Children with Reading Disabilities: Does Dynamic Assessment Help in the Classification?

This study was conducted to determine whether the cognitive performance of reading disabled and poor readers can be separated under dynamic assessment procedures, and whether measures related to dynamic assessment add unique variance, beyond IQ, in predicting reading achievement scores. The sample consisted of 70 children (39 females and 31 males). Within this sample four groups of children were compared: children with reading disabilities (n=12), children with math/reading disabilities (n=19), poor readers (n=14), and skilled readers (n=25). Intelligence, reading and math tests, and verbal working memory (WM) measures were administered (presented under static and dynamic testing conditions). Two important findings emerged: (a) hierarchical regression analyses found that a dynamic assessment measure factor score contributed unique variance to predicting reading and mathematics, beyond what is attributed to verbal IQ and initial scores related to WM; and (b) poor readers and skilled readers were more likely to change and maintain their WM score gained under the dynamic testing conditions than children with reading disabilities or children with a combination of math/reading disabilities. Implications for a valid classification of reading disabilities are discussed. H. LEE SWANSON, Ph.D., is professor, educational psychology and special education, University of California, Riverside. CRYSTAL B. HOWARD is a doctoral candidate, educational psychology, University of California, Riverside. Children with reading disabilities (RD) experience information-processing difficulties on specific cognitive tasks (e.g., Stanovich & Siegel, 1994; Swanson & Siegel, 2001; Torgesen, 2002). These processing difficulties are assumed to be intrinsic to the child; that is, they are not due to instructional or environmental factors (e.g., Shaywitz et al., 1999). Further, RD children’s processing difficulties are reflected in specific academic domains (e.g., reading) that draw upon those processes (e.g., Swanson & Siegel, 2001; Torgesen, 2002). In addition, it is assumed that these specific processing deficits are unexpected given their overall potential (see Fletcher et al., 2002; Stanovich & Siegel, 1994; for a review of assumptions). Given these assumptions, at least two questions emerge. First, how should “potential” be measured? The notion of potential has played a critical role in defining learning disabilities (LD) since the inception of the field (e.g., see Bateman, 1992, for review). Typically, differences between IQ and achievement on standardized Volume 28, Winter 2005 17 tests are viewed as a prototype for representing differences between potential and actual performance (Fletcher, Francis, Rourke, Shaywitz, & Shaywitz, 1992; Shepherd, Smith, & Vojir, 1983). However, a review of the literature suggests that such procedures are invalid for classification purposes (e.g., Fletcher et al., 1992; Hoskyn & Swanson, 2000; Stuebing et al., 2002). For example, the relevance of standardized intelligence measures (e.g., WISC-III) in the diagnostic classification of learning disabilities has been criticized because reading achievement within samples with LD is not predicted by variations (high vs. low) in IQ (e.g., Fletcher et al., 2002; Hoskyn & Swanson, 2000; Stanovich & Siegel, 1994; Siegel, 1989, 1992; Stuebing et al., 2002). Further, several authors (e.g., Brown & Ferrara, 1999; Campione, 1989; Embretson, 1992) have suggested that traditional intelligence tests (i.e., tests that measure unassisted performance on global measures of academic aptitude) provide a poor estimate of general ability. These authors argue that because static or traditional approaches to assessment typically provide little feedback or practice prior to testing, failure often reflects children’s misunderstanding of instructions more that their ability to perform the task. Thus, whether “potential” is adequately captured on traditional IQ measures presents a conceptual problem. One possible alternative or supplement to traditional assessment is to measure a child’s gain in performance when given examiner assistance. Thus, “potential” for learning new information (or accessing previously presented information) is measured in terms of the distance, difference between, and/or change from unassisted performance to a performance level with assistance. Procedures that attempt to modify performance via examiner assistance in an effort to understand learning potential are called dynamic assessment (e.g., see Grigorenko & Sternberg, 1998; Swanson & Lussier, 2001). Although dynamic assessment is a term used to characterize several distinct approaches (see Grigorenko & Sternberg, 1998; Swanson & Lussier, 2001; for a review) it includes two critical features: to determine the learner’s potential for change when given assistance, and to provide a prospective measure of performance change independent of assistance (Embretson, 1987). Unlike traditional testing procedures, score changes due to examiner intervention are not viewed as threatening task validity. In fact, some authors argue that they increase construct validity (e.g., Carlson & Wiedl, 1979; Elliot & Lauchlan, 1997; Swanson, 1992). Although dynamic assessment has been suggested as an alternative to traditional assessment (e.g., Day, Engelhardt, Maxwell, & Bolig, 1997; Jitendra & Kameenui, 1993), there are no published data, to the authors’ knowledge, on whether children with RD are more sensitive than other ability groups to such procedures. Thus, a number of questions need to be addressed if such procedures are to be used to assess RD. For example, can children with RD, when given instructional support on processing tasks, be differentiated in performance from poor and average readers? This question is important because the processing difficulties of children with RD are assumed to be stable compared to other processing abilities (see Swanson & Hoskyn, 1998, for discussion). Thus, if the processing performance of children with RD can be substantially modified and their performance is statistically comparable to that of normally achieving children, the “intrinsic nature” of RD needs to be reexamined. Another question is whether children with RD can be separated from poor readers. This issue is important because assessment practices that rely heavily on psychometric tests for classification of children with RD have not provided, to date, systematic procedures for separating those children who primarily have reading problems related to inadequate or weak instructional support from children who have information processing deficits (Torgesen, 2002). Related to this issue is the finding that the cognitive profile of children with RD cannot always be discriminated from that of generally low-achieving children when using static or traditional assessment (Hoskyn & Swanson, 2000; Stuebing et al., 2002). In summary, the present study had two purposes. First, a determination was made as to whether processing “potential” via dynamic assessment is related to reading achievement. Processing potential is defined as the score obtained with examiner assistance (i.e., gain score) and sustained performance without assistance (i.e., maintenance score). In statistical terms, the question is whether gain and maintenance scores contribute unique variance to reading achievement beyond what is contributed by a traditional intelligence measure. Linking dynamic assessment with reading achievement as well as determining whether “potential” as measured on a commonly used IQ test differs from potential as measured under dynamic testing conditions in the prediction of achievement are important issues if dynamic assessment is to be taken seriously as a valid assessment procedure in the diagnosis of RD. The tasks used in this study for assessing information processing potential were related to working memory (WM), a critical component of major informationprocessing models (e.g., Baddeley & Logie, 1999) that has been found to be seriously deficient in children with RD (e.g., De Beni, Palladino, Pazzaglia, & Cornoldi, 1998; Siegel & Ryan, 1989; Swanson, 1993, 2003; also Swanson & Siegel, 2001, for a comprehensive review).1 All major information-processing models involving skill Learning Disability Quarterly 18 acquisition and learning include the component of WM (e.g., see Daneman & Merikle, 1996, for a review), because it is highly correlated with performance on several academic and language-related tasks, such as vocabulary (e.g., Baddeley, Gathercole, & Papagno, 1998), reading comprehension (e.g., Swanson, 1999), language acquisition (e.g., Baddeley et al., 1998), problem solving (e.g., Kyllonen & Christal, 1990), mathematics (e.g., Bull, Johnston, & Roy, 1999), fluid intelligence (e.g., Engle, Kane, & Tuholski, 1999), and writing (McCutchen, 2000). Correlations between WM and reading or intelligence measures with adult samples are in the range of .55 to .92 (e.g., see Daneman & Merikle, 1996). The standardized test (N=1594) used to measure WM was the Swanson-Cognitive Processing Test (S-CPT; Swanson, 1995a). As indicated by Grigorenko and Sternberg (1998), this is one of the few tests that report validity and reliability data. It is an individually administered battery that is assumed to measure different aspects of WM ability and processing potential. Working memory is defined in this test as concurrent processing and storage activities, whereas potential, via dynamic assessment, is defined as (a) learner performance change relative to initial performance on WM measures when given assistance (gain) and (b) performance change independent of assistance (maintenance). Second, it was of interest to determine whether children with RD can be discriminated via dynamic assessment from children who are poor readers. This is important because several studies (see synthesis of the literature by Hoskyn & Swanson, 2000; Stuebing et al., 2002) indicate that there are no clear psychometric and processing distinctions between poor readers and children with RD. However, the fact that current practices using static measures do not distinguish children with RD from children who are poor readers does not mean it cannot be done. Thus, we examine whether a child’s response to assisted performance provides a frame of reference for separating children who are poor readers from children who are RD.2 Although not related to dynamic assessment, a comprehensive synthesis of the treatment intervention literature indicated that the magnitude of treatment outcomes (effect size) for children with RD was smaller (i.e., they were less responsive) than for poor readers (see Swanson & Hoskyn, 1998, p. 307, for discussion). Based on these findings, it is possible that poor readers will be more responsive to measures of change than children with RD. In summary, the purpose of the present study was twofold: (a) to determine whether dynamic assessment adds unique variance beyond IQ in predicting reading achievement scores; and, (b) to compare children classified as RD with skilled and poor readers on dynamic assessment measures. It was hypothesized that (a) dynamic assessment measures will contribute significant variance in predicting reading and (b) children with RD will be less responsive to dynamic assessment than children who are poor readers.