Learning About Learning Disabilities

Butler

Chapter 1

Classification and Identification of Learning Disabilities

Jack M. Fletcher

Department of Psychology, University of Houston, Houston, TX 77204-5053, USA

Chapter Contents

Introduction

What Is a Learning Disability?

Exclusionary Criteria

Inclusionary Criteria

LD Is an Unobservable Construct

Classification Issues in LD

Categorical versus Dimensional Classifications

Neurological Models of LD

Behavior Disorders Due to Intrinsic Factors

Dyslexia

Demise of the Concept of MBD

Cognitive Models

Emergence of the Concept of LD

LD and US Public Policy

IQ-Achievement Discrepancy

Other Cognitive Discrepancy Approaches

Psychometric Issues Underlying Cognitive Discrepancy Methods

Instructional Models

Low Achievement Methods

Response to Intervention Methods

A Hybrid Approach to LD Identification

Conclusions

Acknowledgment

References

Introduction

What Is a Learning Disability?

Few terms used to identify people with a set of problems producing major difficulties with adaptation to life and society generate as much contention and confusion as the term learning disability (LD). Children with the attributes of a LD in reading have been identified since before the start of the previous century as children with severe reading problems who seemed bright and otherwise normal (Hinshelwood, 1895; Morgan, 1896). As was the situation then, there is consensus among contemporary researchers and practitioners that LDs are brain-based and heritable. However, measuring brain dysfunction and heritability have proven elusive, although people with LDs clearly differ in brain function compared to typically developing people and people with different types of LD (e.g., reading versus math LD) (Fletcher, Lyon, Fuchs, & Barnes, 2007; Gabrieli, 2009). Similarly, there is strong evidence that LDs have partial genetic origins (Pennington, 2009; Plomin & Kovas, 2005), but the effects of individual genes are small and the mode of inheritance fits a multifactorial model similar to that seen in other developmental disorders, such as Attention-Deficit/Hyperactivity Disorder (ADHD) (Willcutt, Pennington et al., 2010).

There is also strong consensus that the core attribute of any conceptual model of LD is unexpected underachievement (Kirk, 1963), largely because people with LD do not learn to read, write, and/or do arithmetic despite the absence of conditions frequently associated with low achievement. Samuel Kirk, often credited with coining the term LD, stated that It is clear that people with LD do not learn to read, write, or do arithmetic despite the absence of conditions that are known correlates of low achievement, such as an intellectual, sensory, or motor disability, emotional and behavioral difficulties, economic disadvantage, and lack of instructional language proficiency. (Kirk, 1963, pp. 2–3). These conditions, which are present in most definitions of LD, are commonly referred to as exclusionary because they represent factors in which low achievement is expected.

Exclusionary Criteria

Defining LD according to the absence of conditions that cause other forms of low achievement has never been satisfactory (Rutter, 1978), with some arguing that definition by exclusion makes efforts to identify LD circular: Stripped of clauses which specify what a learning disability is not, this definition is circular, for it states, in essence, that a learning disability is an inability to learn. It is a reflection of the rudimentary state of knowledge in this field that every definition in current use has its focus on what the condition is not, leaving what it is unspecified and thus ambiguous (Ross, 1976, p. 11). Thus, the classification issue with which researchers and practitioners have wrestled is what makes low achievement unexpected. To address this issue, efforts have been made to identify attributes other than achievement that can be used to operationalize the concept of unexpected underachievement and represent inclusionary criteria.

Inclusionary Criteria

In considering additional criteria for classifying and identifying LD, three overarching models have emerged over the past 125 or so years: Neurological, cognitive, and instructional. The earliest models were neurological because they attempted to identify special signs of brain dysfunction that indicated the presence of LD. These models began to recede in the 1970s and models based on some form of cognitive discrepancy gained prominence. More recently, instructional models based on the idea of using intervention response as an indicator of unexpected underachievement have emerged. The methods are tied to Response to Intervention (RTI) service delivery frameworks used by schools to accelerate academic and behavioral outcomes in all children. Thus, attributes of LD variously considered as indicators of unexpected underachievement include neurological markers and signs, unevenness in cognitive functions, and an inability to respond to instruction that benefits most children.

LD Is an Unobservable Construct

Altogether, classifying LDs, which leads to definitions for identification, involves the application of criteria that include and exclude specific attributes of people hypothesized to represent the construct of LD. As a construct, LD is unobservable, which means that at a latent level, the concept is pure and untarnished by our imperfect efforts to measure it. We can propose key features of the construct, especially the concept of unexpected underachievement, and propose attributes of LD, like low achievement, cognitive discrepancies, and poor instructional response. However, these attributes are hypotheses and must be validated through research (Morris, 1988). We can measure them, but our efforts at measurement will always be imperfect because of measurement error. Thus, no single indicator is likely to be adequately reliable for measuring the different hypothetical attributes of LD.

In itself, the construct indicates that low achievement is a necessary but not sufficient condition for identification of LD because there must be criteria that indicate unexpectedness as well as low achievement. As I discussed above, many would agree that LD should not be invoked when there are other attributes that explain low achievement. As such, LD is one of several factors that produce low achievement in children; it is the unexpected kind of low achievement. Nonetheless, researchers and practitioners disagree on inclusionary criteria and the precise role of different exclusionary criteria. These disagreements are less about the construct of LD at a latent level, but more about how the essential attributes (e.g., unexpected underachievement) are measured.

Because efforts at measurement always have error, there will be imprecision in efforts to measure and indicate any latent variable. The situation is no different than attempts to measure intelligence. Few doubt that there is a latent construct of intelligence. The problem is that there are competing theories and multiple IQ tests that don’t always provide the same conclusion about a person’s IQ. But these are differences in IQ test scores that reflect in part differences in the underlying theory of intelligence that leads to differences in how the tests are constructed and the measurement error of the tests. The construct of IQ is untarnished by our efforts to measure and operationalize it.

In the next sections, I will expand this discussion of conceptual issues in classifying and defining LD and then discuss evidence for the three models of LD in a historical context. I will provide evidence that supports the reality of the LD construct and then discuss efforts to operationalize it from neurological, cognitive, and instructional models. By way of preface, it is important to recognize that LD has neurological, cognitive, and instructional attributes. Deciding among the models are not black and white issues and all contribute to our understanding of LD. However, the ultimate decisions may be pragmatic and guided by how well different models facilitate outcomes given available resources.

Classification Issues in LD

Any discussion of LD seems to assume that LD is represented by discrete groups that can be operationalized and defined. In a neurological model, people with LD are identified because of special signs presumed to indicate brain dysfunction: motor clumsiness, perceptual difficulties, confusion of right and left, difficulty perceiving symbols written on the finger tips, and even specific language problems (e.g., slow naming speed). Cognitive discrepancy and instructional models use psychometric criteria and look for performance below a specified threshold to indicate the presence of an attribute of LD.

Regardless of the model, the most common approach to identifying people with LD for research is to select an achievement measure, establish a threshold for low achievement (e.g., reading score below the 20th percentile) and then compare children who are low achieving on the achievement measure with another group that achieves above the threshold. Such a method would represent a low achievement approach to definition that I have classified as an instructional model of LD. In a cognitive discrepancy method or another instructional method, the attributes might change, but the approach would be the same: contrast groups created by a cut-off point on the measurement tool used to define the attribute of interest. In providing services, similar psychometric approaches are used; children receive services when they score below a specified threshold on a test, show a specified difference in IQ and achievement, or are below the threshold on an assessment of instructional response. When groups are compared on measures not used to define them, such as measures of cognitive function, brain function, or a genetic assessment, the approach taken to define the groups is validated if the groups are significantly different, which in classification research is termed external validity (Skinner, 1981). The fact that (a) children defined as LD using these different psychometric methods differ from typically achieving children and (b) that children with different types of LD (reading vs. math LD) differ on measures of cognitive functions, brain function, and heritability, is strong evidence for the validity of the construct of LD. Moreover, it is rather obvious that there are interactions of the type of LD with the treatment approach: children with reading difficulties improve in reading when they receive a reading intervention, but not a math intervention, and vice versa (Morris et al., 2012).

Categorical versus Dimensional Classifications

The problem with these approaches is the assumption that LD represents a discrete group, representing a categorical classification. A categorical classification is usually appropriate when there are subgroups with firm boundaries and whose members are qualitatively different from one another. Alternatively, if the differences across members of an overarching classification like LD are not qualitatively different, the classification may be dimensional. In a dimensional classification, members are quantitatively different and usually represent an unbroken continuum where specific levels of severity lead to problems with adaptation (Morris, 1988). If there are no qualitative breaks, dichotomizing dimensions leads to unreliability in identification of people around the threshold and reduces power in research studies (Cohen, 1983).

The best examples of dimensional disorders in medicine are problems like obesity or hypertension (Ellis, 1984). Weight and blood pressure are continuous attributes of a human population. When decisions are made to treat a person as being overweight or for high blood pressure, it is because the risk of an adverse outcome is triggered at certain levels of the continuum. This threshold is not firmly fixed, but is usually represented by multiple criteria and will vary depending on different risk characteristics. But decisions to treat are related to indices of outcome and may vary across individuals.

Whether the attributes of LD can be represented as a categorical or a dimensional classification is an open empirical question. But it is very important because subdividing a normally distributed dimension to create categories is an arbitrary process that introduces unreliability into decisions about individuals who may or may not be members of a group. To take simple examples, defining a reading LD as a score below the 20th percentile on a reading test or a difference between IQ and achievement of 15 standard score points (at one point the predominant definition in public policy in North America) are unreliable indicators of individuals who need services because the tests used to assess the cut-off point have small degrees of measurement error and are correlated in the case of IQ and achievement. If we assess individuals across multiple occasions with either a single test or an aptitude-achievement discrepancy, or use different tests that measure the same ability constructs individuals will fluctuate around the cut-off point because of measurement error (Francis et al., 2005; Macmann et al., 1989). This fluctuation is most serious for individual diagnostic decisions and is why defining LD should never rest solely on a single indicator or a battery of test scores. For research comparing groups, this kind of unreliability does not have much effect on the patterns of group differences because individuals around the cut-off point are more similar than different. However, if the attributes are dimensional and a category is introduced, the difference between the groups (effect size) will be smaller and more participants will be needed in the contrasted groups to detect the difference (i.e., the power of the study is reduced) (Cohen, 1983; Markon et al., 2011).

Contrasting groups in studies of LD may be an inefficient way of understanding the relations of different attributes of LD (Doehring, 1978), especially because the methods that emerge for testing of group differences are based on analysis of variance (ANOVA). In a dimensional approach, knowing the correlation of the dependent measures with the independent measures used to form the groups, and the intercorrelations of the dependent measures, would lead to an estimate of the relations (effect sizes) among the independent and dependent variables (e.g., achievement and cognition) that would likely be more reliable across studies and require fewer participants to evaluate (Stuebing et al., 2002).

Why has research and practice on LD relied upon categorical classifications? This has occurred because of policy considerations and the need to identify individuals who might be eligible for services. However, epidemiological and behavior genetic studies of the achievement attributes of LD have supported a dimensional view (Plomin & Kovas, 2005). Although a considerable body of research has examined possible subtypes of LD using some type of theory to create groups or through the application of empirical, exploratory methods like cluster analysis to search for subgroups (see review in Fletcher et al., 2007), these efforts have not been strongly related to treatment outcomes or other external validity indicators except along the broader dimensions of achievement (e.g., reading vs. math disability). Here it is clearer that the dimensions, while correlated, are differentially related to treatment outcomes and other external indices. Nonetheless, researchers for many years have argued that the achievement attributes of LD are normally distributed (Ellis, 1984; Stanovich, 1988) and that differences relate to severity. More recently, Snowling and Hulme (2012) have argued that reading disabilities involving word recognition and comprehension, while correlated, represent distinct dimensions of a broader classification of LD. In our book (Fletcher et al., 2007), we suggested that the evidence supported six subgroups of LD involving reading (word recognition, fluency, and comprehension), math (calculations and problem solving), and probably written expression. The latter could involve either the generation of text (handwriting, spelling) or composition, but research was not clear on distinctions among these written expression components or overlap with other forms of LD. However, treatment needs were clearly different depending on the affected component. Although the language refers to groups, these attributes may be correlated dimensions with no explicit group structure.

Dimensionality also helps us understand how people with LD may also have low achievement in more than one of these domains (e.g., both reading and math LD) and may also meet criteria for other neurodevelopmental disorders, especially ADHD. These are considered co-occurring or comorbid associations in which the person has more than a single problem and are usually not explained as the presence of one problem causing another (Willcutt, Betjemann et al., 2010). Research suggests that people meeting criteria for comorbid disorders show similar cognitive performance to individuals with a single disability (i.e., show characteristics of both a reading disability and ADHD). Behavior genetic studies suggest that there are generalist genes that appear to be involved in reading and math LD, and in ADHD, as well as specific genetic factors related to each disorder in isolation (Plomin & Kovas, 1985; Willcutt, Pennington et al., 2010). Understanding these attributes as correlated dimensions and not as independent categories helps us understand comorbidity.

There is good evidence from cognitive, neurobiological, and treatment studies for the six component classification we proposed (Fletcher et al., 2007), but more research needs to be completed. In particular, the field of LD needs to look at categorical versus dimensional distinctions on other attributes of LD. Instructional response, for example, may lie on a continuum of severity (Fletcher et al., 2011; Vellutino et al., 2006). Understanding these classification issues would also be facilitated by the application of more recent statistical methods for identifying discrete groups, such as latent mixture modeling and other methods commonly utilized in classification research on psychopathology (Ruscio, Haslam, & Ruscio, 2006). I will return to these issues as I review different models for conceptualizing LD.

Neurological Models Of LD

Behavior Disorders Due to Intrinsic Factors

Although a discussion of earlier concepts of LD as a form of brain dysfunction may seem outdated, the conceptualizations that motivated and emerged from these concepts have strongly influenced contemporary concepts of LD. The fundamental issue was the identification of children whose difficulties seemed to be of constitutional origin and not attributable to environmental factors.

Early neurological models initially focused on children with behavioral problems we would now recognize as ADHD and emerged because these children had behavior problems that were unexpected, along with poor school performance. One early paper described children with a disorder of morbid control to represent children with a behavioral pattern characterized by hyperactivity, impulsivity and difficulty with abstraction (Still, 1902). Because this behavioral pattern seemed to be associated with birth complications, other physical anomalies, and occurred more frequently in boys than girls, Still (1902) concluded that the origin was intrinsic to the child.

Continuing the focus on behavioral patterns associated with brain dysfunction, concepts emerged with terms like organic driveness (Kahn & Cohen 1934), minimal brain injury (Strauss & Lehtinen, 1947), and minimal brain dysfunction (Clements, 1962). As the concept of MBD emerged, it was recognized that many of the children had academic problems and the concept was expanded to include reading, math, and writing difficulties.

The concept of MBD was formally defined in 1964 as children of near average, average, or above average general intelligence with certain learning or behavioral disabilities … associated with deviations of function of the central nervous system. These deviations may manifest themselves by various combinations of impairment in perception, conceptualization, language, memory, and control of attention, impulse, or motor function (Clements, 1966, pp. 9–10). This definition of MBD incorporated (almost verbatim) the exclusionary criteria in the subsequent first US statutory definition of LD: The term does not include children who have learning disabilities, which are primarily the result of visual, hearing, or motor handicaps, or mental retardation, or emotional disturbance, or of environmental, cultural, or economic disadvantage (US Office of Education, 1968, p. 34). These exclusionary criteria have been part of every statutory and regulatory definition of LD in the US since 1968.

Dyslexia

A somewhat separate strand represented efforts to understand children with severe reading disabilities. Described initially as word blindness by ophthalmologists (Morgan, 1894; Hinshelwood, 1895), Orton (1928) developed a neurological theory of dyslexia in which problems with reading were part of a broader failure to establish hemispheric dominance for language, such that people with dyslexia saw mirror images of letters and words. Thus, letter reversals and related signs became evidence of dyslexia in much the same way that behavioral patterns and perceptual and motor difficulties became signs of MBD. Subsequent research increasingly focused on characteristics that could be assessed through neuropsychological evaluations, such as right-left confusion, finger agnosia (difficulty appreciating numbers and letters written on the fingers), language and perceptual problems, and motor coordination problems.

As we can see, conceptualizations of MBD and dyslexia reflected a neurological classification. The purpose was to determine the cause of the brain-related disorder in order to identify treatments that would correct the brain dysfunction (Benton, 1975). However, for both dyslexia and MBD, the pool of children was very heterogeneous and there was no agreement on how to define people with the disorders. Theories based on single deficits proliferated and training models for correcting problems seen as direct evidence of MBD or LD (e.g., motor or perceptual skills training) were clearly not efficacious (Doehring, 1978). For example, to diagnose MBD, people were taught to use a checklist of 37 behaviors. If the person had nine of the symptoms, treatment for MBD was indicated almost regardless of whether the problem was academic, cognitive, or behavioral (Peters, Davis, & Goolsby, 1973).

Demise of the Concept of MBD

The neurological model eventually collapsed with the demise of the concept of MBD in the 1980s, reflecting the failure of training programs addressing special signs to generalize to important areas of adaptation (e.g., better reading performance). In addition, medication treatment using stimulants, which are clearly efficacious for problems with impulsivity and hyperactivity, were often recommended because a person showed multiple attributes of the group, but not those for which stimulants appeared particularly useful. With the rise of the formal concept of learning disabilities and federally led efforts to define them in the 1960s, the Diagnostic and Statistical Manual (DSM) III (American Psychiatric Association, 1980) formally separated academic skills disorders involving reading, math, and writing from ADHD, which was a set of problems in the behavioral domain involving inattention, hyperactivity, and impulsivity. The issue now is the comorbidity of different disorders and few would lump together children with these diverse difficulties into a single group (for an exception, see Gilger & Kaplan, 2001). Moreover, neurobiological research into brain function and genetics has flourished in part because criteria for different kinds of LD are specific about the area of academic impairment and separate LD from ADHD.

Cognitive Models

Emergence of the Concept of LD

As the influence of older neurological models began to subside, the role of cognitive factors in LD became increasingly prominent. Instead of conceptualizing LD as a form of brain dysfunction with a set of special signs that might represent qualitative distinctions separating those with and without LD, behavioral scientists and speech and language specialists like William Cruickshank, Helmer Myklebust, Doris Johnson, and Samuel Kirk viewed language and cognitive functions as central to the concept of LD. This early work on the concept of LD (as opposed to MBD or dyslexia, the latter a common and specific form of LD) emphasized unevenness in cognitive functions and a need for cognitive and educational interventions, along with an absence of other conditions associated with low achievement (i.e., the exclusionary criteria).

Thus, Cruickshank, Bice, and Wallin (1957) recommended modifications in classroom environments to reduce distractions for children with academic and behavioral problems even as they evaluated perceptual and motor training programs. At Northwestern University, Myklebust and Johnson evaluated language and perceptual deficits on academic and social functioning of children suggesting both verbal and nonverbal factors in LD. They also introduced scripted interventions remediating these deficient skills (Johnson & Myklebust, 1967). Samuel Kirk developed assessment methods for detecting unevenness in language abilities and as I discussed above, proposed the term learning disabilities at a 1963 conference.

LD and US Public Policy

As this research and advocacy continued, the idea of LD rapidly emerged and made its way into public policy. Based on the work of Kirk and others, it was recognized that children with LD: (a) had learning characteristics that were different from children identified with intellectual or emotional difficulties; (b) demonstrated unexpected problems with achievement given strengths in other areas; and (c) required specialized educational interventions that were not needed for typically achieving children. In addition, reflecting the influence of neurological conceptualizations, it was argued that learning difficulties in this population resulted from neurobiological factors intrinsic to the child rather than environmental factors. Not surprisingly, the formal definition of MBD in 1962 led to reactions by educators and other professionals interested in moving away from etiological concepts. At a meeting convened in 1966 by the US Office of Education, Kirk’s (1963) concept of learning disability was formally defined:

The term specific learning disability means a disorder in one or more of the basic psychological processes involved in understanding or in using language, spoken or written, which may manifest itself in an imperfect ability to listen, speak, read, write, spell, or to do mathematical calculations. The term includes such conditions as perceptual handicaps, brain injury, minimal brain dysfunction, dyslexia, and developmental aphasia. The term does not include children who have learning problems, which are primarily the result of visual, hearing, or motor handicaps, or mental retardation, or emotional disturbance, or of environmental, cultural, or economic disadvantage.

(US Office of Education, 1968, p. 34).

The importance of this definition is easily seen by the fact that it continues to serve as the federal statutory definition of LDs in US policy for special education. It has persisted since Public Law 94–142 was adopted in 1975, which subsequently became known as the Individuals with Disabilities in Education Act (IDEA; U.S. Department of Education, 2004) last revised by the US Congress in 2004. This definition has endured despite the fact that all it really says is that LDs are heterogeneous, associated with problems involving cognitive processing, and are not to be mixed with other disorders and conditions that represent exclusionary criteria.

The 1966 definition of LD and the 1962 definition of MBD are similar, reflecting their common roots in neurological models of LD (Satz & Fletcher, 1980). Both are represented as unexpected disorders not attributable to intellectual difficulties, sensory disorders, emotional disturbance, or economic and cultural diversity. The definitions acknowledged intrinsic factors within a child. However, there are no clearly specified and measurable inclusionary criteria, which became a problem when schools were expected to identify and serve children with LDs. Thus, in 1977, the US Office of Education (1977) published a regulatory definition of LD that included a discrepancy between IQ and achievement as an inclusionary criterion:

… a severe discrepancy between achievement and intellectual ability in one or more of the areas: (1) oral expression; (2) listening comprehension; (3) written expression; (4) basic reading skill; (5) reading comprehension; (6) mathematics calculation; or (7) mathematic reasoning. The child may not be identified as having a specific learning disability if the discrepancy between ability and achievement is primarily the result of: (1) a visual, hearing, or motor handicap; (2) mental retardation; (3) emotional disturbance, or (4) environmental, cultural, or economic disadvantage (p. G1082).

IQ-Achievement Discrepancy

From this regulatory definition, the idea of a cognitive discrepancy between higher IQ and lower achievement as a marker has become instantiated in policy and societal concepts of LD. There was research at the time supporting an IQ-achievement discrepancy model (Rutter & Yule, 1975) that has not held up over time (Fletcher et al., 1998). The Isle of Wight studies showed a qualitative break in the distribution of reading scores suggesting a categorical distinction between children with reading achievement well below IQ (specific reading disability) and reading consistent with IQ (general backwards readers). However, the Isle of Wight study was epidemiological and asked the question of whether there was a specific form of reading failure that stood out against all other forms of reading failure (Rutter & Yule, 1975). This study did not exclude children with brain injury and intellectual disabilities, many of whom emerged as backwards readers (Fletcher et al., 1998). The qualitative break in the distribution was due to the inclusion of children with low IQ scores. To reiterate, most studies find that the achievement attributes of LD are dimensional with no qualitative breaks (Rodgers, 1983; Shalev, Manor, Auerbach, & Gross-Tsur, 2000; Shaywitz et al., 1992; Silva, McGee & Williams, 1985).

At this point in time, it is well-established that cognitive discrepancies based on differences in aptitude and achievement measures do not have strong validity based on two meta-analyses of the literature (Hoskyn & Swanson, 2000; Stuebing et al., 2002). These studies did not identify major differences in the behavioral, cognitive, and achievement characteristics of children who met IQ-achievement discrepancy criteria in reading versus children with reading difficulties whose achievement was consistent with IQ (low achievers, excluding those with intellectual deficiencies). It has also been found that these two subgroups do not differ in the long term development of reading skills (Shaywitz et al., 1999) and that IQ and IQ-achievement discrepancies are at best weak predictors of treatment outcomes (Stuebing et al., 2009; Vellutino et al., 2000). Most recently, Tanaka et al. (2011) found no differences in the brain activation patterns of two different samples of children identified as IQ-achievement discrepant and low achieving when reading real words and pseudowords in a functional brain imaging study (functional magnetic resonance imaging). Fletcher et al. (2007) reviewed evidence questioning the validity of the IQ-achievement classification method using other methods for assessing aptitude (e.g., listening comprehension), other domains of achievement, and for children with language problems.

Other Cognitive Discrepancy Approaches

Given the difficulties with classifications based on IQ-achievement discrepancies, other approaches to operationalizing a cognitive discrepancy method have been proposed. The most prominent uses a pattern of processing strengths and weaknesses across a battery of cognitive tasks (Hale et al., 2008). Depending on the model, a child may be identified as LD because of a strength in cognitive processing, a weakness in achievement, and an achievement weakness related to the processing weaknesses. Thus, children with a word level reading disability may have a strength in nonverbal processing (e.g., matrix reasoning), a weakness in vocabulary, and a weakness in reading comprehension (for which vocabulary weakness are common). Children who show achievement deficits but no processing strength (or processing weaknesses not related to the achievement domain) would not be identified as LD (Hale et al., 2008).

There are multiple methods for operationalizing patterns of strengths and weaknesses (PSW) methods. In contrast to the research on aptitude achievement discrepancy models, there is little evidence for their validity. Given the absence of validity studies, Stuebing et al. (2012) simulated three PSW methods that have been proposed for the identification of LD. Based on the assumptions of each of the three PSW methods, latent data that replicated the reliabilities, intercorrelations, and cut-off points were generated. Observed data were then created so that the agreement of decisions about LD and not LD could be compared. All three methods were stronger for decisions concerning not LD. However, the methods did not identify many as LD and when those with LD were identified, many of the children were not LD at the latent level, suggesting high false positive rates. Cognitive skills are related to LD, but this relation does not mean that assessment of these skills helps identify or plan treatment. A method with a high false positive rate will systematically mismatch children with instruction that might not be beneficial.

In thinking about this issue, consider that LD is related to how the brain functions, which we can assess with functional brain imaging studies. However, a brain imaging study is not recommended for every person evaluated for possible LD. It is simpler to assess people with measures like those used to activate the brain because the changes in brain activity (by definition) are neurophysiological correlates of task performance. In addition, at this point in time, brain imaging studies help us understand the neural mechanisms underlying LD, but don’t have specific implications for intervention (Goswami, 2008).

Psychometric Issues Underlying Cognitive Discrepancy Methods

The problems with the reliability and validity of cognitive discrepancy models of LD are not surprising given what we have learned in this chapter about categorical and dimensional classifications, the reliability of decisions based on cut-off points, and the small amount of measurement error in even the best psychological and achievement tests. Such approaches are inherently low in reliability in identifying individual children as LD or not LD in real and simulated data (Frances et al., 2005; Macmann & Barnett, 1989). This unreliability occurs across different attributes of LD, including assessments of achievement (Francis et al., 2005; Macmann & Barnett, 1989), IQ-achievement discrepancies (Francis et al., 2005; Macmann et al., 1985), patterns of strengths and weaknesses across cognitive tests (Kramer et al., 1987; Stuebing et al., 2012), and even assessments of instructional response (Barth et al., 2008). In the math area, it has been proposed that children who perform below the 10th percentile have a specific learning disability and those between the 10th and 25th percentile are low achieving (Geary, Hoard, Byrd-Craven, Nugent, & Chattavee, 2007). However, because of the imposition of rigid cut-off points, the measurement error of the tests, and the possibility that math skills are an unbroken continuum of severity, individual decisions may not be reliable. A problem with any cognitive model involves the psychometric issues we have described throughout this chapter.

The other major problem with cognitive models is that they are predicated on the assumption that there are treatment implications signaled by the presence of a cognitive deficit or a pattern of strengths and weaknesses. There is little evidence that directly training deficient cognitive skills out of the context of an intervention that involves reading, math, or writing produces achievement gains (Mann, 1979; Torgesen, 2002). Researchers have for many years searched for aptitude by treatment interactions, which have largely not emerged for cognitive patterns, learning styles, and similar efforts to identify child traits that interact with specific forms of treatment (Pashler et al., 2009). The exceptions, as we shall see below, are interactions of specific academic strengths and weaknesses (Connor et al., 2009). None of these concerns should be taken to indicate that cognitive skills are not related to LD because the manifestations of LD in achievement and other functional limitations are clearly associated with specific cognitive difficulties. Using this information for identification and treatment has proven elusive.

Instructional Models

Low Achievement Methods

One alternative to the difficulties presented by cognitive discrepancy models is to focus just on the achievement domain and identify children as potentially LD if they read, write, or do arithmetic below a certain level (Siegel, 1992). This approach simplifies the assessment and identification approach and eliminates any complications from the need to compare two scores, with differences scores known to be less reliable than single indicators (Bereiter, 1967). In addition, variations in different domains of reading, math, and/or writing are clearly related to variations in cognitive processing, neurobiological correlates, and intervention response whether categorical or dimensional methods are used (Fletcher et al., 2007).

The presence of child attribute by treatment interactions is clearly apparent in studies where a group of poor readers receives only school-based reading instruction and extra tutorial math instruction and does not improve in reading compared to children who receive reading tutoring (Morris et al., 2012). In reading, Connor et al. (2009) evaluated reading decoding and comprehension and used a computer algorithm to adjust the amount of instruction in each domain. Helping teachers evaluate strengths and weaknesses in decoding versus comprehension skills and adjust instruction based the assessment led to better outcomes in classrooms receiving this assistance than in classrooms that did not receive this assistance. Low achievement methods are associated with good external validity.

There are also problems with low achievement methods, especially because the group identified with LD solely on the basis of low achievement would include students with other disorders and conditions typically considered exclusionary. Exclusionary criteria could be included in the definition. With this modification, the primary inclusionary criterion would be low achievement and unexpectedness would be indicated by absence of exclusionary criteria. However, this definition is still based primarily on exclusion because there is no inclusionary criterion indicating unexpectedness.

Low achievement methods do not resolve the psychometric issues involved in identifying individual people with LD and make even plainer the issues with cut-off points. What is the threshold for defining low achievement? Policy and different studies vary considerably in the cut-off points used to indicate the possibility of LD. The selection of a cut-off point makes assumptions about prevalence. If the 20th percentile is selected, and we exclude 2% for intellectual disability, and some unknown proportion because of exclusionary criteria (say 3%), the resulting estimate of 15% on a single type of achievement measure seems high for the number of children with LD, especially if the same cut-off point is used across six achievement domains. Certainly, there would be overlap because some would have problems in multiple domains and the measures of achievement are correlated, but the overall prevalence will be much higher than 15%. The cut-off point could be lowered, but because achievement distributions do not have natural breaks, any decision is potentially arbitrary. These issues are fundamental to the concept of LD and reflect social, political, and economic decisions that have not been adequately addressed.

Response to Intervention Methods

Methods for classification, definition, and identification of LD that emanate from response to intervention (RTI) frameworks generate considerable confusion. I find it helpful to separate the RTI service delivery framework from the assessment of intervention response, the latter most often used as an identification criterion. RTI is an approach to delivering services in schools that introduces universal screening for academic and behavioral problems, progress monitoring for students who are at risk, and increasingly intense interventions depending on how well at risk children progress (Fletcher & Vaughn, 2009; Jimerson, VanDerHeyden, & Burns, 2007; VanDerHeyden & Burns, 2010). The interventions are usually organized hierarchically as a set of tiers beginning in the general education classroom through which students pass when progress at less intense intervention levels is not adequate. Some children will continue to struggle despite multiple intense interventions; this intractability to instruction that works with most children may be evidence of a disability. Thus, in identification methods based on RTI frameworks, inadequate response to instruction becomes a key inclusionary criterion and indicates unexpectedness (Fletcher et al., 2007).

Decisions about how well a child is progressing are usually made on the basis of short assessments of key academic skills that are predictors of proficiency (Stecker, Fuchs, & Fuchs, 2005). Thus, in reading, it is common to assess oral reading fluency for a short passage and count the number of words read correctly per minute. This type of curriculum-based measurement (CBM) assessment can be repeatedly administered over the course of a school year or intervention and is predictive of multiple proficiencies in reading. Charts showing growth in fluency over time can be created and compared to benchmarks for the end of the intervention or to a normative standard. Similar methods are used in math and writing. Lack of growth or failure to attain key benchmarks may be indicative of LD and certainly support the idea of a need for more intervention. Such an approach to identification is strongly linked to treatment (Fuchs & Fuchs,