Abstract

Martine A. R. Gijsel, Anna M. T. Bosman, and Ludo Verhoeven

JOURNAL OF LEARNING DISABILITIESVOLUME 39, NUMBER 6, NOVEMBERIDECEMBER 2006, PAGES 558-571

disorders in relatives, specific lan-guage impairment, and gender), cog-nitive factors (i.e., letter knowledge,rapid naming ability, and nonwordrepetition), and teachers' judgments.With respect to cognitive factors, weincluded a letter knowledge test, be-cause letter knowledge has been foundto be the best predictor of reading per-formance (Blaiklock, 2004; Braams &Bosman, 2000;Catts, Fey,Zhang, &Tom-blin/ 2001; Gallagher, Frith, & Snow-ling, 2000; Hammill, 2004; Pennington& Lefly, 2001; Scarborough, 1998). Inaddition to letter knowledge, phono-logical processing skills have also beenshown to correlate with subsequentreading skills. Wagner and Torgesen(1987) distinguished three types ofphonological processing abilities: pho-nological awareness, phonological re-coding in lexical access (long-termmemory), and phonetic recoding tomaintain information in working mem-ory (short-term memory). Althoughmost research has aimed at investigat-ing phonological awareness, Scarbor-

With respect to the prediction of read-ing skills, we explored the relationshipbetween reading performance and riskfactors (i.e., dyslexia, speech-language

Predictive Factors forReading Performance

wants to know which child is at risk forreading problems in order to preventdifficulties and also to take appropriatepreventive measures. Therefore, it isimportant to know the critical predic-tive factors for reading failure. Thesefactors could be useful in a screeningbattery for subsequent reading skills.In this study, we will investigate theimportance of discriminating between.the prediction of reading abilities ingeneral and the prediction of readingdifficulties in particular. In shortt in ad-dition to the relationship betweenreading ability and risk factors, cogni-tive factors, and teachers' judgments,we will explore the discriminatorypower of these predictors.

This study focused on the predictive value of risk factors, cognitive factors, and teachers' judgments in a sample of 462 kindergartnersfor their early reading skills and reading failure at the beginning of Grade 1.With respect to risk factors, enrollment in speech-languagetherapy, history of dyslexia or speech-language problems in the family, and the role of gender were considered. None of these risk fac-tors were significantly related to reading performance. Cognitive factors in this study included letter knowledge, rapid naming ability,and nonword repetition skills. Of these skills, letter knowledge seemed to have the highest correlation with reading. Kindergarten teach-er,s' judgments, including a task assignment scale and teachers' predictions, demonstrated a significant relationship with reading. Finally,to judge whether these predictors could identify reading disabilities, the discriminatory power of all predictors was assessed and ap-peared to be insufficient. Implications for screening purposes are discussed.

Kindergarten Risk Factors,Cognitive Factors, and TeacherJudgments as Predictors of EarlyReading in Dutch

Factors that have been dem-onstrated to correlate with read-ing performance in previous re-

search include risk factors, cognitivefactors, and teachers' judgments. If asignificant relationship between a par-ticular factor and reading performancehas been established, the factor is oftenidentified as a predictor of readingability. Predictors are referred to as con-current predictors if the relationship be-tween the factor and reading is dem-onstrated at the same point in time. Incontrast, in longitudinal predictors, therelationship is demonstrated over acertain time interval. In this case, as-sessments are usually made in kinder-garten in order to predict reading skillsin higher grades. In the current study,we expand on the existing researchand focus on longitudinal predictors ofreading performance in Dutch firstgraders.

For screening purposes, the pre-diction of reading disabilities (RD) ismore important than the prediction ofreading ability in general. A teacher

ough (1998) pointed out that phono-logical awareness tests appear to bemore successful in predicting futuresuperior reading than future readingproblems. Moreover, Blaiklock (2004)suggested that letter knowledge andphonological awareness show substan-tial overlap in the explained amount ofvariance in reading (for relations be-tween letter knowledge and phono-logical awareness in preschoolers, seeJohnston, Anderson, & Holligan, 1996).Therefore, instead of measuring phon-ological awareness, it was decided toassess phonological memory by meansof a nonword repetition test and to testphonological naming by means of arapid serial naming test. We decidedon these tests because little is knownabout the relative impact of these abil-ities on predictions of Dutch students'reading abilities.

Risk FactorsA family history of dyslexia or speech-language disorders in relatives is oftenconsidered to negatively affect stu-dents' performance on language andreading tests. Most studies of geneticfactors in reading performance havebeen concerned with the prevalence ofdyslexia in relatives. Snowling, Gal-lagher, and Frith (2003) showed that66.1% of the students with at least onefamily member with dyslexia experi-enced reading problems at the age of 8.Lewis, Freebairn, and Taylor (2000)also showed a trend for a family his-tory of RD to predict reading failure.Moreover, a family history of readingproblems significantly predicted spell-ing impairment. A family history ofspeech-language disorders as a pros-pective risk factor for reading has beenstudied less frequently. Lewis et a1.(2000) showed a moderate associationbetween a family history of RD andreading impairment at school age ifthis variable was assessed in a dichot-omous fashion (i.e., positive vs. nega-tive). When it is was coded as an ordi-nal variable (Le.,the number of nuclearfamily members affected) no signifi-cant effect emerged.

VOLUME 39. NUMBER 6. NOVEMBER/DECEMBER 2006

A second variable that possiblyconstitutes a risk factor for RD isspeech and language characteristics. Inmost studies, poor language or speechcharacteristics have been diagnosed bymeans of test assessment (see, e.g., deJong, & van der Leij, 1999; Menyuk etal., 1991; Scarborough, 1990). Usually,the results of speech-language tests arerelated to reading performance. An al-ternative way to assess informationabout speech and language character-istics is to investigate the history ofspeech-language therapy. Weiner (1985)found a high incidence of later readingproblems in preschoolers who wereenrolled in language therapy.

Third, differences in gender mayplay a causal role in lowering testperformance. Petersen (2002) demon-strated that although boys showed asignificantly higher score on vocabu-lary and nonverbal IQ at the beginningof kindergarten, they showed lowerscores on reading performance inGrade 2. Badian (1999) indicated thatgirls were significantly better at read-ing comprehension than boys. No suchdifferences were found for listeningcomprehension. Blonk and Bosman(2003) found that teenage girls in thefirst year of secondary school per-formed better than teenage boys onreading and language skills and, moreimportant, that teenage girls withdyslexia scored better than teenageboys with dyslexia. Flynn and Rahbar(1994) tested 708 students in Grades 1and 3 on reading achievement and cat-egorized the students as having severeRD « 10th percentile), RD (11th to 30thpercentile), or typical reading abilities.In the severe RD category, the ratio ofboys and girls was 1.4 to 1 in first gradeand 1.3 to 1 in third grade. In the RDcategory, the number of boys and girlswas approximately the same in eachgrade.

Cognitive FactorsLetter Knowledge. In the do-

main of cognitive abilities, letter-nameknowledge is a factor that has beenstudied frequently in relation to read-

559

ing skills. The combined results of threemeta-analyses demonstrated that to-gether with reading itself and knowl-edge about writing convention, letterknowledge (r = .52) was the best pre-dictor of reading (Hammill, 2004).Other studies of both English- andnon-English-speaking populations haveyielded similar results. De Jong andvan der Leij (1999) demonstrated thehigh predictive value of letter knowl-edge in the Dutch language. In theirtest, they presented five letters usedrelatively frequently in Dutch books.On the productive letter test, both let-ter names and letter sounds were con-sidered correct. The correlation be-tween receptive and productive letterknowledge in kindergarten and worddecoding at the beginning of Grade 1was moderate (r = .39 and r = .51, re-spectively). At the end of Grade 2,however, the correlations between let-ter knowledge in kindergarten andword decoding in Grade 2 had de-creased dramatically. Braams andBosman (2000) also used a letter nam-ing task in kindergarten to predict thereading and spelling ability of Dutchstudents in Grade 1. At the beginningand at the end of kindergarten, stu-dents had to name 20 letters. Letterknowledge tested at the beginning ofkindergarten revealed a moderate cor-relation with reading performance atthe middle of the curriculum inGrade 1 (r = .44),but the correlation be-tween letter knowledge in kinder-garten and reading declined with in-creasing reading experience to .36 atthe end of Grade 1.

Rapid Naming. Another factoroften related to the prediction of read-ing performance is rapid naming (e.g.,Allor, 2002; Blachman, 1984;Cornwall,1992; Hammill, Mather, Allen, &Roberts, 2002;Kirby, Parrila, & Pfeiffer,2003). In rapid naming tests, partici-pants are asked to name a set of items(usually pictures, colors, letters, or dig-its) as quickly and accurately as pos-sible. The strength of the relationshipbetween rapid naming and readingperformance is dependent on the age

560

of the participants. Kirby et aI. (2003)demonstrated that naming speed inhigher grades had much stronger ef-fects on reading than in kindergartenand Grade 1, with the latter gradesshowing significant, albeit weak, ef-fects. De Jong and van der Leij (1999)also reported small effects of rapidnaming ability in kindergarten onreading skills in Grade 1. Van den Bos,Lutje Spelberg, and Eleveld (2004)tested kindergartners on rapid naming(colors and pictures) and visual match-ing, and they studied the predictivepower of these tests for reading abilityin Grade 1.They demonstrated a mod-erate but significant multiple correla-tion (r = .52). Note that this correlationis even higher than the one betweenletter knowledge in kindergarten andreading ability in Grade 1 as reportedby Braams and Bosman (2000). Ham-mill et aI. (2002) also investigated theimportance of rapid naming ability inpredicting word identification. In theirstudy, the correlation between rapidnaming (letters and words) and wordidentification was .52. In a study byCornwall (1992), the correlation be-tween rapid naming and the identifi-cation of individual words was .49 forletters and .19 for colors. It thus seemsthat the strength of the relationship be-tween rapid naming and reading alsodepends on the kind of stimuli used inthe test.

Nonword Repetition. Anotherfactor that is often associated withreading ability is nonword repetition.In nonword repetition tests, partici-pants are asked to repeat a set of non-words. The test items generally obeythe phonological structure of the lan-guage. A number of studies havedemonstrated that students with RDperform relatively badly on a test forrepeating nonwords (Gallagher et aI.,2000; Lewis et al., 2000; Muter &Snowling, 1998;Snow ling, 1981;Snow-ling, Goulandris, Bowlby, & Howell,1986;Snowling et aI., 2003).These find-ings are often explained in terms ofdeficits in phonological working mem-ory, integrity of phonological represen-

JOURNAL OF LEARNING DISABILITIES

tations, or phonological decoding. Inthe study of Lewis et al. (2000), allparticipants were enrolled in speech-language therapy. Reading impairmentin this group of students was associ-ated with their scores on nonword rep-etition in kindergarten. Bishop (2001)also concluded that in children withSLI, deficits in nonword repetition andpoor literacy skills were correlated,suggesting that the same genes wereresponsible for both deficits. Muterand Snow ling (1998) demonstratedthat the strength of the correlation be-tween nonword repetition as a longi-tudinal predictor and reading perfor-mance at the end of Grade 5 wassignificant in a typical population-that is, in children without SLI (rs rang-ing from .31 at the age of 4 to .53 at theage of 6).

Teachers' Perceptionsand PredictionsFinally, we were interested in the pre-dictive value of the perceptions andpredictions of kindergarten teachers.In a study by Kenny and Chekaluk(1993), teachers' questionnaire scoresand teachers' category ratings (ad-vanced, average, or poor reading)showed good predictive value for au-ditory conceptualization scores andreading achievement. Flynn and Rah-bar (1998) demonstrated that teacherscorrectly predicted 64% of poor read-ers and missed 36% of those studentswho failed in reading. In their study,the correspondence between teacherratings and students' scores was low,except for scores in the area of letter-sound knowledge, and it appearedthat a combination of test performancescores and teacher information wasmost effective with respect to the pre-diction of reading failure.

Discriminatory Power ofPredictive Factors

From a scientific point of view, it is in-teresting not only to acquire knowl-edge about predictors of reading per-formance in general, but also to obtain

an accurate prediction of the likelihoodof developing RD. Therefore, we wereinterested in the discriminatory abilityof risk factors (e.g., dyslexia or speech-language disorders in relatives, enroll-ment in speech and language therapy,gender), cognitive factors, and teach-ers' predictions and perceptions. Inother words, we wanted to knowwhether these variables could predict achild's likelihood of developing RD inGrade 1. A moderate or high correla-tion demonstrating the predictive va-lidity of a factor does not imply thatthis factor will discriminate well be-tween students who will develop RDand those who will not.

In the current study, children whoperformed below the cutoff score-that is, below the 25th percentile-inreading after 2 months of formal read-ing instruction were considered chil-dren with RD. Both practical andtheoretical evidence demonstrate thevalidity of early assessment and stabil-ity of reading scores for Dutch studentsthroughout all grades. Verhoeven andvan Leeuwe (2003) categorized 2,873students into five groups at differentreading levels, based on their readingperformance after 3 months of formalreading instruction in Grade 1. Thesestudents showed quasi-stable meanlevels of reading skills through Grade 6for one-syllable CVC words (C =: con-sonant; V = vowel). Bast and Reitsma(1998) reported a similar outcome in280 Dutch students-that is, childrenwho were diagnosed as poor readersafter 3 months of reading instructionremained poor readers during the firstthree grades.

Moreover, the reading-acquisitionrate in consistent alphabetic orthogra-phies such as Finnish (see, e.g., Aro,2004; Lyytinen et aI., 2004) is differentfrom the rates reported for English-alanguage with a highly inconsistent or-thography. In Finnish, one of the mostconsistently spelled alphabetic lan-guages, the majority of children haveacquired accurate decoding skills afterthe first semester inGrade 1. Thus, RDcan be identified much earlier than inEnglish. The same point can be made

for Dutch; the Dutch orthography isalso relatively consistent in its spelling-sound relationships. After a very shorttime (within 4 months, before Christ-mas), children in first grade havelearned all the prototypical grapheme-phoneme correspondences. Thus, aftera relatively short time, children shouldbe fully aware of the alphabet princi-ple. Moreover, Dutch reading curriculastrongly stimulate the awareness of thealphabet principle, because they all ad-here to the phonics principle.

Moreover, Wentink and Verhoe-ven (2001) implemented a protocol forthe early assessment and interventionof RD, which is now used widely inDutch primary schools. They recom-mended assessing reading skills after 6weeks of formal reading instruction.

Risk FactorsWith respect to risk factors, the numberof children with speech-language dis-orders or affected relatives (i.e., withdyslexia or speech-language disor-ders) or the number of boys in an RDsample may indicate the discrimina-tory power of these variables to someextent. If a significant proportion ofthese children at risk is represented inthe lower tail of the reading distribu-tion and a very small proportion is rep-resented in the highest tail, the par-ticular risk factor might discriminatequite well between typical childrenand children with RD. As mentionedearlier, Snowling et a1. (2003) showedthat 66.1% of the students with at leastone family member with dyslexia ex-perienced reading problems at the ageof 8. Scarborough (1989, 1991) alsodemonstrated that family incidence ofRD was an accurate predictor of read-ing ability; outcomes were correctlypredicted for 72.6%to 80.6%of the par-ticipants, depending on the way inwhich family incidence was identified(Scarborough,1989).

Cognitive FactorsLetter knowledge seemed to be a use-ful factor in predicting reading ability.

VOLUME 39, NUMBER 6, NOVEMBER/DECEMBER 2006

But the question is, will letter knowl-edge discriminate between poor read-ers and typical readers? To answer thisquestion, the number of valid positivesand valid negatives versus false posi-tives and false negatives can be calcu-lated. Valid positive rate (hits) refers tothe number of students who were pre-dicted to have RD who did turn out tobe poor readers. False positive rate (falsealarm) refers to the number of studentswho were predicted to have RD butwho turned out to be typical readers.Valid negative rate (correct rejection)refers to the group of students in whichRD were not predicted and not ob-served. Finally,false negative rate (misses)refers to the number of students whowere predicted to be typical readersbut who turned out to have RD. In thisway, the correctness of classification(poor vs. typical readers) based on per-formance on the predictor tests can beevaluated. Muter and Snowling (1998)considered this issue and selected agroup of poor readers (reading accu-racy scores below 25th percentile) andtypical readers (reading scores abovethe 75th percentile) in their sample.They used rhyme detection, phonemedeletion, nonword repetition, and letter-name knowledge as predictors. Scoreswere obtained at the ages of 4,5, and 6years. These four variables did notclassify the students well at all test mo-ments. At the age of 4, no significantpredictor set was found. When testscores were obtained at the age of 5 or6, 80% of the students were classifiedcorrectly when rhyme detection waseliminated. Analyses revealed thatphoneme deletion and nonword repe-tition were the best predictors. Letterknowledge was a significant predictoras well, but its discriminatory powerwas relatively low.These results shouldbe interpreted carefully-first, becausetheir sample size was quite small (only20 children), and second, because highcorrelations do not imply high dis-criminatory power, and the latter is de-sirable for a screening battery.

Hammill et a1. (2002) also as-sessed the discriminatory power oftests. For this purpose, participants

561

were divided into two groups (scoringabove or below the 25th percentile)based on performance variables. Lev-els of agreement (i.e., the score on a testor subgroup of tests correctly identifiesthe level of word identification) werecalculated. Results indicated 26 out of56 false positives (46%). The percent-age of false negatives was 15%;none ofthe test variables in the study (seman-tics, grammar, phonology, rapid nam-ing, and rapid marking) were effectivepredictors of poor reading. With re-spect to nonword repetition, Muterand Snowling (1998)showed that scoreson nonword repetition and phonemedeletion obtained at the age of 5 or 6significantly discriminated betweengood and poor readers.

Teachers' Perceptionsand PredictionsSatz and Fletcher (1988) reviewed a setof major studies regarding teachers'predictions. A comparison between kin-dergarten teachers' predictions and testresults in Grades 1 and 2 revealed thatoverall hit rates were almost identicalfor test results and teacher predictions.Kenny and Chekaluk (1993) studiedthe utility of teachers' perceptionsand test outcomes from kindergartenthrough Grade 2. Teachers completed aquestionnaire and had to categorizethe students into advanced, average,and poor readers. In teachers' classifi-cations, false positive rates (rangingfrom .30 in kindergarten to .13 inGrade 2) were higher than false nega-tive rates (.23, and .06, respectively).This means that the number of stu-dents who were predicted to have RDbut who turned out to be typical read-ers was higher than the number of stu-dents who were predicted to be typicalreaders but who turned out to have RD.

Flynn and Rahbar (1998) pre-sented teachers with a rating scale withbehavioral descriptions reflecting 10kindergarten skills and assessed theirpredictive value for reading skills inthird grade. Teachers correctly classi-fied 64% of the poor readers, in con-trast with a screening test, which

562

yielded 80% valid positives. A combi-nation of teachers' ratings and thescreening test resulted in 88% validpositives. However, the number offalse positives increased from .23(teachers' ratings alone) to .39, whichindicated an overidentification of stu-dents at risk.

In addition to level of agreement,the sensitivity, specificity, and positivepredictive value can be calculated toexplore the practical value of a factor.The sensitivity index reflects the abilityof a test to correctly identify individu-als who have the disorder. The speci-ficity index reflects the ability of a testto correctly identify individuals whodo not have the disorder. The positivepredictive value reflects the proportionof valid positives among all individu-als whom the screening measure iden-tifies as at risk. Petersen (2002)demon-strated a better sensitivity of predictorsfor' advanced reading than for poorword reading.

To sum up, in the present study,we seek answers to the following twoquestions:

1. What is the relationship betweenrisk factors, cognitive factors, andteachers' predictions and percep-tions on the one hand and readingperformance on the other hand?

2. How well do these factors identifystudents who will develop RD?

Method

The data collection in this study com-prised a kindergarten test battery, twoquestionnaires designed for the par-ents and teachers of the kindergart-ners, and a Grade 1 test to assess read-ing skills.

ParticipantsOf the 462 students participating inthis study, 241 were boys (52.2%) and221 were girls (47.8%) from 20 generaleducation primary schools in theNetherlands. All students were first

JOURNAL OF LEARNING DISABILITIES

tested when attending kindergarten,and their mean age was 70.8 months(SD = 4.4). In Grade 1, when they weretested a second time, the mean age was79.1 months (SD = 4.4).

QuestionnairesParents' Questionnaire. This ques-

tionnaire consisted of four questionsconcerning the native language of thechild, the child's enrollment in speechor language therapy, and the presenceor absence of dyslexia or speech-language difficulties in relatives. Thesequestions had to be answered by mark-ing the option yes or no. If the answerwas yes, further information was re-quired (e.g., which of the first-degreerelatives had dyslexia or speech-language problems, or what the child'snative language was if not Dutch). Theparents of all kindergartners received aquestionnaire.

Teacher Questionnaire. This ques-tionnaire consisted of two parts. InPart 1, teachers were asked to writedown the names of students whomthey believed would develop readingor spelling difficulties in Grade 1.Withrespect to the analyses, this variablewas dichotomous (yes vs. no). Part 2was a sub scale of the Task AssignmentScale (Werkhoudingslijst) of van Doorn(1996) and consisted of 10 items con-cerning concentration, motivation, andattitude of the child. The teacher wasasked to make judgments for each ofthe students in their classroom aboutthe frequency of the described behav-ior on a 5-point scale, ranging fromnever to always. Examples are "the childtries to solve problems by himself orherself," "the child needs encourage-ment during work," and "the childstarts working immediately after in-struction." With respect to the analy-ses, scores on task assignment were as-signed to the answers and categorizedeither in the lowest quartile (scoresbelow the 25th percentile) or in thehighest quartile (scores above the 75thpercentile) .

Kindergarten Test Battery

Nonword Repetition Test. TheNonword Repetition Test (Nonwoord Rep-etitietaak; Irausquin, 1999) required therepetition of 22 pronounceable non-words. To control for possible articula-tion errors or hearing problems of thechild, the test was preceded by 15 realwords. Thus, errors in nonword repeti-tion (e.g., substitutions due to certainarticulation errors) that already oc-curred in real word repetition werecounted as correct. In case of hearingdisorders, test assessment was stopped.Performance on real words was not in-cluded in the test score; only repetitionof the nonwords determined perfor-mance. The length of the items in-creased, ranging from one to four syl-lables. Nonword examples are klodaand bledistot. Three practice items pre-ceded the nonwords. The experi-menter pronounced each word andnonword with a piece of paper in frontof the mouth in order to prevent theuse of visual information. The childwas asked to repeat the words andnonwords after the experimenter. Withrespect to the analyses, the test scorewas the number of nonwords correctlyrepeated. The maximum score on thetest was 22.

Rapid Naming Picture Test. TheRapid Naming Picture Test (BenoemtaakPlaatjes; van den Bos, 2004) consisted of5 different pictures (tree, duck, chair,scissors, and bicycle) depicted in fivecolumns of 10 pictures each, yielding atotal number of 50 pictures to benamed. Each picture was presented 10times at random positions. The childwas asked to name all the pictures asfast as possible and as correctly as pos-sible in a vertical direction (i.e., columnby column). The test was preceded bya short training, in which the child hadto name the pictures in the final col-umn. The time to complete the taskand the number of errors were regis-tered. With respect to the analyses, themean time to name one picture (cor-rectly or incorrectly) was computed.

The number of errors was not includedas a variable because few errors weremade.

Rapid Naming Colors Test. TheRapid Naming Colors Test (BenoemtaakKleuren; van den Bos, 2004) consistedof five columns of 10 colored squares(black, yellow, red, green, and blue).The test procedure and the scoringwere exactly the same as with the RapidNaming Picture Test. Note that we didnot use alphanumeric stimuli becausethe students had not received formalreading and writing instruction yet.

Letter Knowledge Test. The Let-ter Knowledge Test considered produc-tive letter knowledge as well as recep-tive letter knowledge. First, productiveletter knowledge was tested. This testconsisted of all 26 letters of the alpha-bet, presented on a card in six rows of4 letters and one row of 2 letters. Theletters were typed in Helvetica regularfont, size 24. If the capital letter had adifferent shape than the lowercase let-ter (which is the case for most letters),the capital letter was presented justbelow the lowercase letter. Thus, all ofthe letters had two presentation forms,except the letter a, which had three dif-ferent representations (a, A, and a).Three different orders of letters werecreated. The child was presented onerow of letters at a time; a sheet of papercovered the other letters. The child hadto name the letters in a horizontal di-rection. Responses were consideredcorrect if the child named either thelowercase letter or the capital lettercorrectly; both letter names and lettersounds were considered correct. Incase of doubt about the production ofa letter, the child was asked to name aword with that letter in the initial posi-tion. Uncertainty only emerged in thedistinction between voiced sounds andtheir voiceless counterparts like s/ zandf/v. When the child did not knowa certain letter, he or she was allowedto guess and continue with the nextletter.

Subsequently, receptive letterknowledge was tested. The child was

VOLUME 39, NUMBER 6, NOVEMBER/DECEMBER 2006

presented with the same card. The ex-perimenter named a letter sound (inrandom order), and the child wasasked to point to that letter on the card.Guessing was permitted. With respectto the analyses, two variables werecomputed, namely, the score on theproductive and the score on the recep-tive letter knowledge test. The testscore was the total number of correctresponses. The maximum score oneach test was 26.

Grade 1Reading TestThe Word Reading Test (Toets WoordenLezen; Wentink & Verhoeven, 2001)consisted of three parts, each of themincluding 10 (C)VC words. The firstpart consisted of well-known wordsthat had been taught in the classroom.The second part consisted of newwords that differed by one letter fromwords that had been taught in theclassroom. In this way, the items re-sembled the words that had alreadybeen taught. The third part consisted ofnew words that differed by two lettersfrom the words taught in the class-room. In this condition, there washardly any resemblance to the wordsthat had already been taught. The childwas asked to read aloud the words asfast and as accurately as possible. Foreach part, the test was stopped whenthe child failed on three consecutiveitems. With respect to the analyses, thenumber of items read correctly and thetime needed to perform each part ofthe test were converted into the num-ber of items read correctly in oneminute. The first part of the test (de-coding well-known words) was notconsidered in the analyses. Therefore,the results of the second and third partare henceforth referred to as WordReading 1 and 2, representing the abil-ity to decode new words.

ProcedureStudents were tested in kindergartenand at the beginning of Grade 1. Inkindergarten, no structural (reading)

563

program was used and, in most schools,there was no structural remediation inkindergarten concerning reading orreading-related skills. In Grade I, themajority of the children were in-structed with Veilig Leren Lezen (Learn-ing to Read Safely), the most widelyused reading program in Dutch schools(Mommers, Verhoeven, & van der Lin-den, 1979, 1994). The emphasis in thismethod is on the structure of the or-thographic system and the relationshipbetween letters and sounds (i.e., phon-ics). Initially, only consistently spelledwords are used. After 4 months of in-struction, the children are familiar withthe main grapheme-phoneme corre-spondence rules. It is a fairly rigid,preprogrammed curriculum, whichimposes a strict day-by-day and week-by-week progression.

In kindergarten, the studentswere tested halfway through the cur-riculum Oanuary-March, 2002; hence-forth Time 1; see Note 1). This tookabout 20 min for each child. The orderof presentation of the tests variedamong students, such that each testwas administered equally often first,second, third, and so forth. At the endof kindergarten (May, 2002), the par-ents' questionnaires were provided tothe teachers with the request to presentthem to the parents. Parents wereasked to complete the list and returnit in 2 weeks. Furthermore, the teach-ers' questionnaires were provided. InGrade I, all students were tested 8weeks after formal reading instructionhad started (Fall 2002; henceforthTime 2). All students completed theWord Reading Test. In Grade I, addi-tional instruction or intervention wasimplemented at the earliest after thefirst test assessment-that is, after 2months of formal reading instruction.This coincided (not accidentally) withthe time that RD were diagnosed.

AnalysisToevaluate the predictive value of cog-nitive factors for reading performance,a structural analysis was performed.

564

Furthermore, a multiple regressionwas carried out to determine the bestcombination of variables to predictreading. To investigate the relationshipbetween risk factors and teachers'judgments, t tests were performed onscores on the Word Reading Test inGrade 1. To investigate the discrimina-tory power of all variables, percent-ages of valid and false positive andnegative outcomes were calculated.Moreover, the sensitivity and speci-ficity indexes were computed. Forthese analyses, the outcome had to belabeled dichotomously (Le., develop-ing RD or not). We defined studentswith RD as students with readingscores below the 25th percentile, andgood readers as students demonstrat-ing reading scores above the 75th per-centile. The reading score was themean number of items read correctly inOl'leminute on Word Reading Test 1 and2. We chose the 25th percentile to rep-resent the group of students that per-formed below standard, because thiscriterion is used in standardized Dutch.Finally, we performed a discriminantfunction analysis in order to establishwhich combination of variables dis-criminated best between poor and typ-ical readers.

Results

Sample Characteristics

The majority of students (95%) hadDutch as their native language; 22 stu-dents (5%) were originally from othercountries (both in and outside Europe)and had Dutch as their second lan-guage or were bilingually educated.The number of reported speech andlanguage problems was 71 (15.4%),whereas 6.3% of the parents of all par-ticipants had coped with speech or lan-guage problems. Finally, 6.9% of theparents reported a history of dyslexiain relatives-that is, a first- or second-degree family member with dyslexia.The lack of response on the question-naires was very low: Of the parents'questionnaires, 400 copies (86.6%)werereturned, and of the teachers' ques-

JOURNAL OF LEARNING DISABILITIES

tionnaires, 439 copies (95%) were re-turned.

Statistics

Because 22 students in our samplewere native speakers of a languageother than Dutch, we first establishedwhether the performance of these stu-dents differed from that of their peerswhose mother tongue was Dutch. Aone-way ANOVA on the mean scoresof all tests in kindergarten and on themean scores of the Word Reading Test inGrade 1 indicated that students whosenative tongue was not Dutch did notdiffer significantly from students withDutch as native language (all ps > .05).Therefore, it was decided to include insubsequent analyses all students whoattended kindergarten and partici-pated in the Word Reading Test at thebeginning of Grade 1. Students whohad to repeat kindergarten or who re-peated Grade 1were excluded. Despitethis exclusion criterion, a few missingvalues were inevitable. To resolve thisissue, we made use of the "nearestneighborhood" method to estimate themissing scores.

Predictive Factors forReading Performance

Table 1 shows the mean scores andstandard deviations on all tests inkindergarten and Grade 1. Scores werecomputed for all students together andfor poor readers and typical readersseparately. Table 2 shows the charac-teristics for the poor reading group andthe typical reading group. To explorethe predictive value of the reportedtests, a structural analysis was per-formed. In this analysis, all variableswere included. Figure 1 depicts allvariables and standardized regressionweights. The fit of the model was good,X2(9,N = 462) = 16.2,P = .06,GFl = .99,AGFl = .97, Nfl = .99, RMSEA = .04.

Next, a stepwise multiple regres-sion was carried out in order to findout whether a combination of factorsincreased the strength of the correla-tion. The dependent variable was the

mean number of words read correctlyon Word Reading Test 1 and 2. All riskfactors, cognitive factors, and teachers'predictions and perceptions were in-cluded as independent variables. Re-ceptive letter knowledge was excludedfrom the analysis because of its strongcorrelation (r = .96) with productiveletter knowledge. The results showedthat productive letter knowledge wasthe best predictor of reading perfor-mance, r = .53, P < .001. A second fac-tor that contributed significantly to thevariance was the rapid naming of col-ors. If this variable was added to letterknowledge, 29% of the variance wasexplained (r = .54, P < .001). Despitethese two factors, no other predictorturned out to be significant. These re-sults support the outcome of the struc-tural analysis. Furthermore, to investi-gate the relationship between riskfactors and teachers' predictions andperceptions and reading ability, t testswere performed. Table 3 shows thatonly the teachers' predictions and per-ceptions significantly discriminatedstudents' scores on word reading.

To further investigate the role ofthe teacher, scores on all tests werecompared between students with pos-itive and negative predictions (seeTable 4) and between students withscores within the lowest quartile ontask assignment and those with scoreswithin the highest quartile (see Table 5).For screening purposes, it would beuseful to know whether the studentswith negative ratings represented thelower tail of the distribution of readingscores. Analyses revealed that this wasindeed the case. About half of the stu-dents who were assigned to the lowesttask assignment group and who werepredicted to develop RD were repre-sented in the 25th reading percentile.The scores of the students without pre-dicted RD were more widely distrib-uted.

Discriminatory Power ofPredictive Factors

To answer our second question con-cerning the discriminatory power of

VOLUME 39, NUMBER 6, NOVEMBER/DECEMBER 2006 565

TABLE 1Mean Scores and Standard Deviations in Kindergarten and Grade 1

All students Lowest reading quartile Highest reading quartile(N= 462) (n = 118) (n = 123)

Gradeltest M SO M SO M SO d

KindergartenRapid NamingColors 1.6 0.5 1.9 .6 1.4 .4 .97Pictures 1.6 0.4 1.8 .4 1.4 .4 .96

Letter KnowledgeProductive 10.9 7.8 5.4 5.3 17.5 6.8 1.99Receptive 10.5 7.4 5.4 4.8 16.7 6.8 1.92

Nonword Repetition 16.8 2.8 16.2 2.9 17.2 3.0 .35

Grade 1Word Reading 1Score 7.5 2.4 4.5 2.3 9.3 1.0Time (see) 54.7 30.7 92.2 25.6 22.4 8.6wpm 13.8 14.1 3.2 2.1 30.4 17.7 2.16

Word Reading 2Score 7.3 2.9 3.4 2.5 9.1 1.4Time (see) 59.9 31.5 94.9 29.6 27.9 11.7wpm 11.9 13.0 2.1 1.7 26.0 17.6 1.91

Word Reading 1/2Score 7.4 2.5 3.9 2.1 9.2 1.1Time (see) 57.3 29.4 93.6 22.6 25.1 9.4wpm 12.8 13.1 2.7 1.6 28.2 16.7 2.15

Note. The scores on Rapid Naming are the number of seconds to name an object (colors/pictures). The scores on Letter knowledge (maximum score ~ 26) andNonword repetition (maximum score is 22) are the number of items responded correctly. Word Reading score is the number of items read correctly (maximumscore ~ 10). Wpm is the number of words read per minute.

aLowest quartile task-assignment group.

TABLE 2Frequency Distribution of Poor-Reading (Lowest Quartile) andSuperior-Reading (Highest Quartile) Groups With Respect to

Risk Factors and Teachers' Predictions

valid positives/ (valid positives + falsenegatives), and the specificity index isthe number of valid negatives/(validnegatives + false positives). These dataare presented in Table 6. The results

Lowest reading Highest readingAll students quartile quartile

Factor/judgment (N= 462) (n = 118) (n = 123)

Risk factorsSpeech/language therapy 71 30 11Dyslexia 32 13 7SLl 29 11 4Boy 241 67 62

Teachers' judgmentsNegative prediction 91 50 5Low task assignment a 127 59 15

scores below the 25th percentile on de-coding new words (Word Reading Test 1and 2). Next, for q.llpredictors, we cal-culated the sensitivity and specificity.The sensitivity index is the number of

the variables, we calculated the num-ber of false positive and valid positiveoutcomes and false negative and validnegative outcomes. Recall that the falsepositive rate reflects the number of stu-dents who were predicted to have RDbut who turned out to be typical read-ers. The valid positive rate refers to thenumber of students who were pre-dicted to have RD who indeed turnedout to be poor readers. The valid nega-tive rate refers to the group of studentsfor whom RD were not predicted andnot observed. Finally, the false nega-tive rate refers to the number of stu-dents who were predicted to be typicalreaders but who turned out to haveRD. These percentages were computedfor all variables. In this way, we wereable to evaluate different predictors ofRD. Concerning the cognitive factors,difficulties were predicted if the scoreon the test was below the 25th per-centile. Recall that RD were defined by

FIGURE 1. Structural model of the relationship between kindergartentest scores and decoding skills in Grade 1 (N = 462). *p < .005.

demonstrated that teachers' predic-tions have more predictive power thansome cognitive factors (i.e., rapid nam-ing and nonword repetition).

Finally, we submitted all variablesto a stepwise discriminant functionanalysis in order to establish whichcombination of variables discrimi-nated best between poor readers (read-ing scores below the 25th percentile)and typical readers. All risk factors,cognitive factors, and teachers' percep-tions were included in the analysis.Productive letter knowledge, teachers'predictions, and rapid naming of col-ors contributed significantly to theclassification (p < .001). The canonicalcorrelation between these variablesand group membership (above or be-low the 25th percentile) was .49. Stan-dardized discriminant function coeffi-cients reflecting the contribution ofeach variable showed that productiveletter knowledge contributed most tocorrect classification; rapid naming ofcolors contributed .36, teachers' pre-dictions .47, and productive letterknowledge -.64. The discriminantanalysis had an overall accuracy rate of70.8%. The number of valid positiveswas 46.3%, whereas 12% of the chilodren were misclassified (i.e., false neg-atives). The number of valid negativeswas 88%, and 53.7% of the childrenwere predicted to develop RD butturned out to be typical readers (falsepositives).

DiscussionOur first question concerned the rela-tionship between risk factors, cogni-tive factors, and teachers' predictionsand perceptions on the one hand andreading ability on the other hand. Riskfactors turned out to playa minor rolein the prediction of the reading perfor-mance of Dutch children in Grade 1.Students without speech-languagetherapy only seemingly outperformedthose students enrolled in speech-language therapy in Grade 1; the dif-ference, however, did not reach signif-icance. In kindergarten, students with

t test

Time 2

t(201) = -6.0, P < .001

t(306) = 8.21, P < .001

t(460) = -.59, P = .56

t(390) = 1.73, P = .08

t(366) = 1.81, P = .07

t(394) = 1.87, P = .06

SD

JOURNAL OF LEARNING DISABILmES

Word-Reading Test 1 and 2

M

10.3 14.013.5 12.5

9.17 7.0713.50 13.36

8.90 7.2713.19 13.17

12.49 13.7813.20 12.38

6.3 6.714.7 13.5

8.3 10.618.9 16.7

_ * ~ Word Reading 1\.14 .97* ~--~

.Ol~

.52* .91~ I1Word Reading 2

.98* .97*

B

TABLE 3Results of t Tests for All Students Concerning Risk Factors

and Teachers' Predictions

~~.77* .87*

Time I

Factor/judgment

Risk factorsSpeech/language therapyYesNo

DyslexiaYesNo

SLIYesNo

GenderBoyGirl

Teachers' judgmentsPredictionNegativePositive

Task assignmentBelow 25th percentileAbove 75th percentile

566

VOLUME 39, NUMBER 6, NOVEMBERIDECEMBER 2006 567

TABLE 5Mean Scores and Standard Deviations on All Tests for Both

Task-Assignment Groups

TABLE 4Mean Scores and Standard Deviations of Students With PredictedReading Difficulties (Negative Prediction) and Students Without

Predicted Difficulties (Positive Prediction) by their Kindergarten Teacher

Students (N = 414)

Task assignment

ttest

t test

so

so

Positive(n = 323)

M

ten test scores for reading performancewas analyzed with the use of structuralequation modeling. Letter knowledgeturned out to be the strongest pre-dictor, followed (remotely) by rapidnaming. The important role of letterknowledge in predicting reading per-

M

Highest quartile(n=121)

so

so

Negative(n = 91)

M

M

1.7 0.5 1.6 .4 t(412) = -3.30, p=.0011.8 0.6 1.6 .5 t(412) = -3.45, P =.001

5.1 5.3 12.4 7.6 t(206) = 10.48, P < .0015.2 4.9 11.9 7.2 t(213) = 10.18, P <.001

15.4 3.0 17.1 2.7 t(412) = 5.38, P < .001

6.8 6.8 15.8 14.6 t(324) = 8.27, P < .0015.7 6.8 13.7 13.4 t(295) = 7.68, P < .0016.3 6.7 14.7 13.5 t(306) = 8.21, P < .001

1.74 .51 1.46 .37 t(231) = -5.02, P < .0011.85 .65 1.42 .36 t(198) = 6.50, P < .001

7.1 6.9 14.5 7.5 t(246) = -8.13, P < .0017.1 6.2 14.0 7.3 t(236) = -8.05, P < .001

16.0 2.9 17.1 2.5 t(246) = -3.23, P = .001

8.9 11.5 19.9 17.6 t(205) = -5.81, P < .0017.7 10.7 17.9 16.8 t(202) = -5.71, P < .0018.3 10.6 18.9 16.7 t(201) = -6.00, P < .001

Lowest quartile(n = 127)

Test

reading performance at the beginningof Grade 1. Scarborough (1998) alsoconcluded in another review that boysare only slightly more at risk for RDthan girls.

With respect to cognitive factors,the relative importance of kindergar-

KindergartenRapid NamingPicturesColors

Letter KnowledgeProductiveReceptive

Nonword Repetition

Grade 1Word-Reading 2Word Reading 3Word Reading 2/3

Prediction

KindergartenRapid NamingPicturesColors

Letter KnowledgeProductiveReceptive

Nonword Repetition

Grade 1Word-Reading 2Word Reading 3Word Reading 2/3

a history of speech-language therapyperformed significantly worse on alltests. The lower performance on letterknowledge and the lower (though notsignificant) reading performance instudents with speech or language dis-orders correspond with the findings ofearlier studies (e.g., Carroll & Snow-ling, 2004; Scarborough, 1990). Notethat most studies include formal speechor language tests, and few have usedenrollment in speech-language ther-apy as an indicative variable for read-ing problems. Further research is re-quired to establish the value of thisvariable in identifying students at risk.We suggest gathering detailed infor-mation about the treatment (kinds ofproblems, duration of the treatment) toincrease the accuracy of the identifica-tion of students at risk.

Furthermore, and surpnsmgenough, the current study provided noevidence for a hereditary factor inreading or speech-language problems:The test performance of students withand students without a family historyof dyslexia or speech-language diffi-culties was statistically equal for allmeasures. This result is in contrastwith a number of studies performedwith English-speaking children (e.g.,Pennington & Lefly, 2001; Scarbor-ough,1989,1991;Snowlingetal.,2003).Scarborough (1998) reviewed severalstudies and concluded that the familyincidence of dyslexia certainly in-creases the risk for RD in a child. How-ever, the degree of risk varies amongstudies and may be caused by the wayin which information was gathered. Inthe current study, parents were inter-viewed by means of a questionnaire. Ifwe had tested the reading ability of theparents and their speech and languageperformance, our results might havebeen different. Scarborough (1989)showed that although self-reportedRD in parents significantly predictedchildren's reading performance, pre-dictions were more accurate when testresults were used. With respect to gen-der, the difference between boys andgirls did not reach significance for

Note. Sensitivity index is the number of valid positives/(valid positives + false negatives), and specificity index is the number of valid negatives/(valid negatives +false positives).aTh,e percentage of valid positives is equal to positive predictive value.

TABLE 6Percentages of Valid and False Positives and Negatives, Sensitivity and Specificity of Risk Factors,

Cognitive Factors, and Kindergarten Predictors

Factor/judgment Valid positives' False positives Valid negatives False negatives Sensitivity Specificity

Risk factorsSpeech/language therapy 42.2% 57.8% 78.2% 21.8% .30 .86Dyslexia 40.6% 59.4% 77.7% 22.3% .15 .93SLI 37.9% 62.1% 75.5% 24.5% .11 .94Gender 27.8% 72.2% 76.9% 23.1% .57 .49

Cognitive factorsRapid NamingPictures 36.5% 63.5% 78.1% 21.9% .36 .79Colors 38.5% 61.5% 79.1% 20.9% .40 .78

Letter KnowledgeProductive 49.2% 50.8% 84.1% 15.9% .56 .80Receptive 50.8% 49.2% 83.9% 16.1% .54 .82

Nonword Repetition 33.3% 66.7% 77.8% 22.2% .39 .73

Teachers' judgmentsPrediction 55.0% 45.0% 83.0% 17.0% .48 .87Task assignment 46.5% 53.5% 82.0% 18.0% .51 .79

568

formance has been found in previousstudies (e.g., Braams & Bosman, 2000;Catts et al., 2001; de Jong & van derLeij, 1999;Gallagher et a1.,2000; Ham-mill, 2004; Pennington & Lefly, 2001).However, the adequacy of this predic-tor might be restricted to the beginningstage of reading. Walsh, Price, andGillingham (1988) demonstrated thatletter-naming speed correlated muchmore strongly with reading achieve-ment among kindergartners than amongstudents in Grade 2. Other studiesyielded similar results, demonstratingdeclining correlations between letterknowledge and reading ability in highergrades (Braams & Bosman, 2000; deJong & van der Leij, 1999;Wesseling &Reitsma, 2000).

The relationship between rapidnaming and reading performance wasmuch weaker, although it also reachedsignificance. The predictive value ofrapid naming has been demonstratedin several studies. However, the pres-ent study only provides weak evidencefor the predictive value of rapid nam-ing in kindergarten. The strength of therelationship in the current study

JOURNAL OF LEARNING DISABILITIES

would probably have been larger if wehad used letters in the test instead ofpictures or colors. However, because ofthe limited letter knowledge of chil-dren in kindergarten, this was not pos-sible at the time of testing. Some re-searchers have suggested that rapidnaming is a better predictor of poorreading than it is of typical reading(Hammill et al., 2002; Meyer, Wood,Hart, & Felton, 1998;Petersen, 2002).

The ability to correctly repeat a setof nonwords was not significantly re-lated to reading skills and thus playeda minor role in the prediction of de-coding skills. The role of nonword rep-etition in identifying students at riskfor RD and the interpretation of the testoutcome are not yet quite clear. In En-glish, Snowling and her colleagues(Gallagher et al., 2000;Muter & Snow-ling, 1998; Snow ling, 1981; Snow linget al., 1986;Snowling et al., 2003) dem-onstrated that poor readers scored sig-nificantly worse on the repetition ofnonwords than controls. However, ef-fects were influenced by phonologicalcomplexity (group differences weremost apparent when four-syllable non-

words had to be repeated). Itmight bethat our selection of items was not ap-propriate, because we included rela-tively many nonwords with a low levelof phonological complexity.

Next, we were interested inwhether kindergarten teachers' predic-tions and perceptions (ratings of taskassignment) would be predictive forreading outcome. Several studies (e.g.,Coleman & Dover, 1993; Flynn & Rah-bar, 1998; Kenny & Chekaluk, 1993;Taylor, Anselmo, Foreman, Schat-schneider, & Angelopoulos, 2000;Teisl,Mazzocco, & Myers, 2001) have dem-onstrated significant relations betweenteachers' ratings and future school per-formance. The results of the currentstudy supported these findings anddemonstrated that test performancesupported the predictions of teachersquite well. Both in kindergarten and inGrade I, the performance of studentswho were expected to develop RD waslower than the performance of stu-dents with positive predictions ofreading ability. This pattern was clearin all tests. The role of kindergartenteachers' predictions may be under-

estimated because students who re-peated kindergarten were excludedfrom the analyses. Those students wereprobably expected to develop RD inGrade 1 with great certainty. Thepower of teachers' judgments was fur-ther demonstrated by the results of thetask assignment scale. Kindergartnerswho were assigned to high levels oftask assignment performed signifi-cantly better (both in kindergarten andGrade 1) than students with moderateor low task assignment. This resultcorresponds with that of Kenny andChekaluk (1993), who reported a sig-nificant relation between teachers' per-ceptions and reading achievement.Flynn and Rahbar (1998) reported highcorrespondence between teacher rat-ings and letter-sound knowledge, butlow correspondences between teacherratings and all other measures (vocab-ulary, syntax, visual discrimination,form copy). In sum, the present find-ings demonstrated the usefulness ofteachers' predictions and ratings forclassifying children at risk for RD.

The second question concernedthe discriminatory ability of predictivefactors. Rather than looking at the abil-ity to predict actual scores, we exam-ined the ability to predict whether achild will fail in reading or not. Thisshould be an import property of ascreening battery. To evaluate this ca-pacity, calculations of the sensitivityindex, specificity index, and positivepredictive value were performed. Forpractical use in a screening battery,these indexes need to reach a mini-mum value of .75 (Hammill et aI.,2002). In our study, only the specificityindex matched this criterion. Thismeans that most of our predictors wereonly able to correctly identify thosestudents who did not develop RD. Forall predictors, the sensitivity was muchlower than the specificity, which is ageneral trend in prediction studies(e.g., Hammill et aI., 2002; Pennington& Lefly, 2001;Petersen, 2002;Schneider& Naslund, 1993; Teisl et aI., 2001; seeScarborough, 1998, for an overview).Thus, in general, the number of falsepositives and false negatives was too

VOLUME 39. NUMBER 6. NOVEMBER/DECEMBER 2006

high. In particular, the large number offalse positives-that is, the percentageof students who were predicted tohave RD but who turned out to betypical readers (ranging from 45% to72.2%)-was troublesome. Severalother studies also demonstrated the oc-currence of too many false positives(e.g., Catts et aI., 2001; Flynn & Rahbar,1998;O'Connor & Jenkins, 1999;Tayloret aI., 2000) or false negatives (e.g.,Coleman & Dover, 1993; Hammill etaI., 2002; Mantzicopoulos & Morrison,1994). In contrast to the findings ofother studies, RD were identified veryearly in Grade I-that is, after 2months of formal reading instruction.However, as noted in the introduction,results from other Dutch studiesdemonstrated strong stability of read-ing scores during primary school (Bast& Reitsma, 1998; Verhoeven & vanLeeuwe, 2003). Thus, the negative re-sults in our study are unlikely to bedue to early assessment. In sum, riskfactors, cognitive factors, and teachers'perceptions were not sufficiently ade-quate on their own to correctly identifythose students who exhibited RD.

The results of a discriminant func-tion analysis, however, demonstratedthat a combination of productive letterknowledge, rapid naming of colors,and teachers' predictions increased theaccuracy of prediction to an overall ac-curacy rate of 70.8%. These results areconsistent with the results of Penning-ton and Lefly (2001),who performed adiscriminant function analysis anddemonstrated that letter-name knowl-edge and rapid serial naming of colorsand objects were most important inpredicting RD.

In conclusion, the results of thisstudy suggest that group membership(RD or not) in Dutch students at the be-ginning of Grade 1 can be moderatelypredicted in kindergarten. Althoughletter knowledge seems to be a strongcorrelate of word reading, it cannot onits own correctly identify students whowill develop RD. To improve the accu-racy of classification, a combination ofvariables is needed (see also Scarbor-ough, 1998).Surprising enough, teach-

569

ers' predictions and perceptions seemat least as effective as cognitive factorsand might, therefore, contribute toaccurate prediction. A combination ofproductive letter knowledge, teachers'predictions, and rapid naming of col-ors accurately classified 71% of thechildren. Unfortunately, however, toomany students who were predicted tohave RD turned out to be typical read-ers, and too many poor readers werenot identified in kindergarten. Theseresults suggest that kindergarten mea-sures moderately predict subsequentreading skills in Dutch children. How-ever, the results might have been dif-ferent if, in addition to letter knowl-edge, the battery had included othermeasures, such as one or more tests ofphonemic awareness and a test of ver-bal memory. In the Netherlands, agreat deal of attention is already fo-cused on the early screening and pre-diction of RD, especially since theimplementation of the Protocol Lees-problem en en Dyslexie (Protocol ReadingProblems and Dyslexia; Wentink & Ver-hoeven, 2001),which includes a check-list for kindergartners. Letter knowl-edge, phonological awareness, andphonemic awareness are part of thischecklist. In addition to the assessmentof letter knowledge, we recommendthe use of kindergarten teachers' pre-dictions. These two measures are lowcost and accurate and require only alimited amount of time. Finally, read-ing skills in Grade 1 have to be as-sessed as early as possible-that is, after2months of formal reading instruction,as recommended by Wentink and Ver-hoeven (2001). After all, the best pre-dictor of future reading is reading itself(see, e.g., Hammill, 2004).

ABOUT THE AUTHORS

Martine A. R. Gijsel is a certified speech andlanguage pathologist and a PhD student in theDepartment of Special Education, RadboudUniversity Nijmegen. Her research interests in-clude prediction and prevention of reading dif-ficulties, and deficiencies in speech and lan-guage development. Anna M. T. Bosman isassociate professor in the Department of SpecialEducation, Radboud University Nijmegen. Her

570

research interests include reading, spelling,dyslexia, intervention and dynamic systems.Ludo Verhoeven is professor in the Depart-ment of Special Education, Radboud UniversityNijmegen. His research interests include read-ing and language development in L1 and L2learners. Address:

AUTHORS' NOTE

We are greatly indebted to all children andteachers of the schools that participated in ourstudy, and we want to express our appreciationto all students who assisted us in collecting thedata. We are also very grateful to Jan vanLeeuwe for his statistical support. Finally, wewant to thank three anonymous reviewers fortheir helpful comments on an earlier version ofthe article.

NOTE

Near the end of kindergarten, children weretested on letter knowledge for a second time. Be-cause correlations were as high as .92 for bothproductive and receptive letter knowledge, how-ever, this measurement was not included in theanalysis.

REFERENCES

Allor, J. H. (2002). The relationships ofphonemic awareness and rapid namingto reading development. Learning Disabil-ity Quarterly, 25, 47-57.

Aro, M. (2004). Learning to read; The effect oforthography. Jyvaskyla, Finland: Univer-sity of Jyvaskyla.

Badian, N. A. (1999). Reading disability de-fined as a discrepancy between listeningand reading comprehension: A longitudi-nal study of stability, gender differences,and prevalence. Journal of Learning Dis-abilities, 32,138-148.

Bast, J., & Reitsma, P. (1998). Analyzing thedevelopment of individual differences interms of Matthew effects in reading: Re-sults from a Dutch longitudinal study.Developmental Psychology, 34, 1373-1399.

Bishop, D. V. M. (2001). Genetic influenceson language impairment and literacyproblems in children: Same or different?Journal of Child Psychology and Psychiatryand Allied Disciplines, 42, 189-198.

Blachman, B. A. (1984). Relationship ofrapid naming ability and language analy-sis skills to kindergarten and first-gradereading achievement. Journal of Educa-tional Psychology, 76,610-622.

JOURNAL OF LEARNING DISABILITIES

Blaiklock, K. E. (2004). The importance ofletter knowledge in the relationship be-tween phonological awareness and read-ing. Journal of Research in Reading, 27, 36--57.

Blonk, A M. T., & Bosman, A (2003, No-vember). Girls are better, even girls withdyslexia?! Gender differences in reading per-formance of teenagers with and withoutdyslexia. Paper presented at the Interna-tional Conference on Dyslexia, Univer-sity of Amsterdam, the Netherlands.

Braams, T., & Bosman, A M. T. (2000). Fo-nologische vaardigheden, geletterdheiden lees- en spellinginstructie. [Phonolog-ical skills, literacy, and reading andspelling instruction]. Tijdschrift voor 01'-thopedagogiek, 39, 199-211.

Carroll, J. M., & Snowling, M. J. (2004). Lan-guage and phonological skills in studentsat high risk of reading difficulties. Journalof Child Psychology and Psychiatry and Al-lied Disciplines, 45, 631-640.

Catts, H. w., Fey, M. E., Zhang, X., &Tomblin, J. B. (2001). Estimating the riskof future reading difficulties in kinder-garten students: A research-based modeland its clinical implementation. Lan-guage, Speech, and Hearing Services inSchools, 32, 38--50.

Coleman, J. M., & Dover, G. M. (1993). TheRISK Screening Test: Using kindergartenteachers' ratings to predict future place-ment in resource classrooms. ExceptionalChildren, 59, 468--477.

Cornwall, A (1992). The relationship ofphonological awareness, rapid naming,and verbal memory to severe reading andspelling disability. Journal of Learning Dis-abilities, 25, 532-538.

de Jong, P. F, & van der Leij, A (1999). Spe-cific contributions of phonological abili-ties to early reading acquisition: Resultsfrom a Dutch latent variable longitudinalstudy. Journal of Educational Psychology,91, 450--476.

Flynn, J. M., & Rahbar, M. H. (1994). Preva-lence of reading failure in boys comparedwith girls. Psychology in the Schools, 31,66-71.

Flynn, J. M., & Rahbar, M. H. (1998). Im-proving teacher prediction of students atrisk for reading failure. Psychology in theSchools, 35,163-172.

Gallagher, A., Frith, U., & Snow ling, M. J.(2000). Precursors of literacy delayamong children at genetic risk of dys-lexia. Journal of Child Psychology and Psy-chiatry and Allied Disciplines, 41,203-213.

Hammill, D. D. (2004). What we know

about correlates of reading. ExceptionalStudents, 70,453--469.

Hammill, D. D., Mather, N., Allen, E. A., &Roberts, R. (2002). Using semantics,grammar, phonology, and rapid namingtasks to predict word identification. Jour-nal of Learning Disabilities, 35, 121-136.

Irausquin, R. (1999). Nonword repetition test.Unpublished test materials, University ofNijmegen, the Netherlands.

Johnston, R. S., Anderson, M., & Holligan,C. (1996). Knowledge of the alphabet andexplicit awareness of phonemes in pre-readers: The nature of the relationship.Reading and Writing, 8, 217-234.

Kenny, D. T., & Chekaluk, E. (1993). Earlyreading performance: A comparison ofteacher-based and test-based assess-ments. Journal of Learning Disabilities, 26,227-236.

Kirby, J. R., Parrila, R. K., & Pfeiffer, S. L.(2003). Naming speed and phonologicalawareness as predictors of reading devel-opment. Journal of Educational Psychology,95, 453--464.

Lewis, B. A., Freebairn, L. A, & Taylor,H. G. (2000). Academic outcomes in stu-dents with histories of speech sound dis-orders. Journal of Communication Disor-ders, 33, 11-30.

Lyytinen, H., Aro, M., Eklund, K., Erskine,J., Guttorm, T., Laakso, M. L., et al. (2004).The development of children at familialrisk for dyslexia: Birth to early school age.Annals of Dyslexia, 54, 184-220.

Mantzicopoulos, P. Y, & Morrison, D.(1994). Early prediction of readingachievement: Exploring the relationshipof cognitive and noncognitive measuresto inaccurate classifications of at-risk sta-tus. Remedial and Special Education, 15,244-251.

Menyuk, P., Chesnick, M., Liebergott, J. w.,Korngold, B., D'Agostino, R., & Belanger,A. (1991). Predicting reading problems inat-risk students. Journal of Speech andHearing Research, 34, 893-903.

Meyer, M. S., Wood, F B., Hart, L. A, &Felton, R. H. (1998). Selective predictivevalue of rapid automatized naming inpoor readers. Journal of Learning Disabili-ties, 31, 106--117.

Mommers, M. J. c., Verhoeven, L., & vander Linden, S. (1979). Veilig leren lezen[Learning to read safely]. Tilburg, theNetherlands: Zwijsen.

Mommers, M. J. c., Verhoeven, L., & vander Linden, S. (1994). Veilig leren lezen[Learning to read safely] (2nd ed.).Tilburg, the Netherlands: Zwijsen.

Muter, v., & Snowling, M. (1998). Concur-rent and longitudinal predictors of read-ing: The role of metalinguistic and short-term memory skills. Reading ResearchQuarterly, 33, 320-337.

O'Connor, R E., & Jenkins, J. R (1999). Pre-diction of reading disabilities in kinder-garten and first grade. Scientific Studies ofReading, 3, 159-197.

Pennington, B.F., & Lefly, D. L. (2001). Earlyreading development in children at fam-ily risk for dyslexia. Child Development,72, 816-833.

Petersen, D. K. (2002). Prediction of poorand superior word reading. Working Pa-pers, 50, 17-32.

Satz, P.,& Fletcher, J. M. (1988). Early iden-tification of learning disabled children:An old problem revisited. Journal of Con-sulting and Clinical Psychology, 56, 824-829.

Scarborough, H. S. (1989). Prediction ofreading disability from familial and indi-vidual differences. Journal of EducationalPsychology, 81, 101-108.

Scarborough, H. S. (1990). Very early lan-guage deficits in dyslexic students. ChildDevelopment, 61, 1728-1743.

Scarborough, H. S. (1991). Antecedents toreading disability: Preschool languagedevelopment and literacy experiences ofchildren from dyslexic families. Readingand Writing, 3, 219-233.

Scarborough, H. S. (1998). Early identifica-tion of children at risk for reading dis-abilities: Phonological awareness andsome other promising predictors. In P.Accardo, A Capute, & B. Shapiro, Specific

VOLUME 39, NUMBER 6, NOVEMBER/DECEMBER 2006

reading disability: A view of the spectrum(pp. 75-119). Timonium, MD: York Press.

Schneider, w., & Naslund, J. c. (1993). Theimpact of early metalinguistic competen-cies and memory capacity on reading andspelling in elementary school: Results ofthe Munich Longitudinal Study on thegenesis of individual competencies. Euro-pean Journal of Psychology of Education, 8,273-287.

Snowling, M. J. (1981). Phonemic deficits indevelopmental dyslexia. Psychological Re-search, 43, 219-234.

Snowling, M., Goulandris, N., Bowlby, M.,& Howell, P. (1986). Segmentation andspeech perception in relation to readingskill: A developmental analysis. Journal ofExperimental Child Psychology, 41, 489-507.

Snow ling, M. J., Gallagher, A, & Frith, U.(2003). Family risk of dyslexia is continu-ous: Individual differences in the precur-sors of reading skill. Child Development,74,358-373.

Taylor, H. G., Anselmo, M., Foreman, A L.,Schatschneider, c., & Angelopoulos, J.(2000). Utility of kindergarten teacherjudgments in identifying early learningproblems. Journal of Learning Disabilities,33, 200-210.

Teisl, J. T., Mazzocco, M. M., & Myers, G. F.(2001). The utility of kindergarten teacherratings for predicting low academicachievement in first grade. Journal ofLearning Disabilities, 34, 286-293.

van den Bos, K P., Lutje Spelberg, H. c., &Eleveld, M. A. (2004). Voorspelling vanleessnelheid van leerlingen in Groep 3

571

met tests afgenomen in Groep 2 en 3.[Prediction of decoding speed in Grade 1with test assessment in kindergarten andGrade 1]. Tijdschrift voor Orthopedagogiek,43, 312-324.

van den Bos, K P. (2004). Test serieel benoe-men en woorden lezen [Test of serial nam-ing and word reading]. Unpublished testmaterials. Groningen: RUG.

van Doorn, F. (1996). Werkhoudingslijst[Task-assignment scale]. Amsterdam:Advies- en Begeleidingscentrum.

Verhoeven, L., & van Leeuwe, J. (2003). On-twikkeling van decodeervaardigheid inhet basisonderwijs [Development of de-coding skills in primary school]. Peda-gogische Studien, 80, 257-27l.

Wagner, R K, & Torgesen, J. K (1987). Thenature of phonological processing and itscausal role in the acquisition of readingskills. Psychological Bulletin, 101, 192-212.

Walsh, D. J., Price, G. G., & Gillingham,M. G. (1988). The critical but transitoryimportance of letter naming. Reading Re-search Quarterly, 23, 108-122.

Weiner, P. (1985). The value of follow-upstudies. Topics in Language Disorders,S,78-92.

Wesseling, R, & Reitsma, P. (2000). Thetransient role of explicit phonological re-coding for reading acquisition. Readingand Writing, 13, 313-336.

Wentink, H., & Verhoeven, L. (2001). Proto-col voor leesproblemen en dyslexie [Protocolfor reading difficulties and dyslexia]. Nij-megen, the Netherlands: Expertisecen-trum Nederlands.