The significance of nonverbal performance in children with developmental language disorder

ABSTRACT

Nonverbal deficits are frequently reported in children with developmental language disorder (DLD). In the new diagnostic criteria of DLD, the previous requirement of normal nonverbal performance has been removed and children with below average and even weak nonverbal skills now fit under the DLD definition. However, the significance of the nonverbal cognitive level, and the connection between nonverbal and verbal skills in these children diagnosed according to the new DLD classification is unclear. In the present study, the significance of nonverbal cognitive level on verbal performance was investigated among preschool-aged children with remarkable deficits in language development. Verbal skills were compared between average, below average, and weak nonverbal cognitive level groups. The connection between nonverbal and verbal skills was evaluated with Pearson correlations, and the covariance structure of the subtests used was modeled with Structural Modelling. The connection between nonverbal cognitive level and verbal skills was clear; weaker nonverbal cognitive levels were associated with lower verbal skills. While receptive language skills and verbal short-term-memory (STM) were the most profound weaknesses, relative strengths emerged for each nonverbal cognitive level group in fluid intelligence, especially in nonverbal reasoning tasks without time limits. In addition, fluid intelligence was strongly linked to verbal understanding and reasoning. These results suggest that the relative strength in nonverbal fluid intelligence with specific weaknesses in receptive language, verbal understanding, and verbal STM could be used as basic factors differentiating children with DLD from those with intellectual disability.

KEYWORDS:

• Developmental Language Disorder (DLD)
• fluid intelligence
• nonverbal performance
• WPPSI-III
• NEPSY-II

Introduction

Changes in the diagnostic criteria of Developmental Language Disorder (DLD)

The definition and diagnostic criteria of DLD, previously known as Specific Language Impairment (SLI), have been changed in the new diagnostic classifications. The focus of the DSM-5 (Diagnostic and Statistical Manual of Mental Disorders, DSM-5, American Psychiatric Association, 2013) and the ICD-11 (International Statistical Classification of Diseases and Related Health Problems, ICD-11, World Health Organization [WHO], 2018) DLD definitions is on the verbal criteria, namely language acquisition and the use of language across modalities. The previous requirement of normal nonverbal level is no longer considered relevant for the DLD diagnosis due to concerns of excluding those with developmental language problems combined with other neurocognitive deficits from speech and language therapy (Bishop, 2014; Messer & Dockrell, 2006; Norbury et al., 2016). The multinational and multidisciplinary Delphi Consensus Study panel CATALISE (Bishop et al., 2016) recommended DLD to be identified regardless of the level of nonverbal ability. The additional neurocognitive difficulties (e.g., attentional, social, or behavioral problems) should be documented and considered when planning intervention strategies (Bishop et al., 2016). However, when associated with a known biomedical condition, such as neurological disease, intellectual disability, or hearing loss, the diagnosis should be DLD associated with the co-occurring condition (Bishop et al., 2016).

Nonverbal performance in children with DLD

Another reason why normal nonverbal skills are not considered critical in the new DLD definitions is that children with DLD have on average significantly lower Performance Intelligence Quotient (PIQ) values than children with typical development (TD; see the meta-analysis of Gallinat & Spaulding, 2014). Nonverbal skills also appear to depend on the subtype of DLD (Reham & Hassnaa, 2020; Saar et al., 2018) and to occasionally progress to a significant decline in school-aged children originally diagnosed with DLD (Botting, 2005; Stothard et al., 1998). In our previous study (Saar et al., 2018), nonverbal cognitive performance of children aged 4 to 6 years diagnosed with DLD was significantly lower than the normal age level. In addition, children with more widespread and severe language comprehension difficulties had significantly lower nonverbal cognitive skills than those with mainly expressive language difficulties.

The deficits and possible decline of nonverbal performance in children with DLD have been explained by a complex and dynamic relationship of verbal and nonverbal development (Botting, 2005). On the other hand, Ullman and Pierpont (2005) proposed that disturbed development of brain structures and neural networks can cause disorders in both linguistic and non-linguistic functions. Similarly, Karmiloff-Smith (2007) suggested an early local disorder to cause atypical development in several parts of the brain and to interfere with specialization and localization of brain functions.

Though nonverbal impairments are evident in children with DLD, they may be inconsistent (DeThorne & Watkins, 2006) and do not concern all children. Conti-Ramsden et al. (2012) reported one third of children with DLD to have a deceleration of nonverbal development over time. Visuospatial memory deficits also concern a minority of children diagnosed with DLD (Archibald & Gathercole, 2006; Vugs et al., 2013).

In a retrospective study of children diagnosed with varying developmental delays at age 3 to 7 years, Liao et al. (2014) grouped the participants according to language skills (DLD, below average language development, and normal language development). The PIQ was significantly lower in the two groups with below normal language development than in the group with normal language development. However, in all three groups, nonverbal performance was significantly better than verbal performance.

The nature of the relationship between verbal and nonverbal cognitive skills is not clear in children with DLD or with more widespread cognitive deficits, such as mixed specific developmental disorder (ICD-10 definition for children with a mixture of specific developmental disorders of speech and language, motor function, and scholastic skills, but in which none predominates sufficiently to constitute the main diagnosis) or borderline intellectual functioning (BIF, ICD-10 definition for children with full IQ level between 71–84, (WHO, 2018). In the new diagnostic criteria, these diagnostic groups fit under the DLD definition. However, measures and analyses of verbal and nonverbal functions in children with DLD and more widespread neurocognitive developmental disorders are commonly based on assessments that focus on specific and often narrow aspects of cognitive functions, and the tests used may have different psychometric properties (DeThorne & Schaefer, 2004; Gallinat & Spaulding, 2014; Miller & Gilbert, 2008). Typically, while only PIQ is used to define the level of nonverbal performance, the subtest profiles are mostly not described.

Pulina et al. (2019) compared cognitive performance of children with BIF to children with TD. They found all WISC-IV indices of children with BIF to be significantly lower than those in children with TD. However, the index profile of children with BIF was uneven with specific weakness in Working Memory (WM) and highest scores in visuo-perceptual performance, regardless of any additional neurodevelopmental disorders such as ADHD or specific learning disorders. In addition to WM deficits shown in several previous studies (Alloway, 2010; Fernell & Gillberg, 2020; Gillam et al., 2021; Pulina et al., 2019; Santegoeds et al., 2021; Träff & Östergren, 2021) problems with executive functions (Alloway, 2010; Erostarbe-Pérez et al., 2022, Fernell & Gillberg, 2020; Träff & Östergren, 2021), processing speed (Santegoeds et al., 2021; Träff & Östergren, 2021), and adaptive skills (Greenspan, 2017; Pulina et al., 2019) are evident in children with BIF. In the longitudinal study of Träff and Östergren (2021), children with BIF received lower scores than children with TD in all cognitive and academic tasks but exhibited similar developmental growth and trends as children with TD, supporting a developmental delay model opposed to an atypical developmental model.

While cognitive skills can be expressed both verbally and nonverbally, robust correlations between verbal and nonverbal scores in children with DLD have been shown in some studies (Durant et al., 2019; Liao et al., 2014; Plym et al., 2021; Saar et al., 2018). To our knowledge, this has not been observed in children with mixed specific developmental disorders or BIF.

Aim of the study

While there is considerable evidence showing weaker nonverbal skills in children with DLD than in children with TD (e.g., Gallinat & Spaulding, 2014; Liao et al., 2014; McGregor et al., 2013; Stothard et al., 1998), nonverbal deficits do not concern all children with DLD (Archibald & Gathercole, 2006; Botting, 2005; Conti-Ramsden et al. (2012) and the DLD group is not homogeneous (Reham & Hassnaa, 2020; Saar et al., 2018). In addition, research on the cognitive performance of children with DLD has mainly been performed from narrow perspectives, and with composite scores (Gallinat & Spaulding, 2014). In the new diagnostic criteria of DLD, the previously used PIQ limit is no longer required and thus children previously diagnosed with more widespread cognitive deficits (such as mixed specific developmental problems or BIF) now fit in the DLD diagnostic category and were included in this study. To our knowledge, there are only a few studies where cognitive subtest profiles of these diagnostic groups have been evaluated (Durant et al., 2019; Peltopuro et al., 2014; Saar et al., 2018) and none where cognitive index or subtest profiles of these diagnostic groups have been directly compared.

The aim of the present study was to evaluate the significance of nonverbal performance level on cognitive subtest profiles and to examine the connections between nonverbal and verbal skills in children with DLD, now also including children with more widespread cognitive deficits. Based on previous findings in children with DLD (Gallinat & Spaulding, 2014; Liao et al., 2014; Saar et al., 2018), we assumed that children with more severe problems in nonverbal performance and reasoning skills would also exhibit more severe problems in verbal performance and reasoning. To study this, we first divided the participants into PIQ-level groups and evaluated the cognitive subtest profile differences between these groups. Secondly, we presumed a robust correlation between verbal and nonverbal performance in children with DLD. To further study the connection of verbal and nonverbal functions in children with DLD, a two-factor structure of the test variables was examined using Structural Methodology (SEM; Kline, 2015; Schumaker & Lomax, 2016).

Data and methods

Test materials and participants

The neuropsychological test materials used in this study were collected from the archived neuropsychological test records of 155 monolingual Finnish-speaking children (117 boys and 38 girls aged 4.0 to 6.9 years; 68 children were aged 4 years, 46 were aged 5 years old, and 20 were aged 6 years). Mean age was 5.3 years.

The participants of the present study had been admitted, assessed, and diagnosed with DLD or with mixed specific developmental disorders by a multidisciplinary team at Helsinki University Hospital Phoniatric Outpatient Clinic in 2010–2013 following the Finnish Current Care Guidelines based on the ICD-10 classification (World Health Organization [WHO], 2018). The exclusive PIQ limit of 70 is suggested for DLD diagnosis in the ICD-10. In the Finnish Current Care Guidelines, a diagnosis “mixed specific developmental disorder” is recommended for children who have DLD associated with PIQ values 70–84. In the present study, we included all children diagnosed with language problems (i.e., standard scores 1.5 SD below the age mean). However, the individual verbal subtest scores of these children varied considerably. Most of our participants had PIQ values ≥85 and received DLD diagnosis. Those with PIQ values <85 and some with PIQ values <70 but with significantly uneven nonverbal subtest profiles were diagnosed with mixed specific developmental disorders (Table 1).

Table 1. Diagnoses and number of participants in each PIQ group.

Diagnosis	PIQ>84	PIQ 70–84	PIQ<70	Total
Expressive Language Disorder	62	15	4	81
Receptive-Expressive Language Disorder	42	20	4	66
Mixed specific developmental disorder including language disorder	0	3	5	8
Combined	104	38	13	155

A phoniatric examination and assessments by a neuropsychologist and a speech-and-language therapist were performed over several appointments on different days. The tests used were chosen by these professional clinicians and were based on the age and symptoms of each child. Results of the logopedic and phoniatric evaluations that were part of the diagnostic process are not included here. Only the archived WPPSI-III (Wechsler Intelligence Scale for Children – Third edition; Finnish version, Wechsler, 2009) and NEPSY-II (Children’s Neuropsychological Research, Korkman et al., 2008) test records from the neuropsychological evaluation were used in the present study. Although the present study did not include data from an age-matched TD group, using subtest standard scores based on the normative data of the Finnish WPPSI-III and NEPSY-II tests allowed some comparison between children with DLD and TD. The data were partly used in our previous study (Saar et al., 2018). However, the present sample, research design, and analyses have not been published before.

The present data contains values of 155 subjects, in 16 test variables for all WPPSI-III subtests, and the NEPSY-II subtests Comprehension of Instructions and Sentence Repetition. We chose to use standard scores of the subtests instead of the raw scores due to the lack of children with TD in the present study and to be able to compare the scores of the children with DLD to the normative scores in the Finnish version of the WPPSI-III test. The material included missing values (21.1%) as the subtests administered to each participant were chosen purely for clinical purposes. Missingness cumulated in three WISC-III subtests (Comprehension, Similarities, and Symbol Search). Three subjects had more than 50% of their scores missing. The median missingness was 25%. Since Little’s test (Little & Rubin, 2002; Little, 1988) did not discard the hypothesis of CMAR (Completely Missing At Random; χ² = 925.27, df = 886, p = 0.175), all 16 variables and 155 subjects were retained in the analyses. Missingness was addressed with Multiple Imputation (MI) using Bayesian technique (Asparouhov & Muthén, 2020; Rubin, 1987; Schafer, 1997). The imputed data were used in all analyses.

PIQ level groups

To evaluate the impact of nonverbal performance level on verbal skills, we first divided the participants into the following three groups based on their PIQ values that were calculated following WPPSI-III instructions using nonverbal subtests Block Design (estimates the ability to analyze and synthesize visual stimuli with two-colored blocks), Matrix Reasoning (estimates analogical and serial reasoning), and Picture Concepts (estimates abstract categorical reasoning): Group 1 (PIQ >84), Group 2 (PIQ 70–84) corresponding to the Finnish Current Care Guidelines PIQ limits separating DLD from mixed specific developmental disorders, and Group 3 (PIQ <70) corresponding to the ICD-10 PIQ limit separating DLD from more widespread developmental delays.

Group sizes of DLD children with average (Group 1) compared to below average (Group 2) and weak (Group 3) nonverbal performance were not balanced (see Table 1), which in statistical analyses across all DLD groups caused a bias toward the average nonverbal performance Group.

Statistical methods

Within each PIQ group, mean values for all WPPSI-III verbal subtests were calculated. These were Information (estimates general knowledge and memory), Vocabulary (estimates word knowledge, concept formation and verbal fluency), Word Reasoning (estimates verbal abstract reasoning and concept formation), Comprehension (estimates social knowledge and verbal reasoning), Similarities (estimates logical thinking and verbal reasoning), and for General Language Index (GLI) subtests Receptive Vocabulary (estimates the ability to find correct responses to spoken words) and Picture Naming (estimates expressive language and word finding skills). The mean values were also calculated for the two NEPSY-II verbal subtests Comprehension of Instructions (estimates the ability to receive, process and give a correct response to verbal instructions) and Sentence Repetition (estimates verbal STM). In addition to the subtests included in the PIQ index, mean values were calculated for nonverbal subtests Picture Completion (estimates perceptual reasoning), Object Assembly (estimates visual-perceptual organization and part-whole relationship), and for the two Processing Speed index subtests Coding (measures speed and accuracy on paper and pencil task) and Symbol Search (estimates perception, recognition, and speed). The significance of the differences between the three PIQ groups were estimated for each verbal and nonverbal subtest using One-way ANOVA. η2 was used as a scale-free measure of the effect size in these analyses.

Further, to evaluate connections between nonverbal and verbal functions, we calculated a Pearson correlation between PIQ and VIQ indices. VIQ was determined following WPPSI-III instructions using verbal subtests Information, Vocabulary, and Word Reasoning. In addition, Pearson correlations between verbal subtests and PIQ and between nonverbal subtests and VIQ were estimated.

Finally, to analyze the similarity of covariances in the different PIQ-levels, to evaluate the measurement invariance (MI), and to summarize and elaborate the correlational results in a two-factor two-group model, we approached the test covariance matrix (16 x 16) using (SEM) techniques in the form of Multi-Group Confirmatory Factor Analysis (Kline, 2015; Schumaker & Lomax, 2016). The main advantages of using SEM techniques over traditional, exploratory multivariate techniques are explicit modeling of measurement errors; estimation of latent (unobserved) variables via observed variables; and comprehensive model testing where a structure can be imposed and assessed as to fit of the data (Smelser & Baltes, 2001). We assumed the 16 subtests to form two measurement models (i.e., latent P [nonverbal performance] and V [verbal] factors), which were free to correlate.

These two measurement models were subjected to SEM congeneric factor analysis to examine the unidimensionality hypothesis. Then the latent P and V factors were analyzed separately and together under progressively demanding constraints (factor loadings, intercepts, error correlations). The nested SEM models (configural, metric, and scale MI) were performed applying multiple imputed (10 times) two-group procedures. This is a standard procedure in modern evaluation of psychological measurements (Millsap, 2011; Van de Schoot et al., 2012, Xu & Traceya, 2017). In this analysis, PIQ groups 2 and 3 were merged to PIQ2 + 3 group due to statistical reasons (see Table 1).

All SEM analyses were performed using Mplus 8 program with multiply imputed matrices. The parameters of the model were estimated using maximum likelihood (ML) estimation (Muthen & Muthen, 1998–2017).

Results

Effects of the PIQ level on cognitive performance in children with DLD

While the PIQ groups were based on the level of their nonverbal skills, all verbal and Processing Speed subtest values were systematically highest for PIQ group 1 and lowest for PIQ group 3 (see Figure 1 and Tables 2 and 3). However, the cognitive subtest profiles of the PIQ–level groups had parallel strengths and weaknesses. Figure 1 also shows the cognitive subtest profile of the combined DLD group, which closely matches those of PIQ groups 1 and 2. Similar weaknesses in the subtest profiles in all PIQ–level groups and the combined DLD group were seen in subtests Word Reasoning, Comprehension, and Sentence Repetition, with relative strengths in nonverbal subtests Matrix Reasoning and Picture Concepts and in Processing Speed subtest.

Figure 1. Disclosure provided: Verbal and nonverbal subtest profiles in PIQ groups 1 (PIQ>84), 2 (PIQ=70–84), and 3 (PIQ<70). Verbal subtests: IN=Information, VC=Vocabulary, WR=Word Reasoning, CO=Comprehension, SI=Similarities, CI=Comprehension of Instructions, and SR=Sentence Repetition. Nonverbal subtest: BD=Block Design, MR=Matrix Reasoning, PCn=Picture Concepts, PCm=Picture Completion, OA=Object Assembly, CD=Coding, and SS=Symbol Search.

Table 2. Verbal subtest mean values in the three PIQ level groups and the total DLD group.

	PIQ group1 N=104	PIQ group 2 N=38	PIQ group 3 N=13	Differences between PIQ groups	Combined DLD group N=155
Sub- test	M (SD) Range	M (SD) Range	M (SD) Range	F (p) η2	M (SD)
IN	5.80 (3.24) 1–15	4.68 (2.91) 1–10	2.08 (1.55) 1–5	9.29 (.001).109 (medium)	5.21 (3.22)
VC	6.64 (2.15) 3–14	6.26 (1.88) 3–12	4.15 (1.34) 2–6	8.75 (.001).103 (medium)	6.34 (2.13)
WR	4.95 (3.37) 1–14	3.59 (3.12) 1–13	1.98 (1.66) 1–4	6.47 (.003).078 (medium)	4.36 (3.32)
CO	4.79 (3.07) 1–9	4.37 (3.11) 1–11	2.18 (1.94) 1–2	4.70 (.030).058 (small)	4.47 (3.08)
SI	7.01 (3.01) 1–11	5.52 (2.86) 1–10	2.93 (2.12) 1–2	13.62 (.001).150 (large)	6.30 (3.15)
RV	6.99 (3.25) 1–14	5.19 (3.46) 1–14	2.50 (2.08) 1–6	13.56 (.001).151 (large)	6.17 (3.48)
PN	6.40 (3.58) 1–16	5.60 (3.55) 1–12	4.68 (3.18) 1–12	1.87 (.174).024 (small)	6.06 (3.56)
CI	5.80 (3.01) 1–13	4.22 (2.51) 1–10	3.41 (2.32) 1–7	7.23 (.001).087 (medium)	5.22 (2.96)
SR	4.87 (3.01) 1–12	4.15 (3.16) 1–11	2.37 (2.25) 1–4	4.63 (.026).057 (small)	4.48 (3.06)

M (mean), SD (standard deviation), N (sample size), range: minimum and maximum scores. Differences between PIQ group variances by F ratio (p) and effect size by η2. Subtests: IN=Information, VC=Vocabulary. WR=Word Reasoning, CO=Comprehension, SI=Similarities, RV=Receptive Vocabulary, PN=Picture Naming, CI=Comprehension of Instructions, SR=Sentence Repetition.

Table 3. Nonverbal subtest mean values in the three PIQ level groups and the total DLD group.

	PIQ group 1 N=104	PIQ group 2 N=38	PIQ group 3 N=13		Combined DLD group N=155
Sub- test	M (SD) Range	M (SD) Range	M (SD) Range	F (p) η2	M (SD)
BD	9.45 (2.37) 5–18	5.34 (2.30) 1–9	2.85 (2.15) 1–8	75.98 (.001).500 (large)	7.89 (3.29)
MR	10.07 2.54) 5–18	6.87 (2.20) 1–12	4.12 (2.24) 1–7	50.08 (.001).397 (large)	8.79 (3.12)
PCn	9.96 (2.14) 4–14	7.11 (2.13) 3–-12	5.13 (1.53) 4–9	48.48 (.001).389 (large)	8.86 (2.66)
PCm	7.75 (3.39) 1–13	5.69 (3.34) 1–11	3.55 (2.78) 1–7	12.67 (0.001).142 (large)	6.89 (3.59)
OA	8.23 (3.44) 1–14	6.18 (3.53) 1–15	2.81 (2.65) 1–5	18.15 (.001).190 (large)	7.28 (3.76)
CD	8.70 (3.25) 4–19	6.25 (2.25) 2–10	4.51 (2.69) 2–10	17.83 (0.001).190 (large)	7.75 (3.31)
SS	8.48 (2.88) 4–15	6.86 (2.80) 4–13	4.50 (2.37) 3–5	15.55 (.004).163 (large)	7.75 (3.09)

M (mean), SD (standard deviation), N (sample size), range: minimum and maximum scores. Differences between PIQ group variances by F ratio (p) and effect size by η2. Subtests: BD=Block Design, MR=Matrix Reasoning, PCon= Picture Concepts, PCom=Picture Completion, OA=Object Assembly, CD=Coding, SS=Symbol Search.

For the verbal subtests, the effect of PIQ level was large on Similarities (F(2,154) = 13.62, p < .001 ; η2 = .150) and Receptive Vocabulary (F(2,154) = 13.56, p < .001 ; η2 = .151); medium on Information (F(2,154) = 9.29, p < .001 ; η2 = .109), Vocabulary (F(2,154) = 8.75, p < .001 ; η2 = .103), Word Reasoning (F(2,154) = 6.470, p = .003 ; η2 = .078), and Comprehension of Instructions (F(2,154) = 7.23, p < .001 ; η2 = .087); and small on Comprehension (F(2,154) = 4.70, p = .030 ; η2 = .058), Picture Naming (F(2,154) = 1.87, p = .174 ; η2 = .024), and Sentence Repetition (F(2,154) = 4.63, p = .026 ; η2 = .057). Bonferroni post hoc tests revealed that all verbal subtests except Picture Naming differed significantly between PIQ groups 1 and 3. However, significant differences between PIQ groups 1 and 2 were only found in subtests Comprehension of Instructions and Receptive Vocabulary. In addition, significant differences between PIQ groups 2 and 3 occurred in subtests Information, Vocabulary, Similarities, and Receptive Vocabulary.

The effect of PIQ level was naturally significant and large on the following three PIQ index subtests: Block Design (F(2,154) = 75.98, p < .001 ; η2 = .500), Matrix Reasoning (F(2, 154) = 50.08, p < .001 ; η2 = .397), and Picture Concepts (F(2,154) = 48.48, p < .001 ; η2 = 0.389). However, a significant and large effect was also seen on Picture Completion (F(2,85) = 12.67, p < .001 ; η2 = .142), Object Assembly (F(2,154) = 18.15, p < .001 ; η2 = .142), and the two Processing Speed subtests Coding (F(2,154) = 17.83, p < .001 ; η2 = .190) and Symbol Search (F(2, 154) = 15.55, p = .004 ; η2 = .163). Bonferroni post hoc tests revealed that while the differences between PIQ groups 1 and 3 were significant in all nonverbal subtests, the difference between PIQ groups 1 and 2 was not significant in subtests Object Assembly and Symbol Search. The difference between PIQ groups 2 and 3 was not significant in Picture Completion and the two Processing Speed subtests.

Correlations between verbal and nonverbal indices and subtests

The correlation between PIQ and VIQ indices in the combined DLD group (see Figure 2) was significant and moderate (r = .368, p < .001 , n = 155). Verbal subtests tended to have positively skewed distributions while nonverbal subtests had fairly symmetric distributions. However, the scatterplot matrix with linear and nonlinear Locally Weighted Scatterplot Smoothing (lowess) lines showed that acceptable linearity was maintained in the correlations.

Figure 2. Scatter plot between VIQ and PIQ in three PIQ level groups.

All the following verbal subtests had significant correlations to PIQ: Receptive Vocabulary (r = .416, p < .001), Comprehension of Instructions (r = .386, p < .001), Similarities (r = .378, p < .001), Information (r = 0.341, p < .001), Word Reasoning (r = .338, p < .001), Vocabulary (r = .297, p < .001), Sentence Repetition (r = .253, p = .003), Comprehension (r = .229, p = .030), and Picture Naming (r = .208, p = 0.012). In addition, the correlation between VIQ and the following other nonverbal subtests except Block Design (r = .126, p = .121) was significant: Symbol Search (r = .437, p = .005), Picture Concepts (r = .413, p < .001), Coding (r = 0.357, p < .001), Picture Completion (r = .333, p < .001), Object Assembly (r = .270, p = .016), and Matrix Reasoning (r = .275, p < .001).

Structural equation modelling

Results of the initial SEM analyses showed that covariance matrices of the V (9 x 9) and P (7 x 7) variables were equal across PIQ1 and PIQ2 + 3 groups. The total sample covariance matrices (16 x 16) across these groups could not be compared due to an excessively large df/N -ratio (see Table 4, rows 1–3). Table 4 rows 4–9 contain one-factor (i.e., V and P factors separately) models of the combined DLD sample and PIQ1 and PIQ2 + 3 groups, which all showed good fit to the data.

Table 4. SEM analysis. Chi-square, df, p, RMSEA (Root Mean Square Error of Approximation), CFI (Comparative Fix Index), and TLI (Tucker-Lewis Index)..

Row	Model	Chi-square	df	p	RMSEA	CFI	TLI
1	Model 1 16 × 16 covariance matrix across groups PIQ1 vs PIQ2 + 3	na*	na*	na*	na*	na*	na*
2	Model 2 9 × 9 covariance matrix across groups PIQ1 vs PIQ2 + 3	30.710	45	.949	.000	1.000	.063
3	Model 3 7 × 7 covariance matrix across groups PIQ1 vs PIQ2 + 3	18.330	28	.918	.000	.000	.316
4	Model 1 V subtests: one factor, combined DLD sample	3.930	27	.168	.041	.982	.976
5	Model 2 V subtests: one factor, group PIQ1	31.55	27	.249	.040	.980	.974
6	Model 3 V subtests: one factor, group PIQ2 + 3	23.80	27	.641	.000	1.000	1.033
7	Model 4 P subtests: one factor, total DLD sample	12.47	14	.569	.000	1.000	1.024
8	Model 5 P subtests: one factor, group PIQ1	10.78	14	.703	.000	1.000	1.189
9	Model 6 P subtests: one factor, group PIQ2 + 3	13.47	14	.490	.000	1.000	1.037
10	Model 1 configural invariance: across groups PIQ1 vs PIQ2 + 3	221.66	208	.246	.029	.963	.957
11	Model 2 metric invariance: across groups PIQ1 vs PIQ2 + 3	224.52	222	.440	.012	.993	.993
12	Model 3 scalar invariance: across groups PIQ1 vs PIQ2 + 3	320.21	238	.000	.067	.778	.776
13	Model 3 scalar invariance: factor means released across groups PIQ1 vs PIQ2 + 3	269.85	236	.064	.043	.909	.907
14	Model 3 scalar invariance: 3 intercepts released across groups PIQ1 vs PIQ2 + 3	258.75	235	.138	.036	.936	.934
15	Model 3 scalar invariance: factor means & 3 intercepts released across groups PIQ1 vs PIQ2 + 3	236.95	233	.416	.015	.989	.989
16	Model 3 scalar invariance: factor means & 3 intercepts & factor variances freed across groups PIQ1 vs PIQ2 + 3	236.46	231	.389	.017	.985	.985
17	Model 4 stringent invariance: across groups PIQ1 and PIQ2 + 3	338.14	254	.000	.065	.773	.785
18	Model 1 two factors: combined DLD sample	151.980	103	.001	.055	.892	.874
19	Model 2 two factors: group PIQ1	118.43	102	.127	.039	.925	.912
20	Model 3 two factors: group PIQ2 + 3	88.01	103	.854	.000	1.000	1.122
21	Model 4 two factors: 6 error correlations released, total DLD sample	104.47	95	.238	.025	.978	.973
22	Model 5 two factors: 4 cross loadings, combined DLD sample	140.95	99	.004	.052	.907	.888
23	Model 6 two factors: 5 subtests removed, combined DLD sample	55.10	53	.395	.016	.994	.992
*	na = not applicable: df/N ratio too high

V= verbal, P= nonverbal. The models contain subtest covariances across PIQ1 and PIQ2 + 3 groups (rows 1–3), one-factor model in the combined DLD sample and in PIQ1 and PIQ2 + 3 groups for V factor (rows 4–6) and P factor (rows 7–9), stringent models of invariance across PIQ1 and PIQ2 + 3 groups (rows 10–17), the final two-factor model in the combined DLD sample and in PIQ1 and PIQ2 + 3 groups (rows 18–20), and testing the two-factor model in the combined DLD sample (rows 20–23).

The nested models of the measurement invariance (i.e., V and P factors together) are presented in Table 4, (rows 10–17). The fit values showed that the model of configural invariance (factor loadings equal) was good and acceptable. Metric invariance model (factor loadings and intercepts equal) was also good, showing strong factorial invariance. The subtest factor loadings were similar in PIQ1 and PIQ2 + 3 groups. The next step of strong measurement (scalar) invariance showed an ill-fit (row 12). Factor means were unequal (row 13) and the first three P factor variables (BD, PCn, and MR) accounted for the significant χ². When their intercepts were relaxed, the model showed a fair fit (row 14). Releasing both factor means and the first three P factor variable intercepts changed the model to be acceptable (row 15). The next model releasing factor means, the first three P factor variable intercepts, and factor variances across PIQ1 and PIQ2 + 3 groups showed a similar good fit (row 16). We can thus claim that the measurements have a metric and partially scalar (weak) invariance.

The behavior of factor loadings remained fairly stable in the nested models. Since the covariances were equal, the metric invariance was good, and the scalar invariance partial, we chose a final model (Figure 3) where the combined DLD group was used without dividing it according to the PIQ level, although the fit of this model was not very good (Table 4, row 18). Other factors supporting this decision were the small number of cases when using divided groups and the amount of missing values, which made some of the models unstable. This model had no cross-loadings, intercepts were not constrained, error correlations were set to zero, and residual and factor correlations were freely estimated. The correlation between the latent V and P composite factors was large (r = .573, p < .001).

Figure 3. Factor loadings in SEM modeling.

We used the following three options in the model trimming (Table 4, rows 21–23): releasing 6 of 120 error correlations, allowing some variable cross loadings, and dropping ill-fitting variables Comprehension and Sentence Repetition from the latent V factor, and Block Design, Symbol Search, and Picture Concepts from the P factor. This improved the fit to very good. However, to avoid over-fitting, the total 16-variable model without these trimmings was chosen as the final model.

In summary, the SEM process led to the following findings: measurement invariance was metric and partially scalar (i.e., the variables measure the same features in PIQ1 and PIQ2 + 3 groups); factor loadings were equal across these two groups; factor means were significantly higher in PIQ1 than PIQ2 + 3 group, but the means of V factor were smaller than the means of P factor in both groups; item intercepts were higher in PIQ1 than PIQ2 + 3 group; factor correlations were high and did not differ between the two groups; and factor variances were similar across these groups. Thus, the SEM results confirm the view obtained with the more traditional linear least-square methods.

Discussion

Several earlier studies have shown that children with DLD tend to have weaker nonverbal performance skills than their peers with normal development (e.g., Gallinat & Spaulding, 2014; Liao et al., 2014; McGregor et al., 2013; Stothard et al., 1998). While the level of overall nonverbal performance is no longer considered critical for DLD diagnosis in the new diagnostic classifications (DSM-5 and ICD-11), the exact nature and significance of the connection between verbal and nonverbal processing in children with DLD diagnosed following the new diagnostic criteria is not clear.

The aim of the present study was to evaluate the significance of nonverbal performance level on cognitive subtest profiles and the connections between nonverbal and verbal skills in children previously diagnosed with DLD and in those diagnosed with mixed specific developmental disorders following the ICD-10 criteria. Following the new diagnostic criteria, both these groups of children would be diagnosed with DLD. To our knowledge, the relations between verbal and nonverbal cognitive performance have not been previously examined in this newly defined DLD group.

Cognitive subtest profiles in children with DLD

Overall, our results indicate a strong connection between nonverbal and verbal cognitive skills in children diagnosed with DLD; the weaker the nonverbal cognitive level, the lower the verbal skills. Although verbal subtest profiles in all our PIQ groups fell below the normative age range (i.e., below −1 SD), the differences in verbal performance were greatest between children with average (PIQ group 1) and poor (PIQ group 3) nonverbal cognitive skills. Our results are partly consistent with those reported by Norbury et al. (2016), showing only minimal differences in the severity of language problems between children with average (−1 SD or better, corresponding to our PIQ group 1) and low average (between −1 SD and −2 SD, corresponding to our PIQ group 2) nonverbal cognitive skills. However, in our study, significant differences in verbal performance between PIQ groups 1 and 2 were evident in subtests related to receptive language, whereas in the Norbury et al. (2016) study, those with low average nonverbal cognitive skills had significantly more severe expressive language deficits. Moreover, PIQ group 3 with poor nonverbal cognitive skills had significantly more severe problems in verbal reasoning and receptive vocabulary than PIQ groups 1 and 2, but not in the expressive language test Picture Naming. The differences between the results in this study and Norbury et al. (2016) are most probably related to selection of participants, selection of nonverbal and verbal tests, and the composite scores used in each study.

Similar to the Conti-Ramsden et al. (2012) study, one third of children in the present study had low-average or poor nonverbal skills. These children resemble those with BIF, who typically score 1 to 2 standard deviations below the normal age mean in overall intelligence and adaptive functioning (Greenspan, 2017; Pulina et al., 2019). In the study by Pulina et al. (2019), school-aged children with BIF had an uneven intellectual profile with significant weakness in the WM index and relative strength in the Perceptual Reasoning index. Children diagnosed with DLD typically also have relative strengths in perceptual reasoning (e.g., Hick et al., 2005; Saar et al., 2018) and in addition to speech and language deficits, have weaknesses in WM and verbal STM (e.g., Alloway et al. 2010; Archibald & Gathercole, 2006; Saar et al., 2018). In accordance with these findings, the most profound weaknesses in each PIQ-level group and the combined DLD group appeared in Word Reasoning, Comprehension, and Sentence Repetition, which are subtests related to verbal reasoning and STM. Relative strengths were evident in Matrix Reasoning and Picture Concepts, which require serial and categorical reasoning, respectively, and in Processing Speed subtests.

The combined DLD group consisted of children with average (PIQ group 1), low-average (PIQ group 2) and poor (PIQ group 3) nonverbal performance level. The subtest profile of the combined DLD group closely resembled that of PIQ group 1, reflecting the fact that most participants belonged to this PIQ-level group. However, the overall level of the combined DLD group subtest profile would probably be lower if the number of children in PIQ groups 2 and 3 were balanced with the number of children in PIQ group 1. The overall cognitive level, typically indexed with full scale IQ (FSIQ), was not evaluated in this study. Generally, FSIQ <70 is diagnostically related to intellectual disability (ID) but the use of FSIC, especially in clinical groups, has been questioned (e.g., Fiorello et al., 2007; Greenspan, 2017). In clinical groups, evaluation of the relative cognitive strengths and weaknesses is considered central for differential diagnosis and planning of interventions (Jankowska et al., 2021; Pulina et al., 2019). In the previous DLD diagnostic criteria, the discrepancy between VIQ and PIQ indices was considered crucial. In the new criteria, PIQ <70 is used to separate those with DLD from those with ID. However, diagnostically, differences and connections between more specific cognitive domains, such as processing speed, executive functions, and adaptive skills (Garret & Gilmore, 2009; Santegoeds et al., 2021) may be more useful than composite IQ scores. Our results suggest that in addition to clear deficits in verbal reasoning and STM, relative nonverbal strengths in processing speed and in subtests Matrix Reasoning and Picture Concepts could be used to separate children with DLD from those with ID. However, comparing our results to other studies is difficult because different tests and subtests were used, and subtest profiles have rarely been studied in children with or without DLD.

Connections between nonverbal and verbal cognitive skills in children with DLD

A strong connection between nonverbal cognitive level and verbal skills was also evident in the correlations and were further confirmed and elaborated with SEM modeling. These findings suggest a significant general connection between verbal and nonverbal cognitive processing in children with DLD.

The VIQ-PIQ correlation of the combined DLD group (r = .368) in the present study closely corresponds to the PIQ-VIQ correlation (r = .373) in the normative data of the Finnish WPPSI-III test (Wechsler, 2009), suggesting a similar relationship between nonverbal and verbal reasoning in preschool-aged children with and without DLD. A significant correlation has also been shown in children aged 5–6 years with DLD between PIQ and a composite language score consisting of pronunciation, fluency, articulation, receptive language, and expressive language (Liao et al., 2014). In addition, PIQ is significantly related to semantics and morphosyntax in children with DLD (DeThorne & Watkins, 2006). Together, these results suggest that in children with DLD, both verbal reasoning and basic language skills are connected to nonverbal reasoning abilities.

The connections of overall nonverbal cognitive performance to different verbal subtests and overall verbal performance to different nonverbal subtests varied in strength but were mainly stronger than those in the Finnish WPPSI-III normative data (Wechsler, 2009). This difference reflects the heterogeneity of our sample that produces larger variance, and consequently higher covariance and correlation values. The PIQ connection in the present study was strongest to subtests Similarities and Receptive Vocabulary (emphasizing verbal understanding) and weakest to subtests Picture Naming, Sentence Repetition, Comprehension (emphasising expressive language), and verbal STM, which are typical weaknesses in children with DLD (Archibald & Gathercole, 2006; Baird et al., 2010; Kenney et al., 2006; Messer & Dockrell, 2006) and consequently do not vary with nonverbal abilities as strongly as verbal comprehension and reasoning skills. On the other hand, all nonverbal subtests other than Block Design had significant correlations with VIQ, indicating a clear verbal impact on nonverbal reasoning in children with DLD. In our previous study (Saar et al., 2018), Block Design was the only nonverbal subtest that did not differentiate subjects with expressive or mixed receptive-expressive DLD but varied independently. Together, these results suggest that in children with DLD, Block Design is the only WPPSI-III subtest that evaluates nonverbal skills independently of the verbal cognitive level. In addition, consistent with our results showing a clear connection between processing speed and verbal reasoning, Miller et al. (2001) found slow processing speed particularly in children with more severe language problems.

The two-factor structure of cognitive functions in children with DLD

A two-factor structure of cognitive functions representing verbal and nonverbal skills has been frequently suggested for typically developing preschool-aged children (e.g., Brito et al., 2011; Peyre et al., 2016; Tideman & Gustafsson, 2004). In other studies, the structure of cognitive functions in children with TD has been found to be more complex (e.g., Plym et al., 2021; Tusing & Ford, 2004) with differing numbers of factors suggested. Tusing and Ford (2004) suggested a five-factor structure of cognitive functions in children with TD aged 4–5 years, including crystallized intelligence, auditory processing, STM, LTM, and one factor related to nonverbal ability. These results differ from those in Plym et al. (2021), where the best-fitting structure for preschool-aged children with TD included factors reflecting verbal abilities, processing speed/short-term memory, visuomotor functions, and visuoconstructive abilities/nonverbal reasoning. The differences between these results are most probably related to the combination of tests used in each study that may have evaluated partly different cognitive functions.

In some studies, the structure of cognitive functions in preschool-aged children (Ottem, 2003) and school-aged children (Gillam et al., 2021) with and without DLD was similar, although again with different factor solutions from study to study. However, the structure of cognitive functions may also differ between preschool-aged children with and without DLD (Plym et al., 2021), showing more homogeneous verbal performance in TD children, and more interconnected nonverbal skills in children with DLD, indicating possible differences in the maturation rate of these cognitive skills between these groups of children.

In the present study, we chose the commonly used two-factor model with latent V (verbal) and P (nonverbal performance) composite factors to evaluate the connections between verbal and nonverbal performance in preschool-aged children with DLD with average versus below average nonverbal performance. Our results showed that in children with DLD, the latent V and P factor variables measured the same features in those with average (PIQ1 group) and below average (PIQ2 + 3 group) nonverbal abilities, and that the fit of the two-factor model in both groups was good. Thus, the structure of cognitive functions, evaluated with all WPPSI-III subtests and NEPSY-II subtests Comprehension of Instructions and Sentence Repetition, can adequately be modeled with a two-factor solution in children previously diagnosed with DLD and in those with mixed specific developmental disorders.

The final two-factor model selected for the combined DLD group exhibited the most generalizable solution with an acceptable fit without the danger of overfitting. In this group, all subtests representing verbal reasoning (Information, Comprehension, Word Reasoning, Vocabulary and Similarities) had high loadings on the latent V factor which closely resembles the verbal factor identified in the factor analysis of the Finnish WPPSI-III normative data (Wechsler, 2009). Also, in line with the Finnish WPPSI-III normative data, the latent P factor was most strongly defined by processing speed with additional impact from serial reasoning indexed by Matrix Reasoning and attention to visual details central in Picture Completion. These latent factors resemble fluid and crystallized intelligence, first described in Cattell’s Gf-Gc model of intelligence (Cattel, 1943) and later expanded (Cattel & Horn, 1978) to include broad abilities of visual processing, short-term apprehension and retrieval, long-term storage and retrieval, and speed of processing. More recently, other broad abilities have been added to the model. However, in a meta-analysis examining the structure of broad abilities, Bryan and Mayer (2020) reported a reasonable fit for a two-factor solution with factors resembling fluid and crystallised intelligence. In the present study, the correlation between the latent V and P factors was large, indicating a strong connection between fluid and crystallised intelligence in children with DLD.

Fluid intelligence in children with DLD

The connection between fluid intelligence and verbal abilities was already presented in Cattel’s investment theory (Baghaei & Tabatabaee, 2015; Cattel & Horn, 1978; Cattel, 1943; Kvist & Gustafsson, 2007; Thorsen et al., 2014), where fluid intelligence is assumed to represent general intelligence and to have a continuous influence on crystallized intelligence (i.e., on the ability to use experience, knowledge, and verbal skills in reasoning). Fluid intelligence is most typically evaluated with tests resembling WPPSI-III subtest Matrix Reasoning (e.g., Ferrer et al., 2009; Ren et al., 2014). In WPPSI-IV (Wechsler, 2012), the new Fluid Reasoning index is composed of Matrix Reasoning and Picture Concepts, subtests significantly connected to verbal reasoning in children with DLD in the present study. These findings are in line with a notion that measures of nonverbal skills might not be totally discrete from or in contrast to verbal skills (Durant et al., 2019). In future studies applying WPPSI-IV, it would be interesting to compare the connection between verbal/crystallised and nonverbal/fluid reasoning in children with and without DLD.

Processing speed in general is considered a central component in all cognitive processing (e.g., Kail, 1994; Leonard et al., 2007), and together with Picture Concepts and Matrix Reasoning, represents nonverbal fluid intelligence in the WPPSI-III test (Wechsler, 2009), the newest Finnish version of the WPPSI test presently available. The strong connection between processing speed and DLD suggested in numerous earlier studies (e.g., Leonard et al., 2007; Miller et al., 2001, Park, Mainela-Arnold et al., 2015; Windsor et al., 2008) was also clear in the present study. The contribution of processing speed and WM to fluid intelligence has been emphasized (Dehn, 2017; Engle et al., 1999; Gillam et al., 2021; Leonard et al., 2007; Montgomery et al., 2018, 2021), and these basic cognitive processes are suggested to develop in concert with fluid intelligence (Fry & Hale, 2000). WM capacity is also central to verbal processing (Lum & Bleses, 2012; Montgomery et al., 2018), and the relationship between WM capacity and language skills has been shown to be stronger in TD children than in children with DLD (Gillam et al., 2021). However, it is not clear whether this difference is inherent in children with DLD or results from difficulties with both memory and language.

Interestingly, Chuderski (2013) found that the relation of WM and fluid intelligence depends on the time allowed for nonverbal tasks, and that the performance of participants with low cognitive capacity approached those with higher capacity when no time limits were applied. In our study, children with DLD had on average weakest nonverbal reasoning scores in subtests with time limits (Block Design, Object Assembly and Picture Completion), suggesting a general slowness in nonverbal reasoning capacity. In addition, both nonverbal reasoning and processing speed were weakest in children with the most severe language problems. Consistent with our results, slow processing speed was observed particularly in children with more severe language problems (Miller et al., 2001), and it has been suggested that nonverbal tests should be presented and evaluated without time limits when diagnosing children with DLD (Park, Miller et al., 2015).

Conclusions

The present results suggest that nonverbal fluid intelligence is strongly linked to verbal understanding and reasoning in children with DLD; the weaker the nonverbal fluid intelligence, the more severe the DLD. On the other hand, following the new DLD diagnostic criteria and including children with below average and weak nonverbal skills, fluid intelligence and processing speed also emerged as relative cognitive strengths in the subtest profiles of these children, regardless of the nonverbal cognitive level. These results suggest that in addition to specific weaknesses in receptive and expressive language, verbal reasoning, STM and WM, relative nonverbal strengths in fluid reasoning and processing speed could be used as diagnostic markers when identifying children with DLD.

Consistent with Park, Miller et al. (2015), we further suggest that nonverbal tests should be presented and evaluated without time limits, especially when the focus is to separate those with DLD from those with ID, as children with ID might not benefit from the extra time allowed for completing the tasks the same way as children with DLD. However, future studies are needed to evaluate the diagnostic significance of this suggestion. Following CATALISE (Bishop et al., 2017) recommendations, adaptive skills should also be evaluated to define the functional impact of verbal and nonverbal deficits and possible differences in children with DLD and ID.

Acknowledgements

This study was approved by the Ethics Committee of the Hospital District of Helsinki and Uusimaa. The authors alone are responsible for the content and writing of this manuscript. The authors are grateful to Dr Johanna Rosenqvist for her valuable comments on the manuscript.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

The author(s) reported there is no funding associated with the work featured in this article.