Joseph K. Torgesen, Florida State University
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 | 8 |
To be considered as an improvement over current diagnostic procedures for children with learning disabilities, any alternative must meet several important criteria. First, it must support identification of children with learning disabilities before their academic failure has progressed to the point that it begins to have motivational/emotional consequences and produce secondary knowledge and skill deficits (Cunningham & Stanovich, 1998). We know enough about the advantages of early intervention to assert that whatever diagnostic criteria are selected, they should facilitate intervention to prevent children with learning disabilities from falling seriously behind their age peers in critical academic skills.
Second, new diagnostic criteria should support the delivery of appropriate instruction to all children, not just those who show an arbitrary level of discrepancy between one set of learning abilities and another. For example, current evidence suggests that all children who have weaknesses in phonological abilities require more explicit instruction in this area in order to learn to read (Foorman & Torgesen, in press). Further, level of discrepancy between general intelligence and phonological ability is not a powerful or unique predictor of how well children will profit from this type of instruction (Torgesen et al., 1999; Vellutino et al., 1996). In other words, one cannot argue that children who do not show a discrepancy between their phonological processing abilities and other cognitive abilities (i.e., general intelligence) do not have a very important learning disability that affects their ability to acquire accurate and fluent word-level reading skills. They clearly have such a disability, but they are excluded from services under procedures that require an aptitude-achievement discrepancy for identification.
Finally, new diagnostic procedures must meet broadly acceptable standards for psychometric reliability and validity. That is, the criteria established should be those that can be measured with reasonable reliability, and they should also be conceptually and empirically consistent with current definitions. If indicators of potential failure are used to identify children at risk for the development of learning disabilities once formal school instruction has begun, these indicators must have sufficient predictive validity to warrant their widespread use in early identification.
I made the point earlier that measures of phonological awareness are not direct measures of intrinsic or constitutionally based psychological processing weaknesses. Rather, individual differences in phonological awareness reflect both the operation of biologically based processing abilities and the learning opportunities to which a child has been exposed. On the one hand, if a child has a weakness in intrinsic phonological processing capability, phonological awareness will be weak in spite of ample preliteracy learning opportunities. On the other hand, if a child's preschool environment does not provide the kind of experiences that stimulate growth of beginning levels of phonological sensitivity and awareness, the emergence of phonological awareness will be delayed even if phonological processing abilities are relatively intact.
Even though measures of phonological awareness do not directly assess an intrinsic processing weakness, they are markers for the presence of a pre-literacy skill that is critical in learning to read. The same could be said for rapid automatic naming tasks; they are markers for a functional capability (arising from an interaction between intrinsic processing capabilities and experience) that is causally related to early reading growth. There is also substantial evidence that simple knowledge of letter-sound relationships in kindergarten, or the ability to "invent" phonetic spellings for words, has the same or even greater predictive power (Mann & Ditunno, 1990; Scarborough, 1998) for later reading growth. Thus, outcomes on these pre-reading skills are markers for early failure to acquire skills that are critical to the process of deciphering print. In essence, variability on these markers reflects the operation of both basic (intrinsic) processing capabilities and learning opportunities in the child's environment. Both the ability to acquire these skills and the actual presence of the skill itself in sufficient strength are predictive of response to future instruction in reading.
The alternative diagnostic scheme proposed here would facilitate early intervention through assessment of reliable and valid predictors of future difficulties acquiring essential academic skills. Measurement of these marker variables would allow us to identify children in need of more powerful instruction in a particular domain. Children would be initially identified for this special preventive instruction because they met some criteria of low performance on these marker (predictor) variables and were not classifiable as mentally retarded. The label learning disabled would not be assigned until some later point in development (perhaps 2nd or 3rd grade, or even later), but in the meantime, every child who was determined to require special instruction in reading, math, or writing on the basis of low performance on these marker variables, and who was not mentally retarded, would be eligible for special instructional services designed to maintain the child's academic growth (e.g., reading or pre-reading skills) within normal limits. During the period of early intervention (and before labels were assigned), response to instruction would be periodically assessed to examine the continuing need for the assignment of at-risk status and the associated special interventions to which the child had been assigned. Thus, every child who was failing to acquire critical pre-academic or academic skills at acceptable levels, and who was not classified as mentally retarded or some other primary classification, would be eligible for special education services under learning disabilities regulations. These children would not be officially labeled as learning disabled until later in development, but would have at-risk status and be eligible for services until their achievement fell within normal limits or they were officially labeled as learning disabled.
At whatever point in development it is judged proper to assign the official label of learning disabled, this designation would be applied to any child who fell below designated levels on measures of the learning outcomes specified in the definition and regulations and who also had general intelligence above some agreed-upon level. This level should probably be the same as the criteria for the diagnosis of mental retardation (i.e., IQ above 70), so that there would be continuity with earlier procedures for determining at-risk status and so that we would not automatically create a category of children critically behind in academic skills but who "fall through the cracks" between diagnostic categories. Additional processing or non-academic cognitive assessments would be part of the diagnostic criteria for learning disabilities only if it is clearly established that they provide information critical to further instruction, or if they predict future academic growth beyond the predictive power of the child's current academic levels.
The diagnostic and classification model I am proposing here is a combination of early assessment of marker variables for academic failure combined with ongoing assessment to determine response to treatment. At this point, I want to be very clear about one thing. This model will not guarantee that only children with intrinsic psychological processing disabilities will be identified as learning disabled. In fact, there is no practical way to do that on a large-scale basis at present. Stanovich and Siegel (1994) make this point in a powerful way when they sum up evidence against using IQ-discrepancy procedures to classify children as learning disabled:
...neither the phenotypic nor the genotypic indicators of poor reading are correlated in a reliable way with IQ discrepancy. If there is a special group of children with reading disabilities who are behaviorally, cognitively, genetically, or neurologically different, it is becoming increasingly unlikely that they can be easily identified by using IQ discrepancy as a proxy for the genetic and neurological differences themselves. Thus, the basic assumption that underlies decades of classification in research and educational practice regarding reading disabilities is becoming increasingly untenable. (p. 48)
Some would argue (Vellutino et al., 1996) that only children who do not respond adequately to well designed instruction can be considered classically learning disabled (in the sense that they have fundamental processing limitations). This is simply not true. For example, failure to respond to interventions could be the result of factors other than intrinsic processing deficits that are either not understood or not measured for each child. The only way to rule this out is to be completely sure one has accurately measured the entire knowledge and skill domains, as well as the motivational and emotional domains, as well as the environmental domains (support for learning outside the immediate instructional situation) that are relevant to achievement in the area being instructed. In a recent study of intensive preventive instruction in early reading skills, we (Torgesen et al., 1999) found that the three best independent predictors of response to the intervention were beginning levels of phonological processing ability, socioeconomic status of the child's parents, and classroom teacher ratings of attention and behavior. Further, the intrinsic processing disabilities that cause academic failure are almost certainly normally distributed in terms of their severity (Shaywitz, Escobar, Shaywitz, Fletcher, & Makuch, 1992). Children with mild intrinsic processing disabilities will respond to more intensive and explicit instruction, and those with more severe problems will respond less well. A good response to excellent instruction does not mean that the child does not have a constitutionally based processing disorder; it just means the particular instruction the child received was powerful enough to compensate for it.
In principle, any methodology that uses response to treatment as a way of classifying children as learning disabled has no greater chance of correctly identifying children with intrinsic learning disorders than do traditional assessment procedures. The children identified by the response-to-treatment method will be those who are most difficult to teach, no matter what the reason. For example, some estimates suggest that the variability in pre-school exposure to literacy learning opportunities can vary by as much as 1,000 hours in children from different home environments (Adams, 1990). If these estimates are close to being correct, and unless we can measure all the effects of the environmental difference before instruction begins, we cannot tell whether problems in responding to an intervention in kindergarten are the result of constitutionally based processing weaknesses or to unspecified weaknesses in the knowledge domains most relevant to the task being learned.
No method of educational or psychological assessment currently available can identify with certainty children who have intrinsically based psychological processing disorders. However, children who continue with severe learning difficulties after several years of appropriate early intervention are the ones most likely to have this kind of enduring learning disability. Thus, within present assessment capabilities, the method most likely to reliably identify the kind of children who are described in widely accepted definitions of learning disabilities involves early identification with process/outcome markers followed by careful monitoring of growth on critical skills in response to appropriate and consistent early interventions. This model clearly implies that our methods of early identification and monitoring will develop and change as we learn more about the developmental course of each of the kinds of learning disability outcomes described in current definitions. If the federal government were to specify which early markers of pre-academic development can be used to identify children for at-risk status to be served under learning disabilities regulations, this set of process/outcome markers would need to be periodically updated as new knowledge about emergent indicators of learning disabilities is developed.
Two immediate points of vulnerability and difficulty with the classification model just presented are current levels of accuracy in identifying children at risk for learning disabilities and problems ensuring that children who are identified as at risk receive appropriate, research-based instruction delivered with sufficient intensity and skill. The model can easily break down, and create many difficulties for schools, children, and families, if identification for preventive instruction is not reasonably accurate and if preventive interventions are not optimal. There are a few facts relevant to these two issues to guide formation of policy in this area. First, if we want to ensure that a very high proportion of children at risk for the most serious reading difficulties (e.g., the bottom 10%) are identified in kindergarten for preventive instruction, we must be prepared to provide preventive instruction to more than 10% of children.
Two kinds of errors can be made when identifying children at risk for future reading failure. False positive errors are made when children who will eventually become good readers score below the cut-off score on the predictive instrument and are falsely identified as at risk. In general, the proportion of this type of error has ranged between 20% and 60%, with an average of around 45% (Catts, 1996; Scarborough, 1998). That is, almost half of the children identified during kindergarten as at risk turn out not to have serious reading problems by the end of first grade. False negative errors occur when children who later exhibit reading problems are identified as not being at risk. Typical percentages of false negative errors range from 10% to 50%, with an average of around 22%. That is, on average, current procedures fail to identify about 22% of children who eventually end up with serious reading difficulties (Catts, 1996; Scarborough, 1998).
In any given study, the relative proportion of false positive and false negative errors is somewhat arbitrary, since it depends on the level of the cut-off score. For example, we (Torgesen & Burgess, 1998) reported a significant reduction in the percentage of false negative errors within the same sample of children by doubling the number of children we identified as at risk. Our goal was to identify, during the first semester of kindergarten, the children most at risk to be in the bottom 10% in word reading ability by the beginning of second grade. When we selected the 10% of children who scored lowest on our predictive tests, our false negative rate was 42% (we missed almost half the children who became extremely poor readers). However, when we identified the 20% of children who scored lowest on our measures, the false negative rate was reduced to 8%. As a practical matter, if schools desire to maximize their chances for early intervention with the most impaired children, they should provide this intervention to as many children as possible. This is less of a waste of resources than it might seem at first glance, because, although many of the falsely identified children receiving intervention may not be among the most seriously disabled readers, almost all of them are likely to be below-average readers (Torgesen & Burgess, 1998).
It is also important to note that prediction accuracy increases significantly the longer a child has been in school. Prediction of reading disabilities from tests given at the beginning of first grade is significantly more accurate than from tests administered during the first semester of kindergarten (Scarborough, 1998; Torgesen, Burgess, & Rashotte, 1996). Given the widely varying range of children's pre-school learning opportunities, many children may score low on early identification instruments in the first semester of kindergarten simply because they have not had the opportunity to learn the skills. However, if pre-reading skills are actively taught in kindergarten, some of these differences may be reduced by the beginning of the second semester of school. Accuracy of identification of at-risk students can potentially be increased to 100% by frequent assessments of critical pre-reading and reading skills during the early elementary years. A model such as that established in Texas using the Texas Primary Reading Inventory (Texas Education Agency, 2000), in which a combination screening/assessment instrument is administered three times a year during kindergarten through 2nd grade will guarantee that any child who falls critically behind in important early literacy skills will be identified for extra supportive instruction.
The examples of assessment issues provided here have focused on reading, because that is the area we know the most about. However, Berninger and her colleagues (Berninger, in press; Berninger, Stage, Smith, & Hildebrand, 2001) have demonstrated the effectiveness of a "3-Tier Model for Prevention and Remediation" that involves early assessment to identify children at risk for difficulties in writing and math. The 3-Tier model is actually quite similar to the model being proposed here, except that it has an additional layer of intervention at the classroom level. In the simpler model I am proposing, I am assuming that classroom teachers are doing all they can to deliver high-quality, research-based instruction to all children, and that they are actively trying to accommodate individual differences in response to their instruction. If this is not the case, there will be far too many children requiring services under the learning disabilities regulations for the system to work effectively (Foorman & Torgesen, in press).
What do we know about the effectiveness of early interventions in preventing serious reading disabilities? We know, for example, that the best preventive interventions tested in research thus far typically reduce the percentage of children who are continuing be at risk for reading failure (defined as falling below the 30th percentile on critical word reading skills) at the end of first or second grade to about 2% to 6% of the population (Torgesen, 1999). We also know a great deal about the characteristics of effective instruction for children with learning disabilities (Foorman & Torgesen, in press; Swanson, Hoskyn, & Lee, 1999; Vaughn, Gersten, & Chard, 2000), and we know that they will frequently require instruction that is much more intensive and systematic than typical children if they are to attain reading levels within the normal range (Torgesen et al., 2001). One of the major challenges for politicians, school administrators, teachers, and parents in the model I have presented would be to ensure that all children who are at risk for learning disabilities receive appropriate and skillful instruction delivered with the right intensity for sufficient periods of time. A further challenge would be to ensure even greater levels of intensity and skill in instruction for children who do not respond successfully to the first layers of intervention.
As an example of what can be accomplished if excellent classroom instruction in reading is supplemented with more intensive instruction for children identified as at risk for reading failure, consider what happened at Hartsfield Elementary School over a period of 5 years (King & Torgesen, 2000). Hartsfield Elementary School serves a mixed population of school children of whom about 65% qualify for free and reduced lunch services and of whom about 65% are minority (primarily African-American). In the first year of the multiyear change project when only partially improved classroom instruction in reading was accomplished, 32% of the children obtained scores below the 25th percentile on a nationally standardized measure of word-level reading skills at the end of first grade. Once classroom instruction was more consistently high-quality and early identification procedures were in place, only 3.7% of the children fell below the 25th percentile at the end of first grade, and only 2.4% fell below this mark in second grade. In the present model, it would be those 2.4% of children who were still struggling to acquire basic reading skills who might be eligible for further assessment and diagnosis as learning disabled.
The classification model being recommended in this paper is a two-stage or two-tier model that combines assessment of marker (predictor) variables with careful and continuous monitoring of children's response to early and subsequent interventions. Initially, children in first grade, kindergarten, or even preschool (depending on accuracy of predictive measures) would be identified for special preventive instruction under learning disabilities regulations if they performed below criterion on predictors of specific academic achievement and were not mentally retarded. These children would be assigned some kind of at-risk status to justify or certify their eligibility for these special services. The pre-academic or academic skills of these children would be assessed periodically (at least three times a year, perhaps more) to determine their continuing need for special services, and any child not being served could be identified for special services by referral and administration of similar tests. Any child not classified with some other primary disability (e.g., mental retardation, visual handicap) who was achieving below criterion on markers for at-risk status would be eligible for services.
Children would not be assigned the formal label of learning disabled until later in elementary school (perhaps 3rd grade or later). After receiving several years of special preventive instruction, a child could be certified as learning disabled if they continued to experience severe difficulties with any of the academic skills specified in the definition and regulations and if they attained a score on a measure of general learning ability above a given level. Any child with continuing severe academic difficulties who was not classified with some other primary disability (e.g., mental retardation, visual disabilities) would be considered learning disabled for purposes of instruction and accommodation. Additional processing or non-academic cognitive assessments would be part of the diagnostic criteria only if it is clearly established that they provide information critical to further instruction, or if they predict future academic growth beyond the predictive power of the child's current academic levels.
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