Kenneth A. Kavale, University of Iowa
Learning Disabilities Summit: Building a Foundation for the Future White Papers
This paper is available in alternative formats: | Download Word | Download pdf |
Pages: | 1 | 2 | 3 | 4 | 5 | 6 | 7 |
The idea of distributional and etiological differences in a population was first proposed in the MR field. At IQ 50, it becomes possible to distinguish between mild and severe MR. Severe MR (about 25% of the MR population) typically represents "clinical" MR in the sense of probably possessing, besides limited cognitive ability, central nervous system pathology and associated disabilities. The larger mild MR group typically shows no neurological signs or associated clinical signs, and represents what is termed "familial" MR (Zigler, 1967). In the severe cases, the pathological factors significantly interfere with intellectual development (see Tarjan, Wright, Eyman, & Keeran, 1973) to such an extent that they distort the IQ score distribution as shown by Dingman and Tarjan (1960). In comparing the IQ distributions of low IQ populations (mild and severe) with those of the general population, there was an indication of an excess of cases ("hump") at the lower end of the distribution. Above IQ 50, there were few discrepancies between expected and actual percentages in the distribution but an excess of cases in the 0-19 IQ and 20-49 IQ ranges. This excess population formed a hump: an additional normal distribution of IQs with a mean IQ of 32 and an SD of 16. Clearly, when compared with IQ levels, the two groups appeared to differ with respect to etiology and clinical manifestations (Jastak, 1967; Weir, 1967).
The qualitative differences between the two MR "populations" became a source of debate and evolved into what was termed the "developmental-difference controversy" (Zigler & Balla, 1982). Generally, "this controversy centers around the question of whether the behavior of those retarded persons with no evidence of central nervous system dysfunction is best understood by those principles in developmental psychology that have been found to be generally applicable in explaining the behavior and development of non-retarded persons, or whether it is necessary to involve specific differences over and above a generally lower rate and asymptote of cognitive development" (p. 3).
Because of the developmental-difference controversy, the related GRB-SRR distinction also became contentious. For example, many studies have failed to find a GRB-SRR bimodal distribution (e.g., Rodgers, 1983; Share, McGee, McKenzie, Williams, & Silva, 1987; Stevenson, 1988). Van der Wissel and Zegers (1985) suggested that no hump was found because it may, in reality, be an artifact resulting from floor and ceiling effects associated with the reading measures used. Using designs where students differed in reading level but were comparable in age (CA design) or comparable in reading level but varied in age (reading-level match design), a number of studies failed to demonstrate that SRR groups (achievement scores below levels predicted by IQ, i.e., discrepant) could be differentiated from a GRB group (depressed achievement not discrepant from IQ) (Fletcher et al., 1989; Fletcher, Francis, Rourke, Shaywitz, & Shaywitz, 1992; Foorman, Francis, Fletcher, & Lynn, 1996; Rispens, van Yperen, & van Duijn, 1991; Share & Silva, 1986; B. A. Shaywitz et al., 1992; L. S. Siegel, 1992; Vellutino, Scanlon, & Lyon, 2000). Consequently, IQ was not a major factor associated with SRR, which was interpreted to mean that SRR was not a discrete entity, but rather
...occurs along a continuum that blends imperceptibly with normal reading ability. These results indicate that no distinct cut-off exists to distinguish children with dyslexia clearly from children with normal reading ability; rather, the dyslexic children simply represent a lower portion of a continuum of reading capabilities (S. E. Shaywitz, Escobar, Shaywitz, Fletcher, & Makuch, 1992, p. 148).
Rutter (1990) suggested that "the crucial test of the SRR hypothesis, however, does not depend on the presence or absence of a hump in the distribution but whether the correlates and outcomes of SRR serve to differentiate the syndrome from GBR" (p. 637). A number of studies have failed to differentiate GRB and SRR groups, however. For example, GRB groups (i.e., no I.Q.-achievement discrepancy) performed no differently on independent measures of reading achievement or on assessments of the cognitive abilities presumed to underlie the ability to learn to read (Fletcher et al., 1994; Francis, Shaywitz, Stuebing, Shaywitz, & Fletcher, 1996; Morris et al., 1998; Share, McGee, & Silva, 1989; Stanovich & Siegel, 1994). With respect to gender differences, the presumption of a disproportionately greater number of boys than girls in SRR groups has not received support (Pennington, Gilger, Olson, & DeFries, 1992; Share et al., 1987; S. E. Shaywitz, Shaywitz, Fletcher, & Escobar, 1990). Finally, SRR groups were presumed to have a poorer educational prognosis than GRB groups (Yule, 1973), but no evidence supports the validity of this assumption (Francis et al., 1996; Share et al., 1989; B. A. Shaywitz et al., 1992; Vellutino et al., 1996).
In a summary of the available evidence, Fletcher et al. (1998) concluded that
Under no circumstances is wholesale use of IQ test for learning disabilities justified. We have shown numerous problems with the discrepancy model, regardless of whether IQ tests or some other measures are used to operationalize the aptitude index. It is not the use of the IQ test that creates the problems with discrepancy. Classifications of children as discrepant versus low achievement lack discriminative validity (p. 200).
It was then suggested that the discrepancy criterion not be part of the LD identification process primarily because "it is not the score on the IQ test that identifies the child as having learning disabilities, but rather the score on the test of academic achievement that identifies the child with LD" (p. 201). Similarly, Aaron (1997) concluded that "a review of research in the area of reading disabilities indicates that classifying poor readers on the basis of a discrepancy formula into LD and non-LD categories lacks validity on both theoretical and empirical grounds" (p. 488). As an alternative, Aaron suggested a more pragmatic approach based on the Reading Component Model that identifies the source of the reading problem for all students and then focuses remedial efforts on that particular source.
The discrepancy criterion for LD identification has thus been seriously challenged, with some anticipating its "impending demise" (Aaron, 1997). One difficulty, however, is in interpreting what that means for LD. The many analyses investigating discrepancy focused attention on the GRB-SRR distinction where, in both cases, the primary problem was an inability to read. Consequently, there was little question about the presence of reading disability (RD), but the presence or absence of LD was not really considered except by implication. Although students with LD are quite likely to manifest reading difficulties, they may not, and this fact makes any generalization from a GRB-SRR comparison suspect. The primary difficulty is conceptual and relates to the fact that if RD and LD are considered equivalent, then the law of parsimony is violated (Swanson, 1991). There appears, however, to be a decided tendency to view LD and RD as the same thing as evidenced by statements such as, "It is time to abandon the discrepancy-based classification of poor readers into LD and non-LD categories and expand the boundaries of LD to include all children who experience difficulties in learning to read" (Aaron, 1997, p. 488). Instead of providing conceptual clarity, such a suggestion would result in even greater confounding between concepts.
The same possible confounding is found with RD itself. The focus on GRB and SRR as discrete groups tends to obscure the fact that almost all students with SRR could be classified as GRB, while half of students with GRB can be classified as SRR (Hinshaw, 1992). Even when considering SRR itself, there are questions about its proper relationship with dyslexia, an RD equally difficult to define with precision (Benton & Pearl, 1978). The many similarities between the conditions raise the question as to whether SRR and dyslexia are the same thing (Yule & Rutter, 1976). Regardless of the answer, discussion about LD seems inappropriate as it is a different (and distinct) phenomenon that may or may not include students with these types of reading problems.
Thus, both LD and RD are complex entities, and eliminating the discrepancy criterion does not appear to be a sensible solution for resolving these complexities. Any suggested alternative, as, for example, in the Reading Component Model proposed by Aaron (1997), does not appear to be a viable solution in any significant sense unless it is also accompanied by a belief that LD is not a legitimate construct. When LD is not considered legitimate, there is a general theme that calls for a cessation of the illegitimate and unnecessary LD labeling, and a focusing instead on the difficulties of some students in learning to read by providing them with effective and responsive interventions (e.g., Christensen, 1992; McGill-Franzen & Allington, 1993; Swerling & Sternberg, 1996). As suggested by Aaron (1997), "When the discrepancy formula disappears from the educational scene, so will the concept of LD. After 40 years of wandering in the wilderness of learning disabilities, we are beginning to get a glimpse of the promised land" (p. 489). Whether or not the disappearance of the discrepancy formula leads to a promised land is certainly moot and would do little to resolve the complex and vexing problems associated with defining LD.
A major roadblock to problem resolution is the lack of a precise description of LD (Kavale & Forness, 2000). Although the description of LD is far from complete, the field has witnessed unprecedented growth and has accomplished this expansion not by using formal, albeit limited, definitions but rather by using a number of singular operational definitions stipulating rules about how a term is to apply in a particular case if specified actions yield characteristic results. Thus, a concept like LD may have a set of operations that define it, and knowing these operations presumably provides complete understanding of the concept (Benjamin, 1955).
For LD, the primary (and often sole) operation has been the application of a discrepancy criterion. Beginning with the USOE (1976) regulations and reaffirmed in proposed operational definitions (e.g., Chalfant & King, 1976; Shaw, Cullen, McGuire, & Brinckerhoff, 1995), discrepancy has emerged as the major means of LD identification. The LD identification process, however, may be more difficult and complicated than it appears to be with the use of a discrepancy criterion. For example, a problem surrounds the theoretical validity of operations. In a scientific sense, an operational definition must bear a logical and rational relationship with the verified theoretical constructs stipulated in the formal definition (Bergmann, 1961). For LD, a problem is created because the formal definition includes no mention of discrepancy (or underachievement) (Kavale, 1993). The resulting lack of congruence between definitions means that essentially two distinct views of LD are being presented: a formal representation and an operational representation.
The lack of correspondence creates a consequential problem: an increased probability that the operational definition may not be justified and may lead to potentially meaningless and insignificant operations that do not meet formal criteria (Deese, 1972). The operations specified may not actually "define" anything but merely state procedures required to test for the presence of the phenomenon to which the operations refer (Kavale, Forness, & Lorsbach, 1991). For example, assume an operational definition of LD that is based on the Learning Disability Coefficient (LDC) whose procedures require a calculation including an individual's white blood cell count multiplied by body weight in ounces, divided by head circumference in centimeters. Although possible to calculate, the LDC would possess little meaning or significance because the available validated knowledge about LD clearly indicates that the LDC does not "fit" any of it.
A less obvious example surrounds the different meanings that may be conveyed when different operational indicators are chosen. For example, discrepancy is defined as the difference between ability and achievement, but any number of ability (i.e., IQ) measures and probably even a greater number of achievement measures might be chosen for comparison. The problem is that when different combinations of measures are used to define discrepancy, it is not at all clear that the assessments are operationally, and thus, definitionally equivalent (Deese, 1972). It may, therefore, be difficult to "make sense" of the calculated discrepancy.
The use of operational definitions is thus neither a simple nor straightforward process but one that requires significant theoretical verification. Unfortunately, the LD field has not achieved the necessary verification primarily because discrepancy was so quickly embraced: "The debate that rages over what LD might be and the lack of consensus over the importance of any given variable is in sharp contrast to the relative unanimity regarding discrepancy. The consensus regarding discrepancy as the primary identification variable for LD has entrenched discrepancy to the point where it now represents the foundation concept for LD diagnosis" (Kavale & Forness, 1994, p. 23). In fact, discrepancy has become a deified concept as evidenced in its ascension to the status of "imperial criterion" (Mather & Healey, 1990) and a reified concept as seen in its elevation to an almost tangible property of students with LD (Kavale, 1987). Such deification and reification do not appear justified given the fact discrepancy itself is a hypothetical construct defined by hypothetical constructs (see Messick, 1981) resulting in the possibility that, in a theoretical sense, discrepancy may be a "fictitious concept" (Hempel, 1952).
The wide embrace of discrepancy has obscured some fundamental considerations. One such consideration surrounds the relationship between discrepancy and LD. With discrepancy often the only criterion used for LD identification, there has been an accompanying assumption that discrepancy represents the operational definition of LD. In reality, "Discrepancy is best associated with the concept of underachievement. This is true now and has historically been the case" (Kavale, 1987, p. 18). In a theoretical context, Shepard (1983, 1989) argued that discrepancy is the operational definition of underachievement. Thus, when a student meets the discrepancy criterion, what is being affirmed is underachievement, not LD. The scientific law of parsimony would suggest that underachievement and LD are not the same thing. To avoid confounding, the proper conclusion when the discrepancy criterion is met is that underachievement has been identified. If it is believed that underachievement is associated with LD (certainly a valid assumption), then discrepancy becomes a necessary but not sufficient criterion for LD identification (Kavale, 1987; Reynolds, 1984-1985).
Within the context of LD identification, discrepancy and the documentation of underachievement should represent only the first step in diagnosis (Kavale & Forness, 1994). Discrepancy is important in the identification process because it establishes a sound theoretical foundation for later LD determination. Although the discrepancy concept possesses psychometric and statistical problems, they have been satisfactorily addressed, and a technically defensible procedure to indicate the presence or absence of underachievement has been achieved. The findings from large-scale investigations appear to have affirmed the relationship between discrepancy and underachievement, and the possibility of reliably differentiating LD (i.e., students who meet the discrepancy criterion) from LA (i.e., students who do not meet the discrepancy criterion). Although critical as the initial step in LD determination, discrepancy should not be elevated to the status of being LD but rather viewed simply as the most useful means for defining underachievement, a necessary part of LD.
With discrepancy placed in proper perspective, attention needs to be directed at what else should be considered in the identification process in order to capture the complex and multivariate nature of LD (Kavale & Nye, 1991). Kavale and Forness (1995) suggested a way the process might proceed. The initial step is the formulation of foundation principles aimed at developing a theoretical framework for elucidating the basic nature of LD. Kavale and Forness (2000) elucidated the process further by proposing an operational definition in the form of a hierarchical scheme where each level depicts a decision point in the determination of LD. The scheme includes five levels where the first includes an ability (IQ)-achievement discrepancy to document the presence or absence of underachievement. The next levels focus on other stipulated criteria (e.g., psychological process deficits, exclusion), and a final LD designation is predicated on a student proceeding through each level. The process ceases if a student cannot meet the requisite criterion at any level. With its initial position, discrepancy provides the foundation and would be further strengthened if the difference score were based on the most reliable total IQ score and total achievement test score. In this way, a too narrowly focused discrepancy, as in, for example, a comparison between a Performance IQ and a Social Studies achievement subtest, would be eliminated. With such a scheme, a more comprehensive view of LD is achieved along with greater confidence in declaring that a student is "truly" LD.
Discrepancy is an important and legitimate concept applied to LD. Beginning with its status as a measure of educational progress, discrepancy evolved into an index of underachievement. Because LD has always been viewed as a construct associated with underachievement, discrepancy became a necessary component of LD. Although subject to debate about statistical and psychometric properties, discrepancy calculation can be made adequate and defensible for use in LD identification. Because of pragmatic reasons, discrepancy has become the primary LD identification criterion, and this emphasis has led to a number of difficulties, most noticeably the failure to appropriately differentiate LD and LA. When viewed properly, discrepancy is a useful component for LD identification and any presumed problems can be resolved satisfactorily. The most important point is that discrepancy not be used alone for LD identification. Discrepancy is the operational definition of underachievement and, when present, reliably and appropriately documents the presence of underachievement, not LD. With the valid assumption that LD and underachievement are not equivalent, the task becomes one of deciding what other factors need to be considered before there is confidence that LD has been determined. When placed in proper context, any arguments about the use of discrepancy for LD determination would cease. It would, therefore, be an error to eliminate discrepancy when considering the best means of defining the LD construct. The task is one of using discrepancy so that it is not LD itself but rather only part of a more comprehensive identification process.
Funding for NRCLD is provided by the U.S. Department of Education, Office of Special
Education Programs. Award No. H324U010004. Opinions expressed herein are those of the National
Research Center on Learning Disabilities and do not necessarily represent the position of the U.S.
Department of Education.
© 2007 National Research Center on Learning Disabilities
Last Updated: November 19, 2007
Contact: nrcld@ku.edu