Predictors of Success and Failure in Reading
Of the many conditions that appear to contribute to successful reading by schoolchildren, among the more important are each child's (1) intellectual and sensory capacities, (2) positive expectations about and experiences with literacy from an early age, (3) support for reading-related activities and attitudes so that he or she is prepared to benefit from early literacy experiences and subsequent formal instruction in school, and (4) instructional environments conducive to learning.
This chapter reviews the evidence concerning the predictors of reading achievement: some measurable characteristic of a child or the child's home, school, or community that has been associated with poor progress in learning to read.1It is critical to distinguish predictors from causes or explanations of reading difficultiespredictors are simply correlates. Nor can predictors be interpreted as suggesting the inevitability of poor reading achievement. To the
1 Some sections of this chapter are based closely on a recent review of prediction research by Scarborough (1998), which provides much more detail about the sources and findings that are the basis for many summary statements presented here.
contrary: the whole point of identifying risk factors is to alert parents, physicians, and teachers to potential obstacles children might face so that effective interventions can be devised and implemented.
In the absence of other (noncorrelational) evidence, therefore, these predictors cannot be considered causes of reading problems but rather as associated conditions implicated in reading difficulty. Nevertheless, the fact that these characteristics correlate with subsequent reading achievement is potentially very useful for identifying children who may be in the greatest need of intervention. Our goal in this chapter is to present ways of identifying who should receive services to prevent reading difficulties.
That an individual or group has been identified as being at risk for reading difficulties has no direct implications for the nature of the appropriate intervention. It is not the case that treating the predictor itself is necessarily the right approach; for instance, if difficulty with letter identification turns out to be a predictor, this does not mean that instruction on letter identification is a sufficient or the best treatment for preventing all reading difficulties (see Adams, 1990). Conversely, the skills that are the focus of treatment may not necessarily be the ones on which the identification of the individual or target group was based. In practice, identification criteria and treatment plans can, and often will, be chosen somewhat independently of each other.
It should be borne in mind while reading this chapter that relationships between effective predictors and reading difficulties are markers only and that other mediating variables, which are not measured in a particular research study, may also correlate with reading difficulties. Again consider letter identification: Scanlon and Vellutino (1996) found a moderately high correlation (r= .56) between letter identification and reading achievement. In this same study, the correlation between number identification and reading achievement was .59. Since these results indicate that both poor letter identification and poor number identification predict reading difficulty, they weaken or at least complicate the hypothesis that either of them is a direct cause of reading difficulty. Both may be marker variables for another factor that goes further to explain both letter and number identification.
When deciding which factors to use to identify children who are at risk for reading difficulties, the main determinant should be the strength of the association. (Of course, other practical matters, such as cost and ease of assessment, also affect assessment decisions.) One way to measure the strength of the relationship between a kindergarten predictor and a later reading score is to compute a "correlation" statistic (symbolized by r), which takes a value of zero when there is no predictive relationship at all and takes a value of 1.0 when there is perfect predictability. In between, the higher the correlation, the stronger the tendency for children who did well on the predictor measure to become good readers, and for children who did poorly initially to end up with lower reading achievement scores later. For example, when reading is measured yearly, correlations between scores in one year with scores in the next year are typically in the .60 to .80 range; in other words, they are quite strong but not perfect. As will be seen, correlations between the best kindergarten predictors and later reading scores are not quite as strong (in the .40 to .60 range) but still provide a great deal of useful predictive information. For other predictors, however, the correlations tend to be lower.
Because correlations summarize the strength of the relationship across the full range of children's abilities, their use is consistent with a dimensional account of individual differences in reading discussed in Chapter 3. Another way to look at the strength of prediction instead reflects the categorical model, which continues to predominate in educational practice. In this approach, an at-risk subgroup of kindergartners is designated based on their scores on the predictor measure, and a reading disability subgroup is identified based on later achievement scores. The percentage of children whose outcome classification was correctly predicted is an overall measure of prediction accuracy. Furthermore, a predictor is said to have high sensitivity if most of the disabled readers had been correctly identified as at risk at the outset and to have high specificity if most nondisabled readers had been classified as not at risk. It is also informative to examine errors of prediction, including false positives (children deemed at risk who did not develop reading problems) and false negatives (those who did not meet the risk criterion but nevertheless had difficulty learning to read).
In what follows we attempt to estimate the degree of risk associated with many kinds of predictor measures, alone and in combination. Sometimes the magnitude of risk can be estimated quite closely on the basis of an abundance of longitudinal findings. For other factors, far less information is available regarding the degree of risk they pose. For each predictor, we describe the average strength of its correlation with future reading achievement and, when possible, estimate the probabilities of prediction errors and correct predictions from studies in which risk status has been examined in relation to outcome classifications.
We have organized this chapter by first considering predictors that are intrinsic to the individual and would be identified by assessing the child. We then move to a discussion of factors identified in the household and then to factors associated with the child's larger environmentthe neighborhood, the school, and the community.
CHILD-BASED RISK FACTORS
Physical and Clinical Conditions
Some primary organic conditions are associated with the development of learning problems as secondary symptoms. That is, the child's reading and more general learning problems are thought to result from cognitive or sensory limitations that follow from the primary diagnosis. These primary conditions include:
· severe cognitive deficiencies,
· hearing impairment,
· chronic otitis media,
· (specific) early language impairment, and
· attention deficit/hyperactivity disorder.
Children with severe cognitive deficiencies usually develop very low, if any, reading achievement. Other factors that are associated with developmental delays in cognitive abilities include severe nutri-
tional deficiency, very low birthweight, fetal alcohol syndrome, lead poisoning, and severe psychopathological conditions that emerge in early childhood.
Hearing impairment or deafness is another condition well documented to be associated with reading difficulty (Conrad, 1979; Karchmer et al., 1978; Waters and Doehring, 1990). Chronic ear infections (chronic otitis media) often lead to intermittent hearing loss during the early years. Concern has thus been raised regarding the effects of this on language development and, later, on reading. For chronic otitis media and reading difficulties, results are mixed. Wallace and Hooper (1997) reviewed 18 studies examining chronic otitis media and reading and noted a modest association between the two for language-based skills such as reading.
Early Language Impairment
Although there is tremendous variability in the rate with which children acquire language during their first four years of life, some children are so clearly behind by age 3 that it arouses concern on the part of their parents, neighbors, preschool teachers, pediatricians, or others. In many such cases, delayed language development is the first indication of a broader primary condition, such as a general developmental disability, autism, hearing impairment, or neurological condition, which is likely to be associated with reading difficulty.
In other cases, however, an evaluation by a speech-language professional results in a diagnosis of "(specific) early language impairment"(ELI) and usually the initiation of a course of therapy designed to stimulate language growth in one or more domains.
There have been more than a dozen follow-up studies of the later academic achievements of children who were clinically identified as having specific early language impairment. In this work, the sampling criteria, the initial skill levels of the children, and the measures of outcome status have not always been well specified and are rarely comparable from study to study; nevertheless, several general trends
are evident. First, between 40 and 75 percent of preschoolers with early language impairment develop reading difficulties later, often in conjunction with broader academic achievement problems (Aram and Hall, 1989; Bashir and Scavuzzo, 1992). Second, the risk for reading problems appears to be lowest among those whose early language weaknesses were relatively mild or were confined to a narrow domain (especially to speech production alone). Nevertheless, some children with only mild-to-moderate language delays, who appear to overcome their spoken-language difficulties by the end of the preschool period, remain at greater risk than other youngsters for the development of a reading difficulty (e.g., Scarborough and Dobrich, 1990; Stark et al., 1984; Stothard et al., in press). Third, regardless of a child's general cognitive abilities or therapeutic history, in general the risk for reading problems is greatest when a child's language impairment is severe in any area, broad in scope, or persistent over the preschool years (e.g., Stark et al., 1984; Bishop and Adams, 1990).
Although good evidence indicates that attention deficit/hyperactivity disorder and reading disability are distinct disorders, they frequently co-occur. Longitudinal follow-up indicates that, from the beginning of formal schooling, reading disability is relatively common in children with inattention problems (31 percent in first grade), becoming even more frequent as the child matures (over 50 percent in ninth gradeS.E. Shaywitz et al., 1994; B.A. Shaywitz et al., 1995a).
A visual impairment is not in itself a predictor of reading difficulty. If not correctable, it makes the reading of printed text impossible, so the visually impaired child must instead learn to read Braille manually. Because Braille notation for English text is alphabetic, and because discovering the alphabetic principle is often the biggest obstacle to children in learning to read, many of the same risk fac-
tors that have been identified for sighted children also presumably apply. Unless these or some other Braille-specific processing difficulties (such as poor manual discrimination) are present, there is probably no higher risk for reading difficulties among blind children than among sighted children, provided that early and adequate instruction in reading Braille is provided.
Developmental Differences in Language and Linguistic Development
Children who are developing normally achieve certain milestones of motor, linguistic, and cognitive development at predictable ages. Children who show delays in language development in particular have been studied to determine whether these early language delays relate to literacy problems later on. As described earlier, clinical follow-up studies of preschoolers who had been diagnosed as having ELI indicate that this diagnosis is associated with considerable risk. Even among children who do not receive an ELI diagnosis, there is tremendous variation in language skills. Only a handful of longitudinal prediction studies have initially assessed children from birth through age 4, in part because of the difficulty of testing children accurately in this age range. The main focus of these investigations has been to describe the development of various linguistic and metalinguistic abilities in very young children and then follow them up through their early school years.
To our knowledge, only one study has directly examined the prediction of reading from language and linguistic developmental differences among infants (Shapiro et al., 1990). A composite measure of infant achievement was found to predict reading status (reading disability or not) with .73 sensitivity (i.e., 73 percent of children with reading disability had been classified initially as at risk) and .74 specificity (i.e., 74 percent of nondisabled readers had been classified as not at risk). Individually, the expressive language milestones made a particularly strong contribution to prediction; including IQ in the composite measure did not improve accuracy. Although not sufficiently accurate for practical use, this degree of predictive success is nevertheless remarkably high, particularly in comparison to
the results emerging from studies predicting reading difficulty from kindergarten (see section below).
Walker et al. (1994) cumulatively monitored mean utterance length and number of vocabulary words produced, two developmentally sensitive aspects of emerging language. The two early-language measures, which were highly intercorrelated, correlated moderately well with reading scores in grades 1 through 3, as did the preschool IQ scores.
Bryant et al. (1989, 1990) tested young children on several phonological awareness measures, as well as IQ. Performance on reading tests was predicted by receptive vocabulary, expressive language ability, receptive language ability, nursery rhyme recitation, and IQ. Correlations of the rhyme-matching measure with later reading were not reported, and this measure was only weakly related to the tests of phonological awareness at 40-55months, the last of which were strongly predictive of reading.
Scarborough (1991) considered several language and IQ measures and reading outcomes at the end of grade 2 for a sample of 62 children, about half of whom had parents and/or older siblings with reading problems. IQ scores correlated moderately with later reading, as did scores on receptive language. Expressive vocabulary skill at age 42 months predicted reading a bit more strongly than did receptive vocabulary scores at the same age. In addition, for a subset of 52 children at age 2.5 years (20 from affected families who became reading disabled; 20 similar in sex, socioeconomic status (SES), and IQ; nonreading disability cases from unaffected families; and 12 who became good readers despite a family history of reading disability), measures of expressive phonological (pronunciation accuracy), syntactic (length/complexity of sentences), and lexical (word diversity) abilities were derived from naturalistic observations of children's language during play sessions with their mothers (Scarborough, 1990). The children who became poor readers were much weaker than the other groups on the syntactic and phonological measures. At ages 3, 3.5, and 4 years, however, only the syntactic differences were evident.
What is most striking about the results of the preceding studies is the power of early preschool language to predict reading three to five
years later. In fact, the correlations between reading and early preschool measures are almost as high as those between kindergarten predictors and reading (see next section).
Predictors at School Entry
Acquired Proficiency in Language
Spoken language and reading have much in common. If the printed words can be efficiently recognized, comprehension of connected text depends heavily on the reader's oral-language abilities, particularly with regard to understanding the meanings of words that have been identified and the syntactic and semantic relationships among them. Indeed, many early research reports called attention to the differences between good and poor readers in their comprehension and production of structural relations within spoken sentences.
Given the close relationship between reading and language, we should expect that normally occurring variations in language differences would be related to speed or ease of the acquisition of reading. Earlier, we reviewed the empirical data indicating that language development in the preschool years is indeed related to later reading achievement and that preschoolers with language disabilities are highly likely to show reading problems as well. Here we consider whether variation in language abilities at the time children typically begin to receive formal reading instruction also relates to variability in reading outcomes.
The ability to retain verbal information in working memory is essential for reading and learning, so it might be expected that verbal memory measures would be effective predictors of future reading achievement. Many prediction studies have included such measures within their predictor batteries. From the results of those studies (Scarborough, 1998), it is clear that, on average, kindergartners' abilities to repeat sentences or to recall a brief story that was just read aloud to them are more strongly related to their future reading achievement than are their scores on digit span, word span,
and pseudo-word repetition measures. Sentence or story recall (r = .45), in fact, compares favorably with other predictors of reading (see Table 4-1).
Lexical and Syntactic Skills
Several kinds of vocabulary measures have been examined as predictors of future reading achievement. On each trial of a ''receptive" vocabulary test, the child must indicate which of several pictures best corresponds to the word (usually a noun, adjective, or gerund) spoken by the examiner. A long series of items of increasing difficulty is available, and testing terminates when the child's vocabulary level is exceeded. As shown in Table 41, in 20 prediction studies the mean correlation between receptive vocabulary scores in kindergarten and subsequent reading scores in the first three grades is .36.
With regard to lexical abilities, one can also examine expressive, rather than receptive, vocabulary, which is also sometimes referred to as "confrontation naming" or simply "object naming." On such tests, the child is shown a series of drawings of objects and is asked to name each one. Compared with receptive tests, these measures place greater demands on accurate retrieval of stored phonological representations of lexical items and on the formulation and production of spoken responses.
To our knowledge, only five kindergarten prediction studies have included confrontation naming measures in the predictor battery, but the magnitude and consistency of the results of those studies suggest that naming vocabulary is a reliable predictor of future reading ability. On average, expressive vocabulary measures are associated (r = .45) with a considerable amount of variance in subsequent reading scores, which compares favorably with the effect sizes for receptive vocabulary and IQ.
Not only the accuracy of name production but also its speed can be measured. Rapid serial naming speed has been shown to correlate with concurrent and future reading ability but not with IQ in several dozen studies of schoolchildren (e.g., Ackerman et al., 1990; Bowers and Swanson, 1991; Cornwall, 1992; Denckla and Rudel, 1976b; Felton et al., 1987; Spring and Davis, 1988; Wolf and Obregon, 1992). Rapid serial naming speed has been found to be
TABLE 4-1 Prediction of Reading Difficulties at School Entry
Factors Identified in the Child
Number of Samples
Strength of Relationship
Verbal memory for stories/sentences
Median r = .49
1. Receptive vocabulary
Median r = .33
2. Confrontation naming
Median r = .49
3. Rapid serial naming
Median r = .40
Receptive language, syntax/morphology
Median r = .38
Median r = .37
Median r = .47
Median r = .42
Early Literacy-Related Skills
Median r = .56
Median r = .53
Concepts of print
Median r = .49
NOTE: Only studies with sample sizes of 30 or more were considered. At least one of the risk factors of interest had to be assessed initially when the children were within about one year of beginning formal schooling in reading, and at least one assessment of reading skills had to be obtained after one, two, or occasionally three years of instruction. If a word recognition measure was used in a prediction study, its correlation(s) with predictors was used; otherwise, a composite reading score or, rarely, a reading comprehension measure was instead accepted as the criterion variable. When more than one correlation value per risk factor was available in a given sample of children (because multiple reading assessments were conducted and/or because multiple measures of the predictor were used), the average correlation for the sample was used for aggregation. To obtain the average correlations across samples, therefore, each contributing sample contributed only one independent observation. SOURCE: adapted from Scarborough (1998).
related to speech production (Kamhi and Catts, 1986; Kamhi et al., 1988). Somewhat weaker associations with reading are obtained when "discrete" naming (response time to name an individual stimulus) rather than "serial" naming is measured, suggesting that the naming speed problems of poor readers involve more than just difficulty in retrieving and producing item names. A full understanding of the relationship between speeded naming and reading remains to be determined.
Studies have also been made of the semantic, morphological, and syntactic skills of kindergartners. Receptive language measures (sentence comprehension) that emphasize the understanding of complex syntactic and morphological forms have been more successful predictors than other (or unspecified) kinds of receptive measures (Table 4-1).
Expressive language (production) measures, which include mean length of utterance, sentence completion, tasks requiring the child to fill in morphological markers, and others, are about equally strongly predictive of reading as receptive language. It should be noted, however, that the goal in these studies has been to predict reading achievement during the first few school grades, when the emphasis is primarily on the acquisition of word recognition and decoding skills rather than on the comprehension of challenging material.
In examining the connection between measures of overall language ability and future reading achievement, the highest average correlation has been found when a broad composite index of language abilities has been used. Since only four studies have taken this approach, these findings can be considered promising but not conclusive at this point.
Phonological awareness, or phonological sensitivity, is the ability to attend explicitly to the phonological structure of spoken words, rather than just to their meanings and syntactic roles. This metalinguistic skill involves treating language as the object of thought, rather than merely using language for communication.
Given the importance of phoneme-letter mapping in the English alphabetic writing system, phonological awareness would be expected to be an excellent predictor of the future reading skills of kindergartners, particularly when the child's appreciation of the subsyllabic or phonemic structure of words is measured. This predictive correlational relationship has been examined in 27 research samples from 24 studies (Table 4-1). On average, phonological awareness (r = .46) has been about as strong a predictor of future reading as memory for sentences and stories, confrontation naming, and general language measures.
When classificatory analyses are conducted, phonological awareness in kindergarten appears to have the tendency to be a more successful predictor of future superior reading than of future reading problems (Wagner, 1997; Scarborough, 1998). That is, among children who have recently begun or will soon begin kindergarten, few of those with strong phonological awareness skills will stumble in learning to read, but many of those with weak phonological sensitivity will go on to become adequate readers (Bradley and Bryant, 1983, 1985; Catts, 1991a, 1996; Mann, 1994; also see discussion of the reciprocity between phonological awareness and reading presented in Chapter 2).
In sum, despite the theoretical importance of phonological awareness for learning to read, its predictive power is somewhat muted, because, at about the time of the onset of schooling, so many children who will go on to become normally achieving readers have not yet attained much, if any, appreciation of the phonological structure of oral language, making them nearly indistinguishable in this regard from children who will indeed encounter reading difficulties down the road.
Acquired Knowledge of Literacy
Even before children can read in the conventional sense, most have acquired some information about the purposes, mechanics, and component skills of the reading task. For some children, opportunities for acquiring this sort of information abound, whereas others have relatively little relevant exposure (McCormick and Mason,
1986). Therefore, by the time children are about to begin school, they vary considerably in how much they already know about books and reading. Researchers have tested children's reading readiness, letter identification, and concepts of print to determine whether differences in these abilities can predict differences in future reading achievement.
Reading readiness is a term used by both researchers and educators to mean accomplishment of skills presumed to be prerequisite to benefiting from formal reading instruction. It is measured by comparing the accomplishments of children in kindergarten, where prereading skills are practiced, with their scores on standardized reading tests in the primary grades. Reading readiness has been shown to have a high correlation with reading ability: children who lack reading readiness at school entry have a harder time learning to read in the primary grades. This has been found in prediction studies since 1950 (Hammill and McNutt, 1980; Scarborough, 1998).
Among the readiness skills that are traditionally evaluated, the one that appears to be the strongest predictor on its own is letter identification. Table 4-1 shows a summary of results for longitudinal studies since 1975 that have included this measure. Just measuring how many letters a kindergartner is able to name when shown letters in a random order appears to be nearly as successful at predicting future reading, as is an entire readiness test.
The prediction of future reading by kindergarten measures of letter identification and other early reading skills is quite substantial, accounting on average for nearly one-third of the variance in reading at grades 1-3. Nevertheless, the predictive accuracy derived from using such readiness measures alone may be lower than desirable for practical purposes. For instance, in Scanlon and Vellutino's very large district-wide sample, letter knowledge was highly correlated with reading test scores and with teacher ratings of reading skill at the end of first grade. The results obtained when letter identification was used to classify kindergartners as at risk versus not at risk are shown in Table 4-2.
TABLE 4-2 Accuracy of Prediction of Grade 1 Reading Status from Kindergarten Letter Identification Differences
Classification of Kindergartners According to Their Letter Identification Skills
A. Stricter kindergarten cutoff
"at risk" (bottom 10 percent)
"not at risk" (top 90 percent)
Grade 1 Reading (Teacher Ratings)
Bottom 20 percent
63 (correctly predicted)
131 "miss" errors
Top 80 percent
37 "false alarm" errors
769 (correctly predicted)
B. More lenient kindergarten cutoff
"at risk" (bottom 25 percent)
"not at risk"(top 75 percent)
Grade 1 Reading (Teacher Ratings)
Bottom 20 percent
118 (correctly predicted)
73 "miss" errors
Top 80 percent
132 "false alarm" errors
677 (correctly predicted)
SOURCE: Adapted from Scanlon and Vellutino (1996)
The upper part of the table illustrates the pattern of prediction errors when a rather strict criterion was adopted, that is, when letter identification scores were used to identify the bottom 10 percent of kindergartners as at risk. When approximately the bottom 20 percent of first graders were designated as having reading difficulties, 83.2 percent of the grade 1 outcomes of the approximately 1,000
children would have been correctly predicted on the basis of letter knowledge. This is far better than chance and better than could be achieved using any other single kindergarten measure, but it still means that a considerable number of prediction errors would occur. Of the 100 kindergartners who would have been identified as most at risk (and who would presumably be targeted to receive intervention), 37 would have turned out not to have reading difficulties. Furthermore, of the 900 children deemed not to be at risk on the basis of letter knowledge, 131 (14.5 percent) would have developed reading problems by the end of first grade. In other words, only about one-third of the children who became the poorest readers would have been selected initially for early intervention.
Table 4-2B also shows that, when a more lenient criterion was used to classify kindergartners, such that 25 percent rather than 10 percent were considered at risk, the "miss" rate would drop to a more acceptable level (10 percent). However, the overall accuracy of prediction would decrease (to 79.5 percent), and the rate of false positives would increase substantially, such that less than half of the at-risk group would be expected to develop reading difficulties.
To increase the accuracy with which kindergartners at greatest risk can be identified, it may be useful to examine other individual risk factors that may provide additional information about how readily a child is likely to learn to read.
Concepts of Print
The term "concepts of print" refers to a general understanding of how print can be used rather than knowledge about specific letters. It has been shown to have a moderate correlation with reading ability in the primary grades. A recent study with even higher correlations used two types of measures related to print: ones related to understanding how print can be used and ones related to the mechanics of the writing system (letter naming or letter-sound correspondences) (Stuart, 1995). It therefore appears promising that this combined approach will be more accurate in identifying children at risk, although more work on developing and validating these test batteries is needed.
Other Factors Measured at School Entry
Researchers have examined a number of other factors to see whether there is a connection between them and future reading achievement. A number of longitudinal studies have examined kindergartners' speech perception or production abilities, as well as visual and motor skills, nonverbal memory, age for grade, and sex. The results suggest that these measures are consistently weak predictors of subsequent reading differences. Likewise, nonverbal IQ scores are poor predictors, but verbal (and overall) IQ is about equivalent in strength to receptive vocabulary and various other language measures.
Prediction Based on Multiple Risk Factors
We have assessed individual child predictors to determine whether any of these factors are sufficiently strongly related to reading difficulties that they can be used to help identify children who should receive prevention, intervention, or remediation. Note that from the research we have different measures that predict more strongly at different ages. Across the age span of birth through grade 3, cognitive deficiencies, hearing impairment, and a diagnosed specific early language impairment have strong associations with future reading difficulties. Low IQ and lack of general language ability in infancy through kindergarten are associated with future reading difficulties. Also, in kindergarten, reading readiness measures, letter identification, concepts of print, verbal memory for stories and sentences, confrontation naming, overall language, phonological awareness, and expressive vocabulary or naming skills are associated with future reading ability.
From our review of child-based factors, it should be clear that many measurable individual differences among children at the outset of schooling are reliably correlated with future reading achievement but that most are not strong enough on their own to provide the level of predictive accuracy that would be desired for practical purposes. For this reason, many researchers have examined the combined effects of several or many predictors (e.g., Badian, 1982, 1994;
Butler et al., 1985; Felton, 1992; Bishop and Adams, 1990; Catts, 1991b, 1993; Horn and O'Donnel, 1984).
Different researchers included very different sets of predictor measures in their kindergarten batteries. Most used are some kinds of index of early print skills, such as letter knowledge, word recognition, concepts of print, teacher ratings, and writing. Unfortunately, the other measures that appear to be the strongest single predictors (phonological awareness, sentence/story recall, confrontation naming, and broad language indices) were rarely assessed in these studies, so their potential contributions to prediction when combined with other variables remain unknown.
In most of the studies, multiple regression analyses yielded measures of the strength of the relationship between kindergarten measures and later reading achievement. On average, 57 percent of the variance in reading scores was accounted for by the analysis. In comparison, the mean effect size for readiness tests alone was considerably lower, indicating that adding other kinds of measures to the traditional readiness tests can effectively strengthen the prediction. Moreover, it is impressive that the average correlation in these studies is about as strong as the year-to-year correlations among reading achievement.
Classificatory analyses were conducted in three studies that had the kindergarten measure as the predictor of second or third grade reading achievement (Badian, 1982; Butler et al., 1985; Felton, 1992). The percentage of children whose reading outcome status (reading disabled or nondisabled) was correctly predicted by kindergarten risk status (based on the predictor battery) ranged from 80 to 92 percent. These prediction analyses tended to achieve specificity (i.e., 80 to 92 percent of nondisabled readers had been classified as not at risk in kindergarten) but somewhat lower sensitivity (i.e., 56 to 92 percent of reading-disabled children had been classified initially as at risk). Negative predictive power ranged from 89 to 99 percent; in other words, on average, the proportion of not-at-risk children who nevertheless developed reading problems was low. Positive predictive power, however, ranged from 31 to 76 percent; that is, the proportion of at-risk children who turned out not to have
reading difficulties was substantial and was not markedly lower than when predictions have been based on individual predictors.
In addition, two recent longitudinal studies are particularly informative about the prediction of reading ability for children with early language impairment, based on their observed differences at about the time of school entry (Bishop and Adams, 1990; Catts, 1991, 1993). In both studies, 50 percent of the variance in reading achievement in the sample could be accounted for by a small set of predictors measured at about age 5. In Catts's study, measures of phonological awareness and rapid serial naming of objects permitted 83 percent of the children's outcomes to be correctly predicted, with a false positive rate of 32 percent and a false negative rate of 13 percent. In the Bishop and Adams sample, the predictor set included IQ and a combination of language ability indices. Clearly, the accuracy of prediction in these samples was lower than in the population-representative samples.
The pattern of classification errors is quite similar across these studies and suggests that a fair number of children who will have reading difficulties do not obtain low enough scores to merit an at-risk designation on the basis of the kinds of kindergarten measures that were used (most typically literacy-specific knowledge, phonological awareness, and IQ). Whether the inclusion of sentence/story recall, naming vocabulary, and broader kindergarten batteries would help to pick up these cases is unknown, but it merits investigation on the basis of the strong bivariate results.
Nevertheless, it is clear that batteries consisting of multiple measures are becoming accurate enough to be very useful for identifying individual children who are at greater risk than their classmates. Close monitoring of these children (including follow-up assessments and observations by their kindergarten teachers) would permit them to receive additional assistance (if it turns out to be needed) as soon as possible, a highly desirable objective. Note, however, that individual testing of all kindergartners, which can be costly, probably has less utility in a school in which a large number of entering students are at risk due to economic disadvantage or other group risk factors, discussed below. In that circumstance, the highest pri-
ority in allocating resources should address the goal of raising the group's overall level of achievement.
FAMILY-BASED RISK FACTORS
In many circumstances, early identification of children who will have reading difficulties might proceed better by considering target groups rather than by assessing individuals. Demographic data suggest that a majority of reading problems tend to occur in children from poor families with little education, although they may of course occur in families that are neither poor nor undereducated. Also, being a member of a family in which reading difficulties have occurred before may also constitute a risk, whether for biological or environmental reasons. We review here a number of factors identifiable at the level of the family to assess their value in identifying children who should receive prevention and intervention activities.
Family History ofReading Difficulties
Are children whose parents or older siblings have exhibited reading problems at greater risk for reading difficulties than are other children of otherwise similar backgrounds? Decades of research on the familial aggregation of reading problems suggest that this is so. Factors identified as family risk factors include family history of reading problems, home literacy environment, verbal interaction, language other than English, nonstandard dialect, and family-based socioeconomic status (SES). It is important to bear in mind, however, that family patterns of reading problems can be attributed either to shared genetic or to shared environmental factors (see Chapter 1).
If a child is diagnosed with a reading disability, there is a higher than normal probability that other family members will also have difficulties with reading (see Finucci et al., 1976; Hallgren, 1950; Gilger et al., 1991; Vogler et al., 1985). The exact probability seems to depend on a variety of factors, including the severity of the child's reading disability. Furthermore, when the parents' diagnosis for reading disability is based on self-report, the family incidence tends
to be lower than when the diagnosis is based on the direct measurement of parents' reading skills (Gilger et al., 1991).
Most studies of familial incidence first diagnose a child with reading disability using a severity criterion that would identify 5 to 10 percent of children who have normal intelligence and have had what for the majority of children is effective education. The investigators then attempt to use a similar severity criterion to diagnose reading disability in the parents. Evidence for the family nature of reading disability is based on parental rates that are substantially above the 5 to 10 percent rate estimated for the population. Scarborough (1998) computed the average rate of reading disability among parents across eight family studies that included a total of 516 families. The rate across studies varied from 25 to 60 percent, with a median value of 37 percent. Thus, all studies found rates for reading disability among parents of reading-disabled children that were considerably higher than expected in the normal population. The median proportion of reading disability among fathers (46 percent) was slightly higher than the median proportion among mothers (33 percent).
A few studies have attempted to estimate the prospective risk to the child when parental disabilities are identified first (Finucci et al., 1985; Fowler and Cross, 1986; Scarborough, 1990). Those prospective studies clearly show that parents' reading disabilities predict a higher than normal rate of reading disabilities in their children (31 to 62 percent versus 5 to 10 percent). Although parental reading disabilities are not completely predictive of their children's reading disabilities, the substantially greater risk at least warrants very close monitoring of their children's progress in early language and literacy development. Results from two predictive studies (Elbro et al., 1996; Scarborough, 1989, 1990, 1991) suggest that whether these children develop reading problems can be predicted from preschool measures of language and literacy skills. If so, it would be potentially affordable to assess that small subset of the population a year or two before kindergarten and to provide intervention to those with the weakest skills. Of course, to do so would require an effective means of persuading parents with a history of reading problems to step forward so that this service could be provided for their offspring.
This sort of recruitment program has never been attempted, so its feasibility is unknown.
Home Literacy Environment
Families differ enormously in the level to which they provide a supportive environment for a child's literacy development. Measures of the home literacy environment itself, therefore, may provide an indication of an individual child's degree of risk for reading difficulties. Hess and Holloway (1984) identified five broad areas of family functioning that may influence reading development. The first four are:
1. Value placed on literacy: by reading themselves and encouraging children to read, parents can demonstrate that they value reading.
2. Press for achievement: by expressing their expectations for achievement by their children, providing reading instruction, and responding to the children's reading initiations and interest, parents can create a press for achievement.
3. Availability and instrumental use of reading materials: literacy experiences are more likely to occur in homes that contain children's books and other reading and writing materials.
4. Reading with children: parents can read to preschoolers at bedtime or other times and can listen to schoolchildren's oral reading, providing assistance as needed.
The fifth area, opportunities for verbal interaction, is presented in the next section. Although conceptually distinct and perhaps analytically useful to consider separately, in practice these areas may be highly interrelated. In addition, home characteristics and social class covary to a degree.
We review results of longitudinal prediction studies that have examined aspects of the home environment during children's early years (birth to about age 5) in relation to the development of literacy knowledge and skills during the preschool years and especially to the children's subsequent academic achievement during the primary
school grades. Few studies have derived overall measures of the quality of the preschool home environment.
Most longitudinal studies have looked at the home environment of children at different ages and have identified contributors to literacy development. Unless otherwise indicated, measures of home variables were derived from parental interviews or questionnaires administered at or shortly before the children entered kindergarten, and reading achievement was measured by standardized tests in the first and/or second grade (e.g., DeBaryshe, 1993; DeBaryshe et al., 1991; Mason, 1980; Mason and Dunning, 1986; Scarborough et al., 1991; Share et al., 1984; Thomas, 1984; Wells, 1985).
In summary, although there is considerable evidence that differences in the home literacy environments of preschoolers are related to subsequent achievement differences, the strength of these correlations has tended to be modest, particularly when measured in large population-representative samples (Bus et al., 1995; Scarborough and Dobrich, 1994). Thus, a preschooler whose home provides fewer opportunities for acquiring knowledge and skills pertaining to books and reading is at somewhat higher risk for reading difficulties than a child whose home affords a richer literacy environment.
Opportunities for Verbal Interaction
The major dimension of variability for measures of verbal interaction in the home is the dimension of quantity. It is now clear that, though poor and uneducated families provide much the same array of language experiences as middle-class educated families, the quantity of verbal interaction they tend to provide is much less (Hart and Risley, 1995). A lower quantity of verbal interaction constitutes a risk factor primarily in that it relates closely to lowered child vocabulary scores, as shown in one large prospective observational study (Hart and Risley, 1995) and in a score of less rigorous studies. Because vocabulary is associated with reading outcomes (see Table 4-1), it seems likely that reduced opportunities for verbal interaction would function as a risk factor. Furthermore, language-rich experiences in the home are typically associated with activities (like book reading, shared dinner table conversations) that themselves show
only modest predictive value. It is possible, too, that the effects of differences in verbal interaction may not show up until after the primary grades, that is, when more high-level comprehension is required.
Home Language Other Than English
When a preschool child's home language is not primarily English, the ease of learning to read printed English is likely to be impeded to some extent, particularly if reading instruction in English begins before the child has acquired oral proficiency in English (see August and Hakuta, 1997). One difficulty in trying to evaluate the degree of risk associated with limited English proficiency is that cultural as well as linguistic differences are also involved and may introduce other kinds of risk factors.
Many Hispanic children with limited English proficiency also have in common that their parents are poorly educated, that their family income is low, that they reside in communities in which many families are similarly struggling, and that they attend schools with student bodies that are predominantly minority and low achieving. Not surprisingly, the other factors that have been proposed to explain the typically low levels of academic achievement among Hispanic students include many that have been cited as contributing to the risk factors facing other minority groups, including low SES (and its many concomitant conditions), cultural differences between the home and school (e.g., regarding educational values and expectations), sociopolitical factors (including past and ongoing discrimination and low perceived opportunities for minorities), and school quality.
In summary, low English proficiency in a Hispanic child is a strong indication that the child is at risk for reading difficulty. That low reading achievement is a widespread problem among Hispanic students even when they are instructed and tested in Spanish, however, indicates that linguistic differences are not solely responsible for the high degree of risk faced by these children and that the role of co-occurring group risk factors, particularly school quality, home literacy background, and SES, must be considered.
Use of a Nonstandard Dialect of English in the Home
Dialect differences among English speakers are widely recognizedfor example, a Boston accent or a Southern drawl. There is ample evidence that listeners make stereotyped judgments about speakers of particular dialects. Of greater concern here, however, is that some dialect differences are viewed by some not as regional variations but as ''incorrect" English, connoting aberrant or delayed language development, poor learning, lazy or sloppy articulation, or even purposeful insolence. Particularly under these conditions, the differences between a young child's dialect and the standard classroom English dialect may become a risk factor for reading difficulties.
With regard to reading instruction in particular, the risk for confusion is considerable. For example, if the teacher is pointing out the letter-sound correspondences within a word that is pronounced quite differently in the child's dialect than in the teacher's, the lesson could confuse more than enlighten. Moreover, teachers who are insensitive to dialect differences may develop negative perceptions of children and low expectations for their achievement, and they may adjust their teaching downward in accord with those judgments.
Although these situations undeniably occur, there are many difficulties in measuring the extent to which they happen and the degree to which their occurrence is correlated with, and may contribute to, poor reading achievement. As is the case for children with limited English proficiency, dialect differences are often confounded with poverty, cultural differences, substandard schooling, and other conditions that may themselves impose very high risks for reading difficulties. Even measuring the phenomena and their relation to achievement is confounded by the risk factor itself (Labov, 1966; Smitherman, 1977; Wolfram, 1991). The knowledge base, therefore, is spotty. Some dialects have been researched more thoroughly than others.
Socioeconomic differences are conventionally indexed by such demographic variables as household income and parents' education and occupation, alone or in some weighted combination. In educational studies, furthermore, the socioeconomic level of a school or district may be estimated by the percentage of the enrollment qualifying for federal lunch subsidies. (For a critique and a discussion of some recommended modifications of current methods of measuring SES, see Entwisle and Astone, 1994). Families rated low in SES are not only less affluent and less educated than other families but also tend to live in communities in which the average family SES is low and tend to receive less adequate nutrition and health services, including prenatal and pediatric care. In other ways, too, low SES often encompasses a broad array of conditions that may be detrimental to the health, safety, and development of young children, which on their own may serve as risk factors for reading difficulties. Teasing apart the various aspects of the environment associated with low SES is virtually impossible, and this should be borne in mind as we discuss some particular risk factors that are linked to poverty.
As far back as Galton's (1874) studies of English scientists, SES has consistently been shown to predict cognitive and academic outcomes (Hess and Holloway, 1984; White, 1982, Pungello et al., 1996). Although reliable, the relationship between SES and reading achievement is more complex than is generally realized. Consider, for example, how the findings of Alexander and Entwisle (1996)that low SES students progress at identical rates as middle and high SES students during the school year, but they lose ground during the summershed light on the relationship between SES and reading achievement.
The degree of risk associated with the SES of the individual child's family differs considerably from the degree of risk associated with the SES level of the group of students attending a particular school. The evidence for this, and its implications for the prevention of reading difficulties among such students, is reviewed here. In an earlier section, we turned our attention to aspects of the home envi-
ronment that may be responsible for the degree of risk posed to the individual child from a low SES home.
In principle, low SES could potentially carry risk for reading difficulty for an individual child and for entire groups of children. That is, low SES is an individual risk factor to the extent that among children attending the same schools, youngsters from low-income families are more likely to become poorer readers than those from high-income families. Low SES is also a group risk factor because children from low-income communities are likely to become poorer readers than children from more affluent communities. Because the former are more likely to attend substandard schools, the correlation between SES and low achievement is probably mediated, in large part, by differences in the quality of school experiences. It is thus not very surprising that the strength of the correlation between SES and achievement is stronger when the unit of analysis is the school than when the unit of analysis is the individual child (Bryk and Raudenbush, 1992, on multilevel measures of school effects).
When the average SES of a school (or district) and the average achievement level of the students attending that school are obtained for a large sample of schools, a correlation between SES and achievement can be calculated using the school as the unit of analysis. In a meta-analytic review of the findings for 93 such samples, White (1982) found that the average size of the correlation was .68, which is substantial and dovetails with the conclusion of the section below that attending a substandard school (which is usually one whose students tend to be low in both SES and achievement) constitutes a risk factor for the entire group of children in that school.
When achievement scores and SES are measured individually for all children in a large sample, however, the strength of the association between SES and achievement is far lower. In White's (1982) meta-analysis, for instance, the average correlation between reading achievement and SES across 174 such samples was .23. Similarly, the correlation was .22 in a sample of 1,4599-year-old students whose scores were obtained through the National Assessment of Educational Progress (NAEP) evaluations (Walberg and Tsai, 1985). In a meta-analysis of longitudinal prediction studies, Horn and
O'Donnell (1984) obtained a correlation that was only slightly higher (.31) between SES and early school achievement.
Similar SES findings were found in population-representative studies in the United States and in other English-speaking countries (e.g., Alwin and Thornton, 1984; Estrada et al., 1987; Richman et al., 1982; Rowe, 1991; Share et al., 1984; Wells, 1985). In other words, within a given school or district, or across many districts within a country, SES differences among children are relatively weak predictors of achievement. Thus, all else being equal, coming from a family of low SES (defined according to income, education, and occupation of the parents) does not by itself greatly increase a child's risk for having difficulty in learning to read after school income level has been accounted for.
We are not saying here that SES is not an important risk marker. What we are saying is that its effects are strongest when it is used to indicate the status of a school or a community or a district, not the status of individuals. A low-status child in a generally moderate or upper-status school or community is far less at risk than that same child in a whole school or community of low-status children.
Analysis of Family-Based Risk Factors
Parents' reading disabilities predict a higher than normal rate of reading disabilities in their children (31 to 62 percent versus 5 to 10 percent). Although parental reading disabilities are not completely predictive of their children's reading disabilities, the substantially greater risk at least warrants very close monitoring of their children's progress in early language and literacy development. Lack of English proficiency for a Hispanic child is a strong indication that he or she is at risk for reading difficulty; however, linguistic differences appear to be less responsible than other co-occurring group risk factors, particularly school quality. In a similar manner, the occurrence of family use of nonstandard dialect and individual family SES covary considerably with factors such as school quality, which is discussed in the next major section of this chapter.
The quantity of verbal interaction in families constitutes a risk factor primarily in that it relates closely to child vocabulary scores.
Findings related to home literacy environments are mixed. Many of the large-scale studies (Walberg and Tsai, 1984; White, 1982) of the correlations between home environment and school achievement have focused primarily on samples of children in elementary school (or older). Because the focus of this report is on the prevention of reading difficulties in young children, it is especially important to consider the different roles that home environment may play at different ages. In particular, the opportunities provided in the home for literacy acquisition during the preschool years may contribute primarily to the child's acquisition of attitudes toward literacy, of knowledge about the purpose and mechanics of reading, and of skills (such as vocabulary growth and letter knowledge) that may facilitate learning when school instruction begins. Once the child has begun to attend school and has started to learn to read, the contributions of home and parents may be somewhat different; assistance with homework, listening to the child's efforts at reading aloud, the availability of resources such as a dictionary and an encyclopedia, and so forth may be particularly important for fostering high achievement in school.
NEIGHBORHOOD, COMMUNITY, AND SCHOOL-BASED RISK FACTORS
As is clear from our discussion of the family-based factors that constitute risks, it is extremely difficult to disentangle the effects of family practices from factors such as the neighborhood where the family lives, the cultural and economic community of which the family is a part, and the school the child attends. In this section, we focus on these issues, noting that more research has addressed schooling rather than environmental risks to reading development.
A school in which students are performing at a much higher (or much lower) level than might be predicted using such standard measures as family SES is often described as an "outlier." Studies of outlier schools have overwhelmingly concentrated on positive outlier schools. Variously referred to as studies of "exemplary schools" (Weber, 1971), "unusually effective schools" (Levine and Lezotte, 1990), and ''high-flying" schools (Anderson et al., 1992), these posi-
tive outlier studies have made important contributions to the field (for a review, see Stringfield, 1994). Of the studies that have examined both positive and negative outlier schools, the largest and longest running has been the Louisiana School Effectiveness Study (Stringfield and Teddlie, 1988, 1991; Teddlie and Stringfield, 1993). Classroom practices in ineffective schools (regardless of community SES) were characterized by significantly lower rates of student time on task, less teacher presentation of new material, lower rates of teacher communication of high academic expectations, fewer instances of positive reinforcement, more classroom interruptions, more discipline problems, and a classroom ambiance generally rated as less friendly (Teddlie et al., 1989).
Stringfield and Teddlie (1991) also conducted detailed qualitative analyses of the 16 case studies. Those analyses added significantly to the quantitative findings. Qualitative differentiations were made at three levels: the student, the classroom, and the school.
At the level of student activities, ineffective schools were found to be different from more effective, demographically matched schools in two ways. First, students' time-on-task rates were either uniformly low or markedly uneven. Time on task is a good predictor of achievement gain (Stallings, 1980). In some schools, very few academic tasks were put before any students, and in other schools there were marked differences in the demands made of students, with only some students being required to make a concerted academic effort. Students in positive outlier schools were more uniformly engaged in academic work.
The second student-level variable was whether tasks were put before the students in what appeared to the students to be an organized and goal-oriented fashion. When interviewed, students at ineffective schools were much less likely to be aware of why they were being asked to do a task, how the task built on prior schoolwork, and how it might be expected to lay a foundation for future work.
At the classroom level, ineffective schools were characterized by a leisurely pace, minimal moderate-to-long-term planning, low or uneven rates of interactive teaching, and a preponderance of "ditto sheets" and other relatively unengaging tasks. One of the most readily observable of the classroom differences was that teachers in
ineffective schools simply failed to cover all of the district-mandated materials by year's end. These students were not being provided equal "opportunity to learn." (For a discussion of the power of opportunity to learn, see Muthen et al., 1991). Finally, ineffective schools were structured such that teachers almost invariably taught in isolation from one another; there was little focus on building a professional knowledge base within the school. An additional factor, class size, is related to achievement (Mosteller et al., 1996).
During the kindergarten year, there is evidence that teacher-child relationships are important for later school achievement. Studies have defined the significant qualities of these relationships (Howes and Hamilton, 1992; Howes and Matheson, 1992). One study used a scale based on these findings that describes teachers' perceptions of different qualities of their relationships with their students (Pianta and Steinberg, 1992). Another study compared results on this scale and readiness tests and found that two global qualities of the teacher-child relationship, dependency or conflict, were related to poor performance (Birch and Ladd, 1997). Dependency is an index of the child's overdependence on the teacher; conflict is an index of friction in the teacher-child relationship. Closeness in the teacher-child relationship was associated with better readiness performance. Closeness is an index of warmth and open communication in the teacher-child relationship.
At the school level, ineffective schools were observed to be different from their demographically matched peers along seven dimensions: (1) they were not academically focused; (2) the school's daily schedule was not an accurate guide to academic time usage; (3) resources often worked at cross-purposes instructionally; (4) principals seemed uninterested in curricula; (5) principals were relatively passive in the recruitment of new teachers, in the selection of professional development topics and opportunities for the teachers, and in the performance of teacher evaluations; (6) libraries and other media resources were rarely used to their full potential; and (7) there were few systems of public reward for students' academic excellence. Similar descriptions of a smaller set of negative outlier schools have been provided by Venezky and Winfield (1979).
In this chapter we have examined information about risk factors to determine what kinds of risk are so strongly related to reading difficulties that they can potentially be used to identify children in need of prevention and early intervention. It is clear that the relationships between risk factors and reading achievement are continuous and probabilistic, not categorical or deterministic. Misleading conclusions can be reached if risk factors are not interpreted in this light. It must always be borne in mind that many children whose language and literacy skills are weak at the outset of schooling become successful readers. A majority, however, do not, giving rise to the correlational evidence we have reviewed. It bears repeating, also, that a causal relationship to reading has been shown for only some, but not all, of the measures that best predict future reading ability. Our review of prediction studies indicates clearly that no single risk factor, on its own, is sufficiently accurate to be of practical use for predicting reading difficulties. In combination, however, measures of various kinds of riskindividual, familial, and demographiccan provide useful estimates of future achievement levels. Although prediction accuracy is far from perfect, errors of prediction can be tolerated as long as children's progress is carefully monitored during kindergarten and beyond. As discussed below, how different school systems can best use the available information about risk indicators must be tailored to their particular needs, goals, and resources.
Group Risk Factors
It is abundantly clear that some groups of children are at risk for reading difficulties because they are affected by any or all of the following conditions:
1. They are expected to attend schools in which achievement is chronically low,
2. they reside in low-income families and live in poor neighborhoods,
3. they have limited proficiency in spoken English, and
4. they speak a dialect of English that differs substantially from the one used in school.
Individual Risk Factors
The evidence also indicates that individual children, whether or not faced with the adverse conditions just mentioned, may be at greater risk than other otherwise-comparable children for reading difficulties for any or all of the following reasons:
1. They are children of parents with histories of reading difficulty;
2. they have acquired less knowledge and skill pertaining to literacy during the preschool years, either through lack of appropriate home literacy experiences and/or as a result of some inherent cognitive limitations;
3. they lack age-appropriate skills in literacy-related cognitivelinguistic processing, especially phonological awareness, confrontational naming, sentence/story recall, and general language ability;
4. they have been diagnosed as having specific early language impairment;
5. they have a hearing impairment; and
6. they have a primary medical diagnosis with which reading problems tend to occur as a secondary symptom.
Practical Use of This Information
Detecting problems early, in order to avoid other problems later on, is the most practical course. The ease, cost, and reliability with which various risk factors can be measured are therefore a central concern.
Many of the group factors named above (e.g., a child is expected to attend a school in which achievement is chronically low, the child lives in a low-income family and neighborhood) are easily accessible measures. When they are present, effective preventions and early interventions can be provided throughout the age span we are addressing in this reportbirth through grade 3.
Pediatric screening tools are effective in identifying children who have severe sensory or developmental impairments (hearing impairment, specific language impairment). When these are present, preventions and early interventions can be provided.
There is less practical utility in conducting population-wide individual screening of infants, toddlers, and preschoolers who have acquired less knowledge and skills pertaining to literacy during the preschool years, either through lack of appropriate home literacy experiences or as a result of some inherent cognitive limitations, or of those who lack age-appropriate skill in literacy-related cognitive-linguistic processing, for the purpose of identifying those who are at greatest risk for reading difficulties. Some screening (i.e., language milestones) is already part of regular well-baby visits; in this case the information could help to define risk, especially when aggregated with other risk factors.
Kindergarten screening, in contrast, has become reasonably accurate when a combination of skills is measured (although the optimal combination is not yet identified). Ideally, screening procedures should be quick and inexpensive; they should identify all or most children who have the specific problem; and they should mistakenly detect none or few children who do not have the problem.
To achieve the goal of preventing reading difficulties, it will not be feasible or appropriate to provide the same sort of intervention to all of these groups and individuals, although some kinds of programs may be of benefit to all. In the next chapter, we review and evaluate the possible approaches that can be taken toward addressing the problems of groups and individuals who have been identified as being at risk.