Predicting Reading Progress in Children Receiving Special Help Nathlie A. Badian Holbrook Public School Holbrook, Massachusetts Children's Hospital Boston, Massachusetts Harvard Medical School Boston, Massachusetts
The main aim of the study was to determine whether performance on readingrelated cognitive processing tasks would help predict reading progress in children receiving special help. The 86 subjects were initially aged six to eight years and most were followed up after two years. When variance due to IQ and age was accounted for, an orthographic processing task, phonological awareness (phoneme deletion), and digit-naming speed were significant predictors of later reading skills. A strength in phonological awareness differentiated initial poor readers who later made excellent gains in reading from poor readers who did not improve. Children whose reading deteriorated had serious weaknesses on tasks of naming speed and confrontation naming. Their poor lexical retrieval skills had a more deleterious effect on later reading than on initial. Indications were that for children diagnosed as poor readers at age six or seven years, prognosis is better for boys, and for garden-variety poor readers, than for dyslexics. Caution was urged in applying the term dyslexic to children in the first two school grades because many of them will be slow starters who do not have a persistent reading problem. Annals of Dyslexia, Vol. 43, 1993. Copyright 9 1993 by The Orton Dyslexia Society ISSN 0736-9387
90
PREDICTINGREADING PROGRESS
91
Early learning disabilities are associated with a poor prognosis for future academic success, and the effects of treatment on reading disability or dyslexia do not engender optimism (Finucci 1986; Schonhaut and Satz 1983). In their review of 18 follow-up studies of children with reading/learning disabilities, Schonhaut and Satz (1983) reported that the outcome was favorable for only four studies, though mixed for two. Finucci (1986) concluded that reading deficits are not completely remediated, though IQ and SES exert a positive effect on remedial success. Age at diagnosis may also contribute. In two of the studies reviewed by Schonhaut and Satz (1983), the youngest children at the time of initial assessment showed the greatest amount of academic progress at follow-up. However, in two other studies the opposite was true. In a more recent nine-year follow-up of a cohort of children, the 10 percent defined as poor readers gained only 0.34 grades in reading per year, in spite of very early identification and help for most of them (Badian 1988). Their yearly progress in reading was only about one-fourth of that made by the group as a whole. A problem in longitudinal studies is that the stability of the diagnosis of reading disability or dyslexia is relatively low (Shaywitz et al. 1992). This finding is not unexpected as even small increments or decrements in reading scores of borderline cases will move them above or below cut-off points, whether the diagnosis is dyslexia, implying a discrepancy between actual and expected reading levels, or merely poor reading; In their longitudinal study of white boys in a Florida county, Satz and his colleagues (Satz et al. 1978) found that the incidence of reading disorders increased after first grade, remained at approximately 12 percent from grades two through four, and then increased to 20 percent between grades 4 and 5. Among the boys in the Satz study, only 6.1 percent with severe reading disability in grade 2 improved by the end of grade 5, and 30 percent of average readers had become problem readers by the end of grade 5. From their study of 414 children followed longitudinally from kindergarten, Shaywitz and her colleagues (1992) reported that only 28 percent of those classified as dyslexic in grade 1 were also classified as dyslexic in grade 3, and that only 17 percent classified in grade I would also be classified in grade 6. The term "slow starters" has been used for children who meet criteria for reading disability or poor reading in the very early stages of reading, but are normal readers at follow-up (Badian 1988; Cox 1987; McGee, Williams, and Silva 1988). In comparisons of slow starters and long-term poor readers, continued poor reading has been found to be associated with paucity of literary experience at home and limited vocabulary (Cox 1987), lower maternal reading ability (McGee, Williams, and Silva 1988), and birth complications and lower SES (Badian 1988). Thus, environmental, and possibly genetic, factors appear to play a
92
DYSLEXIA RESEARCH AND ITS APPLICATIONS
role in determining whether young children who are poor readers will be slow starters or continuing poor readers. Recent research indicates that, rather than being a global language disorder, dyslexia is associated with very specific deficits which are primarily phonological in nature (Catts 1989; Stanovovich 1988; Wagner and Torgesen 1987). Aspects of phonological processing that have been shown to be linked to reading include phonological awareness and phonological recoding in lexical access (Wagner and Torgesen 1987). From their review of studies of phonological processing, Wagner and Torgesen concluded that phonological awareness plays a causal role in the acquisition of reading skills. The relationship of phonological awareness and reading may be reciprocal, however, with reading also contributing to later phonological awareness (Perfetti et al. 1987). Wagner and Torgesen concluded that there is insufficient evidence that phonological recoding in lexical access has a causal relationship to reading. However, phonological recoding, commonly tested by naming speed for letters, numbers, pictures, and colors, has been shown to predict later reading (Badian et al. 1990; Wolf 1984, 1991), and to differentiate between dyslexic and normal readers (Badian 1993; Badian et al. 1991; Bowers, Steffy, and Tare 1988; Denckla and Rude11976; Wolf, Bally, and Morris 1986). Although phonological skills predict later reading, it is not known to what extent they contribute to differential individual progress. It is possible, for example, that phonological processing ability may help determine whether initial poor readers turn out to be slow starters or long-term poor readers. Although research on dyslexia suggests that phonological problems are the primary cause of reading and spelling difficulties, for some children deficits in orthographic processing, in addition to, or instead of, phonological processing, may underlie their poor reading (Stanovich 1988; Stanovich and West 1989). Most studies of children with reading problems have focused on those diagnosed as reading disabled or dyslexic, but there are many "garden-variety" poor readers who do not meet discrepancy criteria for dyslexia because both their reading and their intelligence level are below average (Gough and Tunmer 1986; Stanovich 1988). The validity of the distinction between dyslexic and garden-variety poor readers has been questioned (Stanovich 1988, 1991). Studies of differences between dyslexic and garden-variety poor readers in reading subskills or reading-related cognitive skills have had mixed results (e.g., Aaron 1987; Rutter and Yule 1975; Silva, McGee, and Williams 1985; Taylor, Satz, and Friel 1979). Stanovich (1991) stresses that one of the ways to establish that dyslexia is distinct from garden-variety poor reading is to show that the two types of poor readers differ in their educational prognoses.
PREDICTING READING PROGRESS
93
The main goal of this study was to determine whether differences in early reading-related cognitive processing skills (i.e., phonological and orthographic) would help predict reading outcome in children who were receiving special help because of poor reading. Other issues addressed by the study were whether dyslexic and garden-variety poor readers differ in reading progress, and whether the relationship of phonological awareness to reading is causal or reciprocal.
Method Subjects The 86 subjects (61 boys, 25 girls) were children in one school district tested at age six to eight years and followed up after at least one school year of special help in reading. The initial mean age was 7.3 years (SD 0.8) and mean age at follow-up was 8.9 years (SD 0.9). However, the main focus of the study was on the 71 subjects who were at least eight years old at follow-up. They were initially tested at 7.5 years (SD 0.7) and followed up nearly two years later at 9.3 years (SD 0.6). The other 15 subjects were initially tested at 6.3 years (SD 0.2) and followed up approximately one year later at 7.4 years (SD 0.3). The older and younger groups did not differ on any IQ measure or in initial reading scores. Mean IQs of the total sample were: Verbal 99.1 (SD 11.5), performance 103.0 (SD 13.2), and full scale 100.9 (SD 11.8). Three subjects were black (3.5 percent), two were hispanic (2.3 percent), and the other 94 percent were white. With one exception the primary or only language was English. The verbal IQ of the one exception was at the sample mean, and his reading problems were judged to be unrelated to a general language problem. The subjects lived in a suburban community and were mainly of lower middle to middle socioeconomic status. All subjects were students in regular grade I to 3 classrooms, but each was given small-group help in reading outside the regular classroom by one of three supervised teachers, in most cases for half an hour daily. The same multisensory phonetic approach, using materials designed for dyslexic children, was used by each teacher, and was the main focus of the special help. Procedures At the initial testing we administered to all subjects individually a small battery of reading-related cognitive (phonological, orthographic) processing tasks, as well as reading tests. At follow-up all subjects took the same reading tests again, and 60 of them were also given the phonological and orthographic tasks again. These 60 subjects did not differ
94
DYSLEXIA RESEARCH AND ITS APPLICATIONS
from the other 26 in initial or follow-up age, in verbal, performance, or full scale IQ, or on any reading test. The two groups also did not differ on the initial cognitive processing battery: F(7,78.75) = 1.48, p. 1881. To test the specificity of the relationship of predictive tasks to reading, as advocated by Goswami and Bryant (1990), scores on the Stanford Achievement Test Mathematics Computation and (listening) Vocabulary (Gardner et al. 1983), administered by the school system near the time of follow-up, were used. Neither of these tests involves reading. Measures Raw scores of the cognitive processing tasks (described below) were converted to T-scores (M 50, SD 10) based on age norms. Phonological Awareness
Test of Auditory Analysis Skills (TAAS) (Rosner 1979) In the two practice items and the first three test items the child repeats a word, but omits one syllable. For the remaining ten items the child is asked to omit one phoneme from each word in the initial, medial, or final position. Subjects were administered all items. Local age norms, obtained by testing six heterogeneous grade 1-3 classes (n = 131), were used. The corrected split-half reliability for the ten phoneme deletion items was .84 for the local normative group. Coefficients by age were: .75 (age 6), .92 (age 7), .89 (age 8-9). Phonological Recoding in Lexical Access Boston Naming Test (BNT) (Kaplan, Goodglass, and Weintraub 1983) The BNT consists of 60 line drawings of common objects, one to a page, which the child is asked to name. Semantic cues were given for perceptual errors, but otherwise only the first response was scored. The ceiling was at six consecutive errors. Age norms for normal children aged 6 to 10 years were used (Cohen, Town, and Buff 1988). The corrected split-half reliability coefficient for 102 local referred children aged 5 to 10 years, drawn from the same population as the study sample, was .90. Rapid Automatized Naming (RAN) (Denckla and Rudel 1974) The child names 50 items on each of four charts (Colors, Objects, Numbers, Letters), consisting of five different items repeated ten times at random. The score was the time taken to complete a chart. Scores were normalized so that higher scores meant faster performance. Age norms provided by Denckla and Rudel were used. However, the boys' norms for Colors were also used for girls, as previous analyses have shown that if the separate gender norms for Colors are used, boys score significantly higher than girls (Badian 1993). For the other three
PREDICTINGREADINGPROGRESS
95
subtests separate norms by gender are not provided. No reliability data are available for the RAN subtests.
Orthographic Processing Jordan Left-Right Reversal Test (JL-RRT), Level 1 (Jordan 1980) The child scans rows of upper-case letters and numbers and crosses out any that are not correctly oriented. The total number of errors was converted to a normalized T-score, based on the standardization norms (Table III in manual). Test-retest reliability coefficients for Level I range from .87 (ages 8-9) to .98 (ages 6-6 to 6-11). Reading Measures Woodcock Reading Mastery Tests-Revised, Form G (WRMT-R) (Woodcock 1987) Word Identification (WI) Word Attack (WA) (nonword reading) Passage Comprehension (PC) All subjects were administered WI and WA at each time, but those in the first half of grade I were not administered the initial PC because of their limited reading ability. Age norms provided in the test manual were used to obtain standard scores (M 100, SD 15) and percentile ranks. Data Analyses As boys and girls did not differ in age or IQ, or on the cognitive processing or reading tasks, gender was not a factor in statistical analyses. To answer the main question, correlation coefficients and multiple regression analyses were computed, with the initial cognitive processing test scores of the 71 older subjects and their follow-up reading, arithmetic (Mathematics Computation), and vocabulary scores as variables. Verbal and performance IQs and initial and later ages were controlled in the correlation analyses, and the same variables were forced into the multiple regression analyses before the cognitive processing tasks. To examine the question whether the relationship of phonological awareness to reading is a causal or a reciprocal one, correlations of initial TAAS with later reading variables and of initial reading variables with later TAAS were computed, with IQ and age controlled. The 71 older subjects were divided into groups of early and later good and poor readers. With Word Identification (WI) and Word Attack (WA) subtests as reading criterion measures, poor reading was defined as both WI and WA < 25 percentile for age (standard score < 90) or as either WI or WA < 12 percentile (standard score < 82). Although subjects not meeting criteria for poor reading are referred to as "good
96
DYSLEXIA RESEARCH AND ITS APPLICATIONS
readers," they were only relatively good, as all had been identified as requiring special help, and most of them were below age expectation on either WI or WA or both. The 71 older good and poor readers were compared on the initial and later cognitive processing tasks by means of 2-way analyses of covariance (ANCOVAs), with EARLY and LATER reading status (good/poor) as factors. Full scale IQ was a covariate, as good and poor readers differed initially in performance and full scale IQs, though not in verbal. Although age differences were controlled by using age-based scores, age correlated significantly with initial TAAS (r = 0.53, p < .001), and age was, therefore, an additional covariate for TAAS. The 15 younger subjects were also classified as initial and followup good and poor readers. Percentages of younger and older subjects who had improved in reading at follow-up were compared. To investigate whether dyslexic and garden-variety (G-V) poor readers differ in reading progress, the following criteria were used: A dyslexic subject was a poor reader (as defined earlier) with a verbal IQ > 92 (> 30 percentile) and a mean follow-up WI/WA standard score more than one standard deviation (15 points) less than expected reading, regressed on verbal IQ. A garden-variety subject was a poor reader with a verbal IQ < 92 and a mean follow-up WI/WA standard score less than one standard deviation lower than expected reading. Dyslexic and G-V poor readers were defined by follow-up reading scores, rather than initial, as poor reading at a later age and after receiving help makes the diagnosis more reliable. The reading progress of the dyslexic and G-V groups was examined by means of t-tests. Only subjects who were at least eight years old at follow-up were included in the dyslexic and G-V groups.
Results
Partial correlation coefficients of the initial cognitive tasks with later WI, WA, PC, arithmetic, and vocabulary are shown in Table I for the 71 older subjects. Verbal and performance IQ, and initial and later age, are controlled. The cognitive tasks correlating significantly with later achievement were: WI: RAN Objects (r = .260, p.0459), Jordan (r = .303, p.0126); WA: TAAS (r = 0.252, p.0396); PC: TAAS (r = .410, p.0126), BNT (r = .260, p.0381), RAN Numbers (r = .4~4, p.0002); Vocabulary: BNT (r = .295, p.0201). No cognitive task correlated significantly with arithmetic. As shown in Table I, TAAS had significant partial correlations with two later reading subtests (WA, PC). With the same variables as covariates, no partial correlation of any early reading variable with later
t~
0
~
0
0
~
0
E
II
N
E o
~J
II
Z
The main aim of the study was to determine whether performance on readingrelated cognitive processing tasks would help predict reading progress in children receiving special help. The 86 subjects were initially aged six to eight years and most were followed up after two years. When variance due to IQ and age was accounted for, an orthographic processing task, phonological awareness (phoneme deletion), and digit-naming speed were significant predictors of later reading skills. A strength in phonological awareness differentiated initial poor readers who later made excellent gains in reading from poor readers who did not improve. Children whose reading deteriorated had serious weaknesses on tasks of naming speed and confrontation naming. Their poor lexical retrieval skills had a more deleterious effect on later reading than on initial. Indications were that for children diagnosed as poor readers at age six or seven years, prognosis is better for boys, and for garden-variety poor readers, than for dyslexics. Caution was urged in applying the term dyslexic to children in the first two school grades because many of them will be slow starters who do not have a persistent reading problem. Annals of Dyslexia, Vol. 43, 1993. Copyright 9 1993 by The Orton Dyslexia Society ISSN 0736-9387
90
PREDICTINGREADING PROGRESS
91
Early learning disabilities are associated with a poor prognosis for future academic success, and the effects of treatment on reading disability or dyslexia do not engender optimism (Finucci 1986; Schonhaut and Satz 1983). In their review of 18 follow-up studies of children with reading/learning disabilities, Schonhaut and Satz (1983) reported that the outcome was favorable for only four studies, though mixed for two. Finucci (1986) concluded that reading deficits are not completely remediated, though IQ and SES exert a positive effect on remedial success. Age at diagnosis may also contribute. In two of the studies reviewed by Schonhaut and Satz (1983), the youngest children at the time of initial assessment showed the greatest amount of academic progress at follow-up. However, in two other studies the opposite was true. In a more recent nine-year follow-up of a cohort of children, the 10 percent defined as poor readers gained only 0.34 grades in reading per year, in spite of very early identification and help for most of them (Badian 1988). Their yearly progress in reading was only about one-fourth of that made by the group as a whole. A problem in longitudinal studies is that the stability of the diagnosis of reading disability or dyslexia is relatively low (Shaywitz et al. 1992). This finding is not unexpected as even small increments or decrements in reading scores of borderline cases will move them above or below cut-off points, whether the diagnosis is dyslexia, implying a discrepancy between actual and expected reading levels, or merely poor reading; In their longitudinal study of white boys in a Florida county, Satz and his colleagues (Satz et al. 1978) found that the incidence of reading disorders increased after first grade, remained at approximately 12 percent from grades two through four, and then increased to 20 percent between grades 4 and 5. Among the boys in the Satz study, only 6.1 percent with severe reading disability in grade 2 improved by the end of grade 5, and 30 percent of average readers had become problem readers by the end of grade 5. From their study of 414 children followed longitudinally from kindergarten, Shaywitz and her colleagues (1992) reported that only 28 percent of those classified as dyslexic in grade 1 were also classified as dyslexic in grade 3, and that only 17 percent classified in grade I would also be classified in grade 6. The term "slow starters" has been used for children who meet criteria for reading disability or poor reading in the very early stages of reading, but are normal readers at follow-up (Badian 1988; Cox 1987; McGee, Williams, and Silva 1988). In comparisons of slow starters and long-term poor readers, continued poor reading has been found to be associated with paucity of literary experience at home and limited vocabulary (Cox 1987), lower maternal reading ability (McGee, Williams, and Silva 1988), and birth complications and lower SES (Badian 1988). Thus, environmental, and possibly genetic, factors appear to play a
92
DYSLEXIA RESEARCH AND ITS APPLICATIONS
role in determining whether young children who are poor readers will be slow starters or continuing poor readers. Recent research indicates that, rather than being a global language disorder, dyslexia is associated with very specific deficits which are primarily phonological in nature (Catts 1989; Stanovovich 1988; Wagner and Torgesen 1987). Aspects of phonological processing that have been shown to be linked to reading include phonological awareness and phonological recoding in lexical access (Wagner and Torgesen 1987). From their review of studies of phonological processing, Wagner and Torgesen concluded that phonological awareness plays a causal role in the acquisition of reading skills. The relationship of phonological awareness and reading may be reciprocal, however, with reading also contributing to later phonological awareness (Perfetti et al. 1987). Wagner and Torgesen concluded that there is insufficient evidence that phonological recoding in lexical access has a causal relationship to reading. However, phonological recoding, commonly tested by naming speed for letters, numbers, pictures, and colors, has been shown to predict later reading (Badian et al. 1990; Wolf 1984, 1991), and to differentiate between dyslexic and normal readers (Badian 1993; Badian et al. 1991; Bowers, Steffy, and Tare 1988; Denckla and Rude11976; Wolf, Bally, and Morris 1986). Although phonological skills predict later reading, it is not known to what extent they contribute to differential individual progress. It is possible, for example, that phonological processing ability may help determine whether initial poor readers turn out to be slow starters or long-term poor readers. Although research on dyslexia suggests that phonological problems are the primary cause of reading and spelling difficulties, for some children deficits in orthographic processing, in addition to, or instead of, phonological processing, may underlie their poor reading (Stanovich 1988; Stanovich and West 1989). Most studies of children with reading problems have focused on those diagnosed as reading disabled or dyslexic, but there are many "garden-variety" poor readers who do not meet discrepancy criteria for dyslexia because both their reading and their intelligence level are below average (Gough and Tunmer 1986; Stanovich 1988). The validity of the distinction between dyslexic and garden-variety poor readers has been questioned (Stanovich 1988, 1991). Studies of differences between dyslexic and garden-variety poor readers in reading subskills or reading-related cognitive skills have had mixed results (e.g., Aaron 1987; Rutter and Yule 1975; Silva, McGee, and Williams 1985; Taylor, Satz, and Friel 1979). Stanovich (1991) stresses that one of the ways to establish that dyslexia is distinct from garden-variety poor reading is to show that the two types of poor readers differ in their educational prognoses.
PREDICTING READING PROGRESS
93
The main goal of this study was to determine whether differences in early reading-related cognitive processing skills (i.e., phonological and orthographic) would help predict reading outcome in children who were receiving special help because of poor reading. Other issues addressed by the study were whether dyslexic and garden-variety poor readers differ in reading progress, and whether the relationship of phonological awareness to reading is causal or reciprocal.
Method Subjects The 86 subjects (61 boys, 25 girls) were children in one school district tested at age six to eight years and followed up after at least one school year of special help in reading. The initial mean age was 7.3 years (SD 0.8) and mean age at follow-up was 8.9 years (SD 0.9). However, the main focus of the study was on the 71 subjects who were at least eight years old at follow-up. They were initially tested at 7.5 years (SD 0.7) and followed up nearly two years later at 9.3 years (SD 0.6). The other 15 subjects were initially tested at 6.3 years (SD 0.2) and followed up approximately one year later at 7.4 years (SD 0.3). The older and younger groups did not differ on any IQ measure or in initial reading scores. Mean IQs of the total sample were: Verbal 99.1 (SD 11.5), performance 103.0 (SD 13.2), and full scale 100.9 (SD 11.8). Three subjects were black (3.5 percent), two were hispanic (2.3 percent), and the other 94 percent were white. With one exception the primary or only language was English. The verbal IQ of the one exception was at the sample mean, and his reading problems were judged to be unrelated to a general language problem. The subjects lived in a suburban community and were mainly of lower middle to middle socioeconomic status. All subjects were students in regular grade I to 3 classrooms, but each was given small-group help in reading outside the regular classroom by one of three supervised teachers, in most cases for half an hour daily. The same multisensory phonetic approach, using materials designed for dyslexic children, was used by each teacher, and was the main focus of the special help. Procedures At the initial testing we administered to all subjects individually a small battery of reading-related cognitive (phonological, orthographic) processing tasks, as well as reading tests. At follow-up all subjects took the same reading tests again, and 60 of them were also given the phonological and orthographic tasks again. These 60 subjects did not differ
94
DYSLEXIA RESEARCH AND ITS APPLICATIONS
from the other 26 in initial or follow-up age, in verbal, performance, or full scale IQ, or on any reading test. The two groups also did not differ on the initial cognitive processing battery: F(7,78.75) = 1.48, p. 1881. To test the specificity of the relationship of predictive tasks to reading, as advocated by Goswami and Bryant (1990), scores on the Stanford Achievement Test Mathematics Computation and (listening) Vocabulary (Gardner et al. 1983), administered by the school system near the time of follow-up, were used. Neither of these tests involves reading. Measures Raw scores of the cognitive processing tasks (described below) were converted to T-scores (M 50, SD 10) based on age norms. Phonological Awareness
Test of Auditory Analysis Skills (TAAS) (Rosner 1979) In the two practice items and the first three test items the child repeats a word, but omits one syllable. For the remaining ten items the child is asked to omit one phoneme from each word in the initial, medial, or final position. Subjects were administered all items. Local age norms, obtained by testing six heterogeneous grade 1-3 classes (n = 131), were used. The corrected split-half reliability for the ten phoneme deletion items was .84 for the local normative group. Coefficients by age were: .75 (age 6), .92 (age 7), .89 (age 8-9). Phonological Recoding in Lexical Access Boston Naming Test (BNT) (Kaplan, Goodglass, and Weintraub 1983) The BNT consists of 60 line drawings of common objects, one to a page, which the child is asked to name. Semantic cues were given for perceptual errors, but otherwise only the first response was scored. The ceiling was at six consecutive errors. Age norms for normal children aged 6 to 10 years were used (Cohen, Town, and Buff 1988). The corrected split-half reliability coefficient for 102 local referred children aged 5 to 10 years, drawn from the same population as the study sample, was .90. Rapid Automatized Naming (RAN) (Denckla and Rudel 1974) The child names 50 items on each of four charts (Colors, Objects, Numbers, Letters), consisting of five different items repeated ten times at random. The score was the time taken to complete a chart. Scores were normalized so that higher scores meant faster performance. Age norms provided by Denckla and Rudel were used. However, the boys' norms for Colors were also used for girls, as previous analyses have shown that if the separate gender norms for Colors are used, boys score significantly higher than girls (Badian 1993). For the other three
PREDICTINGREADINGPROGRESS
95
subtests separate norms by gender are not provided. No reliability data are available for the RAN subtests.
Orthographic Processing Jordan Left-Right Reversal Test (JL-RRT), Level 1 (Jordan 1980) The child scans rows of upper-case letters and numbers and crosses out any that are not correctly oriented. The total number of errors was converted to a normalized T-score, based on the standardization norms (Table III in manual). Test-retest reliability coefficients for Level I range from .87 (ages 8-9) to .98 (ages 6-6 to 6-11). Reading Measures Woodcock Reading Mastery Tests-Revised, Form G (WRMT-R) (Woodcock 1987) Word Identification (WI) Word Attack (WA) (nonword reading) Passage Comprehension (PC) All subjects were administered WI and WA at each time, but those in the first half of grade I were not administered the initial PC because of their limited reading ability. Age norms provided in the test manual were used to obtain standard scores (M 100, SD 15) and percentile ranks. Data Analyses As boys and girls did not differ in age or IQ, or on the cognitive processing or reading tasks, gender was not a factor in statistical analyses. To answer the main question, correlation coefficients and multiple regression analyses were computed, with the initial cognitive processing test scores of the 71 older subjects and their follow-up reading, arithmetic (Mathematics Computation), and vocabulary scores as variables. Verbal and performance IQs and initial and later ages were controlled in the correlation analyses, and the same variables were forced into the multiple regression analyses before the cognitive processing tasks. To examine the question whether the relationship of phonological awareness to reading is a causal or a reciprocal one, correlations of initial TAAS with later reading variables and of initial reading variables with later TAAS were computed, with IQ and age controlled. The 71 older subjects were divided into groups of early and later good and poor readers. With Word Identification (WI) and Word Attack (WA) subtests as reading criterion measures, poor reading was defined as both WI and WA < 25 percentile for age (standard score < 90) or as either WI or WA < 12 percentile (standard score < 82). Although subjects not meeting criteria for poor reading are referred to as "good
96
DYSLEXIA RESEARCH AND ITS APPLICATIONS
readers," they were only relatively good, as all had been identified as requiring special help, and most of them were below age expectation on either WI or WA or both. The 71 older good and poor readers were compared on the initial and later cognitive processing tasks by means of 2-way analyses of covariance (ANCOVAs), with EARLY and LATER reading status (good/poor) as factors. Full scale IQ was a covariate, as good and poor readers differed initially in performance and full scale IQs, though not in verbal. Although age differences were controlled by using age-based scores, age correlated significantly with initial TAAS (r = 0.53, p < .001), and age was, therefore, an additional covariate for TAAS. The 15 younger subjects were also classified as initial and followup good and poor readers. Percentages of younger and older subjects who had improved in reading at follow-up were compared. To investigate whether dyslexic and garden-variety (G-V) poor readers differ in reading progress, the following criteria were used: A dyslexic subject was a poor reader (as defined earlier) with a verbal IQ > 92 (> 30 percentile) and a mean follow-up WI/WA standard score more than one standard deviation (15 points) less than expected reading, regressed on verbal IQ. A garden-variety subject was a poor reader with a verbal IQ < 92 and a mean follow-up WI/WA standard score less than one standard deviation lower than expected reading. Dyslexic and G-V poor readers were defined by follow-up reading scores, rather than initial, as poor reading at a later age and after receiving help makes the diagnosis more reliable. The reading progress of the dyslexic and G-V groups was examined by means of t-tests. Only subjects who were at least eight years old at follow-up were included in the dyslexic and G-V groups.
Results
Partial correlation coefficients of the initial cognitive tasks with later WI, WA, PC, arithmetic, and vocabulary are shown in Table I for the 71 older subjects. Verbal and performance IQ, and initial and later age, are controlled. The cognitive tasks correlating significantly with later achievement were: WI: RAN Objects (r = .260, p.0459), Jordan (r = .303, p.0126); WA: TAAS (r = 0.252, p.0396); PC: TAAS (r = .410, p.0126), BNT (r = .260, p.0381), RAN Numbers (r = .4~4, p.0002); Vocabulary: BNT (r = .295, p.0201). No cognitive task correlated significantly with arithmetic. As shown in Table I, TAAS had significant partial correlations with two later reading subtests (WA, PC). With the same variables as covariates, no partial correlation of any early reading variable with later
t~
0
~
0
0
~
0
E
II
N
E o
~J
II
Z
