Introduction
Many studies have shown that the Communicative Development Inventories (CDIs) are reliable and valid checklists for measuring early vocabulary across a wide range of participants (Feldman et al., Reference Feldman, Dale, Campbell, Colborn, Kurs-Lasky, Rockette and Paradise2005; Fenson et al., Reference Fenson, Marchman, Thal, Dale, Reznick and Bates2007; Frank et al., Reference Frank, Braginsky, Yurovsky and Marchman2021). Administering CDIs is a standardised, fast, and cost-effective approach that does not require trained lab assistants, lab visits, or the labour-intensive transcription of speech that is required for analysing naturalistic speech samples. This allows for larger sample sizes which is beneficial, especially when examining demographic effects on vocabulary development. Demographic effects typically only capture a small proportion of the large variance in children’s vocabulary size, and most significantly in the early years (e.g., Eriksson et al., Reference Eriksson, Marschik, Tulviste, Almgren, Pérez Pereira, Wehberg, Marjanovič-Umek, Gayraud, Kovacevic and Gallego2012; Fenson et al., Reference Fenson, Marchman, Thal, Dale, Reznick and Bates2007; Kidd & Donnelly, Reference Kidd and Donnelly2020). Previous studies have identified key demographic predictors – including maternal education, children’s gender, gestational age, birth weight, and multilingualism – but there are uncertainties regarding the duration, direction, and magnitude of these effects on different vocabulary outcome measures across development. One advantage of cohort studies is that we can re-evaluate previous research findings in a large sample with repeated measurements to ensure their robustness and generalisability. Longitudinal data also provide insights into how these predictors unfold over time. In the current study, we make use of the large, longitudinal YOUth cohort study that measured vocabulary between infancy and toddlerhood of over four hundred Dutch children. This cohort with multiple different vocabulary measurements over time provided us with an excellent opportunity to study the direction and magnitude of demographic predictors across early development.
Maternal education
Maternal education is often used as a proxy for socio-economic status (SES). Previous studies often reported positive effects of maternal education on children’s vocabularies measured by CDIs for toddlers (e.g., Feldman et al., Reference Feldman, Dollaghan, Campbell, Kurs-Lasky, Janosky and Paradise2000; Fenson et al., Reference Fenson, Marchman, Thal, Dale, Reznick and Bates2007; but cf. Reese & Read, Reference Reese and Read2000; Kuvač-Kraljević et al., Reference Kuvač-Kraljević, Blaži, Schults, Tulviste and Stolt2021). Mothers with higher educational backgrounds produce a higher quantity (i.e., they speak more) and quality (i.e., they use more diverse language) of speech towards their children, mediating the positive relationship between maternal SES and children’s language development (Hoff, Reference Hoff2003; Huttenlocher et al., Reference Huttenlocher, Waterfall, Vasilyeva, Vevea and Hedges2010). However, studies employing CDIs have frequently observed negative correlations between maternal education and children’s vocabularies during infancy (e.g., Bavin et al., Reference Bavin, Prior, Reilly, Bretherton, Williams, Eadie, Barrett and Ukoumunne2008; Feldman et al., Reference Feldman, Dollaghan, Campbell, Kurs-Lasky, Janosky and Paradise2000; Fenson et al., Reference Fenson, Marchman, Thal, Dale, Reznick and Bates2007; Reese & Read, Reference Reese and Read2000). This early negative effect of maternal education on CDIs is likely driven by a caregiver reporting bias: a negative effect of SES is more often reported for vocabulary comprehension which requires more interpretation by the caregiver than vocabulary production, although a negative effect is sometimes reported for production as well (Bavin et al., Reference Bavin, Prior, Reilly, Bretherton, Williams, Eadie, Barrett and Ukoumunne2008; Reese & Read, Reference Reese and Read2000). In contrast, studies rarely report a negative effect of SES on the gesture scale (Bavin et al., Reference Bavin, Prior, Reilly, Bretherton, Williams, Eadie, Barrett and Ukoumunne2008; Feldman et al., Reference Feldman, Dollaghan, Campbell, Kurs-Lasky, Janosky and Paradise2000; Rowland et al., Reference Rowland, Krajewski, Meints, Łuniewska, Kochańska and Alcock2022). Determining whether a child produces a word or gesture does not require the caregiver to draw inferences about the child’s understanding. In addition, there are fewer expectations from caregivers surrounding children’s gesture development compared to their vocabulary development. On the one hand, caregivers could believe that larger vocabularies are more desirable – leading to over-reporting of their infants’ vocabularies, or because some caregivers have more liberal criteria for word comprehension than others (see Feldman et al., Reference Feldman, Dollaghan, Campbell, Kurs-Lasky, Janosky and Paradise2000; Tomasello & Mervis, Reference Tomasello and Mervis1994 for discussions). On the other hand, caregivers may underestimate what their children already know when their children do not produce many words yet (see Houston-Price et al., Reference Houston-Price, Mather and Sakkalou2007). These findings make it relevant to study the effects of maternal education in large, longitudinal samples throughout the first years of development on a variety of vocabulary measures.
Children’s gender
Many studies have identified small effects of children’s genderFootnote 1. More specifically, girls tend to outperform boys on many vocabulary scales (e.g., Eriksson et al., Reference Eriksson, Marschik, Tulviste, Almgren, Pérez Pereira, Wehberg, Marjanovič-Umek, Gayraud, Kovacevic and Gallego2012; Feldman et al., Reference Feldman, Dale, Campbell, Colborn, Kurs-Lasky, Rockette and Paradise2005; Frank et al., Reference Frank, Braginsky, Yurovsky and Marchman2021; Reese & Read, Reference Reese and Read2000; Reilly et al., Reference Reilly, Bavin, Bretherton, Conway, Eadie, Cini, Prior, Ukoumunne and Wake2009; Zink & Lejaegere, Reference Zink and Lejaegere2002, but cf. Bavin et al., Reference Bavin, Prior, Reilly, Bretherton, Williams, Eadie, Barrett and Ukoumunne2008). Simonsen et al. (Reference Simonsen, Kristoffersen, Bleses, Wehberg and Jørgensen2014) showed that boys are characterised by a less steep increase in receptive vocabulary growth than girls – at least until 20 months of age. Feldman et al. (Reference Feldman, Dollaghan, Campbell, Kurs-Lasky, Janosky and Paradise2000) examined over 2,000 American English children using CDIs and reported lower scores for boys in vocabulary production and vocabulary comprehension across children aged 10–13 months. These differences persisted for older children, except for vocabulary comprehension. Girls have also been found to have larger gesture repertoires than boys based on CDIs (Feldman et al., Reference Feldman, Dollaghan, Campbell, Kurs-Lasky, Janosky and Paradise2000; Germain et al., Reference Germain, Gonzalez-Barrero and Byers-Heinlein2022; Simonsen et al., Reference Simonsen, Kristoffersen, Bleses, Wehberg and Jørgensen2014; Zink & Lejaegere, Reference Zink and Lejaegere2002). These studies suggest that overall, girls have faster developmental trajectories than boys. In contrast, previous studies using naturalistic speech samples or lab-administered tasks of children’s receptive vocabularies typically do not report gender differences in diverse samples (e.g., Huttenlocher et al., Reference Huttenlocher, Waterfall, Vasilyeva, Vevea and Hedges2010; Pan et al., Reference Pan, Rowe, Spier and Tamis-Lemonda2004; Washington & Craig, Reference Washington and Craig1999), although these findings are inconsistent, particularly for children’s expressive language skills where girls tend to outperform boys (e.g., Bornstein et al., Reference Bornstein, Haynes and Painter1998; Frank et al., Reference Frank, Braginsky, Yurovsky and Marchman2021; Qi et al., Reference Qi, Kaiser, Milan, Yzquierdo and Hancock2003; but cf. Bergelson et al., Reference Bergelson, Soderstrom, Schwarz, Rowland, Ramírez-Esparza, Hamrick, Marklund, Kalashnikova, Guez, Casillas, Benetti, van Alphen and Cristia2023). The effect of gender could be small and variable across children’s ages and vocabulary measures, causing inconsistent results across studies.
Gestational duration and birth weight
Some studies suggest that preterm children are at a larger risk of having smaller vocabularies than full-term children (e.g., Foster-Cohen et al., Reference Foster-Cohen, Edgin, Champion and Woodward2007; Guarini et al., Reference Guarini, Sansavini, Fabbri, Alessandroni, Faldella and Karmiloff-Smith2009; Sansavini et al., Reference Sansavini, Guarini, Savini, Broccoli, Justice, Alessandroni and Faldella2011, but cf. Ogneva & Pérez-Pereira, Reference Ogneva and Pérez-Pereira2023). There may be negative effects only in extremely or very preterm children. Kern and Gayraud (Reference Kern and Gayraud2007) found that very preterm (28–32 weeks) and extremely preterm (under 28 weeks) children had smaller vocabulary sizes based on CDIs than moderately preterm (33–36 weeks) and full-term children when they were assessed at 24–26 months of age. However, Pérez-Pereira and Cruz (Reference Pérez-Pereira and Cruz2018) found that gestational age did not affect vocabulary growth in a sample of low-risk preterm children with a wide range of gestational ages and birth weights without other medical complications. Still, a meta-analysis showed that very preterm (under 32 weeks) and/or very low birth weight (under 1500 g) children have persistent language delays (Barre et al., Reference Barre, Morgan, Doyle and Anderson2011). Moreover, differences between preterm and full-term children in gestural and lexical development become increasingly more evident during the first two years of life (Sansavini et al., Reference Sansavini, Guarini, Savini, Broccoli, Justice, Alessandroni and Faldella2011; van Baar et al., Reference van Baar, Ultee, Gunning and Soepatmi2006). Previous studies have to our knowledge not concurrently examined the effects of gestational duration and birth weight, and it remains a question whether these factors influence children’s vocabulary development in a non-clinical sample. It also remains largely understudied whether vocabulary differences between preterm and full-term children are apparent during the first year of life. Therefore, it is relevant to study the effects of gestational age and birth weight in a large, longitudinal sample starting from infancy.
Multilingualism
In many studies examining children’s vocabularies using the CDIs, multilingual children are excluded. CDI norming samples also typically exclude multilingual children, while being multilingual is the norm in most places across the world. Therefore, it is important to assess how multilingualism affects children’s performance on a variety of widely used vocabulary tasks. When assessing only one language, multilingual children have smaller vocabularies than their monolingual peers (Blom et al., Reference Blom, Boerma, Bosma, Cornips, van den Heuij and Timmermeister2020; De Houwer et al., Reference De Houwer, Bornstein and Putnick2014; Hoff et al., Reference Hoff, Core, Place, Rumiche, Señor and Parra2012). De Houwer et al. (Reference De Houwer, Bornstein and Putnick2014) showed using CDIs that monolingual toddlers knew more Dutch words than bilingual toddlers (20 months), but both groups understood and produced the same number of lexicalised meanings. They did not find any differences between monolinguals and bilinguals in vocabulary comprehension or vocabulary production for infants (13 months). A recent study showed that multilingualism does not affect infants’ gesture repertoires either (Germain et al., Reference Germain, Gonzalez-Barrero and Byers-Heinlein2022). Other studies suggest that multilingual toddlers do not have smaller vocabularies than their monolingual peers when they receive at least 60% exposure to the assessed language (Cattani et al., Reference Cattani, Abbot-Smith, Farag, Krott, Arreckx, Dennis and Floccia2014). In our study, we included multilingual children to examine whether their vocabularies are negatively affected when examining only one of their languages using the N-CDIs and PPVT-III-NL.
Research aim
Well-known demographic predictors of language – including maternal education, child gender, gestational age and birth weight, and multilingualism – have been documented extensively, and many researchers accept their influences on language development without further question. However, particularly in the 90s when the American CDIs were first created, these predictors were often studied in smaller samples at one point in time. Longitudinal data provide insights into how these predictors of children’s vocabulary unfold over time. Given the replication crisis in psychology, it is valuable to re-examine the findings in a large, longitudinal sample using different vocabulary measures to ensure their robustness and generalisability. In the present study, we aimed to examine whether key predictors that explain variation in children’s early vocabularies are age-specific and task-specific in a large, longitudinal sample of Dutch children. A limited number of studies have examined these predictors within large, longitudinal samples, while there are uncertainties regarding the duration, directionality, and magnitude of these effects on different vocabulary measures across development. By examining the effects on multiple vocabulary measures in a large sample from infancy to early childhood, we analyse whether the effects of well-known predictors are age-specific and task-specific while keeping the characteristics of the sample constant. This helps us to identify whether widely discussed demographic predictors of vocabulary are robust and generalisable across development – ultimately advancing our understanding of children’s language development.
Methods
Participants
The data for this study are derived from the YOUth cohort study following Dutch children prenatally up to early childhood (Onland-Moret et al., Reference Onland-Moret, Buizer-Voskamp, Albers, Brouwer, Buimer, Hessels, de Heus, Huijding, Junge, Mandl, Pas, Vink, van der Wal, Hulshoff Pol and Kemner2020). The cohort involved repeated measurements at regular intervals. From the cohort, 444 Dutch infants around 10 months of age (230 females, 214 males; age M = 10.6 months; age range = 9.0 – 13.1 months; age SD = 0.9) (hereafter Wave 1) were included in this study. These were all the children in the YOUth cohort study who had participated in the next wave by March 2022. During this wave, the same children were on average 3.4 years of age (range = 2.0 – 6.0 years; SD = 0.8) (hereafter Wave 2). There were approximately one to five years (M = 2.5; SD = 0.8) in between measurement waves, randomly varying per participant. We followed the Code of Ethics of the World Medical Association (Declaration of Helsinki), and all caregivers signed informed consent prior to participating. During Wave 1, children received a Miffy picture book for their participation. During Wave 2, children received a frog umbrella.
In total, 426 mothers filled out the demographics questionnaire including questions on the caregivers’ education. All caregivers provided us with their child’s due date and birth date which we used to calculate the child’s gestational duration. Of this sample, 399 caregivers also provided us with their child’s birth weight in grams. Lastly, 369 caregivers filled out the questionnaire including languages spoken at home. The summary of sample characteristics is shown in Table 1. In this sample, at least 29 children were not growing up as monolingual Dutch speakers. We considered a child monolingual when only Dutch was spoken at home. Given the small number of multilingual children, we did not differentiate the group further based on the children’s estimated time of exposure to Dutch.
Table 1. Sample characteristics including the mean (and standard deviation) for continuous variables or frequency counts (and percentage of sample) for categorical variables
Materials and procedure
NYOUth-CDIs
We administered the NYOUth-CDI 1 – measuring vocabulary production, vocabulary comprehension, and gestures – during Wave 1. The NYOUth-CDI 1 contains the short form of words (Zink & Lejaegere, Reference Zink and Lejaegere2003). We used the short form because it contains only 103 compared to 434 items, which makes this form far less time-consuming to complete. This was desired since caregivers already had to fill out a broad range of questionnaires in the YOUth cohort study. Caregivers were asked to check for each item whether their child understands or speaks the word – also when the child produces synonyms or pronunciation errors. In the NYOUth-CDI 1, we replaced or removed 12 typical Flemish words with synonyms that are more common in Standard Dutch spoken in the Netherlands (e.g., we removed mantel from jas(je) / mantel (“coat”)) to make the lists more suitable for children included in the YOUth cohort study. We included the list containing 65 gestures and actions from the full-length N-CDI-WG (Zink & Lejaegere, Reference Zink and Lejaegere2002) which is typically not administered with short forms. This scale contains “early gestures” including the first communicative gestures (e.g., pointing) and games and routines (e.g., playing peekaboo) and “late gestures” including actions with objects (e.g., eating with a spoon or fork) and pretending to be a caregiver (e.g., pretending to feed a doll). The gesture scale could be more suitable compared to the vocabulary scales for this young age group, as it does not suffer from floor effects and is related to children’s later vocabulary size (Zink & Lejaegere, Reference Zink and Lejaegere2002). The NYOUth-CDIs were emailed to the primary caregiver. The NYOUth-CDIs are fully digitised so caregivers could fill them out online. We scored the lists following the instructions of the manuals (Zink & Lejaegere, Reference Zink and Lejaegere2002, Reference Zink and Lejaegere2003).
The NYOUth-CDI 2 is a combination of the short forms N-CDI 2A (16-30 months) and N-CDI 3 (30-37 months) (Zink & Lejaegere, Reference Zink and Lejaegere2003). We combined the two forms because there was only one measurement wave during the toddler and preschool years in the YOUth cohort study (Wave 2). The combined version resulted in a total number of 207 vocabulary items after removing the overlapping ones. Caregivers were asked to check the items that the child speaks – also in case the child produces synonyms or pronunciation errors. In the NYOUth-CDI 2, we also replaced or removed 26 typical Flemish words with similar words that are more common in Standard Dutch spoken in the Netherlands (e.g., bank instead of zetel/sofa (“couch”)). The CDIs for toddlers (including adaptations in other languages) do not measure vocabulary comprehension or gestures. Most toddlers and older children have already acquired all the gestures resulting in a ceiling effect. Children of this age group are also old enough to participate in a lab-administered task of vocabulary comprehension. Caregivers were instructed to fill the NYOUth-CDI 2 out within four weeks after the administration of the PPVT-III-NL in the lab during Wave 2.
Peabody Picture Vocabulary Task
During Wave 2, we also administered the third version of the Dutch Peabody Picture Vocabulary Task (PPVT-III-NL) which is a lab-administered task of receptive vocabulary (Schlichting, Reference Schlichting2005). The task measures whether a person can match a spoken word to one of the four pictures (i.e., multiple choice). It is designed as a behavioural task in which the participant points to one of the images and the experimenter produces the target words and scores manually. For the YOUth cohort study, we developed a computerised version of the PPVT-III-NL. The experimenter runs a script on a computer with a touch screen where children are provided with recordings of the test items and four pictures on the screen. This controls for differences in speaker pronunciations and minimises the role of the experimenter. Children can use the touch screen to select one of the pictures after the target item has been presented. During the task, items become increasingly more complex. The PPVT-III-NL has a total of 204 items, divided into 17 sets of 12 items. The task terminates when the child makes nine or more errors in one set (“final set”) (see Schlichting, Reference Schlichting2005). The programme automatically subtracts the number of errors from the maximum score (which is the number of the final set * 12 items), resulting in the child’s raw score. During the task, the child’s caregiver was present in the back of the room out of the child’s view. Caregivers were explicitly instructed not to help or communicate with the child.
Validity evidence
Reliability
Due to the modifications we made to the NYOUth-CDIs, we first assessed whether the adapted checklists measure a valid approximation of children’s vocabulary size. First, we examined whether we could find evidence for the reliability of the NYOUth-CDIs. For the NYOUth-CDI 1, we calculated Cronbach’s alpha separately for comprehension (α = .97), production (α = .91), and gestures (α = .89) which represents the consistency of items within each scale. We also calculated Cronbach’s alpha for the NYOUth-CDI 2 word production (α = .99) indicating that the items on the scale measured the same construct. Overall, this indicates that the different items included in the caregiver reports show excellent reliability.
Validity
We also present several types of validity evidence. First, we assessed correlations between the different scales included in the NYOUth-CDI 1 for infants. The results of the correlation tests indicate that for the NYOUth-CDI 1, comprehension was positively correlated with both production, r s(336) = .50, p < .001, and gestures, r(335) = .65, p < .001. Production was also correlated positively with gestures, r s(335) = .47, p < .001. We also examined whether vocabulary production obtained by the NYOUth-CDI 2 shows a concurrent relationship with vocabulary comprehension measured by the lab-administered PPVT-III-NL. The correlation test indicates there is a strong, positive correlation between NYOUth-CDI 2 production and concurrent PPVT-III-NL comprehension scores, r s(292) = .64, p < .001. The relationship is depicted in Figure 1.
Figure 1. The concurrent relationship between NYOUth-CDI 2 production and PPVT-III-NL comprehension.
The last step was to assess longitudinal relations between the different vocabulary measures. We examined whether vocabulary production, vocabulary comprehension, and gestures measured at Wave 1 were correlated with NYOUth-CDI 2 production and PPVT-III-NL comprehension measured at Wave 2. In total, 266 participants completed the NYOUth-CDIs during Wave 1 and Wave 2, and 325 participants completed both the NYOUth-CDI 1 at Wave 1 and the PPVT-III-NL at Wave 2. We ran partial correlations correcting for the varying time interval between the two waves using the ppcor package in R (Kim, Reference Kim2015). We therefore used PPVT-III-NL raw scores which are not yet corrected for age. The results of all (partial) correlation tests are summarised in Table 2.
Table 2. (Partial) correlation table showing the links between the different NYOUth-CDI scales at Wave 1 and Wave 2 and the PPVT-III-NL at Wave 2 while controlling for the varying time interval in between measurement waves
Note: * p < .05; ** p < .01; ***p < .001
The results show that all measures of the NYOUth-CDI 1 were positively correlated with later NYOUth-CDI 2 production scores. Overall, the strengths of the correlations were weak to moderate. We also found that comprehension at Wave 2 (i.e., PPVT-III-NL) only correlated with the gesture scale in Wave 1.
Questionnaires
We collected the previously described characteristics of the sample via digital questionnaires. These included questionnaires on the mother’s demographics (e.g., educational background), which we collected when the mother was 20 weeks pregnant; the child’s birth (e.g., due date, birth date, and birth weight), which we collected shortly after the child’s birth; and the languages spoken at home (including questions about the caregivers’ native language(s) and the language(s) spoken at home), which we collected during Wave 1 concurrently with the NYOUth-CDI 1.
Coding and analyses
All analyses were carried out in R version 4.2.0 (R Core Team, Reference Team2022). For the NYOUth-CDI 1, we calculated “vocabulary production” by summing all vocabulary items for which caregivers ticked the box speaks, “vocabulary comprehension” by summing all vocabulary items for which caregivers ticked the box understands or speaks, and “total gestures” by summing all yes, sometimes, and often responses on the gesture scale. Gestures can be subdivided into two categories: “early gestures” and “late gestures” (Zink & Lejaegere, Reference Zink and Lejaegere2002). The sum of both scales results in the score “total gestures”. We used these raw scores to analyse the data. For the NYOUth-CDI 2, we calculated “vocabulary production” by summing all items that were marked by the caregivers indicating that the child produces the word. For the PPVT-III-NL, we obtained “vocabulary comprehension” through the raw scores which were automatically calculated by the computer script. We coded the highest educational degree obtained on a nine-point scale ranging from 1 = no education to 9 = university degree. We calculated gestational duration in days using the discrepancy between children’s due dates and birth dates and adding or subtracting this from 280 days (i.e., full-term gestation). Caregivers reported their children’s birth weight in grams. Lastly, we determined whether a child was growing up multilingual (i.e., at least one caregiver does not only speak Dutch at home).
We fitted robust generalised linear models using the package robustbase version 0.95-0 (Maechler et al., Reference Maechler, Rousseeuw, Croux, Todorov, Ruckstuhl, Salibian-Barrera, Verbeke, Koller, Conceicao and Anna di Palma2022) following Frank et al. (Reference Frank, Braginsky, Yurovsky and Marchman2021). We used “vocabulary comprehension”, “vocabulary production”, and “gestures” measured by the NYOUth-CDI 1, “vocabulary production” measured by the NYOUth-CDI 2, and “vocabulary comprehension” measured by the PPVT-III-NL as continuous outcome measures. We added children’s ages in weeks, gender (female or male), gestational, birth weight, maternal education, and language status (monolingual or multilingual) as predictors to the models. For categorical predictors, we used dummy coding with the categories containing the largest number of observations (gender: female; language status: monolingual) as reference levels. We centred and scaled children’s age, gestational duration, birth weight, and maternal education. We modelled raw scores instead of normed scores or percentiles. By adding age in weeks as a predictor to the models, all other predictors are independent of the effects of age.
Results
Descriptive statistics
We included 444 participants from the YOUth cohort study. During Wave 1, 338 of these participants completed the NYOUth-CDI 1. There was one participant who did not complete the gestures list; this participant is only excluded from analyses involving gestures. During Wave 2, we had to exclude four participants from the PPVT-III-NL because the children did not participate (n = 2) or the test day had ended prematurely before administering the PPVT-III-NL (n = 2) resulting in no data. We excluded an additional 11 children from any analyses involving the PPVT-III-NL because they did not fully complete the task, resulting in a total of 429 participants. There were 303 participants whose caregivers completed the NYOUth-CDI 2 for Wave 2. The descriptive results of the vocabulary tests are presented in Table 3. The high standard deviations indicate that vocabulary scores are spread out over a wide range, revealing a large amount of individual variability.
Table 3. Descriptive results of the vocabulary measures
a At Wave 1, vocabulary comprehension is measured with the NYOUth-CDI 1. At Wave 2, vocabulary comprehension is measured with the PPVT-III-NL (raw scores).
Demographic effects
During Wave 1, vocabulary comprehension, vocabulary production, and gestures were measured with the NYOUth-CDI 1. The results of the robust regression models for vocabulary outcomes at Wave 1 are presented in Table 4. During Wave 2, vocabulary production was measured with the NYOUth-CDI 2 and comprehension was measured with the PPVT-III-NL. The results of the robust generalised linear regression models for vocabulary outcomes at Wave 2 are presented in Table 5. When examining the effects on all vocabulary outcomes of infants and toddlers, we find one consistent predictor: age in weeks has a positive effect on all collected outcome measures. We expected a robust age-related effect as children’s vocabularies grow fast during the first years of development. Figure 2 shows the effect of children’s age and gender on the different NYOUth-CDI 1 scales measured during infancy.
Table 4. Robust regression results for vocabulary outcomes at Wave 1
Note: * p < .05; ** p < .01; ***p < .001
Table 5. Robust regression results for vocabulary outcomes at Wave 2
Note: * p < .05; ** p < .01; ***p < .001
a The NYOUth-CDI 2 was administered across all children included in Wave 2 (up until 72 months), while the N-CDI 2 is designed for children until 37 months. We also fitted the model including only the children aged until 37 months (n = 251). All results remain unchanged, but the magnitude of the negative effect for males becomes much larger (b = -15.11, SE = 6.75, p =. 027). See the R Markdown file on OSF for the results.
Figure 2. Effects of children’s age and gender on vocabulary comprehension (A), production (B), and gestures (C) measured with the NYOUth-CDI 1.
Except for age in weeks, all other predictors show inconsistent patterns across the different measurement waves and vocabulary outcomes. For infants, we found a disadvantage for boys on gestures measured with the NYOUth -CDI 1 (b = -2.60, SE = 0.72, p < .001), but not on word production (b = -0.30, SE = 0.26, p = .25) or word comprehension (b = -4.42, SE = 2.49, p = .07) during Wave 1. During Wave 2, we found a disadvantage for boys on NYOUth-CDI 2 production (b = -5.65, SE = 2.51, p = .03), but not on the lab-administered PPVT-III-NL at this age (b = -1.06, SE = 1.13, p = .35). Figure 3 shows the effect of children’s age and gender on the NYOUth-CDI 2 and the PPVT-III-NL at Wave 2. Although both measures show a similar increase with age, there is a clear ceiling effect for production measured using the NYOUth-CDI 2.
Figure 3. Effects of children’s age and gender on vocabulary comprehension measured with the PPVT-III-NL (A) and production measured with the NYOUth-CDI 2 (B).
Second, we found a negative effect of maternal education on caregiver-reported vocabulary comprehension for infants (b = -4.31, SE = 1.47, p < .01) and vocabulary production for infants (b = -0.37, SE = 0.15, p < .05), but not on gestures (b = -0.26, SE = 0.37, p = .48). The negative effect of maternal education has shifted to a positive effect on the lab-administered PPVT-III-NL task during Wave 2 (b = 1.70, SE = 0.65, p < .01), but we found no effect of maternal education on caregiver-reported vocabulary production during this wave (b = 0.18, SE = 1.89, p = .92).
Third, we did not find any significant effects of children’s gestational age or birth weight on any of the vocabulary outcomes during Wave 1 or Wave in our sample.
Lastly, we only found a negative effect of multilingualism on the PPVT-III-NL (b = -5.23, SE = 2.50, p < .05), but not on any of the caregiver-reported NYOUth-CDIs.
Discussion
We aimed to examine whether key demographic predictors that explain variation in children’s early vocabularies – maternal education, children’s gender, gestational age and birth weight, and multilingualism – were age-specific and task-specific in this large, longitudinal sample of Dutch children. Apart from a consistent positive effect of children’s ages on all outcomes, we found that none of the other predictors remained constant across the different vocabulary outcomes measured in this study. Below, we address all other factors one by one.
Effect of maternal education shifts over time
We examined the effects of maternal education as a proxy for SES on children’s vocabulary outcomes. First, we found negative effects of maternal education on vocabulary production and vocabulary comprehension measured by the NYOUth-CDI 1. This is in line with previous studies that have also reported negative effects of SES on CDIs filled out for infants – usually for vocabulary comprehension and to a lesser extent for word production (Feldman et al., Reference Feldman, Dollaghan, Campbell, Kurs-Lasky, Janosky and Paradise2000; Fenson et al., Reference Fenson, Dale, Reznick, Bates, Thal, Pethick, Tomasello, Mervis and Stiles1994; Reese & Read, Reference Reese and Read2000). This is possibly caused by a caregiver reporting bias. The latter interpretation is strengthened by the finding that there is no effect of maternal education on NYOUth-CDI 1 gestures or NYOUth-CDI 2 production. This is in line with Rowland et al. (Reference Rowland, Krajewski, Meints, Łuniewska, Kochańska and Alcock2022) who found that the reverse SES effect for infants was far less prevalent in the gesture scale across ten cross-linguistic CDI datasets. Gestures may be more easily observable and do not require as much interpretation, making them less susceptible to reporting biases. Unlike gestures, word production still requires a small amount of interpretation because caregivers are instructed to also check speaks for vocabulary items when their child produces synonyms or production errors. In addition, caregivers may over-report their child’s vocabulary if they think larger vocabularies are desirable. This social stigma is less prominent for children’s gesture repertoires which makes the gesture scale less susceptible to caregiver reporting biases. Lastly, we found a positive effect of maternal education on the lab-administered PPVT-III-NL during Wave 2. This result is in line with previous studies finding that a higher SES, often measured through maternal education, correlates with larger vocabularies (Hoff, Reference Hoff2003; Huttenlocher et al., Reference Huttenlocher, Waterfall, Vasilyeva, Vevea and Hedges2010). This could suggest that an advantage of maternal education only emerges later in children’s development, although an effect on infants’ vocabularies could be obscured by caregiver reporting biases or floor effects.
An alternative explanation to consider is that the reverse SES effect found for caregiver-reported comprehension and production during infancy is real. That would imply that infants of lower SES families start out with larger vocabularies compared to infants of higher SES families. However, we believe that this explanation is unlikely. Previous studies assessing a range of language-related abilities in children, including language processing and early use of gestures, show that children of higher SES families tend to outperform children of lower SES families (e.g., Fernald et al., Reference Fernald, Marchman and Weisleder2013; Rowe & Goldin-Meadow, Reference Rowe and Goldin-Meadow2009). Although the SES difference could be explained by several different reasons, including differences in genetics or the environment, we believe that it is unlikely that the reverse-SES effect can be attributed to a real effect. Nevertheless, this recurrent finding across studies should not be dismissed without further thought, and future studies should examine SES differences in other language-related measures of infants.
Girls have an advantage over boys
The results show that girls have an advantage over boys on NYOUth-CDI 1 gestures and NYOUth-CDI 2 production. Previous studies have also frequently reported an advantage for girls using CDIs (Eriksson et al., Reference Eriksson, Marschik, Tulviste, Almgren, Pérez Pereira, Wehberg, Marjanovič-Umek, Gayraud, Kovacevic and Gallego2012; Feldman et al., Reference Feldman, Dale, Campbell, Colborn, Kurs-Lasky, Rockette and Paradise2005; Frank et al., Reference Frank, Braginsky, Yurovsky and Marchman2021; Reese & Read, Reference Reese and Read2000). The results of our study suggest that the gender difference could start with a difference in children’s gesture repertoires during infancy. Infants’ gestures are known to influence children’s later vocabularies (see Brooks & Meltzoff, Reference Brooks and Meltzoff2008; Colonnesi et al., Reference Colonnesi, Stams, Koster and Noom2010; Rowe & Goldin-Meadow, Reference Rowe and Goldin-Meadow2009). Recently, Germain et al. (Reference Germain, Gonzalez-Barrero and Byers-Heinlein2022) also showed that 14-month-old girls produce more gesture types than boys using caregiver reports. Our results add to this finding by showing that a difference in gestures between boys and girls is already present before their first birthday. The gesture scale could be the only scale that shows enough variability across infants, resulting in sufficient variation to detect the gender effect early on. Our findings are also in line with the hypothesis that gender differences are more prevalent in vocabulary production than vocabulary comprehension (see Bornstein et al., Reference Bornstein, Haynes and Painter1998; Feldman et al., Reference Feldman, Dale, Campbell, Colborn, Kurs-Lasky, Rockette and Paradise2005; Frank et al., Reference Frank, Braginsky, Yurovsky and Marchman2021; Qi et al., Reference Qi, Kaiser, Milan, Yzquierdo and Hancock2003). This could explain the absence of a significant gender effect on the PPVT-III-NL. Another possible explanation for this is that the gender effect on NYOUth-CDIs is the result of a reporting bias. Caregivers could expect that girls are more verbal than boys, influencing how they fill out the vocabulary checklist. However, we suspect that this is unlikely because we also found a significant gender effect on word production during Wave 2. Caregiver reports on word production (rather than comprehension) and toddlers (rather than infants) are less susceptible to reporting biases. Frank et al. (Reference Frank, Braginsky, Yurovsky and Marchman2021) also showed that cross-linguistically, the advantage for girls is more prominent in caregiver reports of word production than word comprehension. This suggests that girls truly have an advantage over boys –– at least in their expressive vocabularies.
No effects of gestational duration and birth weight
We did not find any effects of gestational duration or birth weight on children’s vocabularies in this non-clinical sample. This does not support earlier findings that preterm infants are at risk of having smaller vocabularies later in life than full-term infants (e.g., Foster-Cohen et al., Reference Foster-Cohen, Edgin, Champion and Woodward2007; Guarini et al., Reference Guarini, Sansavini, Fabbri, Alessandroni, Faldella and Karmiloff-Smith2009; Sansavini et al., Reference Sansavini, Guarini, Savini, Broccoli, Justice, Alessandroni and Faldella2011). Nevertheless, some studies suggest that only extremely preterm children (under 28 weeks) and/or children of very low birth weight (under 1500 g) have language delays (Barre et al., Reference Barre, Morgan, Doyle and Anderson2011; Kern & Gayraud, Reference Kern and Gayraud2007). None of the children included in our sample fall under those criteria. Therefore, it is possible that we did not find any differences because gestational duration and birth weight predominantly affect the more extreme cases. Future studies should examine the effects of gestational duration and birth weight in longitudinal samples that include very to extremely preterm children and/or children of very low birth weights – but the results of our study do not provide evidence for the generalisability of these predictors across large, healthy samples.
Multilinguals know fewer words than monolinguals
We lastly examined the effects of children in the Netherlands growing up with more than one language. The results show that monolingual toddlers have larger receptive vocabularies measured with the PPVT-III-NL, but not larger productive vocabularies measured with the NYOUth-CDIs. Given the fact that multilingual toddlers are not exposed to as much Dutch language input as their monolingual peers, and vocabulary development is heavily influenced by the quantity and quality of exposure (Hoff, Reference Hoff2003), we expected multilingual toddlers to have smaller vocabularies when measuring only one of their languages (in line with Blom et al., Reference Blom, Boerma, Bosma, Cornips, van den Heuij and Timmermeister2020; De Houwer et al., Reference De Houwer, Bornstein and Putnick2014; Hoff et al., Reference Hoff, Core, Place, Rumiche, Señor and Parra2012). In the NYOUth-CDIs, caregivers were instructed to also check speaks on vocabulary items when their child produces a synonym. Arguably, these instructions yielded large variability in how multilingual caregivers filled out the checklists. It is plausible that some multilingual caregivers also accepted translations for vocabulary items which could explain the absence of a negative effect of multilingualism on the NYOUth-CDIs. Our sample could also have been too homogeneous because all caregivers who participated in the YOUth cohort study were required to be able to fill out Dutch questionnaires to participate. This resulted in a small number of multilingual children in our sample that may not have been sufficient to detect an effect of multilingualism on caregiver reports, especially given the potential variability in how multilingual caregivers filled out the reports. Lastly, we found no effect of multilingual input on gestures, which is in line with a recent study that did not find an effect of multilingualism on 14-month-old infants’ gestures measured with CDIs (Germain et al., Reference Germain, Gonzalez-Barrero and Byers-Heinlein2022). Even though infants’ gesture repertoires are an early indicator of their later vocabulary size, they are likely independent of specific language exposure and therefore not affected by multilingual language input. Importantly, whether multilingualism affects early vocabulary in one language seems dependent on the type of vocabulary measure (i.e., a lab measurement vs. caregiver report).
Limitations
Although we found some effects of maternal education in the expected directions based on previous studies using socio-demographically diverse samples (Feldman et al., Reference Feldman, Dollaghan, Campbell, Kurs-Lasky, Janosky and Paradise2000), our sample is rather homogeneous and overrepresents highly educated mothers. A lack of diversity makes SES differences less apparent. The results of our study suggest that caregiver reports of infants’ vocabulary comprehension and vocabulary production, but not gestures, are negatively affected by maternal education. According to previous studies, infants of lower SES may be using fewer gestures during caregiver-child interactions (Rowe & Goldin-Meadow, Reference Rowe and Goldin-Meadow2009). Although we did not examine gesture rates, we did not find an effect of maternal education on infants’ gesture repertories. Future longitudinal cohort studies with more diverse samples should re-evaluate whether gesture repertoires show any SES differences and how these may affect the predictive value of gestures in diverse samples.
We also want to draw attention to the large age range of children included in Wave 2. This was a decision made by the YOUth cohort study for reasons orthogonal to the present study. While we controlled for children’s ages in the statistical models, the large age range could have impacted the results. Some demographic effects may explain more variation in the first few years of life, but not at later ages. By grouping all children aged 2 to 5 together, we may have underestimated some of the demographic effects on vocabulary that become weaker predictors across development. In addition, many children in Wave 2 were too old for the N-CDI 2. This could have caused a ceiling effect on production measured by caregiver reports for toddlers. In order to address this possibility, we also fitted all models excluding those children (see OSF), which did not change the results. This tentatively suggests that we can use N-CDIs while sampling a large age range of young children, which is beneficial to longitudinal cohort studies with repeated measurements.
Conclusions
The results of our longitudinal study including over four hundred Dutch children suggest that the effects of widely discussed demographic predictors on children’s vocabularies are dependent on children’s ages and the type of vocabulary task being used. Except for age, none of the predictors remained constant across development or the different measurement tasks. We found a disadvantage for males in infants’ gestures and toddlers’ word production. We found a negative effect of maternal education on infants’ caregiver-reported vocabulary, but a positive effect on the lab-administered receptive vocabulary task. Lastly, we found a negative effect of multilingualism – but only for the lab-administered receptive vocabulary task. The results imply that research findings can be influenced by children’s age or the vocabulary task being used in a specific study. This is important to consider for child language researchers in future studies who aim to explain variation in vocabulary development. One advantage of cohort studies with repeated measurements is to gain better insights into which predictors have temporary or weak effects on development. Given our results, we would recommend researchers to sample diverse groups of children – including a broad age range – and use more than one vocabulary outcome when examining predictors of individual variation to gain a more comprehensive understanding of the duration, directionality, and magnitude of the effects on variation in children’s vocabulary across development. Predictors can differentially affect children’s gesture development during infancy and their expressive and comprehensive vocabularies across development. We also found that effects can shift over time, at least from infancy to toddlerhood. This corroborates that we should examine large, longitudinal samples cross-linguistically to determine the generalisability and robustness of key predictors of children’s language development.
Acknowledgements
We are grateful to all families who participate in the YOUth study. YOUth is funded through the Gravitation program of the Dutch Ministry of Education, Culture, and Science and the Netherlands Organization for Scientific Research (NWO grant number 024.001.003). A complete listing of the study investigators and study management can be found at https://www.uu.nl/en/research/youth-cohort-study/about-us/who-isinvolved. YOUth investigators and management designed and implemented the study and/or provided data but did not necessarily participate in the analysis or writing of this report. This manuscript reflects the views of the authors and may not reflect the opinions or views of the YOUth study investigators or YOUth management.
Data statement
YOUth is a longitudinal study that aims to produce and safely store FAIR and high-quality data. The data can be accessed for both use and verification purposes upon request (see https://www.uu.nl/en/research/youth-cohort-study/data-access). The R script and other materials can be found online: https://osf.io/vj72c/.
Competing interest
The authors declare none.
Open access
