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 |
In a similar comparison of LD and LA groups that also included comparisons with an MR group defined as IQ < 75, Gresham, MacMillan, and Bocian (1996) found an average LD-LA level of differentiation of 61% (compare with the 63% reported by Kavale et al. 1994). The differentiation levels for LD-MR and LA-MR averaged 68.5% and 67.5%, respectively. On achievement measures, LD-LA group comparisons revealed an average ES of 0.39 indicating a 65% level of differentiation and confirmation of the finding that "LD children performed more poorly in academic achievement than LA children" (p. 579). The LD group performed most poorly in reading, where almost 3 out of 4 students with LD could be reliably differentiated from LA students. The large achievement differences in reading between LD and LA groups were affirmed by Fuchs, Fuchs, Mathes, and Lipsey (2000) who found that 72% of the LA group performed better in reading than the LD group . Even larger ES differences were found with more rigorous measures, "suggest[ing] that researchers and school personnel in fact do identify as LD those children who have appreciably more severe reading problems compared to other low-performing students who go unidentified" (p. 95).
Gresham, MacMillan & Bocian (1996) also investigated cognitive ability (IQ) differences among the three groups. As expected, 94% of the LD group could be reliably differentiated from the MR group. The percentage fell to 73% in differentiating LD and LA groups, suggesting greater cognitive ability overlap between these two groups. Gresham et al., however, included an LA group defined differently from both the Ysseldyke, Algozzine, Shinn, & McGue (1982) and B. A. Shaywitz et al. (1992) studies: "Our LA group was closer to what might be considered a 'slow learner' group on the basis of their average-level intellectual functioning relative to the LA groups in [the other] studies" (p. 579). The result was that even though achievement was depressed, it was not discrepant when compared to IQ level. In contrast, the LD group revealed significant discrepancies and was thus properly classified because "Children with LD perform more poorly in reading than LA children, even when the former group has higher cognitive ability" (p. 580). This finding has been confirmed by Short, Feagans, McKinney, and Appelbaum (1986) in an analysis of LD subtypes. In examining reading achievement across five groups, they found that "the joint application of IQ- and age-discrepancy criteria appeared to be useful for distinguishing between seriously disabled students and those who might be more appropriately classified as slow learners or underachievers" (p. 223). In summary, Gresham, MacMillan, & Bocian (1996) concluded that LD, LA, and MR groups "could be reliably differentiated using measures of cognitive ability and tested academic achievement" (p. 580). When LD is defined with an ability-achievement difference criterion, the resulting discrepancy appears to be an appropriate metric that permits reliable differentiation between LD and LA groups.
Although empirical evidence appeared to indicate that LD and LA could be reliably differentiated with a discrepancy criterion, questions about its use continued. One form of questioning focused on IQ and whether it was necessary in defining LD. Beginning with the finding that IQ was not useful in locating students with reading disability (L. S. Siegel, 1988), questions arose about whether or not IQ was a necessary component in definitions of LD (L. S. Siegel, 1989, 1990). A major problem surrounded IQ tests and what they presumably measure. Stanovich (1991b) concluded that "an IQ test score is not properly interpreted as a measure of a person's potential" (p. 10). Yet, "the LD field has displayed a remarkable propensity to latch onto concepts that are tenuous and controversial....The LD field seems addicted to living dangerously" (Stanovich, 1989, p. 487). At a practical level, for example, there was controversy about what type of IQ score should be used in discrepancy calculation. Although it was commonly recommended that performance or nonverbal IQ be used (e.g., Stanovich, 1986a; Thomson, 1982), an equally compelling case could be made for the use of verbal IQ (e.g., Hessler, 1987). Without resolution about what IQ tests actually measure, "IQ is a superordinate construct for classifying a child as reading disabled. Without clear conception of the construct of intelligence, the notion of a reading disability, as currently defined, dissolves into incoherence" (Stanovich, 1991a, p. 272).
The ability-achievement discrepancy criterion treats intelligence and achievement as separate and independent variables, but L. S. Siegel (1989) suggested that this may not be valid because "A lower IQ score may be a consequence of the learning disability, and IQ scores may underestimate the real intelligence of the individual with LD" (p. 471). Further confounding was introduced by findings that the IQ of students with LD may actually decline over time (Share & Silva, 1987; Van den Bos, 1989). If this is a valid finding and also assuming that students remain close to their original reading levels over time (see Juel, 1988), then discrepancies should increase over time, but McLeskey (1992) found a negative association between discrepancy level and CA where "students in the elementary grades were most likely to manifest a severe discrepancy between expected and actual achievement, while high school students were least likely to have such a discrepancy" (p. 18).
A partial explanation may be found in what Stanovich (1986b) termed the "Matthew effect," referring to the Biblical statement (Matthew 13:12) that suggests that each advantage leads to further advantage, or conversely, initial disadvantage multiplies into even more disadvantage. For reading, this means that the poor get poorer: "Children with inadequate vocabularies--who read slowly and without enjoyment--read less, and as a result have slower development of vocabulary knowledge, which inhibits further growth in reading ability" (p. 381). B. A. Shaywitz et al. (1995), however, found no evidence of a Matthew effect in reading but a modest Matthew effect for IQ in a large-scale LD sample. For both IQ and reading, however, "the influence of the regression-to-the-mean effect tends to mask the relatively small Matthew effect" (p. 902) which suggests that the presumed cumulative disadvantage (Matthew effect) really refers to the rate of gain or loss in reading ability compared to initial level (see Walberg & Tsai, 1983). There are thus complex reciprocal relationships between reading ability and cognitive skills that are seen to confound the discrepancy notion because "the logic of the learning disabilities field has incorrectly assigned all the causal power to IQ. That is, it is reading that is considered discrepant from IQ rather than IQ that is discrepant from reading" (Stanovich, 1991b, p. 275).
The problem of confounding is most likely to arise in situations where concepts are defined with dual criteria. For example, although the psychometric characteristic IQ has long defined MR (e.g., Hollingworth, 1926), there was a later decision to include a second criterion in the form of adaptive behavior: the effectiveness and degree to which individuals meet standards of self-sufficiency and social responsibility (Heber, 1959). There was, however, concern over the inclusion of adaptive behavior in the MR definition primarily because of measurement issues (Clausen, 1972; MacMillan & Jones, 1972). Specifically, there were no adequate instruments to evaluate adaptive behavior that made it a psychometric characteristic comparable to IQ. [Of course, this situation was remedied with instruments like the American Association on Mental Retardation (AAMD) Adaptive Behavior Scale and the Vineland Social Maturity Scale]. With only one measure acceptable, there would be no means to evaluate both criteria, and this situation would create the possibility of students identified as MR who did not meet the dual criteria definition as well as students not identified who would meet the definition if appropriate assessments for both criteria were available.
When reliable and valid assessments are not available, clinical judgment was likely substituted but was often equally unreliable, especially in the "milder" regions of MR. With significant impairment in intellectual ability (IQ < 50), the corresponding adaptive behavior was probably equally impaired and not difficult to judge. As the upper limit of the IQ criterion was approached (IQ 70-75), however, the probability that adaptive behavior would correspond similarly decreased and clinical judgment became more problematic.
In defining LD, Kavale and Forness (1985) recommended a dual criteria definition similar to MR that included (a) significant subaverage academic impairment and (b) IQ in the average range. The advantage would be that both criteria can be reliably measured and little clinical judgment would be necessary. The two criteria can be readily compared and decision rules adopted to determine when the obtained difference ("discrepancy") was significant. If an additional exclusion criterion was added, then the identification process would avoid the myriad difficulties surrounding attempts to include other definitional parameters (e.g., psychological process deficits, central nervous system dysfunction) that cannot be reliably assessed. For this reason, IQ remains an important component in LD definition.
DEFINING LEARNING DISABILITY WITHOUT INTELLIGENCE
Even though IQ should be considered a necessary criterion, L. S. Siegel (1989) suggested that the LD field "abandon the use of the IQ test in the definition of learning disabilities....[T]he IQ-achievement deviation definition should be abandoned because of its illogical nature" (p. 477). Stanovich (1989) suggested, however, that such a position might be "too extreme" (p. 489) and "perhaps ends up saying too little about too much" (p. 490). Lyon (1989) concluded that "Siegel has raised some interesting and compelling issues but has confounded her position by taking a narrow conceptual and methodological stance in addressing the relationship between intelligence and the LD definition" (p. 506). Baldwin and Vaughn (1989) suggested that "Siegel's position might be illogical because the reasoning was convoluted and misleading" (p. 513).
Meyen (1989) objected to the suggestion that IQ should be eliminated in the LD definition because "challenging the use of intelligence measures in defining learning disabilities, in essence, questions the efficacy of the category of learning disabilities itself as a means to identify students who warrant special education services" (p. 482). By eliminating IQ, a situation would be created where "we would largely serve low achievers and have no basis for determining whether or not a student is achieving at a reasonable level given his or her ability" (p. 482). The result would be an even more contentious LD-LA differentiation debate. The situation would not be remedied with a different IQ cut-off score which L. S. Siegel (1989) suggested as an alternative solution. In applying the discrepancy criterion in LD identification, there has long been the implicit assumption that IQ is at an average or above level in order to "discriminate between poor achievement that is expected (that is, on the basis of intellectual ability or sensory handicaps) and poor achievement that is not expected (that is, the probable presence of LD)" (Scruggs, 1987, p. 22). With an IQ cutoff of, for example, 75 (a level closer to the MR criterion) less than average academic achievement would be neither unexpected nor unexplained. There may be a need for special education, but such a student would not be properly classified as LD.
The primary difficulty with a lower IQ cutoff score in defining LD would be the potential confounding with MR. The AAMD (see Grossman, 1973) shifted the upper cutoff score for MR from -1 to -2 SD, that is, an IQ level of 70 instead of 85. Grossman (1983) later suggested the IQ cutoff could be as high as 75 since IQ should be viewed as only a rough guideline. Thus, cutoff scores really represent arbitrary statistical decisions rather than being based on scientific classification procedures (Zigler & Hodapp, 1986). Such arbitrary decisions create real dilemmas because they cause widely varying prevalence rates. For example, Reschly (1992) demonstrated that the use of an IQ cutoff of 75 and below results in twice as many individuals potentially eligible than would using IQ 70 and below. In addition, more cases fall in the interval 71-75 than in the entire range associated with mild MR (IQ 55-70). For LD with a 75 IQ cutoff, an additional 22.5% of the population would be eligible (given an "average" IQ level arbitrarily defined at 92.5) with perhaps 3% of this group potentially eligible for either MR or LD. With a discrepancy criterion, eligibility for LD can also be defined in SD units similar to MR (-1 to -2 SD depression) (see Mercer et al., 1996). As with MR, however, the choice of criterion level remains arbitrary and will also affect prevalence: the smaller the required discrepancy, the larger the prevalence. The current high prevalence rate for LD suggests a decision including smaller discrepancy levels, but the resulting LD classifications also suggest an increased probability of confounding with MR.
The consequences of the confounding between LD and MR are seen in large variations across states in prevalence rates with the typical outcome being more LD and less MR than expected (U.S. Department of Education, 1999). Gresham et al. (1996) showed that the percentage of students classified as MR was inversely related to the percentage of students classified as LD (r = -0.24). Thus, states serving a small percentage of students with MR classify a larger percentage of students as LD, and vice versa. It is entirely possible then that students with similar cognitive abilities and disabilities are served in one state as LD and in another as MR (MacMillan, Siperstein, & Gresham, 1996).
Although average or above IQ has been considered a prerequisite for LD, a longstanding view holds that average or above intelligence is not a necessary or desirable criterion (e.g., Ames, 1968; Belmont & Belmont, 1980; Cruickshank, 1977). Support for this view was found in large-scale evaluations of LD populations that have found mean IQ levels in both the low average (IQ 80-90) range (e.g., Koppitz, 1971; Smith, Coleman, Dokecki, & Davis, 1977; J. D. Wilson & Spangler, 1974) and the lower regions of the average (IQ 90-100) range (e.g., Kirk & Elkins, 1975; McLeskey & Waldron, 1990; Norman & Zigmond, 1980). In addition, IQ levels of students with LD tended to be quite variable, and anywhere from 10% to 40% of LD samples were found to have IQ scores falling below 85 (e.g., Gajar, 1980; Kavale & Reese, 1992; Shepard et al., 1983).
To explain why the actual IQ level of students with LD might be below average, Burns (1984) used the bivariate normal distribution to show how LD samples can have average IQ scores well below 100. With the known relationship between IQ and achievement, the average IQ of LD samples will decrease as the correlation between IQ and achievement increases. For example, if cases below a given cutoff for achievement (e.g., z < -1.0) and above a given IQ cutoff (e.g., IQ > 80) are considered while postulating a correlation of 0.50 between IQ and achievement, then the average IQ of a sample on the bivariate normal distribution will be about 93. Piotrowski and Siegel (1986), however, suggested that using the bivariate normal distribution to explain mean IQ levels less than 100 for LD samples may not be appropriate. The primary difficulty was found in the use of fixed achievement cutoff scores regardless of IQ score, as achievement is likely to vary as a function of both MA and CA. For example, a student with an IQ of 80 and achievement z-score of -0.05 would meet the LD discrepancy criterion under the bivariate normal distribution, but, in reality, demonstrate almost no underachievement. Conversely, a student with an IQ of 130 and achievement z-score of -0.95 would in fact be underachieving significantly but would not meet the discrepancy criterion for LD. These problems are compounded further as the correlation between IQ and achievement increases. Finally, the bivariate model requires IQ scores to be normally distributed, but this is unlikely given the finding that the IQ of students with LD reveals less stability over time (Kaye & Baron, 1987; Lally, Lloyd, & Kulberg, 1987).
With a proportion of the LD population showing IQ levels falling more than 1 SD below the mean, this group would, at one time, be considered as having borderline MR (see Heber, 1961). As such, this group would qualify under the rubric "slow learners" and likely manifest generalized academic deficiencies. The essential question: Is this group also LD? In some instances the answer might be affirmative, but the majority of this group would probably exhibit academic deficits across all achievement domains that would run counter to the assumption that students with LD exhibit achievement deficits in one or more (but not all) academic areas. When all academic achievement areas are equally depressed, the notion of specificity, in the sense of the presence of intra-individual differences, would not be achieved, even though the idea that LD results from a circumscribed set of problems that interfere selectively with academic performance has received support (Stanovich, 1986a). Thus, instead of specific LD (as defined in the federal definition), there is a more generalized LD, a concept closer to that defined by MR particularly at the borderline levels.
The "unexpected" failure idea often associated with LD has been the source of other concerns about IQ and LD. When identified as LD, a student presumably possesses average or above IQ and meets the discrepancy criterion which then suggests that the cause of the academic problems cannot be attributed to low intelligence. On the other hand, the academic deficiencies of slow learners should not be surprising because the demonstrated achievement problems are consistent with the lower than average intellectual ability. These differences suggest that the etiology of the two conditions is really not the same, and consequently, LD and LA groups appear to possess quantitative and qualitative differences.
The origins of assumptions about possible qualitative differences between LD and LA can be found in the Isle of Wight epidemiological studies (Rutter & Yule, 1975; see also Rutter & Yule, 1973; Yule, Rutter, Berger, & Thompson, 1974). Essentially, the LA sample of poor readers was differentiated into two groups: general reading backwardness (GRB) and specific reading retardation (SRR). The GRB group was defined as reading below expected CA (i.e., no discrepancy between IQ and achievement) while SRR was defined as reading below grade level predicted from IQ (i.e., the presence of an IQ-achievement discrepancy).
In analyzing the population, Rutter and Yule (1975) found that while IQ scores were approximately normally distributed, reading achievement scores did not show the same normal distribution because, at the lower end of the distribution, there was a "hump" indicating the presence of a greater proportion than the 2.3% expected in a normal distribution. This "hump" contained the SRR group whose problems were viewed as "specific" to the reading process. As Yule et al. (1974) suggested, "Extreme underachievement in reading occurs at appreciably above the rate expected on the basis of a normal distribution and so constitutes a hump at the lower end of the Gaussian curve....There are no grounds for believing that the hump is anything but a true finding, and the finding implies that there is a group of children with severe and specific reading retardation which is not just the lower end of a normal continuum" (p. 10, emphasis in original).
Rutter and Yule (1975) concluded that, in addition to IQ differences, "Reading retardation is shown to differ significantly from reading backwardness in terms of sex ratio, neurological disorder, pattern of neurodevelopmental deficits and educational prognosis. It is concluded that the concept of specific reading retardation is valid" (p. 195). Rutter (1978) later affirmed the GRB-SRR distinction and the possibility of etiological differences particularly as manifested in the minimal brain dysfunction syndrome (Clements, 1966).
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