3.4.1 Assessment Form

The assessment form includes all subscales (i.e., Vocabulary, Rhyme Detection, etc.) and their associated test items. For each subscale there is a brief guide containing a description of the subscale, information about which materials are to be used, and a description of how to perform the assessment task with the child (e.g., rhyming task or language comprehension task) and how it is scored. The child’s response can be manually recorded on paper and subsequently entered in the IT system, or it can be registered directly in the IT system. This requires that the assessor has a computer or tablet at hand; feedback from users indicates that some experience is required to use a computer or tablet for registering the child’s response while at the same time keeping a full focus on the child in order to keep the child engaged in the tasks.

3.4.2 Assessment of the Child on Each Subscale

Four of the assessment subscales require the use of materials for the child to look at; that is, pictures or letters. These subscales are Language Comprehension, Productive Vocabulary, Rhyme Detection, and Letter Knowledge. Examples of these are shown in Figures 1–3.

Figure 1 Example of pictures for Language Comprehension. The child has to point to one of four pictures corresponding to a sentence uttered by the administrator

Figure 2 Example of word elicitation pictures for Productive Vocabulary assessment. Contrasting pictures are used to elicit less tangible or illustratable words such as “drinking”: The woman is eating. What is the man doing?

Figure 3 Example of pictures for a Rhyme Detection item using the three words hus [huː^ʔs] (“house”), mus [muː^ʔs] (“mouse”), and sol [soː^ʔl] (“sun”).

For Language Comprehension, the child is asked to choose one of four pictures corresponding to a sentence uttered by the administrator; for example: Where is the boy being chased by the dog? (see Figure 1).

For Productive Vocabulary, the child is asked to label an element of a picture. Two-thirds of the words are nouns, two words are adverbs (place and time), and the remaining words are approximately equal numbers of adjectives and verbs. For nouns, the child is simply asked What is this? For other categories, where it might not be obvious which of the multiple elements of the picture is to be named, two contrasting pictures are given to help elicit the relevant word; for example: The woman is eating. What is the man doing? (see Figure 2).

For Rhyme Detection, three pictures are shown, and the child is asked to point to the two pictures of objects whose names are words that rhyme. The administrator names the three pictures and asks the child to point to “the two things that sound alike; those that rhyme” (see Figure 3).

For assessment of Letter Knowledge, twelve capital letters are shown in a table, and the child is asked to name each of the letters: I C J F E G O S A L D K. There are twenty-nine letters in the Danish alphabet. Testing all letters would be time consuming, and therefore a subset of letters was identified and used in a previous version of the instrument and carried over to the present version.

The Deletion task has three blocked levels of difficulty that require the child to say what is left when deleting (1) one word-part of a compound (e.g., Say snowman without man), (2) deleting a syllable, (e.g., Say window without dow), or (3) deleting a sound (e.g., Say cold without khh). The deleted part could be either at the beginning or the end of the word. The remaining part of the word would always make a word in its own right (such as snow, win, and old in the English examples).

For the assessment of Print Awareness, the assessor uses a picture book of their own choice (not provided with the test). The book must have both text and pictures. The assessor asks the children sixteen questions related to book reading, generally increasing in difficulty, starting with handing the book to the child and asking What is this? Later questions pertain to where to start reading on a page and pointing out examples of a word, the last word on a page, or the last letter in a given word.

For assessment of the child’s communicative skills, the child is not directly involved. An educator who is familiar with the child from day-to-day care rates how frequently (never, seldom, often, or always) a child would give a particular communicative response in a hypothetical communicative situation. For example, A child or an adult talks to the child on the playground. The child reacts by answering the other child/the adult. More advanced examples include You are talking to the child, but the child seems to not understand. The child asks clarifying questions and At home, the child has heard their parents talk about an experience. The child explains to you what the parents were talking about.

3.5.1 The Results Report

Based on the number of items responded to correctly and the age and gender of the child, the child’s performance can be compared with norms (available to potential IT platform contractors upon request from the Danish Ministry of Education). The resulting report contains two overall summary scores in terms of percentiles: a language summary score and a preliteracy summary score. The language summary score is generated by standardizing the language subscale scores (Vocabulary, Language Comprehension, and Communication), averaging the scores, and subsequently converting them to percentile scores (see Section 3.7.4). Likewise, the preliteracy subscale scores (Rhyme, Print Awareness, Letter Naming, and Deletion) were standardized and averaged to generate the preliteracy summary score. Note, however, that the administration of preliteracy subscales depends on age (see Table 2). Because the only subscale administered to 3-year-olds is Rhyme, no preliteracy summary score is generated for 3-year-olds. The mean score for each of the summary scores is referenced against the norms for the child’s age and gender to produce a language summary percentile score and a preliteracy summary percentile score.

The report also shows the percentile scores of each of the subscales completed by the child to help generate a “language profile” for each child; that is, a profile showing which dimensions of language or preliteracy skills are strong or weak in the child. The two summary scores and the language profile scores also indicate the age- and gender-specific 5th and 15th percentiles. An example of a results report is given in Figure 4 for a fictional child.

Figure 4 Example results report showing percentile scores for the two summary scores and some of the individual subscale scores for a fictional child

The 5th and 15th percentile cut-off points were set to facilitate the best use of later resources. The requirements of later intervention based on the results of individual children were operationalized into three levels of skill development for each subscale as well as for the summary scores for language and preliteracy – (1) “low” (percentiles 1–5), (2) “at-risk” (percentiles 6–15), and (3) “typical” (percentiles 16–100) – for determining the language support and educational needs of each child. Rather than outputting the categories low, at-risk, and typical for each child, the report assigns a corresponding intervention category; namely, “Special intervention” (percentiles 1–5), “Focused intervention” (percentiles 6–15), and “General intervention” (percentiles 16–100).

Although using percentiles has the advantage of yielding comparable scores across ages, we have also consistently found that the concept of percentiles is difficult to communicate to users who are ECEC professionals but not necessarily used to manipulating numbers. However, it is usually possible to explain the concept by beginning with the concept of halves, the upper half and the lower half of scores, then progressing to the concepts of quartiles, deciles, and percentiles.

3.5.2 Interpretation of Results

We recommend in the manual that the lowest-scoring 5 percent of the children should be referred to a professional speech therapist in order to conduct a complete professional pathological speech assessment and potentially to receive specialized intervention. This cut-off point corresponds to the existing mean referral rate to speech therapists across municipalities in Denmark for 3-year-olds at 4.3 percent (Reference ReuschReusch, 2006). Children with scores between 6th and 15th percentile are referred to the ECEC center for focused intervention and a later reevaluation for language delay. It is not within the scope of the instrument and manual to specify appropriate interventions, but the manual does give examples of how to target the specific weaknesses of a child according to their profile (e.g., low scores on Rhyme or Vocabulary). The percentile scores for each subscale may help guide educators in best supporting the child by identifying specific domains within language and preliteracy for which the child needs the most support. The 15th percentile cut-off point is based on findings from studies of language delay, which have reported prevalence rates ranging from 2.3 percent to 19 percent (Reference Nelson, Nygren, Walker and PanoschaNelson et al., 2006). The remaining 85 percent of the children can be regarded as having language development within the normal range. Thus these children are recommended to continue the general or usual instructional activities in the ECEC center.

3.5.3 Interpretation of Results for Bilingual and Multilingual Children

The LA 3–6 was developed specifically to provide a norm-referenced assessment of Danish language development in children learning Danish only. However, a large and growing proportion of children in Denmark speak one or more languages at home in addition to Danish. Therefore, assessing only Danish-language skills does not provide a full picture of the overall language skills of children learning multiple languages. Although it would be highly useful to establish norms for bilingual children – either for both/all of a child’s languages or just the majority language – this is not feasible, because no one set of norms would be sufficient, as discussed in Section 1.6.

With the given limitations of only assessing the majority language – Danish – in bilingual and multilingual children, it is important to give extra attention to the interpretation of their assessment results, as their development is highly variable. First, even though more bilingual than monolingual children score below the 15th percentile cut-off point, a large number of bilingual children do not. For these children, there is little reason to suppose that they have any problems with language development, because they score within the typical language development range for monolingual children even in their second language. Second, educators are advised to base their evaluation of the child’s language development more on preliteracy subscales than on oral language subscales. This is because bilingual children have been found to resemble or even exceed monolingual children with respect to preliteracy skills such as rhyming, deletion, and letter knowledge, presumably because such skills translate better between languages than do oral language skills and/or because the bilingual experience facilitates better metalinguistic and phonological awareness (Reference Hammer, Hoff and UchikoshiHammer et al., 2014; Reference Højen, Bleses and JensenHøjen et al., 2019; Reference Højen, Hoff and BlesesHøjen et al., 2021). Therefore, low scores on, for example, Vocabulary and Language Comprehension should raise less concern if the child scores highly on, for example, Rhyme and Deletion than if the child generally scores low on both oral language and preliteracy subscales. In the former case, it is likely that the child has had too little exposure to Danish to score within the typical range for oral language skills. Educators are encouraged to seek information from the child’s parents about when the child began to be regularly exposed to Danish and the proportion of Danish input in order to best interpret low scores in the language assessment.

3.5.4 Supplementary Questions for Parents of Bilingual and Multilingual Children

To provide additional help with the interpretation of the assessment results of these children, we compiled ten language-independent questions in a questionnaire inspired by Reference RestrepoRestrepo (1998). The questions are primarily for parents of bilingual children with low percentile scores in the language assessment. In these cases, it is crucial, but difficult, to determine whether the low scores are due to general language learning difficulties or due to the child being in the early phase of Danish second-language acquisition. The questions for parents are given in Table 4.

As the wording shows, the questions must be answered with reference to the child’s age, and as such, the questions require some insight from the parents with respect to normal language development. The questionnaire was developed as an addendum to LA 3–6 and has not been validated. Therefore, it should be seen as a tool to help interpret the results of bilingual children when they receive low scores in the LA 3–6 proper, along with any other information available on the child’s mastery of the first language. The questions may be answered for just the minority language spoken at home or for both the minority and the majority language.

3.6.1 Sample

The norming study of the LA 3–6 used baseline (pretest) data from a two-year randomized controlled trial study of the effectiveness of an intervention (The Daycare of the Future, Reference Nielsen, Jensen and BlesesNielsen et al., 2017) targeting language, cognitive skills, and social emotional skills in children aged 0–6 years, funded by the Danish Ministry for Children and Social Affairs. Fourteen municipalities participated. They were selected from a pool of interested municipalities to represent geographically disparate parts of Denmark as well as urban profiles and rural profiles and were thus representative of Danish municipalities. A total of 442 home-based care units and 145 ECEC centers were included, encompassing a total of 10,948 children aged between 0 and 6 years, but only children in the age range 34–71 months who had Danish as their first language were included in the norming study of LA 3–6. The LA 3–6 was administered to children in the included ECEC centers by educational staff over a six-week period in August/September 2014 using the professional online documentation system Rambøll Results. The educational staff had already administered the prior version of the language assessment instrument as part of the national language program but received local introductions to the changes introduced in LA 3–6.

The sample was representative with respect to age, gender, and the distribution of maternal educational background. The mean years of formal education for the mothers of the 3–5-year-olds in the sample was 14.6 years (SD = 2.8) for the year 2014, which was virtually the same as for the general population of mothers (14.6 years, SD = 2.5), though just for mothers of 4–5-year-olds and for the year 2012. Table 5 gives more detailed information about the sample.

An important note is that, as is often the problem with large-sample research conducted in the broader population, there is missing data for both background measures and outcome measures. Therefore, in the description of subscale scores in the following sections, the N for each subscale is given.

3.6.2 Descriptive Statistics for All Subscales

Descriptive statistics for the summary scores of the subscales of LA 3–6 for each age group are shown in Tables 5–18 and Figures 5–11. For data protection reasons, minimum and maximum values have been obscured if certain scores have been obtained by fewer than three individuals. Likewise, histogram bins most often encompass several scores in order not to show scores obtained by fewer than three individuals. To examine the robustness of the results, we randomly divided the recruited children into two groups for each age group, and we report the split sample results. Skewness (degree of asymmetry of the distributions) and kurtosis (degree of outliers/heavy-tailed distribution) are indicated for both whole and split samples and statistically tested for whole samples.

Table 6 and Figure 5 present descriptive statistics and histograms for the Vocabulary subscale. Because there is a change of items across ages for this subscale (Section 3.3), the scores are not directly comparable across the age groups. The histograms are more informative, as they demonstrate that variability in this skill is appropriately captured, the degree of skewness is visualized, and any floor or ceiling effects are detected. Although there is a modest degree of skew – significant right skew for age group 3, significant left skew for age groups 4 and 5 – neither floor nor ceiling effects are apparent.

Figure 5 Distribution of scores on the Vocabulary subscale for children aged 3, 4, and 5 years. Age-dependent items

Table 7 and Figure 6 present descriptive statistics and histograms for the Language Comprehension subscale. Like Vocabulary, the Language Comprehension items change somewhat between ages, and consequently the mean levels cannot be compared. The Language Comprehension histograms show a somewhat higher degree of skewness than for Vocabulary, all left-skewed, and a modest ceiling effect that is comparable across the age groups. The left-skewed distribution is in line with our intention, along with obtaining valid measures across the full range of performance, to obtain special sensitivity at the low end of the scale – or in other words, “err on the side of left-skewness” – if a normal distribution could not be obtained.

Figure 6 Distribution of scores on the comprehension subscale for children aged 3, 4, and 5 years. Age-dependent items

Table 8 and Figure 7 present descriptive statistics and histograms for the Communication subscale. For this subscale, the items do not change with age, and thus the means in Table 7 and the relative “peaks” in Figure 7 demonstrate the degree of change with age. At ages 3 and 4, there is only a modest degree of skewness and little evidence for a ceiling effect. However, at age 5 there is a strong ceiling effect and a corresponding high degree of left-skewness, which need to be taken into account when interpreting scores on this subscale at that age. But again, our priority is sensitivity to skill differences at the low end of scores.

Figure 7 Distribution of scores on the Communication subscale for children aged 3, 4, and 5 years

Table 9 and Figure 8 present descriptive statistics and histograms for the Rhyme subscale. As the items on this subscale do not change with age, the means demonstrate growth in this ability. As rhyme detection is known to reach mastery level in the preschool years in literate societies, there is an expected substantial ceiling effect at age 4 and an even stronger ceiling effect at age 5. Thus, at ages 4 and 5, this subscale is best interpreted as a measure of slow rhyme development.

Figure 8 Distribution of scores on the Rhyme subscale for children aged 3, 4, and 5 years

Table 10 and Figure 9 present descriptive statistics and histograms for the Print Awareness subscale. This subscale is administered only to 4- and 5-year olds. The items do not change with age, and thus the table and figure demonstrate developmental growth. At age 4 there is no evidence for skewness or a ceiling effect, but by age 5 both are apparent, reflecting major advances in this skill across the population at this age.

Figure 9 Distribution of scores on the Print Awareness subscale for children aged 4 and 5 years

Table 11 and Figure 10 present descriptive statistics and a histogram for Letter Knowledge. This subscale is administered only at age 5. Letter knowledge is highly variable in 5-year-old Danish children, with a substantial proportion knowing all letters presented and an almost equally large proportion scoring 0. The shape of the histogram is unusual with its downward trend in the percentage of children as the scores increase and with a dramatic increase in percentage for the maximum score. A similar pattern was seen in an earlier version of the assessment material, with an approximate doubling of the percentage of children who obtained the maximum score of 12 compared to the percentage of children scoring 11. This pattern of scores suggests that letter knowledge may often be learned “in a burst” from modest knowledge to relatively complete knowledge.

Figure 10 Distribution of scores on the Letter Knowledge subscale for children aged 5 years

Table 12 and Figure 11 present descriptive statistics and a histogram for the Deletion subscale, which is administered only to 5-year-olds. For this subscale, there is a substantial floor effect but also variability, suggesting that this skill is still at an early stage of development at age 5. Recall that the test comprises word-part deletion, syllable deletion, and sound deletion, the latter being the hardest. In a future revision, as well as in adaptations for other languages, floor effects could be minimized by omitting the sound deletion subsection of the test.

Figure 11 Distribution of scores on the Deletion subscale for children aged 5 years

3.6.3 Relation of Subscale Scores to Gender and Family Socioeconomic Status

It is well-documented in research that the rate of language development varies as a function of both gender and SES, and this pattern was also found in our sample.

Table 13 shows the gender-related results, gender effect sizes (Cohen’s d, indicating the percentage of a standard deviation for the gender difference), and significance level at age 3. Girls scored significantly higher than boys on three out of four subscales. Only for Rhyme were the scores for boys and girls virtually identical.

At 4 years old, the girls scored significantly higher than the boys on all of the subscales with the largest effect size for Communication and the smallest effect size for Rhyme; see Table 14.

At 5 years old, the girls again scored significantly higher than the boys on all of the subscales with the largest effect size for Letter Knowledge and the smallest effect size for the Deletion subscale; see Table 15.

In summary, at each age assessed here, girls are, on average, more advanced in their language and preliteracy development than boys. The gender difference is least pronounced for skills related to phonological awareness, namely rhyme skills and word-part deletion skills, but it is still significant, with the exception of rhyme skills, in 3-year-olds. Given that boys seem to catch up with girls at later ages (Reference WallentinWallentin, 2009), we have chosen to establish separate norms for boys and girls in order not to over-identify boys (false positives, assigned to either special intervention or focused intervention) and under-identify girls (false negatives, assigned to general intervention) with language impairment or delayed language development.

To assess the degree to which language and preliteracy scores were related to parental SES, we correlated scores with maternal education, paternal education, and household income for each of the three ages. The results are shown in Tables 16, 17, and 18, one table for each age. This analysis serves as a tentative validation of the subtests, given that early language development is typically found to be related to family SES (Reference Fernald, Marchman and WeislederFernald et al., 2013; Reference HoffHoff, 2006; Reference Huttenlocher, Waterfall and VasilyevaHuttenlocher et al., 2010). These relations have also been found to hold for Danish using an earlier version of the present instrument on a different sample (Reference Højen, Hoff and BlesesHøjen et al., 2021).

At all ages, each of the SES variables was significantly correlated with each of the language and preliteracy outcomes. Among the SES variables, the strongest correlations were with maternal education and paternal education and the weakest was with household income; the confidence intervals indicate that the correlation with household income was significantly weaker in most cases.

3.7.1 Internal Consistency

In spite of pilot testing of the subscales during development, they are unlikely to be perfect. With the much larger norming sample compared to the pilot samples, we can get a better evaluation of internal consistency. To determine internal consistency, Cronbach’s alpha coefficients for the subscales for each age group are shown in Table 19 and, in addition, item-total correlations, which indicate the correlation between an individual item and the total score without that item. The alpha values were generally high, indicating high internal consistency. The exception to this is the Comprehension subscale for age 4 and age 5, with alphas of 0.67 and 0.68, respectively. Accordingly, the item-total correlation ranges for that scale are relatively low, but no specific items seem to be dragging the alphas down. However, in future revisions, items from the low end of the range of item-total correlations will have to be replaced. This, of course, will require substantial pilot testing to be sure that the substituted items contribute to higher internal consistency.

3.7.2 Construct Validity

To evaluate overall construct validity for the entire instrument, we first calculated zero-order correlations between all subscales separately for each age group. As outlined in Section 3.5.1, the results report summarizes the subscales in two summary scores; that is, oral language skills (Vocabulary, Comprehension, Communication) and preliteracy skills (Rhyme, Print Awareness, Letter Knowledge, Deletion). The zero-order correlations also help evaluate the extent to which these conceptually defined summary scores were empirically consistent; as a second step in this evaluation, we estimated structural equation models (confirmatory factor analyses), again separately for each age.

Tables 20–22 show the zero-order correlations between all subscales. It is notable that Vocabulary generally has the highest correlations with the widest range of other measures. Specifically, the correlation between Vocabulary and Comprehension was significantly stronger than other correlations, judging by the confidence intervals of correlation coefficients. This suggests that vocabulary is a core skill that is broadly indicative of general language development and should be included in all language assessment instruments that target broad language skills and not just a specific language skill such as phonological awareness.

Focusing on the two summary score constructs, oral language (Comprehension, Vocabulary, Communication) and preliteracy skills (Rhyme, Print Awareness, Letter Naming, Deletion), the correlations between subscales within each construct were generally of similar magnitudes to the correlations between subscales across the two constructs, although slightly higher on average within than across constructs. This raises a question about how separate these theoretically defined constructs really are.

To further examine the extent to which the oral language and preliteracy constructs represent separate factors, we estimated structural equation models for 4-year-olds and 5-year-olds separately. Note that for 3-year-olds, there is just one preliteracy subscale, Rhyme. Also note that the number of subscales administered to 4- and 5-year-olds differs. The models were estimated in Stata using maximum likelihood with missing values and standardized coefficients and values.

Figure 12 shows structural equation models of how the subscales relate to the oral language construct and the preliteracy construct for 4-year-olds and 5-year-olds. Post-estimation of the model fit indicated good fits: Significant chi-square fit indices for both 4-year-olds (χ² = 68.7, p < 0.001) and 5-year-olds (χ² = 13.4, p < 0.001), which is not uncommon in larger samples (Reference Bentler and BonettBentler & Bonett, 1980), but a low root mean squared error of approximation (4-year-olds: 0.024; 5-year-olds: 0.037) and high comparative fit indices (4-year-olds: 0.995; 5-year-olds: 0.982) and high Tucker-Lewis indices (4-year-olds: 0.995; 5-year-olds: 0.982). The subscales were loaded significantly on each construct for both ages (ps < 0.001). There was significant positive covariance between the two summary score constructs (p < 0.001).

Figure 12 Structural equation model of the loadings of subscales on the two summary score constructs for 4-year-olds, left, and for 5-year-olds, right. Goodness of fit for each model was R = 0.84 (4-year-olds) and R = 0.87 (5-year-olds)

Summing up, the good model fits suggest that the conceptually defined constructs are reasonable (though also with a high degree of covariance between the constructs), whereas the similarity of the strengths of pairwise correlations within and across the two constructs suggests that the conceptually defined constructs may not fully reflect genuinely distinct constructs, with the exception of generally strong correlations between the language subscales, Vocabulary and Comprehension. This, in turn, suggests that a total summary score might work just as well and that educators should look at scores on individual subscales to determine areas of language development that need support in that particular child.

3.7.3 External Validity

As there is no gold standard language measure in Denmark, it was not possible to perform an external validation study for all subscales. The external validity of LA 3–6 could only be compared with Danish adaptations of the Peabody Picture Vocabulary Test, 4th edition (PPVT-4; Reference Dunn and DunnDunn & Dunn, 2007) and the Expressive Vocabulary Test, 2nd edition (EVT-2; Reference WilliamsWilliams, 2007). A sample of 133 3–6-year-old children, approximately equally distributed across age groups, participated in this study. Correlations between the instruments are presented in Table 23. Five children did not complete the external tests.

The concurrent correlations of LA 3–6 subscales with PPVT-4 ranged from 0.33 to 0.57 and were highest for Vocabulary, Comprehension, and Print Awareness and lowest for Rhyme. Correlations with the EVT-2 ranged from 0.10 to 0.42 and were highest for Vocabulary and Comprehension. The correlations with the preliteracy subscales were weak. It is noteworthy that the LA 3–6 Productive Vocabulary subscale correlated more strongly with the PPVT-4, which is a receptive vocabulary test, than with the EVT-2, which is a productive vocabulary test. One possible reason for this is a higher tendency for ceiling effects for EVT-2 than for PPVT-4. One might also speculate that the PPVT-4 has higher reliability than the EVT-2, but this is probably not the case; for the English versions, test-retest reliability is slightly higher for the EVT-2 (0.94–0.97) than for the PPVT-4 (0.87–0.93) (Pearson, 2007).

3.7.4 Procedure for Establishing Percentile Scores

The results of the language assessment are reported as percentile scores. The instrument assigns percentile scores on each subscale to children based on their raw score. For this purpose, we developed percentile norms from the representative norming sample. The percentile score corresponding to a raw score value x of a child at age a (in months) is based on all children of the same gender and with an age of +/− 4 months and tested with the same material. For these children, an indicator variable is created indicating whether the child has a score of x or less. Then a logistic regression is fitted in these children with age as the only covariate, and the prediction from this model is recorded. The percentile score is then defined as the average between the predictions obtained for the values x and x−1. Percentile scores above 99 are set to 99. As noted in Section 3.5.1, the subscale raw scores were standardized and averaged to generate the oral language summary score and the preliteracy summary score. These summary scores were, in turn, converted to percentile scores.

The procedure of considering the ages +/− 4 months means that we “borrow data strength” from ages just above and below the age in question. The purpose was to iron out natural fluctuations that cause anomalies in the normative development. For example, it happened regularly that a particular age group, such as 39-month-olds, would have a mean score on a subscale that was lower than the mean score of 40-month-olds. We have no reason to assume a drop in skills from one month to the next, although conceivable; it is much more likely that such a drop occurred in the norming study through sampling error – for example, if our group of 39-month-olds had slightly higher skills than our 40-month-olds and therefore obtained higher scores. By considering neighboring ages using the above procedure, such bumps in age-score relations were ironed out in most cases. In a few cases, however, anomalies persisted, meaning that, for example, a 40-month-old could obtain a higher percentile score than a 39-month-old from the same raw score. In future updates of the LA 3–6, we plan to manually eradicate such anomalies, as is also done for LA 2 (see Section 4.9.5), on the assumption that language development does not go backwards – at least not in predictable steps, let alone in predictable steps for predictable monthly ages.

4.4.1 Development of the Receptive Vocabulary Test

As mentioned, we adapted an existing American computer-based test of receptive vocabulary (the CCT; Reference Friend and KeplingerFriend & Keplinger, 2008; Reference Friend, Schmitt and SimpsonFriend et al., 2012). In this test, children are shown two pictures at a time on a touch screen – for example, a pacifier and a teddy bear – and an adult asks the child to touch one of them; for example, the pacifier. In our adaptation of the CCT, the administrator – with live voice – also asks the child to touch or point to one of two pictures. But our adaptation differs in several ways from the CCT.

An initial change compared to the original CCT concerned stimulus presentation and child response mode. In the original CCT, the pictures are onscreen, and responses are recorded automatically when the child touches a picture within a specified frame on the touch screen. In our adaptation, children point to a picture in a booklet showing the target picture and the foil side by side. This change was in part due to the desire of the commissioning body, the Ministry for Children and Education, for the test situation to resemble shared book reading in order to minimize stress or insecurity in the child. In addition, in a previous Danish iPad-based adaptation of the CCT, which the authors made for a randomized controlled trial of language intervention in 1–2-year-olds, children were reported to often have difficulties touching the pictures illustrated in such a way that the iPad successfully recorded the response. Observed errors in touch included sliding, touching with multiple fingers, or touching too hard or too long, all of which could result in failure to record the response or even shutting down the app.

The above difficulties can be avoided by using a picture book. However, the use of pointing to a picture in a book as a task response has its own challenges. It is not always obvious whether the child has pointed or not. A child may shift their gaze to a picture and even say “There!” without clearly pointing their finger. Our decision, to minimize judgment by the educators, was to instruct them to only record a response if the child clearly points with their hand or finger. The instructions also give suggestions as to how to elicit this type of response from the child.

A second major change from the CCT was to develop separate but overlapping difficulty levels for children under 2 years (i.e., 22–23 months), children aged 2–2½ years (24–29 months), and children over 2½ years (30–34 months). One reason for this was to avoid floor effects in the youngest children and ceiling effects in the oldest children, a particular concern given the expanded age range compared to the CCT, by adding easier and harder words. We also wanted to keep the test relatively short so as not to exceed the general attention span of toddlers. This precluded the inclusion of several additional words to a fixed list for all children as a means to avoid floor and ceiling effects with the expanded age range.

The words chosen for the test were, for a start, the Danish equivalents of the words of the original CCT. To expand the word list with easier as well as harder words, the norms for words of the Danish receptive CDI were used as a source of information on how widely specific words are known by Danish children in the age range of 22–34 months.

As in the original CCT, we aimed for three difficulty levels of words: easy, medium-hard, and hard (thirteen each within each age-dependent version). The words used in the three age-dependent versions overlapped so that, roughly, words that are hard in the 22–23 months version are used as medium-hard words in the 24–29 months version and as easy words in the 30–34 months version. Likewise, words that are medium-hard in the 22–23 months version are easy in the 24–29 months version and words that are hard in the 24–29 months version are medium-hard in the 30–34 months version. Easy words in the 22–23 months version and hard words in the 30–34 months version are not used in other versions. In each age-dependent version, easy, medium-hard, and hard words are mixed together so that children experience random variation in difficulty level, with the intention to minimize fatigue due to being unable to respond or finding the task too easy.

Easy words (e.g., doll) were expected to be familiar to most children in the age range, while hard words (e.g., parachute) were expected to be familiar to few children. These expectations were tested in pilot tests, and the set of items for each version was altered in order to increase or decrease the difficulty level with the aim of obtaining a close to normal distribution of scores. In addition, given the small size and unknown representativeness of the pilot study sample, the choice of words also took into account the above-mentioned CDI data, which were obtained from more than 6,000 children.

It should also be noted that the target words are elicited by showing pictures, and therefore the difficulty of an item is not solely based on how well the child knows a word but also on how well the child can relate the word to the picture. This may vary across children depending on their individual life experiences, but it is also inherently easier to unambiguously illustrate, for example, penguin than park because parks are more diverse and come in many more shapes and sizes than penguins do, or because a child may relate the word park to a specific park (while being unlikely to be familiar with individual penguins).

The composition of the thirty-nine items reflects natural vocabulary development such that in the youngest version, the majority of words are concrete nouns with only a few adjectives or verbs, whereas in the oldest version, the proportion of adjectives and verbs is higher.

4.4.2 Productive Vocabulary and Language Use Subscales

As mentioned in Section 4.3, the entire Productive Vocabulary and the (decontextualized) Language Use scales were taken from an existing instrument, namely the CDI: Educator (Reference Bleses, Jensen and HøjenBleses, Jensen, et al., 2018). The Productive Vocabulary scale is based on educator report of which words on a word list are known by the child according to the educator’s knowledge of the child from day-to-day interactions; that is, words that the educator believes they have heard the child produce. The CDI: Educator is a shortened educator version of the Danish version (Reference Bleses, Vach and SlottBleses et al., 2008a) of the MacArthur-Bates Communicative Development Inventories (Reference Fenson, Marchman and ThalFenson et al., 2007). The scale consists of seventy words, and the same scale is used independently of the child’s age (22–34 months). The words are semantically organized in categories, as shown in Table 26.

The CDI: Educator instrument has an additional scale consisting of five questions pertaining to decontextualized language use; for example, whether the child speaks about past or future events, speaks about people not present, or relates items to a person not present (for example, “Michael’s teddy!”). As noted earlier, this scale became part of LA 2 as well.

4.4.3 Development of the Communication Subscale

The Communication subscale was partly a new development and partly based on the Communication subscale for LA 3–6. The purpose of this scale is to capture a range of pragmatic communication skills, starting from the very basic skills of 2-year-old children who are either beginning to learn Danish as an additional language or who for other reasons have only very basic Danish communication skills. The scale items increase in difficulty to advanced communication skills.

For each item, a “scene” is set, and the educator is asked to use their experience to judge how likely it is that a child would typically produce a target response/behavior. Below are three examples of communicative skills ranging from basic to advanced.

Basic: You are talking to the child while changing a diaper or during a meal. Behavior: The child keeps eye contact with you for at least five seconds at a time.

Medium: The child is playing or eating with other children or adults. Behavior: The child uses words to comment on what is happening without being prompted.

Advanced: The child has heard you tell a story, and the child is retelling that story to another child. Behavior: The child tells the story with enough information and organization that the other child can understand the main points.

The subscale was pilot tested with 100 2-year-olds, and clear ceiling effects were present. Therefore, additional items from the Communication subscale of LA 3–6 were successfully added to increase difficulty (see Section 4.8).

4.5.1 Assessment Form

The assessment form includes all subscales and their associated test items. For each subscale there is a brief guide containing a description of the subscale, information about which materials are to be used, a description of how to perform the Receptive Vocabulary assessment with the child, and information about how to fill out the Productive Vocabulary, Language Use, and Communication subscales, as well as how all subscales are scored. The child’s response may be manually recorded on paper and subsequently entered into the IT system, or it can be registered directly in the IT system. This requires that the assessor has a computer or tablet at hand; feedback from users during pilot tests indicates that some experience is required to use a computer/tablet for registering the child’s response while at the same time keeping a full focus on the child to keep them engaged in the tasks and facilitate their participation while at the same time not helping them too much.

The language assessment instrument can be administered to all children in ECEC; that is, including bilingual children, although only Danish is assessed. For bilingual children, special care must be taken when interpreting the results (see Section 4.6.3). For children not in ECEC, only the test of receptive vocabulary can be administered because completion of the other three subscales requires ECEC educators’ experience with the child’s language use in the ECEC center.

4.5.2 Assessment of the Child on Each Subscale

The administration of the Receptive Vocabulary subscale is described here in a little more detail than the subscales of the LA 3–6 earlier in this Element (as well as for the remaining subscales of LA 2, which do not involve the child). This is because consistency and fidelity in the administration of a test is even more important in the case of very young children.

For the Receptive Vocabulary test, the child is shown two photos next to each other in each of thirty-nine trials. The child’s task is to point out, for example, a sock. The foil (a jacket in Figure 13) is a photo that has a level of difficulty similar to that of the target and additionally – as far as possible – shares some characteristic features (for example, being animals, pieces of clothing, items found in the home, items to put in the mouth, etc.), but priority was given to similarity of difficulty (based on scores for each item on CDI measures; Reference Bleses, Vach and SlottBleses et al., 2008a), which meant that a few items had word pairs with relatively low content similarity, such as balloon–bike.

Figure 13 Example of target (sock) and foil (jacket) used in the Receptive Vocabulary test

The educator is advised to ask for the target immediately when showing the two photos in order that the child does not get too interested in one particular photo, which could make it difficult to determine whether the child pointed based on the question or based on their interest in the photo.

The question for the child is phrased slightly differently depending on whether the target word is a noun, a verb, or an adjective. Nouns: Point to the sock! (Where is the sock?) Verbs: Point to the one that is jumping! (Who is jumping?) Adjectives: Point to the one that is tall! (Which one is tall?) The child may be asked up to three times in slightly different ways as deemed appropriate by the educator to help the child understand the question.

Before the thirty-nine test items, four practice items are administered to teach the child what the task is. If the child does not know what to do, the educator may help the child a single time during the practice by pointing and saying in a lively voice, for example, Theeeere is the baby! If the child correctly points to two out of the four practice items, the test proper is given. If not, the four practice items are repeated, and once again, the requirement is two correct practice items in order to start the test proper. If this requirement is not met, the test session is aborted and should be taken up again after a few days. If the child still does not pass the practice items, it is up to the educator decide when to try again.

The educator is advised to begin the assessment with a “calm” or “low-key” attitude. If educators are too lively from the beginning, they will run out of “headroom” to turn up the enthusiasm if the child begins to lose interest. Each item is scored depending on whether the child points to the target photo, the foil, or does not point. In practice, only correct responses are counted when tallying the score; the reason for technically distinguishing between the wrong response and no response is to enable the research team to perform follow-up analyses to determine whether wrong and no responses are differentially predictive of later skill development, which is a result that has been found for the CCT, from which this test was derived (Reference Hendrickson, Mitsven and Poulin-DuboisHendrickson et al., 2015; Reference Hendrickson, Poulin-Dubois and ZesigerHendrickson et al., 2017).

The three remaining subscales are in questionnaire form and are completed by an educator who knows the child from day-to-day interaction in the childcare facility. The child is not present, which means that the educators use their knowledge and memory of how the child talks.

The seventy-item Productive Vocabulary subscale is completed by the educators indicating for each word item whether they have heard the child produce the word. The word should be checked off as heard even if the child pronounces it incorrectly but, based on the context, clearly attempted to say the word. If, for example, the educator has heard the child say “lean!” while showing their hands after having washed them, it is likely from the context that the child is saying the word “clean” with a reduced consonant cluster.

A similar approach is used for (decontextualized) language use and for the Communication subscale. The educators use their experience and memory of the child’s language use and communication skills in specific concrete, but hypothetical, situations in the everyday (see also Sections 4.4.2 and 4.4.3).

4.6.1 The Results Report

Just as is the case for LA 3–6, the child’s performance on LA 2, based on the number of items responded to correctly, is to be compared with the age- and gender-specific norms. The resulting report contains a percentile score for each of the subscales just as for LA 3–6 (see Section 3.5.1). However, because the norming study included only about fifty children per gender per monthly age, percentile scores are given in 5-percentile ranges; giving percentile scores in steps of 1 would suggest a level of precision not warranted, even when data strength was “borrowed” from the ages just above and below the age in question (see Section 4.9.5).

In addition to percentile scores on the individual subscales, a summary percentile score is given. The summary score is generated by standardizing and averaging the individual subscale scores and comparing them to summary scores in the norming sample. A results report similar to that shown in Figure 4 (see Section 3.5.1) for LA 3–6 is given. Again, the summary score is mainly for administrative and benchmarking purposes, because a child may have specific difficulties in one area of language development but score in the normal range on the summary score. Therefore, educators should focus on the child’s results on the individual subscales to determine whether the child is lacking in specific areas of language development.

To generate a summary score across all subscales was a requirement of the Ministry for Children and Education, because many Danish municipalities intend to use such a score either for benchmarking or to track changes between cohorts of children in the face of, for example, new language support initiatives in ECEC centers or changes in the composition of families living in the municipality. It is inherently difficult for municipalities to distinguish the real causes of year-to-year changes or differences between municipalities. For example, one ECEC center/district/municipality may have higher concentrations of at-risk families than another one, and therefore is likely to have children with lower language assessment scores. However, this ECEC center/district/municipality may actually be doing a better job in terms of supporting children’s language development than an ECEC center/district/municipality whose children have higher language assessment scores mainly because of a higher SES demographic. Likewise, demographics can change over time, making year-to-year comparisons potentially difficult to interpret.

4.6.2 Interpretation of Results

The LA 2 uses the same cut-off points as does LA 3–6. Here we summarize the recommendations based on the child’s percentile score. We recommend that the lowest-scoring 5 percent of the children should be referred to a professional speech therapist in order to conduct a complete professional speech pathological assessment and potentially to receive specialized intervention. Children with scores from 6th to 15th percentile should be referred to the ECEC centers for focused intervention and a later reevaluation for language delay. Again, it is not in the scope of the instrument and the manual to specify appropriate interventions, but the manual advises work on the specific linguistic weaknesses of a child according to their profile (e.g., low scores on Vocabulary) and gives some examples. The percentile scores for each subscale may help guide educators to focus support on the domains in which the child needs the most help. The remaining 85 percent of the children can be regarded as having language development within the normal range. Thus these children are recommended to continue with the general intervention and the usual instructional activities in the ECEC centers.

4.6.3 Interpretation of Results for Bilingual and Multilingual Children

Just as for LA 3–6, the interpretation of the results of LA 2 for bilingual and multilingual children requires special attention because bilingual children generally have slower language development in each language than do monolingual children, even if the child has been learning both languages from birth. If a child has begun learning Danish as an additional language only a few months before the language assessment, it is obvious that they will be likely to receive very low scores. For children learning Danish primarily in ECEC, we recommend that they are not assessed using LA 2 if they have spent less than three months in ECEC.

While a bilingual or multilingual child who receives a percentile score within the normal range for monolingual children (percentiles 16–100) is unlikely to have language development problems, low scores (percentiles 1–15) are harder to interpret. A low score can be indicative either of language acquisition problems in general (which would have resulted in a low score in the first language as well, had it been assessed) or limited second-language input and interaction, meaning that the child is highly dominant in the first language. A combination of the two possibilities – developmental problems and lack of second-language input – is possible as well.

In order to help inform the interpretation of the results of bi- and multilingual children, 229 bilingual children were recruited as an addition to the monolingual norming sample. The results showed that bilingual children did not differ from monolingual children uniformly across the four subscales. The ANOVA analyses and effect size estimates indicated that, compared to monolingual children, bilingual children differed particularly in Receptive Vocabulary (bilinguals = 21.5 vs. monolinguals = 29.6, p < 0.001, d = 1.36), slightly less so in Productive Vocabulary (bilinguals = 21.9 vs. monolinguals = 40.5, p < 0.001, d = 1.15), and still slightly less so on decontextualized Language Use (bilinguals = 3.2 vs. monolinguals = 5.7, p < 0.001, d = 0.93) and the Communication subscale (bilinguals = 30.6 vs. monolinguals = 43.0, p < 0.001, d = 0.93). The – at least nominally – smaller difference between monolingual and bilingual children in productive rather than receptive vocabulary differs from previous research showing that bilingualism affects productive vocabulary more than receptive vocabulary (Reference Gibson, Oller and JarmulowiczGibson et al., 2012; Reference Gross, Buac and KaushanskayaGross et al., 2014; Reference Hemsley, Holm and DoddHemsley et al., 2010; Reference HoffRibot & Hoff, 2014). However, the difference is not easy to interpret because the Receptive Vocabulary scores were obtained from a test whereas the Productive Vocabulary scores were obtained from educator reports. However, taking the differing monolingual-bilingual difference in mean scores between subscales at face value, educators should be less worried when bilingual children receive low scores on the Receptive Vocabulary subscale but more so when they receive low scores on the Language Use and Communication subscales.

In contrast to what was done for LA 3–6, we did not develop a set of questions for nonnative parents in order to gain more information about the children’s bilingual development. However, the educators are advised to discuss language assessment results with the child’s parents in order to help interpret the results, especially if the scores are low. If the parents do not speak Danish to the child in the home, low assessment scores in Danish at this young age are virtually inevitable. On the other hand, if one or both parents, and perhaps siblings, speaks Danish to the child daily and has done so for most of the child’s life, very low Danish scores may be a reason for concern. However, given the short time that bilingual 2-year-olds have typically had to acquire the second language, educators are advised to give more attention to bilingual children’s Danish-language development over time rather than their Danish-language skills at a specific point in time. All else being equal, bilingual children are expected to obtain lower scores than monolingual children. But at the same time, bilingual children should narrow the gap with monolingual children over time. Therefore, it is important to assess low-scoring bilingual children regularly; for example, biannually.

4.7.1 Sample

To recruit children for the norming study, all ninety-eight Danish municipalities were asked to participate; thirteen municipalities chose to participate. The participants came from 135 ECEC centers and home-based ECEC groups from across Denmark. The municipalities represented geographically disparate parts of Denmark, and there were municipalities with urban profiles as well as rural profiles, and they were thus representative of Danish municipalities.

Table 28 shows the number of monolingual boys and girls in the sample. The children were grouped in three age categories: less than 2 years old (22–23 months), 2 to 2½ years old (24–29 months), and more than 2½ years old (30–34 months). Only children with a Danish (i.e., nonimmigrant and monolingual) background were in the norming sample; as was the case for LA 3–6, the purpose was to develop norms for monolingual children. While norms for bilingual and multilingual children would be highly desirable, such norms are extremely difficult to develop because of the multiple factors that may vary within this group of children, as discussed in Section 1.6.

The SES background – as indexed by maternal education – of children in the sample with a Danish background was generally representative of the population, although mothers with high levels of education were slightly overrepresented. When calculating the norms, scores were weighted so as to level the slight SES bias, which, however, made virtually no difference compared with basing norms on unweighted scores (even though SES correlated with the raw scores). However, all scores presented in this Element are raw, unweighted scores. Maternal education levels are shown in Table 29.

4.8.1 Descriptive Statistics for All Subscales

Descriptive statistics for the summary scores of the subscales of LA 2 for each age group – 22–23, 24–29, and 30–34 months old – are shown in Tables 29–35 and Figures 14–17. Although only the Receptive Vocabulary subscale uses different items for the three age groups, the scores and distributions are shown separately for each age group for the other subscales as well in order to illustrate the relationship of age to scores and score distribution. For data protection reasons, minimum and maximum values have been obscured if certain scores have been obtained by fewer than four individuals. Likewise, histogram bins encompass several scores in order not to show scores obtained by fewer than four individuals. As also done for descriptives of LA 3–6 above, we randomly divided the children into two groups for each age group in order to examine the robustness of the data via split sample results.

Table 30 and Figure 14 present descriptive statistics and histograms for the Receptive Vocabulary subscale. Because there is a change of items across ages for this subscale (Section 4.4.1), the scores are not directly comparable across the age groups. The histograms are more informative, as they demonstrate that variability in this skill is appropriately captured, the degree of skewness is visualized, and any floor or ceiling effects are detected. The distributions are left-skewed, and there is a tendency to a ceiling effect in the 22–23 months age group, although few children obtained the two highest scores (38 and 39 are represented in the same bin for data protection reasons). It does suggest, however, that a future revision of the scale should slightly increase item difficulty in order to decrease the ceiling effect. Alternatively, the ceiling effect may be seen as warranting a downward extension of the age span to start at, for example, 20 months in a future revision.

Figure 14 Distribution of scores on the Receptive Vocabulary subscale for the three age groups 22–23, 24–29, and 30–34 months. Age-dependent items

Table 31 and Figure 15 present descriptive statistics and histograms for the Productive Vocabulary subscale. As opposed to the Receptive Vocabulary subscale, the Productive Vocabulary subscale has the same items for all children aged 22–34 months, which is why the mean score increases from age group 22–23 months through age group 30–34 months. As shown in the table and histograms, the distribution goes from nonsignificantly right-skewed in age group 22–23 months to significantly left-skewed in the other two age groups, indicating that this subscale captures the development in these ages well.

Figure 15 Distribution of scores on the Productive Vocabulary subscale for the three age groups 22–23, 24–29, and 30–34 months

Table 32 and Figure 16 present descriptive statistics and histograms for the Language Use subscale. This scale uses the same items for all children and accordingly shows an increase in mean score across the age groups and a more left-skewed distribution of scores with higher age groups. In the oldest children, there is a clear ceiling effect.

Figure 16 Distribution of scores on the decontextualized Language Use subscale for the three age groups 22–23, 24–29, and 30–34 months

Table 33 and Figure 17 present descriptive statistics and histograms for the Communicative Skills subscale. The same items are used for all ages, and accordingly the mean score increases and the distribution becomes left-skewed with higher age groups.

Figure 17 Distribution of scores on the Communicative Skills subscale for the three age groups 22–23, 24–29, and 30–34 months.

4.8.2 Relation of Subscale Scores to Gender and Family Socioeconomic Status

The relation of gender and family SES to language development is well-documented, but less so in toddlers than in older children. However, gender and SES are clearly related to all four subscales of our language assessment instrument.

Table 34 shows that girls obtain significantly higher scores on average on all four subscales. The gender effect size is largest for Productive Vocabulary and Communicative Skills and smallest for Receptive Vocabulary. It is interesting that the gender effect is so different for the two vocabulary measures. However, the gender effect is somewhat smaller for the Receptive Vocabulary subscale than for all of the other three subscales. One possible interpretation is that boys trail girls to a lesser extent in receptive compared to productive language skills (Productive Vocabulary, Language Use, and Communication all being productive/expressive language skills). An alternative interpretation is that boys’ perceived language skills tend to be underestimated by educators (and perhaps others as well). Recall that the Productive Vocabulary, Language Use, and Communication subscales are all completed by the educator based on the educator’s estimation of the child’s skills. However, for LA 3–6, the gender effect sizes for the Communication subscale (completed by the educator) were generally not of larger (or smaller) magnitude than the gender effect sizes for subscales that were direct tests. But clearly, the gender effects are large enough for all four subscales to warrant gender-specific norms.

Table 35 shows the correlations of each subscale with three SES variables: maternal education, paternal education, and household income. The correlations tended to be a little stronger with maternal and paternal education than with household income, but the confidence intervals indicate that the strengths of the correlations were generally not significant between the SES variables. Likewise, there was little difference in the strength between the correlations across the subscales.

Just as for LA 3–6, correlations between the subscale measures of LA 3–6 and SES may serve as a tentative validation of the measures, because previous research has consistently found SES correlations with child language skills. However, a general finding in the literature is that associations between gender and child language outcomes become weaker through childhood (e.g., Reference WallentinWallentin, 2009), whereas SES associations with language and educational outcomes persist or even become stronger through the school ages (Reference SirinSirin, 2005). Therefore, an assessment instrument should not take child SES into account or correct scores for SES just because a lower score is “expected” from a child with a low SES background. Differences in outcomes contingent on SES do not reflect temporary developmental differences like those contingent on gender do, but they may be associated with different learning opportunities in the home, which ECEC may help counter.

4.9.1 Internal Consistency

Cronbach’s alpha coefficients for the subscales for each age group were computed to evaluate internal consistency and are shown in Table 36. The alpha values were high, indicating high internal consistency. The lowest values were for the subscale Language Use, but note that this subscale consists of just five items. Recall also that the Productive Vocabulary and Language Use subscales were taken from an existing instrument; the alpha values obtained in the present norming sample were very similar to those originally reported for the instruments (Productive Vocabulary = 0.98, Language Use = 0.88; Reference Bleses, Jensen and HøjenBleses, Jensen, et al., 2018). The range of item-total correlations, indicating the correlation of each item with the overall scale, suggest that the scales have good consistency. A quite low item-total correlation of a single item in the Receptive Vocabulary test for 30–34-month-olds suggests that this item should be replaced in a future revision of the instrument.

4.9.2 Internal Validity

Zero-order correlations were calculated to estimate associations between pairs of subscales. The correlations of the Receptive Vocabulary subscale had age group partialled out because of the difference in items given across the three age groups. The correlations are shown in Table 37. It is perhaps surprising that the strongest correlation was not that between Receptive and Productive Vocabulary. The correlations were strongest between the subscales completed by the educator. This is likely to be due, at least in part, to shared assessor variance. In other words, a child who, in the educator’s recollection, has a large vocabulary is also likely to be recollected as having advanced decontextualized language use and communication skills. In as far as this explanation is true, it goes to suggest the importance of testing (i.e., with active child participation) in addition to scales relying solely on the educator’s estimation and recollection of child skills.

Another, not incompatible, explanation for stronger correlations between the educator-completed subscales than with the Receptive Vocabulary test is that children are likely to differ in test-taking ability independent of language skills. Some children may play along with the educator’s agenda, the test, while other children may not. In as far as this explanation is true, it suggests the importance of assessments of the child that do not directly involve the child in a test.

As is the case for the results report for LA 3–6, a summary score is computed but mainly for administrative purposes, as noted in Section 4.6.1.

4.9.3 External Validity

To externally validate the four subscales, a subsample of fifty-one children were tested using the Danish adaptation of the Receptive One-Word Picture Vocabulary Test (Reference Martin and BrownellMartin & Brownell, 2011). Correlations of each subscale with raw scores on the Receptive One-Word Picture Vocabulary Test are shown in Table 38. The correlations with the external validation instrument were slightly stronger for the two vocabulary measures than for Language Use and Communication and generally similar to the strongest external validity correlations found for LA 3–6 (Section 3.7.3). This indicates that assessment of 2-year-olds with LA 2 is feasible and no less valid than assessment of older children with LA 3–6.

4.9.4 Test-Retest Reliability

Test-retest data are not available for children in the norming study. However, as noted, the subscales Productive Vocabulary and Language Use are from an existing instrument, the CDI: Educator. These two measures were previously evaluated for test-retest reliability by Reference Bleses, Jensen and HøjenBleses, Jensen, et al. (2018); the measures had been used as pretests and posttests in an early language and math intervention study (Reference Bleses, Jensen and SlotBleses, Jensen, et al., 2020). The measurements were taken approximately seven months apart, which means that – just as for the test-retest evaluation of LA 3–6 – the autocorrelation estimates are conservative because it cannot be assumed that all children’s language skills develop at the same rate over such a relatively long period of time. For Productive Vocabulary, the test-retest correlation was r = 0.68. For Language Use, the correlation was r = 0.54.

These indirect measures of reliability suggest moderate autocorrelations. While it was not possible to provide measures of test-retest reliability for the Receptive Vocabulary subscale and the Communication Subscale, we expect to be able to do so in the future through collaboration on data analysis with municipalities and providers of IT platforms for the instrument.

4.9.5 Procedure for Establishing Percentile Scores

Just as for LA 3–6, we established percentile norms based on our norming sample. Because scores are strongly related to both age and gender, as evidenced in Sections 4.8.1 and 4.8.2, we developed age- and gender-specific norms. The percentile score corresponding to a raw score value x of a child at age a (in months) is based on considering all children of the same gender and with an age of +/− two months and tested with the same material. For LA 2, we used a simpler procedure than the regression procedure used for LA 3–6 (see Section 3.7.4) to determine the score required to be on a given percentile. For LA 2, we smoothed the developmental curve by using the simple mean score for children of that gender at the given age, the two months preceding that age, and the two months following (that is, a five-month moving window). The procedure means that we “borrow data strength” from ages just above and below the age in question in a slightly different way than for LA 3–6. The change of approach was because the regression models did not converge. We are not sure why this is the case; one possibility is that it has to do with the lower N per month in LA 2. However, the purpose was, again, to iron out natural fluctuations causing anomalies in the normative development.