Language Testing
Diagnosis of childhood language impairment typically involves the administration of language tests that are valid and reliable. Unfortunately, not all published assessment tools meet these standards. In addition, for many tools, data are missing to confirm their appropriateness for children from minority racial and ethnic backgrounds and for children reared in impoverished environments. Our research works to address this gap in the literature. Two indices we use to examine the clinical utility of tools are sensitivity and specificity. For both, values closer to 1.00 are ideal; however, values above .80 are considered fair to good. Definitions and formulas for calculating these indices are listed below.
Sensitivity (Se): The percentage of children classified as impaired who are identified by a test as impaired; Se = a/(a + c).
Specificity (Sp): The percentage of children classified as typically developing who are identified by a test as not impaired; Sp = d/(b + d).
Below is a figure that is adapted from a study by Oetting, Cleveland, & Cope (in press). The study focused on the clinical utility of two tools, the Language Comprehension Subtest of the Stanford Binet Intelligence Scale (CSSB; Thorndike, Hagen, & Sattler, 1986) and an experimental nonword repetition task (Dollaghan & Campbell, 1998). We found that the clinical utility of these two tools was best when the tools were combined and when an empirically-derived cutoff score of -.87 standard deviations was used. This derived score corresponded to a CSSB standard score of 91 (-.56 SD) and a NRT score of 71% correct, although with D scores, a child’s score on one tool can be slightly higher or lower than these values as long as the score on the other tool increases or decreases accordingly. As can be seen, Se and Sp were above 80%.
Results from discriminant analysis with CSSB and NRT
