Characterizing the Growth Trajectories of Language-Impaired Children Between 7 and 11 Years of Age

Characterizing the Growth Trajectoriesof Language-Impaired ChildrenBetween 7 and 11 Years of Age

REPORT

Background: A number of different systems have been suggested for classifyinglanguage impairment in children but, to date, no one system has been widely accepted.Method: This paper outlines an alternative system looking for distinct patterns of changein receptive language skills across time, involving a secondary analysis of childrenidentified as having specific language impairment.Participants: The participants were 184 children age-assessed at 3 time points—7, 8,and 11 years of age.Results: The pattern of receptive language development is highly predictable. Thedominant pattern of growth is consistent with declining rates of growth over time for allchildren. The primary way in which the children differ is with respect to their initialseverity. The testing of the 2 classification systems revealed some statistically significantdifferences among the subtypes with regard to the shape of the growth rates, but theeffect sizes associated with these differences were very small. Thus, it is possible toconclude that beyond the dominant pattern of growth, some subtypes of languageimpairment at 7 years of age showed only subtle differences in receptive languagechange across time. The results are discussed in terms of the sample selection and theage of the children who were studied.

KEY WORDS: growth trajectories, language impairment, children

A common problem within both clinical and research work on com-munication disorders concerns the classification of individualsinto subtypes of communication disorder. This interest in clas-

sification, or taxonomy, can be found in all areas of science, whether it wasthe development of the periodic table initially constructed by Mendeleyevor the Linnean system for classifying life forms. In all cases, certainproperties of the natural entities of interest are used in the taxonomicsystem to serve as the basis of the organizational system and become im-portant aspects of the theoretical account. Ultimately, the nature andorganization of these categories, or taxa, according to these propertiesshould reveal important aspects of the phenomena to be explained.

There have been a number of attempts to generate classification sys-tems for developmental language impairment (Aram, Ekelman, &Nation,1984; Bishop & Rosenblum, 1987; Rapin, 1996; Wilson & Risucci, 1986;World Health Organization, 1992), but no one system is commonly used.Most recently, van der Lely and colleagues (van der Lely, 2003; van derLely, Rosen, &McClelland, 1998) proposed a subtype of primary languageimpairment involving just a grammatical deficit, although others havebeen unable to validate this (Bishop, Bright, James, Bishop, & van derLely, 2000). The only available replication of one of these taxonomies that

James LawQueen Margaret University, Edinburgh

J. Bruce TomblinXuyang ZhangUniversity of Iowa

Journal of Speech, Language, and Hearing Research • Vol. 51 • 739–749 • June 2008 • D American Speech-Language-Hearing Association1092-4388/08/5103-0739

739

we were able to detect involved the system developed byRapin and Allen (1987; Rapin, 1996). Conti-Ramsden andcolleagues used cluster analysis to identify six groups ofchildren with language impairment and mapped fiveof them onto Rapin and Allen’s (1987) classification sys-tem (Conti-Ramsden, Crutchley,&Botting, 1997). Thesesystems all share the assumption that the subtypes ofprimary language impairments are qualitatively dis-tinct in their pattern of performance. In contrast, Bishopand colleagues (Bishop, 1994; Bishop et al., 2000) havesuggested that although there may be a core inheriteddisorder, there may not be a set of distinct diagnosticentities but, rather, these may be arrayed in a multi-dimensional space. In concert with this viewpoint, ithas been argued—using data from a population-basedrather than a clinical sample—that rather than strictsubtypes, developmental language disorder may be bestdescribed as regions within at least a three-dimensionalspace containing individual differences with respect tolanguage abilities, including those that comprise lan-guage disorder (Tomblin, Zhang, Catts, Ellis Weismer,& Weiss, 2004). Two classes of children were identi-fied across language-impaired and typically developingchildren. The first of these classes had relatively poorsemantic/syntactic skills and relatively poor phonolog-ical processing/verbal short-termmemory but relativelygood social communication skills. The second class com-prised children with relatively weak social communica-tion skills. It has been suggested that there is probablya single dimension accounting for language with test-specific or language area–specific minor dimensions atdifferent ages (Tomblin & Zhang, 2006). Of particularsignificancewaswhat appears to be a change across timewith the single construct givingway to a two-dimensionalmodel, with distinct vocabulary/sentence use dimensionsemerging in the later years. Recently, Dollaghan (2004)directly tested whether specific language impairment(SLI) represented a categorically different form of lan-guage variation from normal forms or whether it wasa region along a dimension. In this study, a taxometricanalysis was performed on the language scores of 3-and 4-year-old children. In both groups, there was noevidence in support of SLI being a discrete variant ofprimary language impairment but, rather, SLI appearedto fall within the same dimensions as children withnormal levels of language.

One of the key challenges to the development of clas-sification systems for developmental language impair-ment has been their capacity to capture a stable subtypeor trait across time—effectively, their predictive validity.Conti-Ramsden and Botting (1999) evaluated the stabil-ity of their Rapin and Allen (1987) subtypes after 1 year.Although they found a similar set of subtypes amongtheir participants, there was considerable instabilityof membership in these subtypes over time. Stothard,

Snowling, Bishop, Chipchase, and Kaplan (1998) foundconsiderable variability in class membership between SLIand general delay in childrenwhowere followed from5.5to 15–16 years of age.

Part of the value of a clinical taxonomic system isits potential to reveal associated conditions that havedifferent but systematic relationships with the sub-types. Such evidence provides more support for the dis-tinctiveness of the subtypes at both a phenomenologicaland etiological level. There have now been a series ofwell-documented studies that have tracked languageover time and that have examined outcomes associatedwith developmental language impairment. The studiesto which reference is most commonly made are those inDunedin, New Zealand (Silva,Williams, &McGee, 1987);Walthamstow,UnitedKingdom (Richman, Stevenson, &Graham, 1982); Ottawa, Carleton, Canada (Beitchman,Wilson, Brownlie, Walters, & Lancee, 1996; Johnsonet al., 1999) and Iowa, United States (Tomblin et al.,1997). Taken together, these studies have demonstratedhigh levels of co-morbidity among language, literacy,and behavior in clinical populations, such that earlylanguage learning difficulties should best be regardedas a risk factor for subsequent educational underattain-ment and that there is also a high level of continuityacross both the primary school and the teenage years.This literature has also shown that outcomes vary withrespect to certain subtypes of early speech and languageimpairment. Beitchman and colleagues (1996) haveshown that both behavioral and academic outcomesare poorer for children with language problems and,more specifically, childrenwith receptive language prob-lems than for children with speech sound disordersonly. Botting and Conti-Ramsden (2000) found behav-ioral outcomes to be differentiated according to subtypesbased on modality such that children with receptive–expressive impairments didmore poorly than thosewithimpairments restricted to expressive language. Thesedifferential outcomes could be used as evidence that thesubtypes identified in these studies represent differentsubtypes of language or that poor outcomes are simplya function of severity.

Growth Trajectories and TaxonomiesIn recent years, researchers have begun to character-

ize language development and developmental languageimpairment across multiple time points. These methodsallow researchers to characterize the dynamic growthpattern or trajectory as a characteristic of the languageimpairment in its own right. Whether these can truly betermed growth curves depends on the number of datapoints available for analysis. The notion that these de-velopmental trajectories are important features in lan-guage impairment can be found in Karmiloff-Smith’s

740 Journal of Speech, Language, and Hearing Research • Vol. 51 • 739–749 • June 2008

(1998) admonition that the key to understanding devel-opmental disorders is in understanding development—that is, growth and change. Leonard (1998) exemplifiedthis kind of thinking when he suggested that there maybe a number of potential pathways for children identifiedwith language impairments.Normal languagedevelopmentis characterized as a linear function of time, although thereare some suggestions that, at least for some grammat-ical functions, performance levels out as children achievecompetence. Three possible patterns are proposed. Thefirst possibility is that children with language impairmentmay also be characterized as starting at the same point asthe typically developing children, with their scores diverg-ing down across time relative to children with typical de-velopment.This could be characterizedas thedeteriorationhypothesis. A second possibility is that these children de-velop at the same speed as typically developing childrenbut then plateau at a specific point, not developing anyfurther. This could be characterized as the plateau hy-pothesis. Both of these two possibilities would effectivelyrepresent a decrease in the children’s language scoresacross time. A third possibility, whichmight be termed thetracking hypothesis, not considered by Leonard, would bethat language-impaired children are simply slower to takeoff, but once started, their language development would beparallel to that of typically developing children. They donot get better or worse but tend to stay on the same tra-jectory. Rice and colleagues (Rice,Wexler,&Hershberger,1998) have provided data showing this kind of growthpattern with respect to the growth of specific grammat-ical forms and vocabulary in children with and withoutSLI. Similarly, Zhang and Tomblin (2004) have foundlittle heterogeneity in the growth of language in a largenumber of children who varied considerably with respectto language abilities at school entry. These findingswould suggest that there is very little heterogeneity inthe growth characteristics of children with language im-pairment and, furthermore, the growth in children withlanguage impairment, at least in the school years, is quitesimilar to that of typically developing children.

The aim of the present study was to look at thepatterns of language change in a group of children with

significant clinical levels of language disorder who par-ticipated in the original Conti-Ramsden ManchesterLanguage Study (Conti-Ramsden & Botting, 1999). Theemphasis in this article is on the children’s receptivelanguage skills rather than their expressive languageskills or their speech skills. Receptive languagemeasureswere used in this study because they were available forall children, at all time points examined, whereas expres-sive measures were available only for a subset of chil-dren at each time point. Two analyseswere performed inorder to address two questions: (a) Are the growth trajec-tories of these children heterogeneous? and (b) Do thegrowth trajectories of subgroups of these children differ?To address the first question, the growth characteristicsof each member of the group were characterized usinga single language measure across time. The measure inquestion was The Test of Reception of Grammar (TROG;Bishop, 1982), ameasure of comprehension that is widelyused in the clinical research literature. To address thesecond question, the trajectories of subgroups of thesechildren were compared using the same measure. Twosystems of subcategorizationwere used. The first systemcorresponded to a traditional expressive/receptive lan-guage impairment classificationsystemof the typeadoptedin the ICD-10 classification system (World Health Or-ganization [WHO], 1992), and the second system corre-sponded to the Rapin and Allen (1987) classificationsystem outlined previously (see Table 1). In both cases,the groups were operationalized in the first instance bythe authors of the original Conti-Ramsden and Bottingstudy (Conti-Ramsden & Botting, 1999).

MethodThe study was conceptualized as a secondary data

analysis of a relatively large cohort of children identifiedas having primary language impairment. The cohortin question, from the Conti-Ramsden Manchester Lan-guage Study, is the single largest clinically identifiedcohort of children with pronounced speech and languageimpairments.

Table 1. Group characteristics of Conti-Ramsden et al. (1997) clusters and Rapin and Allen (1987) subtypes.

Conti-Ramsden Cluster Description Rapin and Allen equivalent

1 Good articulation but other language skills are poor. Lexical–syntactic deficit syndrome2 Poor word reading. None3 General poor language but with good naming vocabulary. Verbal dyspraxia4 Poor articulation and phonology. Normal language. Phonological programming deficit syndrome5 Articulation is fair, but performance on all other tests is poor. Phonological syntactic deficit syndrome6 Good articulation and reasonable expressive

language but poor receptive language.Semantic–pragmatic deficit syndrome

Law et al.: Growth trajectories and language impairment 741

ParticipantsThe data set comprises longitudinal information

on a subset of a cohort of 242 children with SLI. At age7 years, these children were sampled from the popula-tion attending language units in England. These childrenwere reassessed at 8 and 11 years of age. The childrenhave been followed through to 14 years of age, but it is theearlier phase of their development that is the focus of thisanalysis. This cohort has already been extensively dis-cussed in the literature in terms of its initial character-istics (Conti-Ramsden&Botting, 1999), their outcomesat11 years of age (Conti-Ramsden, Botting, Simkin, &Knox, 2001), the extent to which it is possible to predictoutcomes (Botting, Faragher, Simkin, Knox, & Conti-Ramsden, 2001; Conti-Ramsden & Botting, 1999; Conti-Ramsden, Botting, & Faragher, 2001), and subgroupchildrenacross time (Conti-Ramsden,Crutchley,&Botting,1997). This article only reports results from 196 partic-ipants who had a TROG score on all three time points,and for some analyses, the focuswill be on those 184 par-ticipants for whom there were sufficient data to ap-ply the operationalization of the ICD-10 and both theRapin andAllen (1987) classification andConti-Ramsdenclusters.

Table 2 provides descriptive statistics concerningthe TROG scores. These data show that the compositemeasure systematically increases across the age inter-vals and has a similar variance at each age interval. Thetests used to operationalize the different categories ofspeech and language impairment were the TROG, threesubtests of the British Ability Scales (Number Skills,Naming Vocabulary, and Word Reading; Elliott, 1983),the Goldman–Fristoe Test of Articulation (Goldman &Fristoe, 1986), the Bus Story (Renfrew, 1991), andRaven’sColoured Progressive Matrices (Raven, 1986). The firstmethod of subclassification adopted by Conti-Ramsdenand Botting (1999) might be characterized as compara-ble to the ICD-10 system. In this case, four categorieswere identified—namely, those with normal languagescores (above the 40th percentile on the language mea-sures), those with expressive difficulties only, those with

expressive/receptive difficulties combined, anda categoryof children with “complex/receptive” difficulties. Thesecond, third, and fourth categories were comparable to,although not strictly derived from, the language impair-ment categories identified in the ICD-10. Group 1 wasmade up of children with language scores within the nor-mal range. This included children who had been identi-fied as having significant difficulties at Time 1 but whohad then been transferred to mainstream school andwhose difficulties had not beenpickedup on standardizedlanguage measures at Time 2 and Time 3. Group 2 hadcomprehension within the normal range but poor ex-pressive language skills.Group3hadboth expressive andreceptive language scores outside the normal range.Group 4 had poor expressive language scores but lowerreceptive language scores (i.e., below the 10th percentile).This last group was designated as having complex SLIbecause the children appeared to have relatively strongexpressive language skills.

In the second method of subclassification, the pro-cedure followedwas the same as that described in Conti-Ramsden et al. (1999—that is, the data set was analyzedusing cluster analysis and the results indicated that, inpart at least, the resulting categories reflected thoseidentified by Rapin, 1996). The starting point is thethree clinical groupings described in the preceding par-agraph, each of which is then broken down into twogroups. Thus, the expressive language disorder group isfurther broken down into verbal dyspraxia and phono-logical programming difficulties. The expressive SLIgrouping is broken down into verbal auditory agnosiaand phonological programming deficit syndrome. Thecomplex SLI group (or those in Rapin’s “higher order pro-cessing disorder”) is broken down into lexical–syntacticdeficit syndrome and semantic–pragmatic deficit syn-drome. The Conti-Ramsden Clusters 1–6 correspond tothe sixRapin groupings in the original analysis, althoughConti-Ramsden and colleagues identified a group whosedifficulties appeared to have resolved (language scoresabove the 40th percentile), and Rapin and Allen (1987)identified a group with verbal auditory agnosia, whichwas not detected in the Conti-Ramsden sample. In theoriginal analysis, psychometric data were supported byteacher report. The six clusters are provided in Table 1.

ResultsThe first question to be addressed concerned the lan-

guage growth characteristics of the group as awhole andthe extent to which there were differences in the mag-nitude of the growth among the children in this sample.Table 2 provides the descriptive statistics for the TROGscores for the total group at each of the three observationintervals. As expected, the central tendencies reflected

Table 2. Means, medians, and standard deviations of the ReceptiveLanguage score (The Test of Reception of Grammar [TROG]; Bishop,1982) at 7, 8, and 11 years of age.

Variable

Age

7 years 8 years 11 years

M 10.10 12.18 15.16Mdn 10 12 16SD 3.46 3.21 2.99n 196 196 196

742 Journal of Speech, Language, and Hearing Research • Vol. 51 • 739–749 • June 2008

by the means and medians increased across the lon-gitudinal study, and the variance in TROG reducedsomewhat with age. The linear increase in the scoreswas found to be significant, F(1, 367) = 587.39, p < .0001,when these data were analyzed usingmixed linearmodelwithin SAS.

The extent to which there were individual differ-ences can be seen, however, in Figure 1. This figurecontains the individual growth functions for every 10thindividual when the participants were ordered from thelowest initial score to the highest. Therefore, these 18 in-dividuals represent a stratified sampling of individualsaccording to the severity of their language problem atage 7 years. These individual growth patterns reveal adominant trend for the language skills of all individualsto increase. There are some variations in the manner ofthis growth that result in some trends being linear,whereas others appear to have an accelerating exponen-tial or decelerating logarithmic shape. The extent to

which these shapes are reliable, however, cannot be de-termined, as theymay simply reflectmeasurement errorat different time points. At least four time points wouldbe needed to detect nonlinear patterns.

Receptive Language Changefor the Expressive/Receptive Subtypesof the ICD-10 Classification System

Although the data did not allow us to test the extentto which individual patterns of language change dif-fered, the data did allow for tests of differential growthrate among predefined groups of children. Themean testperformance for each group and at each data point areprovided in Table 3.

In each case, the children followed the predictedincremental increase in language skill across time. Thegrowth trajectory for each group is then plotted in Fig-ure 2. Groups 1–3 always followed the same pattern, withGroup 1 performing better than Groups 2 and 3. Despitethe lower receptive language scores at age 7 years,Group4functioned close to Group 2 at 8 and 11 years of age.

To test whether these groups had similar growthcharacteristics, the language scores were submitted to amixedmodel within SAS. This analysismethod providesa means to perform multilevel analysis. When the dataare longitudinal such as the data in this study, this anal-ysis can be considered a growth curve analysis, in whichage was a repeated measure (random effect) and groupwas a between-subjects measure (fixed effect). Becauseonly three time points were available, only a linearmodelwith an intercept and linear slopewas used. This analysisshowed that across all groups, there was a significanteffect of age,F(1, 364) = 382.72,p< .0001, T2 = 0.16. There

Figure 1. Receptive language scores across three observationintervals for every 10th child sampled systematically from the lowestinitial score to the highest initial language level, thus providing arepresentation of the range of initial language ability.

Table 3. Means and standard deviations of the Receptive Languagescore (TROG) for the four language subgroups comparable to thosein ICD-10 at 7, 8, and 11 years based on clusters identified inConti-Ramsden, Crutchley, & Botting (1997).

Group

Age M (SD)

7 years 8 years 11 years

1 (N = 44) 13.66 (2.20) 14.71 (2.20) 17.74 (1.79)2 (N = 47) 12.32 (1.78) 13.49 (2.40) 15.43 (2.92)3 (N = 84) 7.44 (2.17) 10.33 (2.72) 14.06 (2.96)4 (N = 9) 7.44 (2.19) 12.67 (3.12) 15.78 (2.82)

Note. Group 1 = Language normal group; Group 2 = Expressivelanguage impairment group; Group 3 = Expressive/receptive languageimpairment group; Group 4 = Complex specific language impairmentgroup (poor expressive language skills but lower receptive language skills[below the 10th percentile]). ICD-10 = International Statistical Classificationof Diseases and Related Health Problems (10th rev.).

Law et al.: Growth trajectories and language impairment 743

was also a significant difference in overall language sta-tus (intercept) for the different groups, F(3, 180)= 74.09,p < .0001, T2 = 0.28. Finally, the Age × Group interactionwas also significant, F(3, 364) = 23.86, p < .0001, T2 =0.028. A follow-up test to examine the interaction showedthat of the six contrasts, Groups 1 and 2 were not sig-nificantly different in the growth rate andGroups 3 and 4were not significantly different. All other contrasts in thegrowth rate were significant.

Although the mixed model provided some evidenceof reliable differences between some of the subgroups,this fixed-model effect size also showed that these sig-nificant between-group differences resulted from verysmall effect sizes as reflected in the T2 values of 0.04.This estimate of T2 used a method described by Winer,Brown, and Michels (1991; pp. 406–407) and assumes afixed effect of analysis of variance (ANOVA), whereas thesignificance test was based on a mixed model. To date,there is no method available for computing effect sizesfor mixed models; however, the fixed-effect approxima-tion should be quite close. This constitutes a small effect(Cohen, 1988), accounting for only 4% of the variance.Thus, these findings underscore the relative homogene-ity of the growth characteristics of these subgroups. Thegroups did differ with respect to their overall severity asshown by the significant main effect and the large effectsize. Taken together, these results indicate that most of

the differences between the subgroups are found at thegeneral level of language proficiency and that, except fora small amount of the variation, the progress across timefor these four subgroups is similar.

Receptive Language Trajectoriesfor the Rapin and Allen (1987)Classification System

The mean scores for each of the six Rapin and Allen(1987) clusters are presented in Table 4, and the pat-terns of language growth for each of the clusters areshown in Figure 3. The pattern is similar to that de-tected in the earlier analysis in that each of the groupsfollow a similar incremental increase in language skillsacross time. A mixed-model ANOVA was again used totest whether the clusters differed with respect toreceptive language status across the three observationalintervals. This analysis showed that across all groups,there was a significant effect of age, F(1, 362) = 377.42,p < .0001, T2= 0.19. The clusters also differed with re-spect to their overall language status, F(5, 178) = 24.34,p < .0001, T2 = 0.19. Furthermore, the Age × Groupinteractionwas significant,F (5, 362) = 3.15, p < .01, T2 =0.006. A follow-up test to examine the interaction showedthat of the 15 contrasts, 3 were significant at an alphalevel of p < .01. Specifically, Cluster 4 had a growth ratesignificantly lower than Clusters 1, 3, and 5.

An inspection of the plots in Figure 3 shows thatthese differences are largely due to the negative growthfor Cluster 4 from age 7 to age 8 years. Cluster 4 rep-resents a group with normal language abilities but with

Figure 2. The average trajectories for Receptive Language skills acrossthe three age points for the language-normal group (Group 1),expressive language impairment group (Group 2), expressive/receptive language impairment group, (Group 3), and complexSLI group (Group 4).

Table 4. Mean (and standard deviation) of the Receptive Languagescore (TROG) for the Rapin and Allen (1987) classification systemat three age points based on clusters identified in Conti-Ramsdenet al. (1997).

Group

Age

7 years 8 years 11 years

1 (N = 44) 9.93 (3.25) 12.39 (2.54) 15.20 (2.76)2 (N = 14) 13.14 (3.06) 15.57 (1.45) 17.64 (1.15)3 (N = 21) 9.52 (2.54) 12.10 (2.83) 15.57 (3.03)4 (N = 17) 15.24 (1.52) 15.18 (2.74) 17.59 (1.84)5 (N = 71) 8.31 (2.68) 10.63 (2.83) 14.14 (1.83)6 (N = 17) 11.88 (2.71) 13.71 (3.00) 16.12 (3.31)

Note. Group 1 = Good articulation but other language skills poor;Group 2 = Poor word reading; Group 3 = General poor language butwith good naming vocabulary; Group 4 = Poor articulation and phonology,and normal language; Group 5 = Articulation is fair but performanceon all other tests is poor; Group 6 = Good articulation and reasonableexpressive language but poor receptive language.

744 Journal of Speech, Language, and Hearing Research • Vol. 51 • 739–749 • June 2008

phonological impairment. Thus, their performance ontheTROG is close to ceiling. This near-ceiling performancereduces the variance in this group and also the amount ofchange over time. The variance of themeasure plays a rolein determining whether or not the growth rate for onecluster is significantly different from that of other clus-ters. The differences between groups with small var-iances tend to be significant. This may explain why thegrowth rate for Cluster 4 is significantly lower than forClusters 1, 3, and 5 but not for Cluster 2, which has alarger variance for the measure at age 7 years. As wasfound for the earlier subgroups, this analysis revealedthat despite there being a significant difference in thedevelopment pattern of some of the clusters, the mag-nitude of these differences as reflected in the T2 value ofthe interaction was very small. In contrast, the maineffect between the groups was larger, again indicatingthat the principal dimension distinguishing the clusterswas the severity of their receptive language scores atinitial assessment.

The data from this study showed that there was adominant pattern of development consisting of a grad-ually declining rate of growth for these childrenwith clin-ical levels of language impairment. When this pattern

was fit to a linear model, 30.3% of the total variancewas accounted for by a single set of intercept and slopeparameters. Of the remaining variance, nearly 61.6%(100% – 30.3% = 69.7%) was due to individual differ-ences in intercepts, and the remaining 38.4% may beattributed to the individual difference in the growthrate and the variance unaccounted for by the linearmodel. The intercept reflects differences in severity atinitial assessment. Although systems concerned withsubtypes can include differences in severity, typicallythis form of individual difference (i.e., in intercept) isviewed as evidence of dimensional quantitative var-iation. The individual growth rate differences and theaccounted errors can be seen from the samples ofgrowth found in Figure 1. It is within this nearly 26.8%[i.e., 69.7% × (1.000 – 0.616)] of the variance that qual-itative differences in the pattern of growth that wouldcontribute to subgroups may be found. It might be pos-sible to assume that if there were no measurement er-ror, an ideal taxonomy with the correct growth functionthat would capture individual differences in growthpatterns would result in an effect size of this magni-tude. It is not possible, from these data, to extrapolatethis growth function to the population of languagelearners as a whole, although it is important to pointout that whether a child had a clinical level of languagedifficulty did not account for class membership in theTomblin et al. (2004) analysis. Thus, the pattern of de-velopment in language impairment may indeed reflectthat in the typically developing population.

DiscussionThis study examined the growth characteristics of

language in a large clinical sample of children who havebeen described in prior research (Conti-Ramsden &Botting, 1999; Conti-Ramsden et al., 2001). Existingwork by Conti-Ramsden and colleagues has providedevidence for subgroups based on profiles of languageperformance across language measures at each timepoint. Stability of group membership across observationintervalswas not found. These findings suggest that thissample of children with language impairment is hetero-geneous and this heterogeneity is captured, to somedegree, by categories similar to those initially identifiedby Rapin and Allen (1987). Given the potential hetero-geneity of this sample, considerable variability in growthpatterns was anticipated across time, and this variabilitywas expected to be associated with the diagnostic sub-groups found by Conti-Ramsden and colleagues. Indeed,these authors indicated that 45% of the children in thesample moved across subgroups between Time 1 andTime 2 (Conti-Ramsden & Botting, 1999), and this move-ment represented substantial changes in the children’s

Figure 3. The average trajectories for Receptive Language skillsacross the three age points for the profile of Conti-Ramsden et al.(1997) classification system containing the following six groups:(1) Good articulation but other language skills poor; (2) Poorword reading; (3) General poor language but with good namingvocabulary; (4) Poor articulation and phonology, and normallanguage; (5) Articulation is fair but performance on all other testsis poor; and (6) Good articulation and reasonable expressivelanguage but poor receptive language.

Law et al.: Growth trajectories and language impairment 745

profiles. Such heterogeneity would be particularly likelyif these subgroups had different etiologies that wouldhave different effects on the rate and pattern of languagegrowth.

It is within this context of the remaining 26.8% var-iance not accounted for by the dominant model that theRapin and Allen (1987) and ICD-10 (WHO, 1992) tax-onomic systems can be considered, as this would rep-resent the maximum amount of variance that thesesystems could account for beyond that having to do withseverity. Indeed, both systems were able to capture asignificant amount of this variance. The estimated effectsizes for these significant results, however, were quitesmall (.01 and .04, respectively). The failure to obtain astronger relationship between these systems of subtyp-ing and growth over time may be due to their poorreliability as shownbyConti-Ramsden et al. (2001). If so,onewould have to questionwhether these subgroups arereflective of substantial unique etiological effects. Thefailure to find strong evidence for systematic differencesin growth for these taxonomic systems could also be ex-plained by the fact that their differences are due tononlinear aspects of growth that were not incorporatedinto this analysis. An examination of Figures 3 and 4shows, however, that the shapes of the groupmeans acrosstime are quite similar both within and between the twotaxonomic systems. Therefore, it seemsunlikely that theuse of a linear model for analysis obscured importantgroup differences.

The substantial similarity in growth trajectories be-tween individuals and within groups formed by existingdiagnostic systems is similar to other studies that haverecently examined growth in school-aged children withand without specific language impairment. The find-ings of Rice and colleagues (Rice, 2004; Rice, Wexler, &Hershberger, 1998) andTomblin et al. (2004) suggest thatthe growth trajectories of measures of both vocabularyand grammar in children with SLI were similar to thoseof typically developing children. Indeed, the main dif-ferences between groups in both studies was their earlylanguage skills, not the pattern of their subsequent per-formance. As Rice (2004) noted, the acquisition of dif-ferent aspects of the language system appears to beexplained in the first instance by the timing of the onsetof acquisition, but subsequently, the growth trajectoriesfollow the same course in all children, although the exactpatternmay differ for different speech and language skills.In general, this would support the notion that these chil-dren are delayed but not qualitatively different fromtypically developing children, supporting a finding in theliterature that goes as far back as the work of Moreheadand Ingram (1976).

The conclusion from thepresent data is that, in termsof the models of change proposed by Leonard (1998) and

discussed in this article, the pattern appears to be similarfor the group as a whole. Although some individual chil-dren may appear to catch up and others appear to fallbehind, the overall pattern appears to be one of consis-tency. These data do not allow direct comparison with acomparable group of typically developing learners. Thosethat seem to have normal language scores at later stagesin the cohort could hardly be classified as typically de-veloping learners, given the fact that they had been re-ceiving special educational provision at the initial stagesof the original Conti-Ramsden Manchester LanguageStudy. However, data from the Iowa Child LanguageStudy (Tomblin&Zhang, 2006) suggest that this patternof slowing trajectory, arguably to asymptote, is charac-teristic of all children (Zhang&Tomblin, 2004). It is truethat these findings reflect performance on a specific setof standardized language measures, and there may bedifferent patterns for other skills that are not included.It is also true that the patterns of change are different forsome individuals but, across the group as a whole and atleast as far as receptive language skills are concerned, itappears that the tracking hypothesis better accounts forthe data for the receptive language skills of the group asa whole than does either the deterioration or plateauhypotheses. The same pattern might not be found forspeech skills or expressive language skills.

Methodological LimitationsThe information obtained from this study must be

understood within the context of the limitations inher-ent in the data set used. As noted initially, this studyrepresents a secondary analysis of data that had beencollected for a different purpose than that of growthcurve analysis. The value of this data set derives from itsextensive sampling of a clinically served population.This being said, certain features of this data set were notideal.

Limited repeated observations. To obtain effectivetrajectories, it is important to consider the number ofobservations and the intervals between the time points.Growth curve modeling involves the estimation ofparameters in the model. In the case of growth analysis,a minimum of three time points is needed to estimatethe intercept and slope. With three observation inter-vals, it is not possible to test more complex models thatcontain nonlinear effects or to test for heterogeneity ofgrowth parameters. It is also desirable that the intervalsoverwhich the observations aremade allow for sufficienttime for change to occur. Willett, Singer, and Martin(1998) talk about “sensibly spaced intervals.” In thepresent data sets, therewas only 1 year between the per-formance of the children between 7 and 8 years of age(i.e., the first two data points) in terms of both age and

746 Journal of Speech, Language, and Hearing Research • Vol. 51 • 739–749 • June 2008

ability. Data that spanned growth over a larger range ofdevelopment would have been desirable.

Unidimensional representation of language. In ad-dition to a limitation in the number of observations, thelanguage change data were restricted to receptive lan-guage skills. These skills are undoubtedly salient interms of the long-term outcome for children with lan-guage impairment; however, in general, other charac-teristics may be related to language change that play asignificant part and that could be monitored in the sameway. For example, it is possible that comparable ex-pressive language trajectories for grammar or vocabu-lary might be characterized by greater variability,although such an interpretation would be called intoquestion by the data from the Iowa Child LanguageStudy (Tomblin & Zhang, 2006). This analysis was notpossible from the available data.

An alternative to this approach would be a person-oriented approach. The person approach, as described byVon Eye and Bergman (2003), assumes that multipleand somewhat unique factors influence developmentand, therefore, the individual is also part of a systemandcannot reasonably be seen as being outside the context ofthat system. To accomplish such person-level analyses,it is necessary to have multiple measures of the traits ofinterest in order to perform cluster analysis with man-ifest variables or latent class analysis. The data avail-able for this study did allow for such analyses.

Attrition. It is important to comment on the attritionin the study in the process of identifying sufficient casesto plot the trajectories. In doing this, the analysis ef-fectively lost approximately 25% of the participants. Ofcourse, if this dropout was random, it need not be aproblem. However, this explanation is unlikely to be thecase, and it may be that this would affect the outcomes.For example, those children with the poorest languageskills might have been more likely to refuse specifictasks than children with less severe language impair-ments. Their removal from the analysis, consequently,would have reduced the number of a specific subset ofchildren, perhaps those with language difficulties whowere more closely associated with lower nonverbal IQscores.

ConclusionsThe present data support the contention that that

there is unlikely to be a set of diagnostic elements thatcan provide a mutually exclusive classification systemat any one given time (Bishop, 1994).Moreover, our resultssuggest that whether this is possible or not is probablythewrong question. Rather, it is necessary go beyond thenotion of disorders at single time points and see them as

being on temporal continua. The challenge is to identifypredictive models that can be replicated with sufficientstatistical power to sustain the confidence of researchersand clinicians. The data presented here represent a firststep in this process.

This analysis suggests that secondary analysis,which has not been a feature of the literature in thefield of speech-language pathology, offers considerablepotential for addressing salient research questions es-pecially by using different cohorts to address the sameissues. Given the consistency of the results present inthis article, it could be anticipated that these findingswould hold over other cohorts. However, it would beworth going back to those cohorts that are large enoughto provide comparable data, namely those in Dunedin,New Zealand, and Ottawa, Carleton, Canada. It wouldalso be appropriate to test for theoretically and clinicallyderived trajectories in the data to examine whetherthere are meaningful latent classes, which can be iden-tified and again replicated across cohorts and with dif-ferent composite measures. This will lead to furtherdiscussions of the dimensionality of language impair-ment, which can only be addressed in large-scale cohortstudies.

AcknowledgmentsWe would like to thank Gina Conti-Ramsden for allowing

access to the data for the purposes of these analyses. TheConti-Ramsden Manchester Language Study was funded byGrants AT251[OD] and Dir/28 from the Nuffield Foundationin the United Kingdom.

ReferencesAram, D., Ekelman, B., & Nation, J. (1984). Preschoolerswith language disorders: 10 years later. Journal of Speechand Hearing Research, 27, 232–244.

Beitchman, J. H., Wilson, B., Brownlie, E. B., Walters,H., & Lancee, W. (1996). Long term consistency in speech /language profiles. 1: Developmental and academic out-comes. Journal of American Academy of Child Psychiatry,35, 804–825.

Bishop, D. V. M. (1982). The Test of Reception of Grammar(2nd ed.). Manchester, United Kingdom: University ofManchester.

Bishop, D. V. M. (1994). Is specific language impairment avalid diagnostic category? Genetic and psycholinguisticevidence. Philosophical Transactions of the Royal Societyof London, 346, 105–111.

Bishop, D. V. M., Bright, P., James, C., Bishop, S. J., &van der Lely, H. K. J. (2000). Grammatical SLI: A distinctsubtype of developmental language impairment? AppliedPsycholinguistics, 21, 159–181.

Bishop, D. V. M., & Rosenblum, D. (1987). Classification ofchildhood language disorders. InW. Yule &M. Rutter (Eds.),

Law et al.: Growth trajectories and language impairment 747

Language development and disorders (pp. 16–42). Oxford,England: MacKeith.

Botting, N., & Conti-Ramsden, G. (2000). Social andbehavioral difficulties in children with language impair-ment. Child Language Teaching and Therapy, 16, 105–120.

Botting, N., Faragher, B., Simkin, Z., Knox, E., &Conti-Ramsden, G. (2001). Predicting pathways of specificlanguage impairment: What differentiates good and pooroutcomes. Journal of Child Psychology and Psychiatry, 42,1013–1020.

Cohen, J. (1988). Statistical power analysis for the behavioralsciences. Hillsdale, NJ: Erlbaum.

Conti-Ramsden, G., & Botting, N. (1999). Classification ofchildren with specific language impairment: Longitudinalconsiderations. Journal of Speech, Language, and HearingResearch, 42, 1195–1204.

Conti-Ramsden, G., Botting, N., & Faragher, B. (2001).Psycholinguistic markers for specific language impairment(SLI). Journal of Child Psychology and Psychiatry, 42,741–748.

Conti-Ramsden, G., Botting, N., Simkin, Z., & Knox, E.(2001). Follow-up of children attending infant languageunits: Outcomes at 11 years of age. International Journalof Language and Communication Disorders, 36, 207–219.

Conti-Ramsden, G., Crutchley, A., & Botting, N. (1997).The extent to which psychometric tests differentiate sub-groups of children with SLI. Journal of Speech, Language,and Hearing Research, 40, 765–777.

Dollaghan, C. A. (2004). Taxometric analyses of specificlanguage impairment in 3- and 4-year old children. Journalof Speech, Language, and Hearing Research, 47, 464–475.

Elliott, C. D. (1983). British Ability Scales. Windsor, UnitedKingdom: nferNelson.

Goldman, R., & Fristoe, M. (1986). Goldman–Fristoe Test ofArticulation. Circle Pines, MN: American Guidance Service.

Johnson, C. J., Beitchman, J. H., Young, A., Escobar, M.,Atkinson, L., Wilson, B., et al. (1999). Fourteen-yearfollow-up of children with and without speech/language im-pairments: Speech/language stability and outcomes. Journalof Speech, Language, and Hearing Research, 42, 744–761.

Karmiloff-Smith, A. (1998). Development itself is the keyto understanding developmental disorders. Trends in Cog-nitive Sciences, 2, 389–398.

Leonard, L. (1998). Children with specific language impair-ment. Cambridge, MA: MIT Press.

Morehead, D. M., & Ingram, D. (1976). The developmentof base syntax in normal and linguistically deviant children.In D. M. Morehead & A. E. Morehead (Eds.), Normal anddeficient child language (pp. 209–238). Baltimore: Univer-sity Park Press.

Rapin, I. (1996). Developmental language disorders: A clinicalupdate. Journal of Child Psychology and Psychiatry, 37,643–656.

Rapin, I., & Allen, D. A. (1987). Developmental dysphasiaand autism in preschool children: Characteristics andsubtypes. In J. Martin, P. Fletcher, R. Grunwell, & D. Hall(Eds.), Proceedings of the first international symposiumon specific speech and language disorders in children(pp. 20–35). London: Afasic.

Raven, J. C. (1986). Coloured Progressive Matrices. London:H. K. Lewis.

Renfrew, C. (1991). The Bus Story: A test of continuous speech.Oxford: Author.

Rice, M. L. (2004). Growth models of developmental languagedisorders. In M. L. Rice & S. F. Warren (Eds.), Develop-mental language disorders: From phenotypes to etiologies(pp. 207–240). Mahwah, NJ: Erlbaum.

Rice, M. L., Wexler, K., & Hershberger, S. (1998). Tenseover time: The longitudinal course of tense acquisitionin children with specific language impairment. Journalof Speech, Language, and Hearing Research, 41,1412–1431.

Richman, N., Stevenson, J., & Graham, P. J. (1982).Pre-school to school: A behavioral study. London: AcademicPress.

Silva, P. A., Williams, S., &McGee, R. (1987). A longitudinalstudy of children with developmental language delay atage three: Later intelligence, reading, and behaviour prob-lems. Developmental Medicine and Child Neurology, 29,630–640.

Stothard,S.E., Snowling,M.J.,Bishop,D.V.M.,Chipchase,B. B., & Kaplan, C. A. (1998). Language impaired pre-schoolers: A follow-up into adolescence. Journal of Speech,Language, and Hearing Disorders, 41, 407–418.

Tomblin, J. B., Records, N., Buckwalter, P., Zhang, X.,Smith, E., & O’Brien, M. (1997). Prevalence of specificlanguage impairment in kindergarten children. Journalof Speech, Language, and Hearing Research, 4, 1245–1269.

Tomblin, J. B., Zhang, X., Catts, H., Ellis Weismer, S. E., &Weiss, A. (2004). Dimensions of individual differencesin communication skills among primary grade children.In M. Rice (Ed.), Developmental language disorders:From phenotypes to etiologies (pp. 53–76). Mahwah,NJ: Erlbaum.

Tomblin, J. B., & Zhang, X. (2006). The dimensionality oflanguage ability in school-age children. Journal of Speech,Language, and Hearing Research, 49, 1193–1208.

van der Lely, H. (2003). Do heterogeneous SLI deficits needheterogeneous theories? SLI subgroups, G-SLI, and theRDDR hypothesis. In Y. Levy & J. Schaeffer (Eds.), Towardsa definition of specific language impairment (pp. 109–134).Mahwah, NJ: Erlbaum.

van der Lely, H., Rosen, S., & McClelland, A. (1998). Evi-dence for a grammar-specific deficit in children. CurrentBiology, 8, 1253–1258.

Von Eye, A., & Bergman, L. R. (2003). Research strategies indevelopmental psychopathology: Dimensional identity andthe person-orientated approach. Development and Psycho-pathology, 15, 553–580.

Willett, J. B., Singer, J. D., & Martin, N. C. (1998). Thedesign and analysis of longitudinal studies of developmentand psychopathology in context: Statistical models andmethodological recommendations. Development and Psy-chopathology, 10, 395–426.

Wilson, B. C., &Risucci, D. A. (1986). Amodel for the clinicalquantitative classification—Generation 1—Application tolanguage disordered preschool children. Brain and Lan-guage, 27, 281–309.

748 Journal of Speech, Language, and Hearing Research • Vol. 51 • 739–749 • June 2008

Winer, B. J., Brown, D. R., & Michels, K. M. (1991).Statistical principles in experimental design. New York:McGraw-Hill.

World Health Organization. (1992). International Statisti-cal Classification of Diseases and Related Health Problems(10th rev.). Geneva: Author.

Zhang, X., & Tomblin, J. B. (2004, June). Lack of individ-ual difference in the language growth rate from kindergartento eighth grade. Presentation at the Society for Researchin Child Language Disorders, Madison, Wisconsin.

Received April 7, 2006

Revision received March 29, 2007

Accepted October 21, 2007

DOI: 10.1044/1092-4388(2008/052)

Contact author: James Law, Department of Language andCommunication Science, Queen Margaret University,Queen Margaret University Drive, Edinburgh EH21 6UU,United Kingdom. E-mail: [email protected].

Law et al.: Growth trajectories and language impairment 749