Colorado Reading Project: Longitudinal Analyses J. c. DeFries University of Colorado Boulder, Colorado

Extensive psychometric test data were obtained from two independent samples of reading-disabled and control children: 70 probands and 75 controls tested on two occasions over an average interval of 4.2 years, and 35 probands and 22 controls tested on three occasions over an average interval of 8.6 years. When composite measures of reading performance and symbol-processing speed were subjected to mixed-model multivariate analyses of variance, significant effects due to group (reading-disabled versus control) and time (i,e., test session) were obtained in both samples, and a significant group-by-time interaction was obtained for the sample tested on three occasions. In general, rates of change in reading performance are highly similar for reading-disabled and control children. However, with regard to symbol-processing speed, differences between the two groups increase as a function of age. Although no evidence was obtained for differential longitudinal stability of either composite measure in reading-disabled and control children, results of a multiple regression analysis suggest that reading deficits during middle childhood are highly predictive of later reading problems, even into early adulthood. Finucci (1986) recently reviewed follow-up studies of developmental dyslexia and concluded that, in general, reading and spelling deficits are not completely remediated. However, as previously noted by Schonhaut and Satz (1983), follow-up studies of learning-disabled children vary markedly in terms of the ascertainment criteria employed, sample size, comparison groups, follow-up intervals, and outcome results. Moreover, differences between the developmental profiles of longitudinal samples This work was supported in part by a program project grant from the NICHD (HD-11681). The invaluablecontributions of staff members of the many Colorado school districts and of the familieswho participated in the study are gratefullyacknowledged. I also thank Dr. Robin P. Corley for statisticalanalyses and RebeccaG. Miles for expert editorial assistance. Annals of Dyslexia,Vol. 38, 1988. Copyright ©1988by The Orton DyslexiaSociety ISSN 0474-7534 120

COLORADO R~ADINC PROJECT

121

of reading-disabled and control children may differ for different measures (DeFries and Baker 1983). Thus, additional longitudinal studies of individuals diagnosed as being reading disabled during childhood are clearly warranted. LaBuda and DeFries (in press) recently summarized results from a follow-up study in which extensive psychometric test data were obtained from two independent samples of reading-disabled and control children: (1) a sample of 70 reading-disabled and 75 control children tested on two occasions over an average interval of 4.2 years, and (2) a sample of 35 reading-disabled and 22 control children tested on three occasions over an average interval of 8.6 years. For both samples, differential developmental functions were observed for the individual tests. Although reading-disabled children manifested deficits in reading performance at each age, the rates of improvement were highly similar for the affected and control groups. In contrast, for measures of symbol-processing speed, differences between the groups were greater at the later ages. In a previous analysis of data from reading-disabled and control children tested on two occasions (DeFries and Baker 1983), a hierarchical multiple regression analysis was employed to assess longitudinal stability. Rather than analyzing individual test scores, composite measures of reading performance and symbol-processing speed were subjected to regression analysis. Results of this preliminary analysis suggested that the long-term stability of the reading performance measure may be less for reading-disabled children than for controls; however, that hypothesis was not explicitly tested. Thus, the primary objectives of the present study were two-fold: (1) to assess differences between the developmental profiles of reading-disabled and control children for composite measures of reading performance and symbol-processing speed obtained on either two or three occasions; and (2) to test the hypothesis of differential longitudinal stability of these measures in the reading-disabled and control groups.

Methods Subjects Subjects were participants in the Colorado Family Reading Study or a subsequent study of reading-disabled and control children only (DeFries 1985). Referral criteria for the probands included a reading achievement level of one-half of grade level expectancy or lower, chronological age between 7.5 and 12 years, and an IQ score of 90 or above, no known emotional or neurological problems, and no uncorrected visual or auditory acuity deficits. Controls were matched to probands on the basis of age (within six months), gender, grade in school, and home neighborhood. Families were middle to upper-middle class Caucasian and English was the primary language spoken in the home. The sample of reading-disabled children tested on two occasions consisted of 54 boys and 16 girls, whereas the control sample contained 55

122

THEORETICAL SEEDS

boys and 20 girls. The average ages at testing on the first and second occasions were 10.! and 14.3 years, respectively. The sample tested on three occasions included 30 reading-disabled boys, five reading-disabled girls, and 22 control boys, with average ages at testing on the three occasions being 9.2, 14.6, and 17.8 years. Tests Subjects who participated in the Colorado Family Reading Study were individually administered a two-hour psychometric test battery by a trained examiner (DeFries and Decker 1982). At follow-up these subjects were readministered several psychometric tests from the initial test battery plus the Wechsler Intelligence Scale for Children (Revised) or the Wechsler Adult Intelligence Scale (Revised). The latter battery was also administered on each occasion to those subjects ascertained as part of the subsequent study of reading-disabled and control children only. Of the various tests administered on each occasion, five define two performance measures (reading performance and symbol-processing speed). The reading measures are subtests of the Peabody Individual Achievement Test (Dunn and Markwardt 1970): Reading Recognition, Reading Comprehension, and Spelling. The measures of symbol-processing speed are the Wechsler Intelligence Scale for Children (Wechsler 1974) Coding Subtest Form B and the Colorado Perceptual Speed Test: Rotatable Letters and Numbers (DeFries et al. 1981). Analyses A matrix of pooled correlations among the five psychometric tests was estimated from data partitioned by group (reading-disabled or control children) and test occasion. In order to compute composite scores, this matrix was subjected to principal component analysis with Varimax rotation, and factor score coefficients were estimated. Composite reading performance and symbol-processing speed scores were then computed for each individual on each occasion by summing cross products of the subject's five standardized test scores and the corresponding factor score coefficients. Mean differences in composite scores between groups and occasions were maintained by expressing each subject's test score as a deviation from the grand mean of the variable and then dividing that difference by the pooled within-group standard deviation. For those subjects tested on two occasions, composite measures of reading performance and symbol-processing speed were subjected to a mixed-model multivariate analysis of variance with two between-subjects factors (group and gender), one within-subjects factor (time, i.e., occasion), and all possible interactions. This analysis yields multivariate tests of significance for the effects of group, gender, time, and their interactions, as well as univariate tests of these effects for each of the two composite measures. For those ~labjects tested on three occasions, the model did not include a main effect or any interaction involving gender because

COLORADO READING PROJECT

123

no female control subjects were included in that sample. As noted in the following section, no main effect or interaction involving gender was significant for the sample tested on two occasions. In order to test the hypothesis of differential longitudinal stability as a function of group membership (i.e., reading disabled versus control), composite scores of subjects tested on two occasions were subjected to a hierarchical multiple regression analysis (Cohen and Cohen 1975) employing the following regression model: C 2 = B I G + B2C1 + B3(GxCI) + A,

where a child's expected score at retest (C2) is expressed as a function of a dummy variable representing group membership ( - 0.5 and + 0.5 for the reading-disabled and control grou~s, respectively), child's composite score obtained on the first test occasion (C1), and their product (GxC1). B1 is the regression of follow-up test score on group membership, a measure of the group difference observed on the second test occasion. B2 is the regression of follow-up score on initial score, a measure of longitudinal stability. B3is the partial regression of child's retest score on the product of the d u m m y variable representing group membership and child's initial test score, a measure of differential longitudinal stability in the two groups. The significance of the main effects and interaction is tested sequentially. B1 and B2 are estimated from data for C1, C2, and G during the first step, and ClxG is added to the equation during a second step. The change in the squared multiple correlation between steps yields a significance test for differential longitudinal stability. For the sample tested on three occasions, composite scores obtained on occasion three were predicted from scores on occasions one and two, the d u m m y variable representing group membership, and the two-way interactions involving group and time (i.e., GxC1 and GxC2).

Results Principal Component Analysis Two principal components with eigen values of 3.02 and 0.94 were retained for rotation, and loadings of the five test scores on these rotated components are presented in Table I. As may be seen, the first component correlates highly with Reading Recognition, Reading Comprehension, and Spelling, whereas the second correlates highly with WISC Coding and Colorado Perceptual Speed. This factor structure has been found to be robust across various group, gender, generation, and test-retest samples administered these psychometric tests (e.g., DeFries and Baker 1983; DeFries, Vogler, and LaBuda 1986). Multivariate Analyses of Variance Average reading performance and symbol-processing speed composite scores for the reading-disabled and control subjects tested on ei-

124

THEORETICAL SEEDS

Table I Varimax Rotated Principal Component Loadings from Longitudinal Data Symbol-Processing Test Reading Speed Reading Recognition 0.86 0.21 Reading Comprehension 0.87 0.18 Spelling 0.79 0.29 WISC Coding 0.14 0.92 Colorado Perceptual Speed 0.37 0.82 Percent common variance 0.58 0.42 ther two or three occasions are shown in Figure 1. A particularly striking feature of these results is the marked similarity in the developmental profiles of the two independent samples. Results of the mixed-model multivariate analyses of variance of these data are summarized in Table II. Multivariate and univariate F ratios are tabulated for each main effect (group, time, and gender) and for the group-by-time interaction. F ratios for the other possible interactions are not tabulated because none is significant. As shown in Table II, a significant multivariate main effect is present for group in both samples IF(2, 140 and 2, 54) = 77.32 and 47.15, respectively, both with p < 0.001]. Inspection of the means in Figure I and the univariate F-ratios in Table II indicates that the reading-disabled group is significantly impaired on both reading and symbol-processing speed in both samples. Improvement in performance across the 4.2-year interval in the sample tested on two occasions and across the 8.6-year interval for the sample tested on three occasions is indicated by the highly significant multivariate main effects for time IF(2, 140 and 4, 218) = 260.45 and 121.70, respectively, both with p < 0.001]. As may be seen from Figure I and the univariate F-ratios in Table II, these developmental changes in performance occur for both reading and symbol-processing speed. However, the significant multivariate group-by-time interaction [F(4, 218) = 4.62, p = 0.001] for the sample tested on three occasions suggests that the rate of improvement differs for the reading-disabled and control groups. As may be seen in Figure 1, the rate of improvement in symbol-processing speed is less for the reading-disabled group than it is for controls, especially for the sample tested on three occasions. The univariate interaction for this measure is highly significant [F(2, 110) = 9.54, p < 0.001] for the sample tested on three occasions and approaches significance for the sample tested on two occasions [F(1, 141) = 2.97, p = 0.09]. In contrast, the rate of improvement in reading performance is very similar for the two groups in both samples. These results again provide evidence that differences between the developmental profiles of reading-disabled and control children differ for the different measures. Although the developmental functions of the two independent samples (i.e., those tested on either two or three occasions)

COLORADOREADINGPROJECT

125

READING

CONTROLS

2-

//~OBANDS

1////

~, o

Colorado reading project: Longitudinal analyses.

Extensive psychometric test data were obtained from two independent samples of reading-disabled and control children: 70 probands and 75 controls test...
550KB Sizes 0 Downloads 0 Views