Stanovich, Keith E. (1986). Matthew effects in reading: Some consequences of individual differences in the acquisition of literacy. Reading Research Quarterly, 22, 360-407.
The Matthew Effects are not only about the progressive decline of slow starters, but also about the widening gap between slow starters and fast starters. In reading, the rich get richer and the poor get poorer. This report presents a framework for conceptualizing development of individual differences in reading ability that emphasizes the effects of reading on cognitive development and on "bootstrapping" relationships involving reading. It uses the framework to explain some persisting problems in the literature on reading disability and to conceptualize remediation efforts in reading.
To synthesize the ever-growing body of literature on individual differences in the cognitive skills related to reading is difficult because of the plethora of relationships that have been found. Good and poor readers have been compared on just about every cognitive task that has ever been devised, and group performance differences have been observed on a large number of these tasks (see, for example, Carr, 1981; DeSoto & DeSoto, 1983; Mitchell, 1982; Palmer, MacLeod, Hunt, & Davidson, 1985; Share, Jorm, Maclean, & Matthews, 1984; Singer & Crouse, 1981; Stanovich, 1982a, 1982b, 1986). Mounds of correlations and significant differences have been found. There is, then, at least one sense in which it can be said that we do not lack empirical evidence. The problem is in deciding what it all means.
The aim of this paper is to attempt to clarify the literature by drawing attention to some alternative ways of interpreting relationships between cognitive processes and reading ability. These alternative interpretations have all been discussed before by numerous authors (e.g., Bryant & Bradley, 1985; Byrne, 1986; Chall, 1983; Donaldson, 1978; Ehri, 1979; Morrison & Manis, 1982), but their implications have not been fully explored, nor have they been brought together within a coherent framework. This review presents such a framework and, in addition, a model of the development of individual differences in reading achievement and related cognitive processes that seems to follow logically from it.
Problems with the Existing Evidence
For many years, research on individual differences was plagued by the failure to carry out thorough process analyses on the experimental tasks employed. Thus, it was rarely possible to ascribe any cognitive specificity to an observed group difference. This problem has partially been alleviated due to the general influence of a paradigmatic assumption of cognitive psychology: that performance on any single task is the result of the simultaneous or successive operation of many different information-processing operations. However, it took a long time for reading disability researchers to accept an implication of this assumption: that one could not merely observe a difference on, for example, a perceptual task, and then announce that "visual processing" was the key to reading failure, based on one's introspection about what the task tapped. It was sometimes hard to understand that no matter how large the performance difference observed on a single task, such an outcome represented not the end, but instead the beginning of a careful task analysis that one hoped would reveal the cognitive locus of the difference. The rise and fall of many of the popular hypotheses in the dyslexia literature mirrors this belated realization (see Vellutino, 1979).
Beyond the issue of inferring the appropriate process difference from task performance lies an even more vexing problem: that of inferring causation. After observing a performance difference in a purely correlational study and carrying out the appropriate task analysis, we are still left with the question of whether the processing difference thus isolated causes variation in reading achievement, whether reading achievement itself affects the operation of the cognitive process, or whether the relationship is due to some third variable. Also, there is the possibility of reciprocal causation: that there are causal connections running in both directions.
Complicating the picture even further is the possibility that the causal connections between variation in reading achievement and the efficiency of various cognitive processes may change with development. This possibility has been strongly emphasized by some researchers (e.g., Chall, 1983; Satz, Taylor, Friel, & Fletcher, 1978), but has been inadequately reflected in much research on individual differences in the cognitive skills of reading. For example, it is possible that some relationships are developmentally limited-that individual differences in a particular cognitive process may be a causal determinant of variation in reading achievement early in development, but at some point have no further effects on the level of reading efficiency. In this case, a correlation between reading achievement and the efficiency of a cognitive process may obtain in adults because the efficiency of the cognitive process determined the ease with which the individual traversed earlier stages of the reading process-stages that laid the foundationfor the presentlevel ofreadingability-but further progress is dependent on the development of processes other than the one in question. A residual correlation between the efficiency of the process and reading level remains as a remnant of a causal connection present during an earlier developmental stage.
The vast literature on individual differences in the cognitive processes of reading will only be fully understood when we are able to determine which performance linkages reflect causal relationships, which are developmentally limited, which are the result of third variables, which enter into relationships of reciprocal causation, and which are consequences of the individual's reading level or reading history. Achieving such a classification will be easier if it is recognized that certain relationships may change status at different levels of reading development. In this review some tentative classifications for some of the cognitive processes that have received considerable attention in recent research will be hypothesized. In order to provide a context for these hypotheses, I will first present a brief outline of a preliminary (and incomplete) model of the development of individual differences in reading skill.
A Model of the Development of Individual Differences in Reading
Evidence is mounting that the primary specific mechanism that enables early reading success is phonological awareness: conscious access to the phonemic level of the speech stream and some ability to cognitively manipulate representations at this level. Although general indicators of cognitive functioning such as nonverbal intelligence, vocabulary, and listening comprehension make significant independent contributions to predicting the ease of initial reading acquisition, phonological awareness stands out as the most potent predictor (Share et al., 1984; Stanovich, Cunningham, & Cramer, 1984; Stanovich, Cunningham, & Feeman, 1984a; Tunmer & Nesdale, 1985). Indeed, phonological awareness tasks often correlate more highly with early reading acquisition than do omnibus measures such as general intelligence tests or reading readiness tests (Mann, 1984; Share et al., 1984; Stanovich, Cunningham, & Cramer, 1984; Stanovich, Cunningham, & Feeman, 1984a; Zifcak, 1981).
Of course, although the strength of these correlations serves to draw attention to phonological awareness, it is not proof that variation in awareness is causally connected to differences in the ease of initial reading acquisition. Proving causation requires much stronger evidence, and this evidence is much less plentiful than the purely correlational data. However, a growing body of data does exist indicating that variation in phonological awareness is causally related to the early development of reading skill. This evidence is of several different types. First, there are several studies showing that measures of phonological awareness predict reading ability even when the former are assessed very early in development (Bradley & Bryant, 1983, 1985; Fox & Routh, 1975; Share et al., 1984; Williams, 1984). Secondly, Tunmer and Nesdale (1985) reported a contingency analysis of their first-grade data which indicated that phonemic segmentation skill was a necessary, but not sufficient, condition for reading acquisition (see also Perfetti, Beck, & Hughes, 1981). In addition, the results of some recent longitudinal studies where cross-lagged correlational methods and/or structural equation modeling have been employed have led to the conclusion that early skill at phonological awareness leads to superior reading achievement (Perfettiet al., 1981; Torneus, 1984). Evidence supporting this conclusion also comes from reading-level match designs. When 10-year-old disabled readers perform worse on phonological tasks than nondisabled 6-year-old children reading at the same level (e.g., Bradley & Bryant, 1978), it is somewhat more difficult to argue that the latter are superior because they have had more reading experience. Last, and of course most convincing, are the results of several studies where phonological awareness skills were manipulated via training, and the manipulation resulted in significant experimental group advantages in reading, word recognition, and spelling (Bradley & Bryant, 1983, 1985; Fox & Routh, 1984; Olofsson & Lundberg, 1985; Torneus, 1984; Treiman & Baron, 1983).
It shouldbe noted that several of the studies cited above have also supported Ehri's (1979, 1984, 1985) position that reading acquisition itself facilitates phonological awareness (see also Perfetti, 1985; Perfettiet al., 1981; Wagner & Torgesen, in press), so that the situation appears to be one of reciprocal causation. Such situations of reciprocal causation can have important "bootstrapping" effects, and some of these will be discussed in this review. However, the question in this section is not which direction of causality is dominant. The essential properties of the model being outlined here are dependent only on the fact that a causallink running from phonological awareness to reading acquisition has been established, independent of the status of the opposite causal link.
Many researchers have discussed the reasons phonological awareness is important in early reading acquisition (see Gough & Hillinger, 1980; Liberman, 1982; Perfetti, 1984; Williams, 1984). A beginning reader must at some point discover the alphabetic principle: that units of print map onto units of sound (see Perfetti, 1984). This principle may be induced; it may be acquired through direct instruction; it may be acquired along with or after the build-up of a visually-based sight vocabulary - but it must be acquired if a child is to progress successfully in reading. Children must be able to decode independently the many unknown words that will be encountered in the early stages of reading. By acquiring some knowledge of spelling-to-sound mappings, the child will gain the reading independence that eventually leads to the levels of practice that are prerequisites to fluent reading. The research cited above appears to indicate that some minimal level of explicit phonemic awareness is required for the acquisition of the spelling-to-sound knowledge that supports independent decoding.
It is apparently important that the prerequisite phonological awareness and skill at spelling-to-sound mapping be in place early in the child's development, because their absence can initiate a causal chain of escalating negative side effects. Biemiller (1977-1978; see also Allington, 1980, 1983, 1984) has documented how extremely large differences in reading practice begin to emerge as early as the middle of the first-grade year. In October, the children in the three most able groups in his sample read a mean of 12.2 words per child per reading session, the children in three average ability groups read 11.9 words per child per reading session, and the children in the two least able groups were not reading. By January, the mean for the most able groups was 51.9, for the average ability groups, 25.8, and for the least able groups, 11.5. In April the respective means were 81.4, 72.3, and 31.6. This of course says nothing about differences in home reading, which would probably be at least as large. Thus, soon after experiencing greater difficulty in breaking the spelling-to-sound code, poorer readers begin to be exposed to less text than their peers.
Further exacerbating the situation is the fact that poorer readers often find themselves in materials that are too difficult for them (Allington, 1977, 1983, 1984; Bristow, 1985; Forell, 1985; Gambrell, Wilson, & Gantt, 1981; Jorgenson, 1977). The combination of lack of practice, deficient decoding skills, and difficult materials results in unrewarding early reading experiences that lead to less involvement in reading-related activities. Lack of exposure and practice on the part of the less skilled reader delays the development of automaticity and speed at the word-recognition level. Slow, capacity-draining word-recognition processes require cognitive resources that should be allocated to higher-level processes of text integration and comprehension (LaBerge & Samuels, 1974; Perfetti, 1985; Stanovich, 1980). Thus, reading for meaning is hindered, unrewarding reading experiences multiply, and practice is avoided or merely tolerated without real cognitive involvement. The downward spiral continues- and has further consequences.
The better reader more rapidly attains a stage of proficiency where decoding skill is no longer the primary determinant of reading level. As word recognition becomes less resource-demanding by taking place via relatively automatic processes of visual/orthographic access, more general language skills become the limiting factor on reading ability (Chall, 1983; Sticht, 1979). But the greater reading experience of the better reader has provided an enormous advantage even here. Reading itself is an important contributor to the development of many language/cognitive skills. For example, much vocabulary growth probably takes place through the learning of word meanings from context during reading (Nagy & Anderson, 1984; Nagy, Herman, & Anderson, 1985; Sternberg, 1985). Similarly, much general information and knowledge about more complex syntactic structures probably also takes place through reading itself (Donaldson & Reid, 1982; Mann, 1986; Perfetti, 1985, pp.172-173, 195). In short, many things that facilitate further growth in reading comprehension ability - general knowledge, vocabulary, syntactic knowledge - are developed by reading itself. The increased reading experiences of children who crack the spelling-to-sound code early thus have important positive feedback effects. Such feedback effects appear to be potent sources of individual differences in academic achievement (Walberg, Strykowski, Rovai, & Hung, 1984).
Paring Down the Number of Causal Relationships
It will be argued here that these bootstrapping effects of reading experience and other secondary effects have been inadequately considered in the extensive literature on individual differences in the cognitive processes of reading. Although it might seem that a consideration of the effects of these reciprocal relationships would complicate our models, it actually has great potential to clarify reading theory. If only a few of these reciprocal effects control a large portion of the variance in reading ability, we will be able to exercise parsimony elsewhere. Such a consideration will