The Irrelevance of IQ to the Definition of Learning Disabilities: Some Empirical Evidence Jan Rispens, Tom A. van Yperen, and Gijs A. van Duijn
The relevance oflQ to the definition of learning disabilities is a much-debated issue. In this article, the effect of not using IQ in the identification of children with reading disabilities is demonstrated. Two classification procedures, differing in their use of IQ, are compared. The first conclusion is that abandoning IQ in classification has a very limited impact on the number of children identified as reading disordered. Our data demonstrate that, if IQ is used, more high IQ children are classified. Another finding pertains to the effect of a restriction oflQ range. The number of children classified as reading disordered is a function of the IQ range.
T
he idea of discrepancy is an important feature of the definition of learning disabilities (Siegel, 1989a, 1989b). An analysis of the literature reveals that several procedures are available to measure discrepancy (e.g., Epps, Ysseldyke, & Algozzine, 1983; Lindgren, De Renzi, & Richman, 1985; Reynolds, 1981; Rispens & Van Yperen, 1990). These measurement procedures differ widely, due to (a) a lack of consensus about the relevance of the idea of a discrepancy (Stanovich, 1986), (b) differences in the operationalization of the concept of a discrepancy, and (c) the existence of a host of psychometric problems (Willson & Reynolds, 1984). In the case of the classification of a child with a reading disability, a discrepancy implies that the actual reading score (Yobs.) lags behind the expected reading performance (Yexp.): Yobs. < Yexp. The operationalization of this discrepancy formula is difficult: It is unclear what type of reading task (word recognition, comprehension, or both) is most appropriate to measure actual reading performance. Another problem is the cutoff: the magnitude of the minimal difference between actual and expected performance that is supposed to reflect a significant discrepancy. Finally, an important issue is how to estimate the expected reading performance. With respect to this problem, the existing measurement procedures can be divided into two broad categories, depending on the use of IQ to estimate Yexp. In models
of Category 1, the reading performance (a score on a reading test) is compared with the average score of the child's age group. In other words, the average score of the age group is used as an estimation of Yexp. In these models, IQ is not used. However, in models of Category 2, IQ plays a key part. These models are based on the correlation between IQ and the score on a reading test, with a correction for regression. Several formulas to predict Yexp. are available (Reynolds, 1981). In a recent article, Siegel (1989a) argued that, for several reasons, IQ is irrelevant to the definition of learning disabilities. She proposed dropping IQ from the classification procedure. The only criterion to classify a child as reading disordered is a performance below a certain score on a reading test. Therefore, SiegePs article can be interpreted as a plea in favor of the exclusive application of Category 1 models. SiegePs (1989a) article prompted extensive discussion in the field. It is interesting to note that, in that discussion, not much attention was paid to possible practical consequences of SiegePs suggestion. However, it is well known that the application of different measurement procedures affects the number of children classified as learning disabled (Epps et al., 1983; Forness, Sinclair, & Guthrie, 1983). It is conceivable that SiegePs (1989a) proposal to abandon IQ will result in an increase of the number of children classified as reading disordered.
434
Another effect could be that the IQ range of the group of children identified as learning disabled will be affected and will contain more low IQ children. Siegel (1989a) raises another question with respect to the IQ issue, namely, the problem of the cutoff. It is not uncommon, both in clinical practice and in research, to exclude children with an IQ < 85 from the classification procedure. Siegel questioned this restriction of IQ range. However, the effect of different IQ cutoffs on the number of children classified as reading disordered was not demonstrated. In the present article, we present some empirical evidence with respect to these two questions. Our data stem from an extensive Dutch research project aimed at the improvement of Axis II of the Multiaxial Classification Scheme (MAS), a classification system of childhood disorders (Rutter, Shaffer, & Sturge, 1979). Axis II contains a number of specific developmental disorders, including specific reading disorders (or dyslexia). In order to improve the reliability of the system, we tried to develop explicit diagnostic criteria based on a clear description of each of the disorders. In the MAS, a reading disorder is defined in terms of a discrepancy; therefore, we had to answer the question of whether IQ has to be included in the procedure to measure a discrepancy. In a pilot study, we tested the effect of the application of different classification procedures (with or without IQ) on the number of children classified as reading disordered. Accordingly, our data shed some light on the questions raised by SiegePs (1989a) article. The study is summarized below. (An extensive report can be found in Rispens & Van Yperen, 1990.)
METHOD Procedure IQs and reading scores of 399 children were collected. The discrepancy between Yobs, and Yexp. was computed for each of these children. Two classification procedures were applied: one with IQ and one without IQ; then the number of children identified as reading disabled was determined. This classification Journal of Learning Disabilities
Downloaded from ldx.sagepub.com at Bobst Library, New York University on June 9, 2015
served research purposes only; therefore, it is conceivable that a number of children classified as reading disordered in this study have not (yet) been identified and labeled as such in their school.
Subjects
when both WR and RC are used to compute a discrepancy, the difference is only . 3 % . Table 1 reveals that both models identify nearly the same number of children. Therefore, it seems safe to conclude that dropping IQ from the definition of learning disabilities will result in a relatively small increase or decrease (depending on the type of reading test) of the number of children classified as reading disordered. Our second question pertains to possible effects of Siegel's (1989a) proposal regarding the IQ range of the group of children classified as reading disordered. In a number of scattergrams, we plotted the IQ and the z scores of the reading tests of those children who were classified as reading disordered by application of the two models. Figure 1 depicts one of these scattergrams: IQ and reading comprehension. The scattergram shows that, after the application of Model 1 in the RC condition and then Model 2 in the RC condition, 32 children were classified as reading disordered; 26 of them were identified by both Model 1 and Model 2. Four children were classified by Model 1 (in which the IQ is not used) and not by Model 2; two of them were in the lower IQ range. On the other hand, there were two poor readers who gained a classification when the IQ-based Model 2 was applied; they were not classified by Model 1. We conclude that the overlap between the two models is considerable: In the case of the RC condition, 93% of the children identified by Model 2 were identified by Model 1, as well; in the WR condition, this percentage was 92; in the WR + RC condition, the overlap was somewhat smaller: 88%. This concurrence of both models is another demonstration of the fact that dropping IQ from the identification procedure has a limited impact on the number of children
Seventy-one regular elementary schools, randomly selected from a large town and its surrounding villages located in the eastern part of the Netherlands, were invited to participate in the study. Operationalization. Two models to Nearly half of them (32 schools) decided measure a discrepancy were applied. In to cooperate. Only first and second Model 1, a reading score is converted graders were involved in the study. Noninto a z score and compared with the promoted children, children from nonaverage score of the age group (z = 0). A Dutch families, and children with an IQ discrepancy exists when the individual z of 80, and I Q > 8 5 . This implies that in, for example, the IQ>80 condition, all children with an IQ80 can gain a classification, Model 1 results in 4.5% of our sample classified in the WR condition. In the case of a cutoff of IQ>70, this percentage increases to 6.3% in the WR condition. Our data demonstrate that this problem of IQ cutoff has important practical consequences. It is interesting to note that the choice
436 Downloaded from ldx.sagepub.com at Bobst Library, New York University on June 9, 2015
of the cutoff has at least as much impact on the number of children that gain a classification as dropping IQ from the procedure to measure a discrepancy. For example, in the WR condition and with a cutoff of IQ = 85, Model 1 results in a classification of 3.0% and Model 2 in 3.8%, a difference of .8%. However, if we change the cutoff to IQ = 80, the percentage of children classified as reading disordered increases by 1.5% (Model 1) or 1.2% (Model 2); the difference between the two models remains very small (.5%). In other words, lowering the IQ cutoff results in an increase of the number of children classified as reading disordered (as could be expected). But this effect holds for both models to the same degree. For that reason, it can be argued that the determination of the cutoff is of more importance than the use of IQ in the procedure to measure a discrepancy. Another interesting finding from the present study is that the type of reading task turns out to be a very important variable. Table 3 demonstrates that, although application of a word recognition test or a comprehension test results in classification of nearly the same number of children, the overlap between both groups is far from perfect. For example, Model 1 in the WR condition results in the identification of 25 children with a discrepancy; in the RC condition it results in the identification of 30 subjects. However, this RC group of 30 does not include all 25 children selected in the WR condition. Only 17 of them are classified in both the RC and the WR condition. In the case of Model 2, the overlap was 16 out of 26 (WR) or 28 (RC). From these data it can be concluded that the type of reading task is an important variable. Although there is nearly no effect on the number of children that are classified, our data suggest that different children are identified as a function of the type of reading task that is used in the procedure to measure a discrepancy.
DISCUSSION The results of our study can be summarized as follows: Dropping IQ from the procedure to Journal of Learning Disabilities
ABOUT THE
Table 2 Effect of Restriction of IQ Range on the Percentage of Children Classified as Reading Disabled (/V = 399) Model 2
Model 1
Lower IQ limit IQ > 85 IQ > 80 IQ > 70
Word recognition
Reading comprehension
Word recognition
Reading comprehension
3.0 4.5 6.3
3.5 5.0 7.5
3.8 5.0 6.5
3.8 5.0 7.0
Table 3 Effect of Type of Reading Task
j
Reading comprehension Nondiscrepant Discrepant
Model 1
Model 2
Word recognition Nondiscrepant Discrepant
Word recognition Nondiscrepant Discrepant
361 13
8 17
361 12
10 16
AUTHORS
Jan Rispens received his PhD from the University of Utrecht in 1974. He is currently professor of special education at the University of Utrecht, with a specific interest in dyslexia research. Tom A. van Yperen received his PhD from the University of Leiden. He is assistant professor of special education at the University of Utrecht. His research focuses on the classification of developmental disorders. Gijs A. van Duijn is a specialist in research methodology. He is currently at the University of Leyden. Address: Jan Rispens, University of Utrecht, Faculty of Social Sciences, Department of Child Studies, PO 80.140, 3508 TC Utrecht, The Netherlands.
AUTHORS'
NOTE
The authors gratefully acknowledge the cooperation of Dr. J.J. Dumont, University ofNijmegen, who made available the reading scores and IQ data of the sample.
REFERENCES
measure a discrepancy has a very limited effect on the number of children classified as reading disabled. If different classification models (with or without IQ) are applied to the same data, nearly the same number of children are identified. However, abandoning IQ has an effect on the IQ range: If IQ is not used in the measurement procedure, more lower IQ children will be included and a number of higher IQ children will be excluded. We also demonstrated the importance of the choice of the IQ cutoff. The number of children classified as reading disordered is a function of the IQ cutoff. Finally, we demonstrated the importance of the type of reading task: Different children are classified if different reading tasks are used in the classification procedure. Our study has a number of limitations that may affect the results. In the first place, it is important to note that we considered a discrepancy between Yobs, and Yexp. as significant if z< - 1.65. However, it is possible to argue in favor of another z value. Another important limitation stems from the fact that cases with an IQ
The relevance oflQ to the definition of learning disabilities is a much-debated issue. In this article, the effect of not using IQ in the identification of children with reading disabilities is demonstrated. Two classification procedures, differing in their use of IQ, are compared. The first conclusion is that abandoning IQ in classification has a very limited impact on the number of children identified as reading disordered. Our data demonstrate that, if IQ is used, more high IQ children are classified. Another finding pertains to the effect of a restriction oflQ range. The number of children classified as reading disordered is a function of the IQ range.
T
he idea of discrepancy is an important feature of the definition of learning disabilities (Siegel, 1989a, 1989b). An analysis of the literature reveals that several procedures are available to measure discrepancy (e.g., Epps, Ysseldyke, & Algozzine, 1983; Lindgren, De Renzi, & Richman, 1985; Reynolds, 1981; Rispens & Van Yperen, 1990). These measurement procedures differ widely, due to (a) a lack of consensus about the relevance of the idea of a discrepancy (Stanovich, 1986), (b) differences in the operationalization of the concept of a discrepancy, and (c) the existence of a host of psychometric problems (Willson & Reynolds, 1984). In the case of the classification of a child with a reading disability, a discrepancy implies that the actual reading score (Yobs.) lags behind the expected reading performance (Yexp.): Yobs. < Yexp. The operationalization of this discrepancy formula is difficult: It is unclear what type of reading task (word recognition, comprehension, or both) is most appropriate to measure actual reading performance. Another problem is the cutoff: the magnitude of the minimal difference between actual and expected performance that is supposed to reflect a significant discrepancy. Finally, an important issue is how to estimate the expected reading performance. With respect to this problem, the existing measurement procedures can be divided into two broad categories, depending on the use of IQ to estimate Yexp. In models
of Category 1, the reading performance (a score on a reading test) is compared with the average score of the child's age group. In other words, the average score of the age group is used as an estimation of Yexp. In these models, IQ is not used. However, in models of Category 2, IQ plays a key part. These models are based on the correlation between IQ and the score on a reading test, with a correction for regression. Several formulas to predict Yexp. are available (Reynolds, 1981). In a recent article, Siegel (1989a) argued that, for several reasons, IQ is irrelevant to the definition of learning disabilities. She proposed dropping IQ from the classification procedure. The only criterion to classify a child as reading disordered is a performance below a certain score on a reading test. Therefore, SiegePs article can be interpreted as a plea in favor of the exclusive application of Category 1 models. SiegePs (1989a) article prompted extensive discussion in the field. It is interesting to note that, in that discussion, not much attention was paid to possible practical consequences of SiegePs suggestion. However, it is well known that the application of different measurement procedures affects the number of children classified as learning disabled (Epps et al., 1983; Forness, Sinclair, & Guthrie, 1983). It is conceivable that SiegePs (1989a) proposal to abandon IQ will result in an increase of the number of children classified as reading disordered.
434
Another effect could be that the IQ range of the group of children identified as learning disabled will be affected and will contain more low IQ children. Siegel (1989a) raises another question with respect to the IQ issue, namely, the problem of the cutoff. It is not uncommon, both in clinical practice and in research, to exclude children with an IQ < 85 from the classification procedure. Siegel questioned this restriction of IQ range. However, the effect of different IQ cutoffs on the number of children classified as reading disordered was not demonstrated. In the present article, we present some empirical evidence with respect to these two questions. Our data stem from an extensive Dutch research project aimed at the improvement of Axis II of the Multiaxial Classification Scheme (MAS), a classification system of childhood disorders (Rutter, Shaffer, & Sturge, 1979). Axis II contains a number of specific developmental disorders, including specific reading disorders (or dyslexia). In order to improve the reliability of the system, we tried to develop explicit diagnostic criteria based on a clear description of each of the disorders. In the MAS, a reading disorder is defined in terms of a discrepancy; therefore, we had to answer the question of whether IQ has to be included in the procedure to measure a discrepancy. In a pilot study, we tested the effect of the application of different classification procedures (with or without IQ) on the number of children classified as reading disordered. Accordingly, our data shed some light on the questions raised by SiegePs (1989a) article. The study is summarized below. (An extensive report can be found in Rispens & Van Yperen, 1990.)
METHOD Procedure IQs and reading scores of 399 children were collected. The discrepancy between Yobs, and Yexp. was computed for each of these children. Two classification procedures were applied: one with IQ and one without IQ; then the number of children identified as reading disabled was determined. This classification Journal of Learning Disabilities
Downloaded from ldx.sagepub.com at Bobst Library, New York University on June 9, 2015
served research purposes only; therefore, it is conceivable that a number of children classified as reading disordered in this study have not (yet) been identified and labeled as such in their school.
Subjects
when both WR and RC are used to compute a discrepancy, the difference is only . 3 % . Table 1 reveals that both models identify nearly the same number of children. Therefore, it seems safe to conclude that dropping IQ from the definition of learning disabilities will result in a relatively small increase or decrease (depending on the type of reading test) of the number of children classified as reading disordered. Our second question pertains to possible effects of Siegel's (1989a) proposal regarding the IQ range of the group of children classified as reading disordered. In a number of scattergrams, we plotted the IQ and the z scores of the reading tests of those children who were classified as reading disordered by application of the two models. Figure 1 depicts one of these scattergrams: IQ and reading comprehension. The scattergram shows that, after the application of Model 1 in the RC condition and then Model 2 in the RC condition, 32 children were classified as reading disordered; 26 of them were identified by both Model 1 and Model 2. Four children were classified by Model 1 (in which the IQ is not used) and not by Model 2; two of them were in the lower IQ range. On the other hand, there were two poor readers who gained a classification when the IQ-based Model 2 was applied; they were not classified by Model 1. We conclude that the overlap between the two models is considerable: In the case of the RC condition, 93% of the children identified by Model 2 were identified by Model 1, as well; in the WR condition, this percentage was 92; in the WR + RC condition, the overlap was somewhat smaller: 88%. This concurrence of both models is another demonstration of the fact that dropping IQ from the identification procedure has a limited impact on the number of children
Seventy-one regular elementary schools, randomly selected from a large town and its surrounding villages located in the eastern part of the Netherlands, were invited to participate in the study. Operationalization. Two models to Nearly half of them (32 schools) decided measure a discrepancy were applied. In to cooperate. Only first and second Model 1, a reading score is converted graders were involved in the study. Noninto a z score and compared with the promoted children, children from nonaverage score of the age group (z = 0). A Dutch families, and children with an IQ discrepancy exists when the individual z of 80, and I Q > 8 5 . This implies that in, for example, the IQ>80 condition, all children with an IQ80 can gain a classification, Model 1 results in 4.5% of our sample classified in the WR condition. In the case of a cutoff of IQ>70, this percentage increases to 6.3% in the WR condition. Our data demonstrate that this problem of IQ cutoff has important practical consequences. It is interesting to note that the choice
436 Downloaded from ldx.sagepub.com at Bobst Library, New York University on June 9, 2015
of the cutoff has at least as much impact on the number of children that gain a classification as dropping IQ from the procedure to measure a discrepancy. For example, in the WR condition and with a cutoff of IQ = 85, Model 1 results in a classification of 3.0% and Model 2 in 3.8%, a difference of .8%. However, if we change the cutoff to IQ = 80, the percentage of children classified as reading disordered increases by 1.5% (Model 1) or 1.2% (Model 2); the difference between the two models remains very small (.5%). In other words, lowering the IQ cutoff results in an increase of the number of children classified as reading disordered (as could be expected). But this effect holds for both models to the same degree. For that reason, it can be argued that the determination of the cutoff is of more importance than the use of IQ in the procedure to measure a discrepancy. Another interesting finding from the present study is that the type of reading task turns out to be a very important variable. Table 3 demonstrates that, although application of a word recognition test or a comprehension test results in classification of nearly the same number of children, the overlap between both groups is far from perfect. For example, Model 1 in the WR condition results in the identification of 25 children with a discrepancy; in the RC condition it results in the identification of 30 subjects. However, this RC group of 30 does not include all 25 children selected in the WR condition. Only 17 of them are classified in both the RC and the WR condition. In the case of Model 2, the overlap was 16 out of 26 (WR) or 28 (RC). From these data it can be concluded that the type of reading task is an important variable. Although there is nearly no effect on the number of children that are classified, our data suggest that different children are identified as a function of the type of reading task that is used in the procedure to measure a discrepancy.
DISCUSSION The results of our study can be summarized as follows: Dropping IQ from the procedure to Journal of Learning Disabilities
ABOUT THE
Table 2 Effect of Restriction of IQ Range on the Percentage of Children Classified as Reading Disabled (/V = 399) Model 2
Model 1
Lower IQ limit IQ > 85 IQ > 80 IQ > 70
Word recognition
Reading comprehension
Word recognition
Reading comprehension
3.0 4.5 6.3
3.5 5.0 7.5
3.8 5.0 6.5
3.8 5.0 7.0
Table 3 Effect of Type of Reading Task
j
Reading comprehension Nondiscrepant Discrepant
Model 1
Model 2
Word recognition Nondiscrepant Discrepant
Word recognition Nondiscrepant Discrepant
361 13
8 17
361 12
10 16
AUTHORS
Jan Rispens received his PhD from the University of Utrecht in 1974. He is currently professor of special education at the University of Utrecht, with a specific interest in dyslexia research. Tom A. van Yperen received his PhD from the University of Leiden. He is assistant professor of special education at the University of Utrecht. His research focuses on the classification of developmental disorders. Gijs A. van Duijn is a specialist in research methodology. He is currently at the University of Leyden. Address: Jan Rispens, University of Utrecht, Faculty of Social Sciences, Department of Child Studies, PO 80.140, 3508 TC Utrecht, The Netherlands.
AUTHORS'
NOTE
The authors gratefully acknowledge the cooperation of Dr. J.J. Dumont, University ofNijmegen, who made available the reading scores and IQ data of the sample.
REFERENCES
measure a discrepancy has a very limited effect on the number of children classified as reading disabled. If different classification models (with or without IQ) are applied to the same data, nearly the same number of children are identified. However, abandoning IQ has an effect on the IQ range: If IQ is not used in the measurement procedure, more lower IQ children will be included and a number of higher IQ children will be excluded. We also demonstrated the importance of the choice of the IQ cutoff. The number of children classified as reading disordered is a function of the IQ cutoff. Finally, we demonstrated the importance of the type of reading task: Different children are classified if different reading tasks are used in the classification procedure. Our study has a number of limitations that may affect the results. In the first place, it is important to note that we considered a discrepancy between Yobs, and Yexp. as significant if z< - 1.65. However, it is possible to argue in favor of another z value. Another important limitation stems from the fact that cases with an IQ
