Classification of Children With

Specific Language Impairment:

Longitudinal Considerations
Gina Conti-Ramsden
Nicola Botting
Centre for the Study of
Language Impairments
University of Manchester
Manchester, United Kingdom

This paper reports on the longitudinal results of a large project involving 242
seven-year-old children attending language units in England. Following our work
outlining 6 subgroups of children with language impairment (Conti-Ramsden,
Crutchley, & Botting, 1997), we examine the stability of the 6 subgroups of
children with specific language impairment already identified, using data
collected from the same children at age 8 years. The findings suggest there is
considerable stability in the patterns of difficulties delineated by the classification
system involving 6 subgroups. Poorer stability was evident in the classification of
the children across time with 45% of children moving across subgroups. The
membership stability of the proposed classification system was very similar to that
found when the children were classified into 3 subgroups following another wellknown system (Rapin, 1996). The findings are discussed with particular reference
to issues surrounding the classification of children with SLI.
KEY WORDS: SLI, classification, (longitudinal) stability

here has been much discussion both in the literature and in current practice regarding the heterogeneous nature of children with
specific language impairment and the usefulness of the term specific language impairment (SLI; Leonard, 1987; Leonard, 1991; Miller,
1996). On the one hand, the debate has centered on the specific or otherwise nature of the language problem these children have (Aram, 1991;
Johnston, 1991; Tomblin, 1991). On the other hand, authors have attempted to find subgroups of children with language impairment in an
attempt to address the heterogeneity issue. Rapin and Allen (1987) for
example, outlined six subgroups of children referred for speech and language problems on a clinical basis. Bishop (1994), Bishop and Edmundson
(1987), and Wilson and Risucci (1986) have also produced reports on the
differing groups of children who may all be described as having SLI. In
a previous paper, we also outlined six subgroups of children who were
attending language units using a hybrid technique involving (a) statistical clustering technique on standard assessment scores and (b) teacher
opinion (Conti-Ramsden, Crutchley, & Botting, 1997). Five of these
groups were closely related to those of Rapin and Allen (1987): Cluster 1
(lexical-syntactic deficit syndrome), Cluster 3 (verbal dyspraxia), Cluster 4 (phonologic programming deficit syndrome), Cluster 5 (phonological-syntactic deficit syndrome), and Cluster 6 (semantic-pragmatic deficit syndrome). It needs to be noted that the terminology used for the
cluster names in this paper is that of Rapin and Allen (1987).
Most of the research that has examined the different profiles of

Journal of Speech, Language, and Hearing Research Vol. 42 11951204 October

1999 American
Speech-Language-Hearing
Conti-Ramsden
& Botting:
Classification of ChildrenAssociation
With SLI
1092-4388/99/4205-1195

Downloaded From: http://jslhr.pubs.asha.org/ by a University of Queensland User on 07/17/2015

Terms of Use: http://pubs.asha.org/ss/rights_and_permissions.aspx

1195
1195

children with SLI has done so on a cross-sectional basis. That is, subgroups have been identified from a group
of children varying in age and for whom data are available at a single point in time. An exception to this is the
study by Bishop and Edmundson (1987) which examined the prediction of outcome based on standardized
tests at 4 years of age. Although there is some merit in
classifying children at one point in time (in that it may
provide a description of potential difficulties and may
be generalizable to other samples and ages), it is of interest to clinicians and researchers to know how reliable the classification system is over time. Given that
standardized tests are not perfectly reliable, some variation may just reflect random noise, that is, a particular child may have a different profile from Time A to
Time B not because of any real change in the child but
because of fluctuations in test scores from extraneous
reasons. Therefore, the issue of the stability of subgroup
membership over time is an important element in the
categorization process.
Consequently, the present study attempts to add to
our knowledge of subgroups of children with SLI by
studying a large cohort of 8-year-old children with SLI
for whom data were also available at 7 years of age
(Conti-Ramsden, Crutchley, & Botting, 1997). The investigation addressed the following questions:
1. What is the stability of the six subgroups of children in the original sample aged 7 years in terms of the
types of profiles of difficulties delineated by the classification system (stability of test score patterns)?
2. To what extent do children with SLI remain in
their original cluster grouping after 1 year, and to what
extent do they move across clusters (stability of childrens
classification)?
3. Is the membership stability of our six subgroups
of children similar to that of another well-known classification system (Rapin, 1996)?

Methods
Participants
A group of 242 children with specific language impairment (SLI) took part in the study. They were recruited from 118 language units attached to English
mainstream schools. Language units are languagebased classrooms for children with SLI. In order to be
placed in a language unit, children generally have to
fulfill a number of criteria. Most units in England require children to have statements of special educational
needs (or to be undergoing assessment for one) that
details their difficulties and the professional input they
require including intensive speech and language treatment. Specific criteria for entry varies from Educational
1196

Authority to Educational Authority across England. Unit

criteria generally require that the child fall within the
normal range on nonverbal cognitive measures; that,
although minor associated physical, emotional, or behavioral difficulties may be present, the language disorder must be the childs primary problem (usually established by discrepancy between the childs language
assessment and cognitive assessment); that the child
would find it difficult to cope in mainstream education
even with support and needs a full-time placement in a
structured small group setting; and that children for
whom English is a second language should demonstrate
the same language difficulties in any other language(s)
they speak. Unit provision is designed to prepare children with language impairments to return to mainstream schools. Thus, most children spend at least part
of their school week in mainstream classes.
It was established (via telephone inquiries to all
primary school-age language units) that throughout
England approximately 500 Year 2 children (6;6 to 8;0
years) were spending 50% or more of the school week in
such language units. The 242 participants who formed
the study cohort represented a randomized sample of
these children. Over 100 language units were visited
and roughly half the eligible children in each unit were
sampled. The cohort consisted of 186 boys and 56 girls
(girls forming 23.1% of the cohort). Twenty-six children
in the study were additionally exposed to a language (or
languages) other than English at home (for further details see Crutchley, Botting, & Conti-Ramsden, 1997).
These 242 children were studied and, based on teacher
opinion and results of standardized assessments, were
classified into six clusters or subgroups of language impairment (Conti-Ramsden, Crutchley, & Botting, 1997).
For the purposes of this investigation, the children were
seen again 1 year later for Time 2 of the project. At Time
2, 234 of the 242 children originally assessed continued
to participate (age range 7;5 to 8;9 years). The socioeconomic background of the children participating in the
study resembled the distribution found in the general
population. The indicator used was income per household which averaged approximately 18,000 per year in
1996/1997 in England (Office of National Statistics,
1998). In our sample, 53.1% of the participants came
from households earning below 18,000 per year and
46.9% of the participants came from households earning more than 18,000 per year. Furthermore, there were
no statistically significant differences across clusters in
family income at either stage of the study (Time 1 and
Time 2).

Measures
The test battery used for this study was intended to
reflect a range of language-based and educational skills,

Journal of Speech, Language, and Hearing Research Vol. 42 11951204 October 1999

Downloaded From: http://jslhr.pubs.asha.org/ by a University of Queensland User on 07/17/2015

Terms of Use: http://pubs.asha.org/ss/rights_and_permissions.aspx

using a selection of well known standardized assessments.

All standardized tests were administered at Time 1 (age
6;6 to 7;9 years) and at Time 2 (age 7;5 to 8;9 years).

Standardized Tests
In total, seven language assessments were administered to the children as part of this study, and these
are described fully in the report of the Time 1 results
(Conti-Ramsden, Crutchley, & Botting, 1997). Briefly,
the assessments were as follows: the Goldman Fristoe
Test of Articulation: Single Word Level (Goldman &
Fristoe, 1986), the British Ability Scales: Number Skills
(Elliot, 1983), the British Ability Scales: Naming Vocabulary (Elliot, 1983), the British Ability Scales: Word Reading (Elliot, 1983), the Illinois Test of Psycholinguistic
Ability: Grammatic Closure (Kirk, McCarthy, & Kirk,
1968), the Test for Reception of Grammar (Bishop, 1982),
and the Renfrew Bus Story (Renfrew, 1991).
In addition to these language and educational assessments, children completed the Ravens Coloured Matrices (Raven, 1986), a measure of nonverbal cognitive ability. A total of 233 children were able to complete the
nonverbal cognitive assessment; 206 children scored above
the 15.9 percentile (above 1 standard deviation), 24 children performed between the 2.5 and 15.9 percentile (between 1 and 2 standard deviations below the mean),
and only 3 children performed below the 2.5 percentile,
suggesting these 3 children had more global delays.

Analyses and Results

Stability of Test Score Patterns
In order to examine the stability of clusters over
time, all Time 2 data on the six tests used to originally
cluster the children at Time 1 were entered into a new
cluster analysis (with six clusters specified). This excluded the ITPA Grammatic Closure Test which was
not used in the cluster analysis at Time 1 because there
was a higher proportion of missing data than on the
other tests. The cluster analysis was done using the Kmeans cluster function of SPSS for Windows (see
Anderberg, 1973 for a discussion of cluster analysis).
The aim was to examine the degree of match between
the two independent cluster analyses. For Time 2, complete information on test data was available for 207 of
the 234 children participating in the study (27 children
had missing data on one or more tests making them
unsuitable candidates for clustering). Time 1 and Time
2 clusters are both shown in Figure 1. There appears to
be a good match between Time 1 and Time 2 clusters
for Subgroups 1 (lexical-syntactic deficit syndrome), 3
(verbal dyspraxia), 4 (phonologic programming deficit
syndrome), 5 (phonological-syntactic deficit syndrome),

and 6 (semantic-pragmatic deficit syndrome) and not a

good match for Cluster 2 (normal group). Tables 1 and
2 show the mean percentiles and 95% confidence intervals for Time 1 and Time 2 clusters. An examination of
the means and 95% confidence intervals for Time 1 and
Time 2 clusters revealed no statistically significant differences between 5/6 test means for Cluster 1 (lexicalsyntactic deficit syndrome) and Cluster 6 (semanticpragmatic deficit syndrome); 4/6 test means for Cluster
5 (phonological-syntactic deficit syndrome); and 3/6 test
means for Cluster 3 (verbal dyspraxia) and Cluster 4
(phonologic programming deficit syndrome). For Cluster 2 (normal group), there were significant differences
in the means for all 6 test means examined.

Stability of Childrens Classification

The movement of children between clusters is shown
in Table 3. Cluster data for Time 1 and Time 2 were
available for 201 children. Fifty-five percent (111/201)
of the children remained in the same clusters, and this
proportion is reflected in each cluster considered separately, kappa = 0.43, confidence intervals = 0.34 to 0.51.
Consequently, 45% of children moved clusters from Time
1 to Time 2, suggesting relative instability in the
childrens cluster membership across time.

Movement of Children Across Clusters:

Addressing Test-Retest Reliability
As emphasised previously, whereas cluster profiles
remained stable across time (i.e., the means of tests for
each cluster at Time 2 were not significantly different
from Time 1) a substantial number of children moved
cluster membership due to individual change in test
scores. It is important to address the question of whether
this fluidity of cluster membership might be the result
of poor test-retest reliability in the measures used. We
therefore examined the major shifts of group (those
moves made by 8 or more children) to determine the
degree of test score change. Moves in cluster membership are not usually caused by a change in performance
on all measures, but rather by a significant clinical shift
in one or two areas of language skill and secondarily by
lesser changes in other assessment scores. Thus we identified the tests in each movement group that appeared
to represent the most change. Table 4 below lists the
mean, minimum, and maximum changes in score on the
test(s) identified as the major cause of movement for
each cluster. These can be compared with the overall
cohort changes presented in Table 5.
As can be seen, the moves are substantial and represent real clinical changes in profile, especially compared
to whole-cohort changes. In most groups, even the most

Conti-Ramsden & Botting: Classification of Children With SLI

Downloaded From: http://jslhr.pubs.asha.org/ by a University of Queensland User on 07/17/2015

Terms of Use: http://pubs.asha.org/ss/rights_and_permissions.aspx

1197

Figure 1. Cluster Profiles for Time 1 and Time 2

conservative changes in score are quite large and are

all in one direction. This is not so clear for the children
moving from Cluster 6 (semantic-pragmatic deficit syndrome) to Cluster 2 (normal group). These data support the argument made previously that Cluster 2 children are difficult to include in a classification of language
impairment because their profiles appear normal on
average on the measures used in this study.
It is also important to note that the changes that
result in cluster membership shifts are not in the same
direction for each movement group. For example those
moving from Cluster 1 (lexical-syntactic deficit syndrome) to Cluster 5 (phonological-syntactic deficit syndrome) all show a marked decrease in articulation skill
for age, whereas those moving from Cluster 6 (semanticpragmatic deficit syndrome) to Cluster 2 (normal
group) all show an increase in articulation score. If score
1198

changes simply were due to test-retest reliability difficulties, we would expect firstly to see a random pattern
of change (in both directions for each subgroup). Or if
the test showed a systematic positive bias for scores to,
say, improve on second testing, then one might expect
that systematic rise in scores to be reflected across all
subgroups in the same increasing direction. As already
mentioned, neither of these patterns were the case in
data obtained from this study. Instead, strong directional
shifts were seen in different subgroups, some all increasing, others all decreasing. Secondly, the magnitude of
the changes observed should be relatively small, assuming that these tests are reliable to some extent, which
clinical experience and, in some cases, reported reliability coefficients (see below) suggest they are. In contrast,
the changes observed in this study were large. Thirdly,
each movement group showed at least one test where

Journal of Speech, Language, and Hearing Research Vol. 42 11951204 October 1999

Downloaded From: http://jslhr.pubs.asha.org/ by a University of Queensland User on 07/17/2015

Terms of Use: http://pubs.asha.org/ss/rights_and_permissions.aspx

Table 1. Cluster percentile means and 95% confidence intervals for Time 1.
Cluster
Group

Test for Reception

of Grammar

BAS Number
Skills

BAS Naming
Vocabulary

BAS Word
Reading

Goldman Fristoe
Test of Articulation

Renfrew
Bus Story

1 n = 52

15.6
11.8 to 19.5

9.9
7.3 to 12.6

32.4
25.8 to 39.0

14.8
11.3 to 18.3

78.9
74.2 to 83.6

15.2
11.7 to 18.7

2 n = 16

40.0
27.8 to 52.2

50.9
36.9 to 65.0

54.7
45.7 to 63.6

28.2
20.2 to 36.2

64.9
52.8 to 77.0

52.3
41.2 to 63.4

3 n = 29

15.1
10.3 to 19.9

12.7
7.4 to 18.1

62.7
57.5 to 67.8

11.3
6.3 to 16.3

14.1
8.9 to 19.2

14.6
8.2 to 20.9

4 n = 23

54.7
47.0 to 62.3

26.8
18.0 to 34.5

36.6
27.3 to 45.8

16.8
9.9 to 23.8

23.1
15.1 to 31.1

43.3
34.8 to 51.7

5 n = 84

9.4
7.2 to 11.5

8.6
6.2 to 10.9

11.7
9.6 to 13.8

11.2
8.3 to 14.2

19.0
15.7 to 22.3

8.7
6.9 to 10.5

6 n = 25

30.8
22.7 to 38.8

14.9
10.3 to 19.5

52.0
42.5 to 61.5

66.3
60.0 to 72.6

78.4
68.1 to 88.7

19.4
13.1 to 25.7

Note. All tests showed a significant variation of mean scores across clusters at p < 0.0001

Table 2. Cluster percentile means and 95% confidence intervals for Time 2 new cluster analysis (n = 207 children).
Cluster
Group

Test for Reception

of Grammar

BAS Number
Skills

BAS Naming
Vocabulary

BAS Word
Reading

Goldman Fristoe
Test of Articulation

Renfrew
Bus Story

1 n = 30

13.7
9.2 to 18.2

3.5
2.0 to 5.0

33.8
26.6 to 41.0

12.9
8.0 to 17.9

84.0
77.8 to 90.3

14.3
8.7 to 19.8

2 n = 25

58.0
48.9 to 67.1

19.3
12.5 to 26.1

75.8
69.3 to 82.5

39.4
29.1 to 35.3

87.5
79.5 to 95.6

39.4
30.5 to 48.3

3 n = 38

30.7
15.3 to 32.2

17.4
10.5 to 24.4

69.7
65.1 to 74.3

12.0
8.2 to 15.8

21.5
16.4 to 26.6

17.7
11.3 to 24.2

4 n = 19

44.2
33.4 to 54.8

21.8
13.8 to 29.7

26.0
18.4 to 33.6

12.9
8.2 to 17.5

54.8
41.0 to 68.7

55.1
45.7 to 64.6

5 n = 77

11.7
8.6 to 14.9

7.2
5.0 to 9.6

20.2
16.6 to 23.8

7.7
5.5 to 9.8

16.0
12.9 to 19.0

8.3
6.4 to 10.3

6 n = 18

26.8
16.5 to 37.0

14.7
8.6 to 20.9

53.2
39.8 to 66.5

68.7
60.2 to 77.3

60.3
46.0 to 74.6

18.8
10.8 to 26.

Note. All tests were significantly different from each other across clusters at p < 0.001.

Table 3. Movement of children across clusters from Time 1 to Time 2 (n = 201 children).
Time 2 clusters

Time 1
clusters

cluster 1
cluster 2
cluster 3
cluster 4
cluster 5
cluster 6

cluster 1

cluster 2

cluster 3

cluster 4

cluster 5

cluster 6

18

1
1
9
1

3
8

4
3
15
4
11

5
4

8
2

10
2
51
1

3
1

1
2
11

Conti-Ramsden & Botting: Classification of Children With SLI

Downloaded From: http://jslhr.pubs.asha.org/ by a University of Queensland User on 07/17/2015

Terms of Use: http://pubs.asha.org/ss/rights_and_permissions.aspx

1199

Table 4. Change of test score in movement groups.

Movement
group (n)

Test of
interest

Minimum
change

Maximum
change

1 to 5 (8)
3 to 5 (10)
5 to 1 (9)

GF-Articulation*
BAS Naming V*
GF-Articulation*
BAS Naming V
BAS Naming V*
GF-Articulation*
BAS Naming V
Bus Story
TROG

22 percentiles
6 percentiles
+23 percentiles
2 percentiles
+14 percentiles
+0 percentiles
3 percentiles
20 percentiles
13 percentiles

66 percentiles
37 percentiles
+91 percentiles
+57 percentiles
+67 percentiles
+59 percentiles
+46 percentiles
+62 percentiles
+65 percentiles

5 to 3 (11)
6 to 2 (8)

Mean
change (SD)
41 (13) percentiles
24 (12) percentiles
+55 (28) percentiles
+20 (19) percentiles
+45 (18) percentiles
+9 (21) percentiles
+18 (19) percentiles
+27 (26) percentiles
+20 (24) percentiles

Note. All children moved in same direction.

Table 5. Whole cohort changes on tests.

Test
GF-Articulation
BAS Naming Vocabulary
BAS Word Reading
BAS Number Skills
Bus Story
TROG

Minimum change
66 percentiles
60 percentiles
41 percentiles
48 percentiles
58 percentiles
40 percentiles

all children moved uniformly in a given direction

(marked * on the table), a pattern not predicted by
measurement error.
Unfortunately, not all the tests used report adequate
test-retest information (means and SD at both time
points) making it difficult to determine their reliability
boundaries. In addition, numbers of children moving
clusters were too small to allow statistical analysis
(which may have confirmed that the moves across subgroups were unlikely to be caused by random measurement error). However, it is highly unlikely that any standardized test would have such poor reliability that all
scores at Time 2 would show these patterns of change
from Time 1 by chance or through measurement error.
Reported test-retest reliability was available for the BAS
number skills test at 0.95. In addition, the GoldmanFristoe test and the BAS word reading test, which appear to be two major factors in cluster movement at this
age, report a 0.95 median agreement for errors in production of words and a 0.97 test-retest reliability coefficient respectively. It has been pointed out (Wilcox &
Morris, 1995) that at certain ages relatively small
changes in performance on tests such as the GoldmanFristoe may lead to large changes in percentile score
(i.e., improvement on 4 or 5 items may result in 30 to
40 percentile changes). However, this is the nature of
standardized tests. The percentile score represents agerelated information for the very reason that raw scores
can be misleading. It may be that, at a certain age, a
1200

Maximum change
+92 percentiles
+81 percentiles
+95 percentiles
+70 percentiles
+63 percentiles
+65 percentiles

Mean change (SD)

+3 (26) percentiles
+9 (24) percentiles
0.5 (12) percentiles
2 (12) percentiles
+3 (17) percentiles
+5 (18) percentiles

four-item change represents a significant clinical improvement, whereas, at another age, the same increase
in raw performance leads to little change in normalized
scores. Consequently, it is possible that relatively small
changes in articulation may have shifted children from
Cluster 1 (lexical-syntactic deficit syndrome) to Cluster
5 (phonological-syntactic deficit syndrome) and from
Cluster 6 (semantic-pragmatic deficit syndrome) to Cluster 2 (normal group). Nonetheless, it needs to be
pointed out that cluster analysis compares scores on all
the tests used for clustering, of which articulation is one,
albeit important, factor.
Thus, taking all the above evidence into consideration, it appears that the majority of movements of children across clusters were due to genuine clinical change
in language profile.

Membership Stability: Comparison

With Rapins (1996) System
The starting point for the comparison with Rapin
(1996) was theoretical in nature. It had been found that 5
of our 6 clusters matched very closely the subgroups proposed by Rapin and Allen (1987). In 1987, Rapin and Allen
proposed six subgroups of SLI: lexical-syntactic deficit
syndrome, verbal auditory agnosia, verbal dyspraxia,
phonological programming deficit syndrome, phonological-syntactic deficit syndrome, and semantic-pragmatic
deficit syndrome. One of our clusters, Cluster 2 (normal

Journal of Speech, Language, and Hearing Research Vol. 42 11951204 October 1999

Downloaded From: http://jslhr.pubs.asha.org/ by a University of Queensland User on 07/17/2015

Terms of Use: http://pubs.asha.org/ss/rights_and_permissions.aspx

group), had no match with Rapin and Allens categories

and was composed of children who appeared to be performing within the normal range (the childrens test
scores for language measures were above the 40th percentile and for word reading above the 28th percentile).
There were 16 children in Cluster 2 (normal group),
and these children were not included in the next step of
the analysis. In addition, Rapin and Allen (1987) had a
further subgroup in their categorization system, that
is, children with verbal auditory agnosia.
No such children were found in our cohort. Furthermore, it is also the case that auditory agnosia is no longer
considered SLI (e.g., it is usual for auditory agnosia to
be prevalent with seizure disorders). Thus, it is not surprising that we found no such children in our cohort of
children attending language units. Consequently, there
were 5 common subgroups of children (again, the terminology in parenthesis is that of Rapin and Allen): our
Cluster 1 (lexical-syntactic deficit syndrome), our Cluster 3 (verbal dyspraxia), our Cluster 4 (phonologic programming deficit syndrome), our Cluster 5 (phonological-syntactic deficit syndrome), and our Cluster 6
(semantic-pragmatic deficit syndrome). It is also important to point out that we are not in agreement with Rapin
and Allens labelling of Cluster 3 as verbal dyspraxia.
The conceptualization of verbal dyspraxia as a subgroup
of SLI is not currently accepted. Verbal dyspraxia is considered a motor speech disorder rather than a language
disorder, although expressive language scores may well
be affected. Our Cluster 3 children had good naming vocabulary but were poor at all other tests used in the study.
Furthermore, we find the labelling of different profiles
of SLI as deficit syndromes problematic and prefer to
describe the subgroups as different profiles of SLI.
Rapin (1996) furthered her classification work by
proposing that the subgroups of children identified could
be grouped together into three larger, clinical groupings.
These subgroups or bands of impairment involved expressive language disorders (composed of verbal dyspraxia subgroup and phonological programming deficit
syndrome; our Clusters 3 and 4), expressive-receptive
language disorders (composed of verbal auditory agnosia and phonologic-syntactic deficit syndrome; our Cluster 5 only), and higher order processing disorders (composed of lexical-syntactic deficit syndrome and
semantic-pragmatic deficit syndrome; our Clusters 1 and
6). Our five clusters were grouped into three broader
subgroups or bands of impairment, following Rapins
procedure. For this stage of the analysis, the following
terminology was used to refer to these bands of impairment: expressive SLI, expressive/receptive SLI, and complex SLI, partially following Rapin. Where the present
terminology differs is in the naming of the third band.
Rapin suggests the term higher order processing disorders for this group of children, whereas the present study

refers to these children as having complex SLI. The dissatisfaction with the term higher order processing disorder comes from the fact that none of the tests used in
this study examined higher order processes per se. Furthermore, there is no consensus in the literature that
the children in question have as a common underlying
problem difficulties in higher order processing (Bishop,
1997). Finally, this group of children was composed of
children who had lexical, syntactic, semantic, and pragmatic difficulties in the absence of phonological disorders, hence the choice of the term complex SLI to describe the extent of these childrens problems.

Stability of Rapins (1996) Banding: Test

Score Patterns and Childrens Classification
Table 6 presents band means and 95% confidence
intervals for Time 1 and Time 2. There appeared to be a
very good match between Time 1 and Time 2 profiles of
difficulties with no statistically significant differences
between 5/6 test means for expressive SLI and complex
SLI, and 4/6 test means for expressive/receptive SLI.
This suggests that the classification of children into
bands of impairment delineated similar profiles of difficulties at Time 1 and Time 2.
Table 7 presents the results for the stability of
childrens classification into the three bands of impairment. Sixty-five percent of children stayed within the
same band from Time 1 to Time 2, Kappa = 0.46, 95%
confidence intervals = 0.35 to 0.58. Examination of these
data showed almost complete overlap between the confidence intervals of this stability kappa (0.35 to 0.58),
and the stability kappa obtained for the classification
involving the six subgroups of children (0.34 to 0.51).
Overall, the above data suggest that no matter which
classification system is used, the stability levels obtained
are similar. Both systems delineate stable patterns of
difficulties across time and both systems are remarkably unstable in the classification of children with SLI
from one year to the next.

Discussion
Main Findings of the Study
The present investigation found considerable stability from Time 1 to Time 2 in the patterns of difficulties (as evidenced by test scores) delineated by the ContiRamsden, Crutchley, and Botting classification system
(1997). In contrast, subgroup membership stability was
poorer with 45% of the children moving across subgroups
from Time 1 to Time 2. Examination of the movement of
children across subgroups revealed that the moves represented substantial and real clinical changes in profile

Conti-Ramsden & Botting: Classification of Children With SLI

Downloaded From: http://jslhr.pubs.asha.org/ by a University of Queensland User on 07/17/2015

Terms of Use: http://pubs.asha.org/ss/rights_and_permissions.aspx

1201

Table 6. Band percentile means and 95% CI for Time1 and Time 2.
Test for Reception
of Grammar

BAS Number
Skills

BAS Naming
Vocabulary

BAS Word
Reading

Goldman Fristoe
Test of Articulation

Renfrew
Bus Story

E-SLI

32.6
25.7 to 39.5

18.9
13.8 to 24.1

51.1
45.1 to 57.1

13.7
9.7 to 17.8

18.1
13.5 to 22.6

27.3
20.9 to 33.6

ER-SLI

9.4
7.2 to 11.5

8.6
6.2 to 19.9

11.7
9.6 to 13.8

11.2
8.3 to 14.2

19.0
15.7 to 22.3

8.7
6.9 to 10.5

Complex-SLI

20.6
16.7 to 24.5

11.6
9.3 to 13.9

38.8
33.1 to 44.5

31.5
25.2 to 37.8

78.7
74.2 to 83.22

16.6
13.5 to 19.6

E-SLI

35.2
29.3 to 41.0

18.9
13.7 to 24.2

55.1
48.4 to 61.8

2.3
9.4 to 15.2

32.6
25.7 to 39.5

30.1
23.2 to 37.2

ER-SLI

11.7
8.6 to 14.9

7.3
5.0 to 9.6

20.2
16.6 to 23.8

7.7
5.5 to 9.8

16.0
12.9 to 19.0

8.3
6.4 to 10.3

Complex-SLI

18.6
13.7 to 23.5

7.7
4.9 to 10.6

41.1
34.1 to 48.1

33.9
24.9 to 42.9

75.2
68.0 to 23.5

15.9
11.5 to 20.4

Bands
Time 1

Time 2

that could not be explained solely in terms of measurement error.

In addition, the stability of the proposed classification system, both in terms of patterns of difficulties and
membership of a subgroup, was very similar to that
found when children were classified into three subgroups
following another well-known system (Rapin, 1996).
Both classification systems produced stable patterns of
difficulties across time and both systems were noticeably unstable in the classification of individual children
from Time 1 to Time 2. In this sense, the Rapin (1996)
classification system did not provide any added information that was not already available in the classification system proposed by Conti-Ramsden, Crutchley, and
Botting (1997). The Rapin (1996) classification system
is simply a less specific, broader classification system
into three subgroups for the clinical population of children with SLI.

The Stability of Language

Impairment Subtypes
The present investigation has demonstrated that
movement across subgroups of language impairment is
not simply a consequence of test error. The picture that
emerges is both interesting and complex: profiles of difficulties in language impairment are stable even though
individual children may be moving across subgroups.
Why might this be the case?
SLI is not a unitary, static condition but a dynamic
difficulty that evolves with developmental time (ContiRamsden & Adams, 1995; Miller, 1996). Thus, it is possible that, as signs of SLI change with time for individual
1202

children, the relations among those signs retain some

predictability resulting in those signs covarying in a limited number of ways. In the present investigation, a clustering technique was used. The clustering technique
assigns children to subgroups taking into consideration
the performance of the children on all the six measures
used in the study. Having said this, in the present investigation two signs or areas of difficulty appeared to
be particularly involved in individuals changing across
subgroups from one year to the next, that is, phonology
and vocabulary (see Table 4). When changes occurred
and children moved to a different subgroup, the profiles
of such children became similar to the profiles of children belonging to another subgroup. Consequently, although children moved across subgroups, patterns of
difficulties remained stable. For example, children who
moved from Cluster 5 (all areas of language affected) to
Cluster 3 (good naming vocabulary but poor on all other
tests used) were children who improved significantly in
their expressive vocabulary. These children no longer
had problems across all areas of language like other
Table 7. Cluster based classification of children into bands: stability
from Time 1 to Time 2.
Time 1

Time 2

Expressive

Expressive/
Receptive

Complex

Expressive

25

13

Expressive/
Receptive

12

51

11

31

Complex

Journal of Speech, Language, and Hearing Research Vol. 42 11951204 October 1999

Downloaded From: http://jslhr.pubs.asha.org/ by a University of Queensland User on 07/17/2015

Terms of Use: http://pubs.asha.org/ss/rights_and_permissions.aspx

children in Cluster 5. Now these children had problems

across all language areas except for expressive vocabulary. Interestingly, there was already a cluster of children who had exactly those characteristics (i.e., Cluster
3). To reiterate, what is interesting about these findings
is that changes in childrens linguistic profiles did not
produce a whole new set of subgroups, instead, childrens
profiles of difficulties continued to fall into a limited
number of patterns already observed at Time 1.
The exception to the above was Cluster 2children
who appeared to be performing within the normal range
and arguably should not be included in the classification of children with SLI. Nonetheless, Cluster 2 children may well be a group of children whose language
difficulties have resolved by the time they participated
in the study (age 7 at Time 1). This interpretation is
supported by the work of Bishop and Edmundson (1987).
These investigators found that approximately 37% of
children who were found to have SLI at the preschool
age of 4 years no longer presented with SLI at 5;6 years;
that is, they appeared to have recovered. Following this
line of argument, it is expected that a proportion of children with SLI may no longer present with SLI profiles
at later ages. In the present investigation the percentage of children with SLI who appeared to have recovered (Cluster 2 children) was 7% at Time 1 (7 years)
and 12 % at Time 2 (8 years).
Thus, patterns of language strengths and weaknesses change with developmental time and no doubt
with the influence of other important factors such as
intervention. What is interesting is that it appears that
there are only a limited number of profiles of language
strengths and weaknesses in children with SLI. In the
present study, five such patterns of strengths and weaknesses have been identified in children with SLI who
are 7 to 8 years of age (excluding Cluster 2, normal
group children, as discussed earlier). In the following
discussion, the five patterns of difficulties will be described without naming the clusters. As noted previously,
Rapin and Allens (1987) terminology is problematic. It
is thought premature to identify in a definite manner
the clusters found in this study with Rapin and Allens
(1987) terminology. With this in mind, the profiles of
the children in this study are described.
Cluster 1 children are children with SLI who have
difficulties with comprehension of grammar, word reading, and retelling a story in the context of good phonology
and adequate expressive vocabulary. Cluster 3 children
are children who have problems with comprehension of
grammar, word reading, phonology, and retelling a story
in the context of good expressive vocabulary. Cluster 4
children were very similar to Cluster 3 children except
that their scores were better across the tests used, and
their expressive vocabulary was not as good as that of

children in Cluster 3. Cluster 5 children performed

poorly on all the tests used and appeared to have difficulties across all areas of language. Finally, Cluster 6
children had difficulties retelling a story in the context
of good phonology, good expressive vocabulary, good word
reading, and adequate comprehension of grammar.
These childrens difficulties were mainly evident at the
discourse level when they were required to retell a story.
It also needs to be noted that, in any classification
exercise, the resulting subgroups are necessarily dependent on the measures used in the investigation and the
psychometric properties of the measurement instruments used. This is particularly relevant in two ways.
First, the specificity of measurement within certain language construct domains necessarily restricts our ability to test discrete constructs such as semantics, syntax, or pragmatics. For example, it is hard to measure
pragmatics, and current tests may collapse across many
dimensions of syntax. Second, detection of stability may
be obscured by psychometric properties of measurement.
For example, relatively small changes in the GoldmanFristoe Test can result, at certain ages, in large changes
in percentile performance. Nonetheless, it is also important to emphasise that the five identified patterns
or subgroups of SLI are based on a study that has a
number of strengths. The present investigation involved
a large sample of children (N = 242) which represented
a random sample of 50% of all 7-year-olds attending language units in England. Although there are advantages
to studies with methodologies that specify a priori and in
detail the criteria used for selecting subjects as belonging
(or not) to SLI, the reverse also has its own advantages.
In this investigation, a large cohort of children receiving
speech-language services in language units were identified and then the question what sorts of problems do these
children have? was asked. In addition, the study was
uniform in terms of age (7 years at Time 1 and 8 years
at Time 2). This is a major methodological advantage
because developmental effects and changes are not confounded in the classification system. Most previous studies (Miller, 1996; Rapin & Allen, 1987; Wilson & Risucci,
1986) have had mixed age groups; therefore developmental and age related changes were confounded.

Concluding Remarks
The results of the present investigation lend further weight to the argument that children with SLI fall
into distinct subgroups. The present investigation also
adds an important caveat to the argument. Although
profiles of language strengths and weaknesses in the
form of distinct subgroups appear to be stable over time,
individual childrens language strengths and weaknesses
change with time. Consequently, a large proportion of
children move to different subgroups resulting in poor
Conti-Ramsden & Botting: Classification of Children With SLI

Downloaded From: http://jslhr.pubs.asha.org/ by a University of Queensland User on 07/17/2015

Terms of Use: http://pubs.asha.org/ss/rights_and_permissions.aspx

1203

membership stability from one year to the next (for children aged 7 and 8 years). Thus, it appears that SLI is a
dynamic condition that changes with developmental
time and very likely with the influence of other important factors such as intervention. At present, we are in
the process of seeing these children again at the age of
11 years. With this longer time scale, it may be possible
for future research to begin to provide some information on patterns of change and development of different
subgroups of children with SLI.

Acknowledgments
The authors gratefully acknowledge the support of the
Nuffield Foundation for Grant AT251 [OD], Educational
Transitions of Language-Impaired Children. We would also
like to thank Alison Crutchley for help with data collection,
Dr. Brian Faragher for statistical advice, and the schools and
language units who gave their time and facilities for our
assessments.

Goldman, R., & Fristoe, M. (1986). Goldman-Fristoe Test

of Articulation. Circle Pines, MN: American Guidance
Association.
Johnston, J. R. (1991). The continuing relevance of cause: A
reply to Leonards specific language impairment as a
clinical category. Language, Speech, and Hearing Services
in Schools, 22, 7580.
Kirk, S. A., McCarthy, J. J., & Kirk, W. D. (1968). Illinois
test of psycholinguistic abilities. Chicago: University of
Illinois Press.
Leonard, L. B. (1987). Is specific language impairment a
useful construct? In S. Rosenberg (Ed.), Advances in
applied psycholinguistics: Disorders of first-language
development (vol. 1, pp. 139). New York: Cambridge
University Press.
Leonard, L. B. (1991). Specific language impairment as a
clinical category. Language, Speech, and Hearing Services
in Schools, 22, 6669.
Miller, J. F. (1996). The search for the phenotype of
disordered language performance. In M. Rice (Ed.),
Toward a genetics of language (pp. 297314). Mahwah,
NJ: Lawrence Erlbaum Associates.

References

Office for National Statistics. (1998). Family Spending

1996/1997. London: The Stationary Office.

Anderberg, M. R. (1973). Cluster analysis for applications.

New York: Academic Press.

Rapin, I. (1996). Developmental language disorders: A

clinical update. Journal of Child Psychology and Psychiatry, 37, 643656.

Aram, D. M. (1991). Comments on specific language

impairment as a clinical category. Language, Speech, and
Hearing Services in Schools, 22, 8488.
Bishop, D. V. M. (1982). Test for reception of grammar. Age
and Cognitive Performance Research Centre, Department
of Psychology, University of Manchester: Author.
Bishop, D. V. M. (1994). Is specific language impairment a
valid diagnostic category? Genetic and psycholinguistic
evidence. Philosophical Transactions of the Royal Society
Lond. B., 346, 105111.
Bishop, D. V. M. (1997). Uncommon understanding. Hove,
UK: Psychology Press.
Bishop, D. V. M., & Edmundson, A. (1987). Languageimpaired four-year-olds: Distinguishing transient from
persistent impairment. Journal of Speech and Hearing
Disorders, 52, 156173.
Conti-Ramsden, G., & Adams, C. (1995). Transitions from
the clinic to school: The changing picture of specific
language impaired children from pre-school to school age.
Journal of Clinical Speech and Language Studies, 5, 111.
Conti-Ramsden, G., Crutchley, A., & Botting, N. (1997).
The extent to which psychometric tests differentiate
subgroups of children with SLI. Journal of Speech,
Language, and Hearing Research, 40, 765777.
Crutchley, A., Botting, N., & Conti-Ramsden, G. (1997).
Bilingualism and language impairment in children
attending language units. European Journal of Disorders
of Communication, 32, 267276.
Elliot, C. D. (1983). British ability scales. Windsor, UK:
NFER-Nelson.

1204

Rapin, I., & Allen, D. A. (1987). Developmental dysphasia

and autism in preschool children: Characteristics and
subtypes. In J. Martin, P. Fletcher, P. Grunwell, & D. Hall
(Eds.), Proceedings of the first international symposium on
specific speech and language disorders in children (pp. 20
35). London: AFASIC.
Raven, J. C. (1986). Coloured progressive matrices. London:
H. K. Lewis & Co.
Renfrew, C. (1991). The bus story: A test of continuous
speech. Oxford, UK: Author.
Tomblin, B. J. (1991). Examining the cause of specific
language impairment. Language, Speech, and Hearing
Services in Schools, 22, 6974.
Wilcox, K. A., & Morris, S. R. (1995). Speech outcomes of
the language-focused curriculum. In M. L. Rice & K. A.
Wilcox (Eds.), Building a language-focused curriculum for
the pre-school classroom (pp. 171180). Baltimore, MD:
Paul Brookes Publishing.
Wilson, B., & Risucci, D. (1986). A model for clinicalquantitative classification. Generation 1: Application to
language-disordered pre-school children. Brain and
Language, 27, 281309.
Received October 24, 1997
Accepted February 17, 1999
Contact author: Gina Conti-Ramsden, Centre for the Study
of Language Impairments, Centre for Educational Needs,
School of Education, University of Manchester, Manchester M13 9PL, England.
Email: Gina.Conti-Ramsden@man.ac.uk

Journal of Speech, Language, and Hearing Research Vol. 42 11951204 October 1999

Downloaded From: http://jslhr.pubs.asha.org/ by a University of Queensland User on 07/17/2015

Terms of Use: http://pubs.asha.org/ss/rights_and_permissions.aspx