ages 4 and 5 years were reviewed for information on 10 psychometric criteria.
Conclusion: This article proposed levels of accuracy expected for a diagnostic measure to be considered acceptable. They proposed a test with 80% accuracy in determining language disorder be considered “fair” and a test with over 90% accuracy be considered “good.” The authors stated that accuracy below 80% would produce misidentifications at “unacceptably high rates.” Only four of the 21 tests reviewed provided at least 6 of the 10 psychometric measures established by McCauley and Swisher (1984) and only one, the SPELT-II, reached acceptable (fair) levels of accuracy in determining the presence of a language disorder. Only one test of 21 could identify the presence of a language disorder with acceptable accuracy, assuming that it is free of bias and that the normative sample is representative of the child being tested.
Relevance to the Field: This article demonstrated that despite the 10 years that had passed between the publication of McCauley and Swisher (1984) and this article, the vast majority of commercially available norm-referenced tests did not provide psychometric measures deemed necessary in order to establish a test as valid. It is important to note that even when a test reaches what Vance and Plante defined as “acceptable” levels of accuracy, 80%, 2 in every 10 children will still be misidentified at that level. Clinicians who rely on standardized scores alone are guaranteeing significant levels of misdiagnosis of the children they intend to help. This results in typically developing children being placed in special education while language impaired children continue to struggle in class without the help they need. According to the authors, this is a particular problem when school systems have specific, arbitrary cutoff points to determine normal versus disordered language. For example, a school in the Tuscon area used arbitrary cutoff scores that systematically underidentified LD children on the TACL-R and TELD-2 and overidentified TD children on the CM-TOL and SPELT-II. No matter which test was administered, the students’ performance was never appropriately assessed. Additionally, the use of two standardized tests in combination actually increased the chances of misdiagnosis for both LD and TD. This has serious implications in school systems that require multiple tests to diagnose a child. It is the responsibility of the clinician to evaluate their assessment materials for validity and to develop their clinical judgment regarding the children they work with.
Vance, R., & Plante, E. (1994). Selection of preschool language tests: A data-based approach. Language, Speech, and Hearing Services in Schools, 25, 15-24.