Other forms of discrepancy

The problems with models based on aptitude-achievement discrepancies are well known. Two major meta-analyses have shown that effect sizes on measures of achievement and cognitive functions are in the negligible to small range (at best) for the comparison of groups formed on the basis of discrepancies between a composite (Full Scale) IQ and reading achievement versus groups of poor readers without an IQ-discrepancy (Hoskyn & Swanson, 2000; Stuebing et al., 2002). Moreover, other validity studies have not found that groups of poor readers formed on the basis of the presence and absence of composite IQ-achievement discrepancies differ in long-term prognosis (Francis et al., 1996; Silva et al., 1985), response to intervention (see Fletcher et al., 2001; Jimenez et al., 2003; Stage et al., 2003), or neuroimaging correlates (see Lyon, Fletcher, & Barnes, 2003; but also see Shaywitz et al., 2003, which shows differences in groups varying in IQ, but not due to IQ-discrepancy). Studies of genetic variability show negligible to small differences related to IQ-discrepancy models that may reflect regression to the mean (Pennington et al., 1992; Wadsworth et al., 2001). Despite the evidence showing weak validity when comparisons of underachievers based on IQ-discrepancy versus poor reading with no discrepancy (low achievers), alternatives based on discrepancy models continue to be proposed. Finally, similar empirical evidence can be cited for LD in math and language (Fletcher et al., 2002; Mazzocco & Myers, in press), which is not surprising given the problems with the underlying psychometric model.

The problems with the underlying psychometric model have been systematically outlined since IQ-discrepancy was first put into federal regulations in 1975 (Christensen, 1992). Most recently, Francis et al. (in press) showed that alternative models based on discrepancy are not viable, reflecting the underlying psychometric difficulties with any discrepancy model. However, there is also empirical evidence pertaining to validity. In the Stuebing et al. (2002) meta-analysis, 32 of the 46 studies had a clearly defined aptitude measure. Of these studies, 19 used Full Scale IQ, eight used Verbal IQ, four used Performance IQ, and one study used a discrepancy of listening comprehension and reading comprehension. In addition, Fletcher et al. (1994) and a study not eligible for inclusion in Stuebing et al. (Stanovich & Siegel, 1994) both evaluated the validity of different discrepancy models by systematically manipulating Full Scale IQ, Verbal IQ, and Performance IQ. Fletcher et al. (1994) also included a listening comprehension-reading comprehension discrepancy model, which continues to be proposed as an alternative identification model for children with reading disabilities (Joshi, in press).

Not surprisingly, these different operationalizations of discrepancy models did not yield results that were different from those apparent when a composite IQ measure is utilized. As Fletcher et al. (1994) showed, the effect of using different aptitude measures in a regression discrepancy model is to simply shift the slope upwards or downwards depending on the correlation of the aptitude and achievement measures. Thus, for example, using Verbal IQ produces a much steeper regression line than Performance IQ because Verbal IQ is much more highly correlated (.70) with reading outcomes than Performance IQ (.58). This has the effect of shifting individuals who are at the edges of the regression cutpoint on one aptitude measure to either a discrepancy or low achievement subgroup when a different aptitude measure is employed. In a sample where there is control for variation in the cutpoint used to identify people with LD, the overlap in terms of who is identified as an underachiever is substantial, usually exceeding 80% (Fletcher et al., 1994). However, exactly who becomes discrepant or low achieving will depend on the slope of the regression line. As the changes in identification reflect fluctuations around the regression cut off, it is not surprising that effect sizes comparing poor readers with and without IQ-discrepancies are uniformly low across these different models. In Fletcher et al. (1994), the use of cognitive tests that were not part of the definition produced effect sizes that were generally below .10. Stanovich and Siegel (1994) found some differences on cognitive measures outside the language domain based on the use of Verbal IQ versus Performance IQ, although the magnitude of differences fluctuated. Neither Fletcher et al. (1994) nor Aaron et al. (1988) were able to demonstrate major differences between discrepancy and low achievement groups formed on the basis of listening comprehension-reading comprehension.

The real problem is the idea that a discrepancy model will produce differences between children with different forms of underachievement. A discrepancy model cannot possibly produce a clearly unique set of underachievers. None of the eight studies in Stuebing et al. (2002) that systematically explored Verbal IQ found any differences on measures closely related to reading, such as phonological awareness, findings also recently reported by Stage, Abbott, Jenkins, and Berninger (2003). As Francis et al. (in press) have shown, the reliability of classification models based on any form of discrepancy is not adequate to produce acceptable validity. Altogether, models based on aptitude-achievement discrepancies do not appear to identify a unique group of underachievers, and therefore do not adequately operationalize the construct of LD.

Previous Page | Next Page
(Introduction) | (Low achievement models)