<?xml version="1.0" ?> <tei> <teiHeader> <fileDesc xml:id="0"/> </teiHeader> <text xml:lang="en"> <p>Reading researchers generally agree<lb/> that most poor readers experience consid-<lb/>erable difficulty in lower-level word rec-<lb/>ognition processes, and particularly in<lb/> translating letters into sound (phonological<lb/> recoding). However, there is far less agree-<lb/>ment as to the origins of phonological re-<lb/>coding difficulties. The current research<lb/> examines the relative contributions of<lb/> phonological analysis and verbal working<lb/> memory to fourth-grade children's word<lb/> reading difficulties.<lb/></p> <head>Phonological Analysis and Word Reading<lb/></head> <p>The association between reading<lb/> achievement and phonological analysis<lb/> skills is now well established (e.g., <ref type="biblio">Bryant<lb/> & Goswami, 1987</ref>). Although evidence from<lb/> training studies indicates that phonological<lb/> analysis does contribute directly to the read-<lb/>ing process (e.g., <ref type="biblio" >Cunningham, 1990; Lund-<lb/>berg, Frost, & Petersen, 1988</ref>), the precise<lb/> nature of poor readers' phonological analysis<lb/> difficulties and their relation to the reading<lb/> process is not yet clear.<lb/></p> <p>Most researchers view phonological<lb/> analysis as contributing directly to the de-<lb/>velopment of efficient word recognition<lb/> skills. The alphabetic writing system is a<lb/> phonological transcription of spoken lan-<lb/>guage, with individual graphemes or graph-<lb/>eme combinations generally representing<lb/> language at the phonemic level. Unfamiliar<lb/> words are identified through a phonological<lb/> recoding mechanism, in which print is trans-<lb/>lated into a phonological representa-<lb/>tion through the application of implicit<lb/> grapheme-phoneme conversion rules <ref type="biblio">(Colt-<lb/>heart, 1978)</ref> and "large-unit" rules. These<lb/> large-unit rules group letters into units cor-<lb/>responding to subsyllabic onsets and rimes<lb/> (e.g., <ref type="biblio" >Bowey, 1990; Treiman & Chafetz,<lb/> 1987; Treiman, Coswami, & Bruck, 1990</ref>).<lb/> The onset is the optional initial consonant<lb/> or consonant cluster (e.g., /br/ in broom).<lb/> The rime is the vowel nucleus and optional<lb/> final consonant or consonant cluster (e.g., /u/<lb/> in blue and /en/ in rain).<lb/></p> <p>Poor readers' word decoding problems<lb/> are thought to stem from difficulties in ana-<lb/>lyzing the sound structure of spoken lan-<lb/>guage (e.g., <ref type="biblio" >Stanovich, 1986; Vellutino &<lb/> Scanlon, 1989</ref>). This impairs their compre-<lb/>hension of the alphabetic principle underly-<lb/>ing English orthography so that they are less<lb/> likely to induce the print-sound correspon-<lb/>dences required for efficient phonological<lb/> recoding. Poor readers must therefore rely<lb/> more heavily on the acquisition of a sight<lb/> vocabulary. Word recognition through direct<lb/> visual access is slower than efficient phono-<lb/>logical recoding in relation to unfamiliar<lb/> words, as evidenced by the fact that, for rela-<lb/>tively unfamiliar words, irregular words take<lb/> longer to name than regular words for both<lb/> children <ref type="biblio">(Backman, Bruck, Hebert, & Sei-<lb/>denberg, 1984)</ref> and adults <ref type="biblio">(Andrews, 1982;<lb/>Seidenberg, Waters, Barnes, & Tanenhaus,<lb/> 1984)</ref>.<lb/></p> <p>However, there are alternate interpreta-<lb/>tions of the association between phonologi-<lb/>cal analysis and reading achievement. For<lb/> instance, deficits in phonological analysis<lb/> may partly reflect impaired verbal working<lb/> memory, an impairment also believed to<lb/> contribute substantially to difficulties in de-<lb/>coding unfamiliar words <ref type="biblio">(Baddeley, 1986;<lb/> Shankweiler & Crain, 1986)</ref>. Alternately, un-<lb/>derlying phonological coding impairments<lb/> may produce deficits in phonological analy-<lb/>sis, verbal working memory, and decoding<lb/> <ref type="biblio">(Vellutino & Scanlon, 1989)</ref>.<lb/></p> <head>Working Memory and Word Reading<lb/></head> <p>Currently, the most prominent theory of<lb/> working memory is that of <ref type="biblio" >Baddeley (1986)</ref>.<lb/> Baddeley's model consists of a limited-<lb/>capacity attentional control system, the cen-<lb/>tral executive, assisted by two slave systems,<lb/> the visuospatial sketchpad and the articula-<lb/>tory loop. The visuospatial sketchpad con-<lb/>structs and maintains visuospatial images,<lb/> and operates independently of the articula-<lb/>tory loop. This memory subsystem appears<lb/> to be unimpaired in poor readers (see <ref type="biblio">Jorm,<lb/> 1983; Vellutino & Scanlon, 1982</ref>). The artic-<lb/>ulatory loop is used in verbal working mem-<lb/>ory tasks and comprises two components, a<lb/> speech-based phonological input store and a<lb/> rehearsal process. Information in the phono-<lb/>logical store fades rapidly unless it is main-<lb/>tained by rehearsal.<lb/></p> <p>The working memory account of word<lb/> reading difficulties has at different times at-<lb/>tributed poor readers' problems to three dif-<lb/>ferent aspects of working memory function;<lb/> the rehearsal process, the phonological store<lb/> component, and the central executive.<lb/></p> <p><ref type="biblio">Baddeley (1979)</ref> proposed that children<lb/> who were unable to use tiie articulatory loop<lb/> to maintain material in the phonological<lb/> store would find it difficult to recall the<lb/> sounds of letters and word fragments that<lb/> had already been decoded while simulta-<lb/>neously decoding the remaining letters,<lb/> hence overloading the central executive. Ef-<lb/>ficient rehearsal is inferred from the word-<lb/>length effect, whereby word span is greater<lb/> for short words than for long words. How-<lb/>ever, available evidence suggests that good<lb/> and poor readers show equally strong word-<lb/>length effects <ref type="biblio">(Baddeley, 1986)</ref>.<lb/></p> <p>A number of studies carried out by the<lb/> Haskins group reported stronger phonologi-<lb/>cal similarity effects (inferior spans for rhym-<lb/>ing materials than nonrhyming materials) in<lb/> good readers than in poor readers of the<lb/> same age (e.g., <ref type="biblio">Shankweiler & Liberman,<lb/> 1976</ref>). Attenuation of the phonological simi-<lb/>larity effect is interpreted by <ref type="biblio" >Baddeley<lb/> (1986)</ref> to reflect impaired processing in the<lb/> phonological store component of the articu-<lb/>latory loop. <ref type="biblio">Shankweiler and Liberman<lb/> (1976)</ref> interpreted attenuated phonological<lb/> similarity effects as indicating degraded<lb/> phonological representation of information<lb/> within the phonological store. However, it<lb/> can also be argued that degraded phonologi-<lb/>cal representation of information held<lb/> within the phonological store may produce<lb/> an increased phonological similarity effect<lb/> <ref type="biblio">(Hall, Wilson, Humphreys, Tinzmann, &<lb/> Bowyer, 1983)</ref>. Furthermore, the attenuated<lb/> phonological similarity effect is not consis-<lb/>tently observed (e.g., <ref type="biblio">Baddeley, 1986; Hall<lb/> et al., 1983; Johnston, Rugg, & Scott, 1987</ref>),<lb/> and appears to be an artifact of lower verbal<lb/> spans in poorer readers. Doubts regarding<lb/> the association between the phonological<lb/> similarity effect and reading skill led <ref type="biblio" >Badde-<lb/>ley (1986)</ref> to speculate that the lower verbal<lb/> memory spans of poor readers may refiect<lb/> a lesser capacity of the phonological store.<lb/> Phonological store capacity can be assessed<lb/> by verbal span tasks, even though they do<lb/> incorporate a rehearsal component.<lb/></p> <p><ref type="biblio">Shankweiler and Crain (1986)</ref> argued<lb/> that reading skill is most strongly correlated<lb/> with performance on verbal working mem-<lb/>ory tasks that involve additional processing<lb/> by the central executive. Such tasks typically<lb/> require verbal material to be held in the ar-<lb/>ticulatory loop, but reorganized in some way<lb/> for recall. <ref type="biblio">Daneman and Carpenter (1980)<lb/></ref> also argued that working memory capacity is<lb/> best measured by tasks that divide capacity<lb/> between storage and processing. <ref type="biblio">Baddeley<lb/> (1986, p. 245)</ref> argued that measures such as<lb/> these refiect central executive processing.<lb/> Thus, he claimed that selective decreases in<lb/> performance on complex verbal span tasks<lb/> in the elderly (e.g., <ref type="biblio">Talland, 1965</ref>) refiect de-<lb/>clines in central executive processing ca-<lb/>pacity.<lb/></p> <p>To summarize, the literature reviewed<lb/> so far reveals considerable theoretical con-<lb/>troversy concerning the phonological defi-<lb/>cits underlying poor word reading. Some<lb/> view phonological analysis as contributing<lb/> directly to decoding difficulties; others view<lb/> phonological analysis as reflecting verbal<lb/> working memory abilities, at least in part.<lb/> Still others view both phonological analysis<lb/> and verbal working memory as indepen-<lb/>dently contributing to decoding skill.<lb/></p> <head>Pseudoword Reading as a Component of<lb/> Word Recognition<lb/></head> <p>If the slower development of word read-<lb/>ing skills reflects the contribution of phono-<lb/>logical analysis and/or working memory dif-<lb/>ficulties, then parallel deficits should be<lb/> observed in tests examining the phonologi-<lb/>cal recoding path to word recognition. Dif-<lb/>ficulties in the operation of this path are<lb/> most obviously detected by tests of pseu-<lb/>doword reading. Since pseudowords, by<lb/> definition, have not been previously seen,<lb/> they cannot be recognized through direct vi-<lb/>sual access but must be decoded by transla-<lb/>tion of letters and letter groups into sound.<lb/></p> <head>Reading-Level Match Designs<lb/></head> <p>Within the same age group, children dif-<lb/>fering in reading ability will differ in many<lb/> other aspects of performance, which may or<lb/> may not be related to reading achievement.<lb/> Thus, associations between reading and<lb/> other abilities within samples of the same<lb/> chronological age reveal little about the po-<lb/>tential causes of reading difficulty. Research<lb/> has increasingly used reading-level match<lb/> designs, in which poor readers are compared<lb/> to younger children reading at the same<lb/> level. The essential logic of such studies is<lb/> this: If poor readers perform worse than<lb/> younger children of equivalent reading abil-<lb/>ity on any task, then the skills underlying<lb/> that task can be viewed as potential causes<lb/> of reading difficulties. It has been argued<lb/> that such designs enable potential causes to<lb/> be differentiated from consequences of read-<lb/>ing difficulties (e.g., <ref type="biblio">Bryant & Goswami,<lb/> 1986</ref>).<lb/></p> <p>Reading-level designs should help clar-<lb/>ify the association between phonological<lb/> analysis, verbal working memory, and word<lb/> decoding skills. Poor readers should perform<lb/> at lower levels than younger children of<lb/> equivalent word identification skill in abili-<lb/>ties that are directly contributing to their<lb/> slower rate of reading progress. However,<lb/> previous studies using reading-level designs<lb/> have not yielded consistent results in rela-<lb/>tion to any of these component phonological<lb/> abilities.<lb/></p> <p><ref type="biblio">Bradley and Bryant (1978)</ref> reported that<lb/> backward readers made more errors than<lb/> younger reading-level controls in three pho-<lb/>nological oddity tasks. Two of the oddity<lb/> tasks (detecting the odd word out in series<lb/> like nod red fed bed) assessed rime sensitiv-<lb/>ity. The third assessed sensitivity to allitera-<lb/>tion (detecting the odd word out in a series<lb/> like sun see sock rag). <ref type="biblio">Bruck and Treiman<lb/> (1990)</ref> reported that poor readers scored<lb/> lower than reading-level controls on some<lb/> phonological analysis tasks. Their data sug-<lb/>gest that poor readers may experience rela-<lb/>tively greater difficulty with tasks requiring<lb/> phonemic analysis than with tasks requiring<lb/> subsyllabic analysis (i.e., analysis of sylla-<lb/>bles into onsets and rimes).<lb/></p> <p>However, other studies have failed to<lb/> detect any differences between less skilled<lb/> readers and reading-level control groups in<lb/> phonological analysis skill. <ref type="biblio">Stanovich,<lb/> Nathan, and Vala-Rossi (1986)</ref> found no<lb/> differences between groups of third-and<lb/> fifth-grade children matched on a group-<lb/>administered test of reading comprehension<lb/> in speeded rhyme production or speeded<lb/> phonological oddity tasks. Similarly, <ref type="biblio" >Stano-<lb/>vich, Nathan, and Zolman (1988)</ref> found no<lb/> differences on a speeded rhyme production<lb/> task among third-, fifth-, and seventh-grade<lb/> children performing at the same level on a<lb/> group-administered test of reading compre-<lb/>hension. <ref type="biblio">Beech and Harding (1984)</ref> also<lb/> found that poor 9-year-old readers and 7-<lb/>year-olds matched on word identification<lb/> skill performed at the same level on phone-<lb/>mic segmentation, rhyme production, rhyme<lb/> recognition, and phonological oddity tests.<lb/> <ref type="biblio">Vellutino and Scanlon (1989)</ref> found no dif-<lb/>ferences between poor sixth-grade readers<lb/> and second-grade reading-level controls on<lb/> phonemic segmentation performance.<lb/></p> <p>Similarly inconsistent findings have<lb/> been reported from reading-level designs<lb/> examining verbal memory span. Although<lb/> <ref type="biblio">Cohen, Netley, and Clarke (1984)</ref> reported<lb/> lower verbal spans in less skilled readers<lb/> than in younger reading-level controls, most<lb/> studies using reading-level designs have re-<lb/>ported no difference <ref type="biblio" >(Beech & Harding,<lb/> 1984; Johnston, 1982; Stanovich etal., 1988)</ref>.<lb/> <ref type="biblio">Johnston et al. (1987)</ref> observed lower word<lb/> spans in poor readers with lower IQ scores<lb/> than younger controls, but observed no dif-<lb/>ference between average ability poor read-<lb/>ers and reading age controls.<lb/></p> <p>Given the inconsistent results from<lb/> reading-level designs investigating phono-<lb/>logical analysis and verbal working memory,<lb/> it is hardly surprising that inconsistent<lb/> findings have emerged from reading-level<lb/> designs investigating pseudoword reading<lb/> skills. The most common finding is that the<lb/> relative difficulty of pseudowords and real<lb/> words is the same for less skilled readers<lb/> and reading-level controls <ref type="biblio">(Beech & Har-<lb/>ding, 1984; Bruck, 1988; Kochnower, Rich-<lb/>ardson, & DiBenedetto, 1983; Stanovich et<lb/> al., 1986, 1988; Treiman & Hirsh-Pasek,<lb/> 1985; Vellutino & Scanlon, 1989)</ref>. However,<lb/> several studies have reported more marked<lb/> difficulties with pseudowords in less skilled<lb/> readers relative to younger reading-level<lb/> controls <ref type="biblio">(Baddeley, Ellis, Miles, & Lewis,<lb/> 1982; Frith & Snowling, 1983; Olson, Wise,<lb/> Conners, Rack, & Fulker, 1989)</ref>. Mixed re-<lb/>sults were observed within the study by<lb/> <ref type="biblio">Szeszulski and Manis (1987)</ref>.<lb/></p> <p><ref type="biblio">Stanovich et al. (1986)</ref> interpreted the<lb/> different outcomes of studies using reading-<lb/>level designs in terms of the type of poor<lb/> reader studied. They argued that poor read-<lb/>ers of average IQ have specific phonologi-<lb/>cal deficits, whereas "garden-variety" poor<lb/> readers, whose reading problems are de-<lb/>fined only in relation to age, perform like<lb/> younger children. They claimed that most<lb/> studies finding differences favoring younger<lb/> reading-level controls have investigated<lb/> samples matched on IQ (e.g., <ref type="biblio">Baddeley et<lb/> al., 1982; Bradley & Bryant, 1978</ref>). In con-<lb/>trast, most studies reporting no differences<lb/> between groups matched for reading level<lb/> have not equated groups for IQ (e.g., <ref type="biblio" >John-<lb/>ston, 1982; Stanovich etal., 1986,1988; Trei-<lb/>man & Hirsh-Pasek, 1985; Vellutino & Scan-<lb/>lon, 1989</ref>). This explanation also accounts for<lb/> the <ref type="biblio">Beech and Harding (1984)</ref> results, if<lb/> their choice of Raven's Progressive Matrices<lb/> as an IQ-matching measure is criticized (see<lb/> <ref type="biblio">Stanovich et ai., 1986</ref>). However, not all<lb/> studies reporting differences matched sam-<lb/>ples for IQ (e.g.. <ref type="biblio">Frith & Snowhng, 1983;<lb/> Johnston et al., 1987; Olson et al., 1989</ref>).<lb/> Furthermore, Johnston et al. found inferior<lb/> letter span performance only in the group of<lb/> poor readers not matched for IQ, and Olson<lb/> et al. reported that IQ-reading discrepancies<lb/> were not associated with deficits in pseu-<lb/>doword reading.<lb/></p> <p>An alternate explanation for the discrep-<lb/>ancies in outcomes of these studies lies in<lb/> the inherent conservatism of reading-level<lb/> designs that do not attempt to control for the<lb/> effects of regression to the mean. In such<lb/> studies, the direction of regression effects is<lb/> reversed for less skilled and skilled groups<lb/> <ref type="biblio">(Campbell & Stanley, 1963)</ref>. To obtain suf-<lb/>ficient young skilled readers, it is necessary<lb/> to sample from the upper end of the reading<lb/> distribution of younger children. Thus, the<lb/> skilled younger group is likely to regress to-<lb/>ward the mean of their age-mates and score<lb/> lower on a second reading test than an older,<lb/> less skilled group, which is likely to have<lb/> regressed toward the mean of older readers.<lb/> The conservatism of reading-level designs<lb/> in which no attempt has been made to con-<lb/>trol for regression effects depends on the<lb/> level of reading impairment studied and the<lb/> reliability of the reading measure used for<lb/> sample selection.<lb/></p> <head>The Present Study<lb/></head> <p>The present study was designed to elu-<lb/>cidate the associations among word read-<lb/>ing ability, phonological analysis, verbal<lb/> working memory, and phonological recod-<lb/>ing (pseudoword reading). A three-group<lb/> reading-level design was used, with less<lb/> skilled fourth-grade readers compared to<lb/> both age-matched peers of superior reading<lb/> skill and younger children matched for read-<lb/>ing achievement in phonological analysis,<lb/> verbal working memory, and pseudoword<lb/> reading skills.<lb/></p> <p>Children were allocated to reading<lb/> groups on the basis of their performance on<lb/> two parallel forms of the Word Identification<lb/> subtest of the Woodcock Reading Mastery<lb/> Tests <ref type="biblio">(Woodcock, 1987)</ref>. The use of two<lb/> forms of screening test minimized the possi-<lb/>bility of regression effects within this study<lb/> <ref type="biblio">(Hall & Humphreys, 1982)</ref>. No attempt was<lb/> made to match groups on intelligence. Not<lb/> only does such matching reduce the general-<lb/>izability of results <ref type="biblio">(Hall & Humphreys,<lb/> 1982)</ref>, it fails to match for either component<lb/> cognitive abilities <ref type="biblio">(Hall & Humphreys,<lb/> 1982)</ref> or level of cognitive development<lb/> <ref type="biblio">(Backman, Mamen, & Ferguson, 1984; Stan-<lb/>ovich et al., 1986)</ref> and is susceptible to re-<lb/>gression effects <ref type="biblio">(Hall & Humphreys, 1982)</ref>.<lb/></p> <p>In relation to phonological analysis<lb/> skills, this research also sought to determine<lb/> whether less skilled readers experience par-<lb/>ticular difficulty in tasks requiring sensitiv-<lb/>ity to phonemes relative to tasks refiecting<lb/> sensitivity to onsets and rimes. The use of<lb/> phonological oddity tasks that were identical<lb/> except in the level of phonological analysis<lb/> required enabled potential differences in<lb/> phonological analysis performance to be un-<lb/>ambiguously interpreted.<lb/></p> <p>Verbal working memory was assessed<lb/> by the Digit Span subtest of the WISC-R<lb/> <ref type="biblio">(Wechsler, 1974)</ref>. The choice of this test per-<lb/>mitted some exploration of the locus of poor<lb/> readers' verbal span problems. The Digits<lb/> Forward subtest can be viewed as a rela-<lb/>tively pure measure of articulatory loop ca-<lb/>pacity. In the Digits Backward subtest mate-<lb/>rial must be maintained in the articulatory<lb/> loop but scanned constantly so that items<lb/> can be repeated backward. This task may<lb/> thus be viewed as a measure of overall ver-<lb/>bal working memory capacity, refiecting<lb/> central executive capacity in addition to ar-<lb/>ticulatory loop capacity. If deficits in phono-<lb/>logical analysis refiect underlying deficits in<lb/> verbal working memory, then group differ-<lb/>ences in phonological analysis skill should<lb/> be paralleled by group differences in Digit<lb/> Span performance. If working memory dif-<lb/>ferences primarily reflect central executive<lb/> processing capacity, then stronger group dif-<lb/>ferences would be expected on the Digits<lb/> Backward subtest than the Digits Forward<lb/> subtest. If group differences primarily re-<lb/>flect articulatory loop capacity, then similar<lb/> group differences would be observed in<lb/> both Digit Span subtests.<lb/></p> <p>Pseudoword reading (assessing phono-<lb/>logical recoding skill) was measured by the<lb/> Word Attack subtest of the Woodcock Read-<lb/>ing Mastery Test <ref type="biblio" >(Woodcock, 1987)</ref>. The in-<lb/>clusion of the pseudoword reading measure<lb/> enabled a preliminary test of causal hypoth-<lb/>eses regarding the contribution of phonolog-<lb/>ical analysis and verbal working memory<lb/> to phonological recoding skills. Observed<lb/> group differences in phonological analysis<lb/> and/or verbal working memory should be<lb/> paralleled by group differences in pseu-<lb/>doword reading accuracy. Additional corre-<lb/>lational analyses provided an assessment of<lb/> the strength of the associations among<lb/> phonological analysis, verbal working mem-<lb/>ory, and phonological recoding performance<lb/> measures.<lb/></p> <head>Method<lb/></head> <head>Subjects<lb/></head> <p>Children from second-and fourth-grade<lb/> classes of a large primary school served as<lb/> subjects in this experiment. This school was<lb/> part of a Catholic school system, and was<lb/> located in a predominantly middle-class<lb/> suburb of Brisbane, Australia. All children<lb/> tested were native English speakers; none<lb/> was bilingual. Testing took place midway<lb/> through the school year.<lb/></p> <p>The experimental design required 48<lb/> children: 16 skilled and 16 less skilled<lb/> fourth-grade children, and 16 second-grade<lb/> children matched to the less skilled fourth-<lb/>grade group on Word Identification perfor-<lb/>mance. To minimize regression effects.<lb/> Word Identification performance was as-<lb/>sessed twice, using Forms G and H of the<lb/> Word Identification subtest of the Revised<lb/> Woodcock Reading Mastery Tests <ref type="biblio">(Wood-<lb/>cock, 1987)</ref>. Children had to be assigned to<lb/> the same group on both testing occasions to<lb/> be included in the experiment. Characteris-<lb/>tics of the three experimental groups are pro-<lb/>vided in <ref type="table">Table 1.</ref><lb/></p> <figure type="table"> TABLE 1<lb/> CHRONOLOGICAL AGE, AND PPVT-R VOCABULARY AGE AND STANDARD SCORES IN THE<lb/> THREE EXPERIMENTAL GROUPS<lb/> SKILLED<lb/> SKILLED<lb/> LESS SKILLED<lb/> FOURTH GRADE<lb/> SECOND GRADE<lb/> FOURTH GRADE<lb/> GROUP<lb/> Mean<lb/> SD<lb/> Mean<lb/> SD<lb/> Mean<lb/> SD<lb/> Chronological age (months)<lb/> 111.69<lb/> 5.86<lb/> 88.31<lb/> 2.68<lb/> 109.37<lb/> 4.81<lb/> Vocabulary age (months)<lb/> 133.63<lb/> 23.20<lb/> 100.63<lb/> 11.15<lb/> 104.12<lb/> 12.39<lb/> PPVT-R standard score<lb/> 114.19<lb/> 12.89<lb/> 110.50<lb/> 9.72<lb/> 95.63<lb/> 7.95<lb/> Word identiflcation (G)"<lb/> 514.81<lb/> 8.42<lb/> 467.94<lb/> 12.48<lb/> 467.63<lb/> 11.63<lb/> Word identiflcation (H)='<lb/> 516.00<lb/> 6.61<lb/> 470.31<lb/> 14.11<lb/> 470.94<lb/> 13.52<lb/> Estimated reading age"<lb/> 140.69<lb/> 13.54<lb/> 94.19<lb/> 6.25<lb/> 94.25<lb/> 6.07<lb/> " Woodcock Word Identification scores are the W-scores (Woodcock, 1987). The U.S. norms for the Woodcock test<lb/> consistently overestimate the Queensland grade level and reading age scores. For example, halfway through fourth<lb/> grade, where, by definition, a reading grade level of 4.5 would be expected, the mean cbronological age of Queensland<lb/> children is approximately 110 months. According to the Woodcock norms, at the age of 110 months a grade level of<lb/> 4.0 would be expected. Thus, at this chronological age level, the Woodcock norms overestimate Australian reading<lb/> age by approximately 6 months. Similar comments have been made in relation to the appropriateness of U.S. norms<lb/> for Canadian children (e.g., Backman et al., 1984; Snart, Dennis, & Brailsford, 1983). Australian reading age scores<lb/> have been estimated by subtracting 6 months from the Woodcock norms.<lb/></figure> <head>Materials<lb/></head> <p>Five published tests were adminis-<lb/>tered: the Revised Peabody Picture Vocabu-<lb/>lary Test <ref type="biblio">(Form M; Dunn & Dunn, 1981)</ref>,<lb/> the Word Identification (Forms G and H)<lb/> and Word Attack (Form H) subtests of the<lb/> Revised Woodcock Reading Mastery Tests<lb/> <ref type="biblio">(Woodcock, 1987)</ref>, and the Digit Span sub-<lb/>test of the WISC-R <ref type="biblio">(Wechsler, 1974)</ref>.<lb/></p> <p>In addition to these published tests, two<lb/> pairs of phonological oddity tasks were de-<lb/>vised. Each item in these tasks consisted of<lb/> a triplet of three-phoneme words, one of<lb/> which was phonologically "odd." For in-<lb/>stance, given the triplet/is/i man dish, man<lb/> is the odd word out, phonologically speak-<lb/>ing. Each task consisted of three practice<lb/> trials and 12 test trials. Within each task, the<lb/> order of the odd word was systematically<lb/> varied so that it occurred equally often in<lb/> each position. Furthermore, within each<lb/> block of three trials, each item position (first<lb/> word, second word, third word) was correct<lb/> once.<lb/></p> <p>Onset and rime oddity tasks. —The first<lb/> pair of tasks required children to classify<lb/> words on the basis of rime or onset units (see<lb/> Appendix A). The first task tested children's<lb/> ability to classify words on the basis of the<lb/> rime unit (e.g., in the sequence deck neck<lb/> fit, fit is the odd word out). The second task<lb/> tested their ability to classify items on the<lb/> basis of the onset unit (e.g., in the sequence<lb/> flow sky flea, sky is the odd word out).<lb/></p> <p>Phoneme oddity. —This pair of tasks<lb/> tested whether or not children were able to<lb/> classify words on the basis of purely phone-<lb/>mic differences (see Appendix B). In both<lb/> tasks, all three words shared the singleton<lb/> onset unit, but the odd word differed in<lb/> terms of one of the two phonemes in the<lb/> rime unit. In these tasks, oddity could be<lb/> detected only through phonemic analysis. In<lb/> the first task, the odd word differed from the<lb/> remaining two in terms of the medial pho-<lb/>neme, the vowel of the rime unit (e.g., bed<lb/> bit bin). In the second task, the odd word<lb/> differed in terms of the word-final phoneme<lb/> (e.g., big bag bet).<lb/></p> <head>Procedure<lb/></head> <p> Children were tested individually by<lb/> two experimenters (MC and SR) in five ses-<lb/>sions spread over 10 weeks. Because the first<lb/> two sessions involved 101 children, they<lb/> were spread over a period of 6 weeks. In the<lb/> first session, MC administered the Form C<lb/> Word Identification subtest, and in the sec-<lb/>ond session SR administered the Form H<lb/> Word Identification subtest. In the third ses-<lb/>sion, MC administered Form M of the Re-<lb/>vised Peabody Picture Vocabulary Test and<lb/> the phonological oddity tasks. The fourth<lb/> session, given by SR, consisted of the Digit<lb/> Span test and a further task unrelated to the<lb/> present research. In a final session, SR ad-<lb/>ministered the Word Attack subtest.<lb/></p> <p>Within each reading group, half of the<lb/> children received the onset/rime oddity<lb/> tasks before the two phoneme oddity tasks.<lb/> Within each of these subgroups, the order in<lb/> which the two component tasks was given<lb/> was counterbalanced.<lb/></p> <p>The oddity tasks were preceded by an<lb/> introduction that focused on singleton on-<lb/>sets (alliteration). The focus on alliteration<lb/> was judged to be suitable for children of this<lb/> age, and task-neutral, given that some chil-<lb/>dren began with onsets, others with rimes,<lb/> and others with phoneme units. After a<lb/> preliminary discussion of alliteration, based<lb/> on familiar tongue-twisters, the concepts of<lb/> similar-and different-sounding words were<lb/> introduced and checked. When it was estab-<lb/>lished that children understood that alliter-<lb/>ated words sounded similar and that nonal-<lb/>Iiterated words sounded different, the first<lb/> experimental oddity task was introduced.<lb/> The child's task was to select the odd word<lb/> out from each set of three words. Each of<lb/> the four oddity tasks was preceded by three<lb/> practice trials, in which the child received<lb/> corrective feedback and encouragement<lb/> wherever necessary. All practice and test<lb/> oddity items were presented orally. Because<lb/> of the occasional need to repeat the items,<lb/> tape-recorded presentation was not used.<lb/></p> <head>Results<lb/></head> <head>PPVT-R Performance<lb/></head> <p>Preliminary analyses of PPVT-R perfor-<lb/>mance were conducted to investigate under-<lb/>lying differences in general verbal ability<lb/> that could have influenced subsequent re-<lb/>sults. Analyses of PPVT-R standard score<lb/> and vocabulary age measures yielded sig-<lb/>nificant group effects, F(2,45) = 14.32, p <<lb/> .001, and F(2,45) = 19.33, p < .001, respec-<lb/>tively (see <ref type="table">Table 1</ref>). For this and all subse-<lb/>quent analyses, post hoc tests of group dif-<lb/>ferences used Newman-Keuls post hoc tests<lb/> (a = .05). The less skilled fourth-grade<lb/> group obtained lower standard scores than<lb/> both the skilled fourth-grade and the skilled<lb/> second-grade groups. The two skilled reader<lb/> groups did not differ in vocabulary standard<lb/> scores. Importantly, the less skilled fourth-<lb/>grade group was equivalent to the skilled<lb/> second-grade group on vocabulary age. Both<lb/> groups scored lower than the skilled fourth-<lb/>grade group. Thus, subsequent differences<lb/> between groups equivalent in reading level<lb/> cannot be attributed to attained levels of re-<lb/>ceptive vocabulary development.<lb/></p> <head>Phonological Oddity Performance<lb/></head> <p>Phonological oddity scores were sub-<lb/>jected to both quantitative and qualitative<lb/> analyses. Following traditional analyses of<lb/> variance on (transformed) scores, individu-<lb/>als' performance on these tasks was assessed<lb/> in relation to two criteria, "above chance"<lb/> and 90% correct performance (see below).<lb/> The latter analyses provide an assessment of<lb/> children's mastery of particular phonological<lb/> analysis skills.<lb/></p> <p>Because the phonological oddity data<lb/> were strongly skewed, raw scores were<lb/> transformed, using a "refiect and square<lb/> root" transformation <ref type="biblio">(Tabachnick & Fidell,<lb/> 1989)</ref> to normalize the distributions of both<lb/> subsyllabic and phonemic oddity scores. A<lb/> 3 (group) X 2 (level) analysis of variance<lb/> showed significant main effects of group,<lb/> F(2,45) = 17.39, p < .001, and level, F(l,45)<lb/> = 89.04, p < .001. All between-group<lb/> comparisons were significant. The skilled<lb/> fourth-grade readers performed better than<lb/> the skilled second-grade readers, who in<lb/> turn performed better than the less skilled<lb/> fourth-grade readers. As expected, children<lb/> scored higher on onset and rime oddity tasks<lb/> than on the two phonemic oddity tasks. The<lb/> group X level interaction was not signifi-<lb/>cant, F < 1. The same pattem of results was<lb/> observed in an analysis of untransformed<lb/> scores. <ref type="table">Table 2</ref> provides descriptive statis-<lb/>tics pertaining to raw phonological oddity<lb/> scores.<lb/></p> <p>A one-way analysis of group effects in<lb/> transformed phonological oddity scores<lb/> (combining snbsyllabic and phonemic odd-<lb/>ity tasks), with PPVT-R standard scores en-<lb/>tered as a covariate, showed a significant co-<lb/>variate effect, F(l,44) = 16,01, p < .001, The<lb/> group effect remained significant, F(2,44) =<lb/> 9.47, p < .001.<lb/></p> <p>Although measures of overall perfor-<lb/>mance allow sensitive tests of group differ-<lb/>ences and comparisons of the relative diffi-<lb/>culty of particular tasks, they do not permit<lb/> an assessment of the mastery of particular<lb/> phonological analysis skills. Further analy-<lb/>ses of children's performance investigated<lb/> the frequencies with which children at-<lb/>tained two levels of mastery on the four pho-<lb/>nological oddity tasks. Data from each sub-<lb/>task have been presented separately to<lb/> demonstrate the consistency of the results<lb/> obtained.<lb/></p> <p>Children's performance was first classi-<lb/>fied as pass or fail, based on whether they<lb/> achieved scores significantly above those ex-<lb/>pected on the basis of random guessing.<lb/> Civen a .33 probability of correct perfor-<lb/>mance by chance, over 12 test items a child<lb/> must obtain at least eight correct answers to<lb/> be credited with having scored significantly<lb/> above chance (p < .05), The frequencies of<lb/> children passing each of the four experimen-<lb/>tal oddity tasks are shown in the upper half<lb/> of <ref type="table">Table 3</ref>. In each group, more children<lb/> scored at this pass level than expected by<lb/> chance, on all tasks (p < .05, by second-<lb/>order applications of the binomial theorem).<lb/></p> <p>These "pass" scores indicated that all<lb/> groups did well on the onset and rime oddity<lb/> tasks. However, the phoneme oddity tasks<lb/> showed marked differences in attainment<lb/> between the groups. Most of the skilled<lb/> fourth-grade readers passed both phoneme<lb/> oddity tasks. Less than half of the less skilled<lb/> fourth-grade readers passed either phoneme<lb/> oddity task. The skilled second graders per-<lb/>formed at an intermediate level on the pho-<lb/>neme oddity tasks.<lb/></p> <p> Thus, a further level of mastery was also<lb/> examined, that of 90% correct performance.<lb/></p> <figure type="table">TABLE 2<lb/> PERFORMANCE ON THE PHONOLOGICAL ODDITY TASKS IN THE THREE READING GROUPS<lb/> SURSYLLABIC<lb/> M<lb/> SD<lb/> PHONEMIC<lb/> M<lb/> SD<lb/> Skilled fourth grade<lb/> 23,31<lb/> 1.25<lb/> Skilled second grade<lb/> 21.63<lb/> 1.86<lb/> Less skilled fourth grade<lb/> 19.75<lb/> 3.26<lb/> NOTE.—The maximum score for each set of tasks was 24.<lb/> 20.81<lb/> 2.29<lb/> 17.50<lb/> 4.15<lb/> 14.63<lb/> 3.42<lb/></figure> <figure type="table">TABLE 3<lb/> FREQUENCIES OF CHILDREN SCORING AT TWO LEVELS ON THE PHONOLOGICAL ODDITY TASKS<lb/> Group<lb/> Onset<lb/> Scoring above chance;<lb/> Skilled fourth grade<lb/> 16<lb/> Skilled second grade<lb/> 16<lb/> Less skilled fourth grade<lb/> 13<lb/> Scoring above 90%:<lb/> Skilled fourth grade<lb/> 13<lb/> Skilled second grade<lb/> 9<lb/> .Less skilled fourth grade<lb/> 5<lb/> NOTE.—Maximum of 16 per group.<lb/> Rime<lb/> Medial<lb/> Phoneme<lb/> Final<lb/> Phoneme<lb/> 16<lb/> 16<lb/> 14<lb/> 15<lb/> 11<lb/> 8<lb/> 16<lb/> 12<lb/> 6<lb/> 11<lb/> 5<lb/> 1<lb/> 14<lb/> 11<lb/> 7<lb/> 8<lb/> 6<lb/> 0<lb/></figure> <p>The frequencies of children reaching this<lb/> level of mastery, shown in the lower half of<lb/> <ref type="table">Table 3</ref>, reveal clear task and group differ-<lb/>ences. These data reveal extremely high lev-<lb/>els of performance by the skilled fourth<lb/> graders on the onset and rime oddity tasks,<lb/> and reasonable performance by the reading-<lb/>level matched groups. The greater difficulty<lb/> of the phoneme oddity tasks is apparent for<lb/> all groups. None of the less skilled fourth-<lb/>grade readers had yet reached this level of<lb/> mastery on both phoneme oddity tasks.<lb/></p> <head>Verbal Working Memory<lb/></head> <p>Digit Span performance was analyzed<lb/> through a 3 (group) x 2 (task) analysis of<lb/> variance, revealing significant main effects<lb/> of group, F(2,45) = 11.10, p < .001, and task,<lb/> F(l,45) = 10,08, p < .01 (see <ref type="table">Table 4</ref>).<lb/> Newman-Keuls tests showed that the skilled<lb/> fourth-grade group performed better than<lb/> both other groups. The less skilled fourth-<lb/>grade readers performed at the same level as<lb/> the skilled second-grade readers. Children<lb/> performed better on the Digits Forward than<lb/> on the Digits Backward test. The group x<lb/> task interaction was not significant, F(2,45)<lb/> = 1.03, p > .05. A one-way analysis of co-<lb/>variance showed that PPVT-R standard<lb/> scores did not significantly covary with Digit<lb/> Span scores, F < 1, w^hile the group effect<lb/> remained significant, F(2,44) = 10.51, p <<lb/> .001.<lb/></p> <head>Pseudoword Reading<lb/></head> <p>Finally, an analysis of Word Attack<lb/> W-scores revealed a significant group effect,<lb/> F(2,45) = 47.53, p < .001. All between-<lb/>group contrasts were significant. The skilled<lb/> fourth-grade readers performed best (mean<lb/> W-score of 516.56, SD = 8.91), followed by<lb/> the skilled second-grade readers (mean —<lb/> 498.25, SD = 7.63) and the less skilled<lb/> fourth-grade readers (mean = 486,50, SD =<lb/> 9.70). Analysis of covariance performed on<lb/> Word Attack W-scores revealed a significant<lb/> effect of the covariate, PPVT-R standard<lb/> scores, F(l,44) = 22.85, p < ,001, with the<lb/> group effect remaining significant, F(2,44) =<lb/> 36.12, p < .001.<lb/></p> <head>Correlational Analyses<lb/></head> <p>Two sets of correlations were computed<lb/> among the following measures: PPVT-R<lb/> Vocabulary Age (in months), PPVT-R Stan-<lb/>dard Scores, Digits Forward (raw). Digits<lb/> Backward (raw). Digit Span (raw), combined<lb/> onset and rime oddity (transformed scores),<lb/> phoneme oddity (transformed scores), and<lb/> Word Attack (W-scores). These correlations<lb/> were computed for two overlapping sam-<lb/>ples, the 36 fourth-grade children (including<lb/></p> <figure type="table">TABLE 4<lb/> PERFORMANCE ON THE DIGIT SPAN TEST IN THE THREE READING GROUPS<lb/> Digits forward<lb/> Digits backward<lb/> SKILLED<lb/> FOURTH GRADE<lb/> M<lb/> 6.69<lb/> 5.75<lb/> SD<lb/> 2.02<lb/> 1.88<lb/> SKILLED<lb/> SECOND GRADE<lb/> M<lb/> 5.13<lb/> 3.63<lb/> SD<lb/> 1.41<lb/> 1.26<lb/> LESS SKILLED<lb/> FOURTH GRADE<lb/> M<lb/> 5.19<lb/> 4.75<lb/> SD<lb/> 1.33<lb/> 1.13<lb/></figure> <figure type="table">TABLE 5<lb/> CORRELATIONS OBTAINED AMONG EXPERIMENTAL MEASURES IN 36 FOURTH-GRADE CHILDREN (Above the<lb/> Diagonal) AND IN THE 32 CHILDREN READING AT A LOWER LEVEL (Underlined, Below the Diagonal)<lb/> Measure<lb/> D<lb/> H<lb/> A. PPVT-R Vocabulary<lb/> age<lb/> B. PPVT-R Standard<lb/> score<lb/> 60*<lb/> C. Digits forward<lb/> -.jj^<lb/> D. Digits backward<lb/> 03<lb/> E. Digit span<lb/> -.05<lb/> F. Onset + rime<lb/> oddity<lb/> -.05<lb/> G. Phonemic oddity ... —.10<lb/> H. Word Attack<lb/> -.33<lb/> 97*<lb/> 14<lb/> 28<lb/> 28<lb/> 29<lb/> 28<lb/> 18<lb/> .28<lb/> .31<lb/> .09<lb/> .75*<lb/> .02<lb/> .06<lb/> -.05<lb/> .14<lb/> .13<lb/> .21<lb/> .73*<lb/> .19<lb/> .10<lb/> -.05<lb/> .28<lb/> .29<lb/> .80*<lb/> .76*<lb/> .14<lb/> .11<lb/> -.06<lb/> .39 +<lb/> .43*<lb/> .33<lb/> .39*<lb/> .46*<lb/> .54*<lb/> .51*<lb/> .37+<lb/> .39+<lb/> .30<lb/> .30<lb/> .38+<lb/> .66*<lb/> .46*<lb/> .52*<lb/> .52*<lb/> .37+<lb/> .32<lb/> .44*<lb/> .69*<lb/> .77*<lb/> NOTE.—The data transformation used for the phonological oddity scores produces scores that are themselves<lb/> negatively correlated with the raw scores. Transformed oddity scores have thus been multiplied by — 1 to preserve<lb/> the sense of this analysis. High scores on all measures indicate superior performance.<lb/> + p < .05.<lb/> *p< .01.<lb/></figure> <p>an additional four children with intermedi-<lb/>ate levels of Word Identification scores) and<lb/> the 32 children from the reading-level<lb/> matched groups. The correlation matrices<lb/> are presented in <ref type="table">Table 5</ref>.<lb/></p> <p>Both subsyllabic and phonemic oddity<lb/> performance were consistently associated<lb/> with superior pseudoword reading perfor-<lb/>mance in both samples, whereas verbal span<lb/> measures were associated with pseudoword<lb/> reading only in the fourth-grade sample (see<lb/> <ref type="table">Table 5</ref>). Inconsistent associations between<lb/> verbal span and phonological oddity mea-<lb/>sures were observed in the fourth-grade<lb/> sample; in the children reading at upper<lb/> second-grade level, the correlations be-<lb/>tween these two aspects of performance<lb/> were uniformly nonsignificant. Because of<lb/> restrictions in the range of reading level<lb/> within the latter sample, estimates of the<lb/> strength of association among experimental<lb/> measures derived from that sample are con-<lb/>servative. Nevertheless, differences in pat-<lb/>terns of significant correlations among<lb/> verbal span, phonological oddity, and pseu-<lb/>doword reading measures within that sam-<lb/>ple confirm the suggestions reached through<lb/> the analyses of variance, namely, that pseu-<lb/>doword reading performance is more<lb/> strongly related to phonological analysis<lb/> than to verbal span performance.<lb/></p> <head>General Discussion<lb/></head> <p>The current study compared the perfor-<lb/>mance of less skilled fourth-grade readers<lb/> with that of two other groups, an age-<lb/>matched group and a reading-level control<lb/> group, on a range of phonological skills. Our<lb/> sample of less skilled readers was delayed<lb/> in the acquisition of phonological analysis<lb/> and pseudoword reading skills, but not in<lb/> verbal working memory.<lb/></p> <p>The use of two initial tests as the basis<lb/> for subject-group assignment minimized the<lb/> possibility of regression effects. No attempt<lb/> was made to match groups for general verbal<lb/> intelligence (see introduction). As estimated<lb/> from PPVT-R standard scores, the two<lb/> groups of skilled readers were equivalent in<lb/> age-corrected vocabulary development, but<lb/> the attained level of receptive vocabulary<lb/> was equivalent for the less skilled fourth-<lb/>grade readers and the skilled second-grade<lb/> readers. The fact that the samples were not<lb/> matched for intelligence permits our results<lb/> to be generalized to unselected samples of<lb/> skilled and less skilled readers, while the<lb/> equivalent level of receptive vocabulary in<lb/> the two groups of central interest means that<lb/> observed differences in other tasks cannot<lb/> be attributed to differences in familiarity<lb/> with individual words used within those<lb/> tasks.<lb/></p> <p>The less skilled fourth-grade readers<lb/> scored lower than both the chronological age<lb/> and reading-level control groups on both<lb/> subsyllabic and phonemic oddity tasks. This<lb/> finding extends those of <ref type="biblio">Bradley and Bryant<lb/> (1978)</ref>. As predicted, onset and rime oddity<lb/> tasks were easier than phonemic oddity<lb/> tasks for all three groups, as in previous work<lb/> with beginning readers <ref type="biblio">(Bowey & Francis,<lb/> 1991; Bruck & Treiman, 1990; Helfgott,<lb/> 1976; Kirtley et al., 1989)</ref>. These results<lb/> were supported by qualitative analyses.<lb/> Most children performed above chance in<lb/> both subsyllabic oddity tasks, but perfor-<lb/>mance was consistently lower in the phone-<lb/>mic oddity tasks. Less than half of the less<lb/> skilled fourth-grade children scored above<lb/> chance in the phonemic oddity tasks. The<lb/> degree of mastery of the oddity tasks dif-<lb/>fered across the groups. Under half of the<lb/> less skilled fourth-grade readers reached the<lb/> 90% correct level even in the onset oddity<lb/> task, and none scored consistently at this<lb/> level in both phonemic oddity tasks. Al-<lb/>though none of the less skilled readers had<lb/> yet truly mastered the phonemic oddity<lb/> tasks, there was no evidence that the relative<lb/> difficulty of phonemic and subsyllabic odd-<lb/>ity tasks differed between this group and the<lb/> reading-level controls.<lb/></p> <p>Although the between-group differ-<lb/>ences in phonological analysis skill were<lb/> small in magnitude (see <ref type="table">Table 2</ref>), it appears<lb/> that small deficits in phonological analysis<lb/> translate into noticeable differences in de-<lb/>coding skill. Our group of less skilled read-<lb/>ers could not be considered severely dis-<lb/>abled. Thus, the current study represents a<lb/> conservative test of between-group differ-<lb/>ences in phonological analysis skill. Never-<lb/>theless, the group effect explained 36% of<lb/> the total variance in phonological oddity<lb/> performance (as determined by a one-way<lb/> analysis of variance on transformed overall<lb/> phonological oddity scores; <ref type="biblio">Keppel, 1973</ref>).<lb/></p> <p>Both the current study and that of <ref type="biblio">Brad-<lb/>ley and Bryant (1978)</ref> observed lower levels<lb/> of phonological analysis in less skilled read-<lb/>ers than in younger children matched for<lb/> reading achievement. We observed similar<lb/> results for pseudoword reading, replicating<lb/> several previous studies <ref type="biblio">(Baddeley et al.,<lb/> 1982; Frith & Snowhng, 1983; Olson et al.,<lb/> 1989)</ref>, but not others (e.g.. <ref type="biblio">Beech & Harding,<lb/> 1984; Stanovich et al., 1986, 1988; Treiman<lb/> & Hirsh-Pasek, 1985; Vellutino & Scanlon,<lb/> 1989</ref>).<lb/></p> <p>The differences between the two<lb/> reading-level matched groups in perfor-<lb/>mance on the phonological oddity and pseu-<lb/>doword reading tasks were not paralleled by<lb/> verbal working memory performance. The<lb/> skilled second-grade and the less skilled<lb/> fourth-grade groups were equivalent in<lb/> Digit Span performance. The latter finding<lb/> replicates the bulk of past work <ref type="biblio">(Beech &<lb/> Harding, 1984; Johnston, 1982; Stanovich et<lb/> al 1988; and cf. Cohen et al., 1984)</ref>. The<lb/> current finding in relation to Digit Span per-<lb/>formance is notable in that it was achieved<lb/> alongside significant differences between<lb/> reading-level matched groups in phonologi-<lb/>cal analysis and pseudoword reading skills.<lb/> This finding does not refiect a lack of statisti-<lb/>cal power; the mean Digit Span scores of the<lb/> two reading-level groups were not even in<lb/> the predicted direction.<lb/></p> <p>According to <ref type="biblio">Stanovich et al. (1986)</ref>, dif-<lb/>ferences between poor readers and younger<lb/> reading-level controls are consistently ob-<lb/>served only when the groups have been<lb/> matched on IQ. However, this argument<lb/> does not fully explain either previous or cur-<lb/>rent results (see introduction). Although no<lb/> IQ matching was attempted in the present<lb/> study, less skilled fourth-grade readers<lb/> scored lower than reading-level controls on<lb/> both phonological oddity and pseudoword<lb/> reading tasks (see also <ref type="biblio">Bradley & Bryant,<lb/> 1978; Frith & Snowling, 1983; Olson et al.,<lb/> 1989</ref>).<lb/></p> <p>It appears more likely that the discrep-<lb/>ancies in reading-level design studies exam-<lb/>ining phonological analysis and pseudoword<lb/> reading skills refiect the inherent conserva-<lb/>tism of this design, with regression effects<lb/> tending to produce Type II errors. The prob-<lb/>ability of regression effects is inversely re-<lb/>lated to the reliability of the measure used to<lb/> match groups for reading level. Regression<lb/> effects may be countered by the careful<lb/> choice of highly reliable tests for initial<lb/> subject-group assignment and by double ini-<lb/>tial testing. In this regard, we should note<lb/> that the correlation between forms G and H<lb/> of the Word Identification subtest of the<lb/> Woodcock Reading Mastery Tests in the<lb/> present study was extremely high, r(99) =<lb/> .97.<lb/></p> <p>These methodological arguments are<lb/> bolstered by purely logical ones. The find-<lb/>ing that less skilled readers perform worse<lb/> than reading-level controls on both phono-<lb/>logical analysis and pseudoword reading fol-<lb/>lows automatically from findings that phono-<lb/>logical analysis contributes directly to word<lb/> reading skill (e.g., <ref type="biblio">Cunningham, 1990;<lb/> Lundberg et al., 1988</ref>). Since younger<lb/> skilled readers are by definition developing<lb/> faster than less skilled older readers, then<lb/> samples from each of the two populations<lb/> matched for reading level at time 1 will dif-<lb/>fer in word identification at time 2. If phono-<lb/>logical analysis contributes to word identi-<lb/>fication, then it is at least in part their<lb/> superior level of phonological analysis skill<lb/> at time 1 that enables the younger children<lb/> to make faster progress from time 1 to time 2<lb/> in word identification. It follows that inferior<lb/> performance by less skilled readers relative<lb/> to younger reading-level controls does not<lb/> necessarily imply a qualitative difference<lb/> between those readers (see also <ref type="biblio">Bryant &<lb/> Coswami, 1986</ref>).<lb/></p> <p>Our results suggest that less skilled<lb/> readers are delayed in the acquisition of<lb/> phonological analysis and phonological re-<lb/>coding skills (assessed by pseudoword read-<lb/>ing accuracy) that may be essential in the<lb/> development of efficient word reading. This<lb/> delay does not appear to refiect underlying<lb/> differences in verbal working memory. The<lb/> view that the inferior word reading ability of<lb/> the less skilled readers refiects underlying<lb/> differences in phonological analysis and re-<lb/>coding skills is supported by the correla-<lb/>tional data which showed that phonological<lb/> analysis, but not verbal working memory,<lb/> was consistently associated with pseudo-<lb/>word reading. In the conservative correla-<lb/>tional analysis of data from the 32 children<lb/> reading at upper second-grade level, verbal<lb/> span measures were not associated with ei-<lb/>ther phonological analysis or pseudoword<lb/> reading. More reliable estimates of the<lb/> strength of the association between phono-<lb/>logical analysis and pseudoword reading are<lb/> obtained from the fourth-grade subsample.<lb/> Within this sample (n = 36), the correlation<lb/> between total phonological oddity (over the<lb/> four subtasks) and Word Attack performance<lb/> was ,80.<lb/></p> <p>Relative to reading-level controls, less<lb/> skilled readers appear to experience the<lb/> same relative deficit in both subsyllabic and<lb/> phonemic analysis skills. It appears that<lb/> both of these skills develop partly as a conse-<lb/>quence of reading instruction (see <ref type="biblio">Bowey &:<lb/> Francis, 1991</ref>), although early instruction in<lb/> these skills enhances reading development<lb/> <ref type="biblio">(Cunningham, 1990; Lundberg et al., 1988)</ref>.<lb/> Given the greater difficulty of phonemic<lb/> analysis, it may be educationally useful to<lb/> consolidate subsyllabic phonological analy-<lb/>sis skills and encourage poor readers to use<lb/> "large unit" analogies in the decoding of un-<lb/>familiar words, at least in the first instance.<lb/> This suggestion is consistent with recent<lb/> work indicating that beginning readers find<lb/> it relatively easy to recognize words taught<lb/> with orthographic rime units highlighted<lb/> <ref type="biblio">(Wise, Olson, & Treiman, 1990)</ref> and that<lb/> poor readers can be taught to pronounce<lb/> new words by analogy to known words with<lb/> similar orthographic rimes (<ref type="biblio">Baron, 1979</ref>; see<lb/> also <ref type="biblio">Coswami, 1986</ref>). Such strategies may<lb/> foster early reading success and avoid many<lb/> of the cumulative effects of reading failure<lb/> (see <ref type="biblio">Stanovich, 1986</ref>).<lb/></p> </text> </tei>