Monitoring oral language development in young children: psychometric evaluation of a learning progress assessment

1 Introduction

The continuous monitoring of progress in oral language interventions is essential to ensure that each child receives the most effective and timely support for their specific language development needs. Strong language skills are fundamental for active participation in social life, and early language development before school entry is considered a key predictor of later text and reading comprehension (Catts et al., 2016; Ennemoser et al., 2012). Research indicates that children with poor language skills (−1 SD) (Viesel-Nordmeyer et al., 2019) and those with German as a second language (Kempert et al., 2016) require increased assistance. Therefore, providing timely and effective support is crucial to prevent potential negative developmental trajectories. From a neurodevelopmental perspective, early oral language development draws on multiple, partly domain-general neural systems involved in perception, attention, and sensorimotor integration (Bates, 1992; Kuhl, 2010). Studies of vulnerable populations, such as children born prematurely, further highlight that subtle early constraints in these systems can have cascading effects on later language and learning. This underlines the importance of early assessment for tailoring support (Charollais et al., 2013).

It is not only essential to identify the need for language intervention as early as possible (Sachse and Buschmann, 2018), but also to monitor the effectiveness of interventions throughout the process (Deno, 2003). This is particularly important because evidence supporting the general effectiveness of an intervention does not guarantee that every individual child will benefit equally from it (Suchodoletz, 2010). By identifying when an intervention is not working at the early stages, practitioners can promptly adjust their approach to ensure that each child receives tailored support for their specific language development needs.

Learning Progress Assessments (LPA) are a valuable tool for monitoring the effectiveness of interventions (Deno, 2003). They provide actionable data for making data-based decisions and evaluating intervention success. Internationally, several meta-analyses have reported positive effects on student achievement when teachers use progress monitoring strategies (Visible Learning, 2024). From a broader perspective, LPAs are a central element of data-based decision making and formative assessment cycles, in which diagnostic data are used iteratively to inform instructional decisions, adapt support and evaluate their effects (e.g., Deno, 1985; Espin et al., 2012; Souvignier and Förster, 2025). In the Anglo-American early childhood sector, this logic has been incorporated into multi-tiered systems of support, where screening and regular progress monitoring guide data-based decisions about the intensity of language and learning support in a collaborative problem-solving process (Buysse and Peisner-Feinberg, 2010; see also Bradley et al., 2005; Vaughn and Fuchs, 2003). However, for oral language in German kindergartens, there is still a lack of time-efficient, easy-to-administer tools that can be used for such systematic progress monitoring in everyday practice.

In German-speaking countries, LPAs are already available, primarily in schools, for written language and mathematics (see Gebhardt et al., 2021). However, a significant research gap exists in the domain of oral language development, particularly in early childhood education (Mahlau et al., 2014). Although progress-monitoring tools for the phonetic, morphological, and syntactic levels have been developed and used as part of the Rügener Inklusionsmodell-a German inclusion model based on the response-to-intervention (RTI) approach (Mahlau et al., 2014)-these tools remain unpublished and inaccessible for broader use.

Furthermore, well-established diagnostic tests for assessing language skills are often unsuitable for use in LPAs due to their extensive scope, high time requirements, and the lack of parallel test versions. To address this research gap, an oral language LPA was developed was developed specifically for systematic progress monitoring in German kindergarten settings.

1.1 Sentence repetition as a global measure of language skills and a tool for progress assessment

Following the logic of curriculum-based measurement as a general outcome measure (Deno, 2003), Learning Progress Assessments can be constructed around robust indicators that represent the targeted competence as a global measure. Sentence repetition tasks are particularly promising as global measures for assessing language skills for several reasons. In such tasks, children are presented with sentences that are grammatically correct but do not make sense semantically (e.g., “During the day, the cap drives on the donkey.”) and are asked to repeat them.

Firstly, sentence repetition tasks function as comprehensive global measures of language skills. Klem et al. (2015) stated, “Sentence repetition is best seen as a reflection of an underlying language ability factor rather than as a measure of a separate construct with a specific role in language processing.” This view is consistent with Tomblin and Zhang (2006) who found out that a common language ability factor explains a substantial part of children's performance across lexical, grammatical and narrative measures in school-age children. These findings are further supported by studies suggesting that performance in sentence repetition tasks is influenced by vocabulary and phonological working memory (Goldammer et al., 2010) as well as familiarity with morphosyntax and lexical phonology (Polišenská et al., 2015). Thus, sentence repetition appears to be a valuable tool for language assessment because it draws upon a wide range of language processing skills. This view aligns with neurodevelopmental accounts that describe language as drawing on the interaction of multiple, partly domain-general systems, including perceptual, cognitive, and sensorimotor processes (Bates, 1992; Kuhl, 2010; Charollais et al., 2013). These systems are all engaged when children perceive, encode, and reproduce complex sentences.

Moreover, sentence repetition tasks have already been employed in various diagnostic tests, such as assessing morphological and syntactic skills (Esser and Wyschkon, 2010) or serving as a screening tool to identify children at risk for written language disorders (Schöler and Brunner, 2008). These tasks are also well-established methods for identifying developmental language disorders (DLD) (see Rujas et al., 2021). International research likewise supports sentence repetition as a sensitive clinical marker of language impairment in monolingual populations (e.g., Conti-Ramsden et al., 2001; Stokes et al., 2006) and in bilingual children, where sentence repetition tasks have been shown to discriminate children with specific language impairment across different language pairs, including German (Marinis and Armon-Lotem, 2015; Abed Ibrahim et al., 2018). Importantly, research indicates that sentence repetition tasks do not disadvantage bilingual children with DLD compared with monolingual children with DLD (e.g., Tuller et al., 2018).

Secondly, sentence repetition tasks seem to be highly practical tools for progress assessment. They are short, easy to administer, and can be efficiently evaluated by counting the number of correctly repeated words. Furthermore, it is theoretically possible to construct an infinite number of parallel tests based on sentence repetition tasks, enabling continuous and reliable progress monitoring. Additionally, linguistic abilities develop interdependently across different levels (e.g., semantic-lexical and morpho-syntactic abilities), meaning that improvements in one area manifest in enhanced sentence performance. As a result, sentence repetition tasks provide a comprehensive approach to assessing language skill development.

Thirdly, sentence repetition tasks offer an effective way for monitoring progress during interventions. Interventions for children with language support needs aim to improve their overall language competence. For example, vocabulary-focused interventions are typically embedded within grammatical structures rather than conducted as isolated vocabulary training. Consequently, improvements are expected to occur across multiple levels of language development. These advancements can be effectively captured using a global measure like sentence repetition. Moreover, there is initial evidence that sentence repetition tasks can be sensitive to developmental change over time and may be used as progress-monitoring tools. For example, Wood and Hoge (2017) reported that a sentence repetition task captured growth in English morphosyntax from kindergarten to first grade in Spanish–English dual language learners.

1.2 Construction of the LPA for oral language

To leverage these advantages, a specific tool was developed to monitor progress in oral language development trough sentence repetition tasks. 11 parallel tests were constructed using the German vocabulary pool from the ChildLex corpus (Schroeder et al., 2015). The sentences in these parallel tests were based on current research into grammatical development of German-speaking children aged 4–9 years who are developing language skills typically (Ulrich, 2017).

Each parallel test consisted of 16 sentences with increasing grammatical complexity, starting with four short, meaningful sentences and concluded with semantically meaningless sentences containing 10–12 words. This structure was designed to assess a broad spectrum of linguistic abilities by evaluating both current language proficiency and target skill levels (Gebhardt et al., 2021). The sentences are presented to the child in a standardized, child-directed speaking style with natural prosody but without exaggerated emphasis on individual words, in order to avoid providing additional cues beyond the linguistic structure of the sentence and to reduce examiner-related variability in prosodic cues, not to imply that prosody is irrelevant for sentence processing. Scoring is conducted by awarding one point for each correctly repeated word. This results in a maximum score of 134–137 points per parallel test. The test itself takes approximately 5 min to administer, while scoring requires an additional 10–15 min.

Word-level evaluation is particularly useful because it enables linguistic progress to be detected in both semantic-lexical and grammatical areas. Improvements in vocabulary are by the ability to confidently retrieve the desired word from long-term memory. Improvements in grammar can also be recognized at word level. In German, for example, this can be seen in the use of the dative case with the article “nach dem Essen” (after eating), the adaptation of the adjective to the noun “ein mutiger Löwe” (a brave lion) or the verb conjugation in tense “rennt/ist gerannt/rannte” (run/has run/ran). A key feature of the German language is that many changes in tense and case are marked at the word level. In contrast, using a two- or multi-level rating scale for the whole sentence (e.g., correct/incorrect or correct/partially correct/incorrect) can lead to reduced sensitivity and specificity (Bishop and Thompson, 2023). Such global sentence scales may fail to capture subtle but meaningful improvements, particularly in children with developing language skills, and they may not differentiate well between qualitatively different error patterns within the same sentence category. More differentiated sentence-level approaches that rate specific target structures or error types have therefore been proposed in the literature (e.g., Komeili and Marshall, 2013; Hamann and Abed Ibrahim, 2017; Marinis and Armon-Lotem, 2015; Leclercq et al., 2014). However, such approaches are not well suited to the purpose of the present LPA, which requires a simple and time-efficient scoring procedure for repeated use in everyday practice.

1.3 Preliminary validation of the LPA in a pilot study

In a pilot study, the LPA for oral language was tested with both a cross-sectional (n = 19) and longitudinal sample (n = 19) of children with DLD in the first and second grades (Besca et al., 2023). The parallel tests, examined cross-sectionally, demonstrated high comparability, with good internal consistency ranging from 0.90 to 0.94 and reliable discriminative power (0.25 to 0.82). The majority of discriminative power values were above 0.6, with only a few values below 0.3. Validity analyses revealed strong positive correlations between the LPA for oral language and conventional diagnostic tests, particularly in vocabulary (r = 0.75^**), grammar (r = 0.73^**), and language comprehension (r = 0.67^**) in contrast, correlations with working memory (r = 0.22) and mathematical abilities (r = 0.09/−0.03) were low. Additionally, preliminary evidence of sensitivity to change was observed in the longitudinal sample at group level with F(6, 108) = 2.494; p = 0.27; η² = 0.12) (Besca et al., 2023). Sensitivity to change is a critical criterion for learning progress assessments, as it demonstrates the ability of the tool to detect developmental changes over time.

However, this pilot study was conducted with a very small sample of school-aged children with DLD only. As no typically developing preschool children were included, these preliminary findings cannot be generalized to 4–6-year-olds in early childhood education or used to draw firm conclusions about the LPA's sensitivity in this age range.

1.4 Open questions and further steps to evaluate the LPA for oral language

The preliminary findings suggest promising results; however, they are based on a small sample size and further studies are needed with a larger and more diverse sample, including participants from different age groups and linguistic backgrounds. In order to provide robust evidence that the LPA is a reliable and valid tool for assessing language development, further examination of its psychometric properties is essential. This should include a comprehensive analysis of reliability, validity, and sensitivity, as recommended by Bishop and Thompson (2023).

Reliability will be examined through test-retest reliability, calculated for both the overall sample and individual age groups, to account for potential age-related variations. Additionally, the intraclass correlation coefficient (ICC) will be calculated to provide further insights into reliability. To evaluate the validity of the LPA, selected language measures will be used that are specifically designed to provide differentiated results for distinct age groups (see 3.2). These measures focus on capturing general linguistic development rather than specific training outcomes, aligning with the overarching objective of the LPA. Factorial validity will also be analyzed to confirm that all parallel tests load onto a single common factor, as suggested by Wilbert and Linnemann (2011).

To assess the sensitivity of the LPA, a longitudinal implementation of parallel tests is necessary. This will enable an evaluation of the tool's ability to detect changes over time, both in general language development and as a result of targeted interventions (Bishop and Thompson, 2023). Sensitivity to change is a critical aspect of the LPA, as its primary goal is to measure developmental progress in language skills (Gebhardt et al., 2021).

The LPA must also account for individual differences in starting levels and growth trajectories (Jungjohann et al., 2021). This is particularly important, as the development of vocabulary and grammar during childhood varies significantly among individuals, influencing both educational needs and the effectiveness of interventions. For example, by the age of three, children's productive vocabulary typically ranges from 300 to 500 words and expands to an estimated 4,000–8,000 words by the end of second grade (Kauschke, 2023). Vocabulary growth continues at a rapid pace until approximately age 12. Similarly, findings from the GED 4–9 study show that grammatical development progresses well into primary school, contrary to earlier assumptions that it is largely complete by the age of four (Ulrich, 2017).

Children acquiring German as a second language exhibit even greater variability in their language development. Influential factors include the quality and quantity of language input as well as the age of onset in learning German (Chilla, 2020). These variations highlight the importance of analyzing learning progress separately for different proficiency levels and linguistic backgrounds, which would enable the development of more tailored and effective educational interventions. These examples illustrate that children's language trajectories are shaped by multiple interacting factors, leading to considerable variability in both initial levels and growth.

2 Aims of the study and hypotheses

This study aimed to replicate and extend previous findings on the newly developed LPA for oral language as a measure of language development in children aged 4–6 years by examining the psychometric properties that are particularly relevant to LPAs. Specifically, the study focused on the following objectives: first, using a cross-sectional sample, we analyzed the fundamental scale statistics and the comparability of difficulty levels across all parallel tests, ensuring that no further adjustments to the current scale construction were necessary. Second, with a longitudinal sample, we gathered additional data on the (retest-) reliability and validity of the instrument. Additionally, we assessed its sensitivity to change, ensuring that any observed changes were not merely the result of test-retest effects. The study is centered around the following hypotheses:

• (1) Homogeneity of the LPA. We hypothesize that the LPA for oral language will exhibit a sufficiently homogeneous level of difficulty across parallel tests, as suggested in the pilot study (Besca et al., 2023). Furthermore, we expect the LPA to demonstrate a one-dimensional structure, reflecting the assumption that performance across the parallel sentence repetition tasks can be represented by a common language ability factor (Klem et al., 2015). In the context of this study, one-dimensionality is understood as a statistical property of the test scores rather than a theoretical claim about the dimensionality of language. Specifically, we assume that children's performance on the sentence repetition tasks reflects a broad oral language factor shaped by the joint contribution of lexical, morphosyntactic and phonological processing.
• (2) Reliability and validity. Based on the results of the pilot study, we expect to observe high test-retest reliability. For validity, we hypothesize that the LPA for oral language will show strong positive correlations with convergent language measures (e.g., vocabulary and grammar) and lower correlations with divergent measures (e.g., working memory, intelligence, mathematical ability, socio-economic status) (Besca et al., 2023) compared to core language measures such as vocabulary and grammar.
• (3) Sensitivity to change. The LPA for oral language specifically aims to capture language development in children aged 4–6 years. We expect the LPA to be sensitive to growth, supported by prior findings in samples of school children with DLD (Besca et al., 2023). To ensure that the observed language progress reflects genuine development rather than an artifact of repeated testing, we will compare the results of an experimental group (EG), in which learning progress is assessed through multiple measurements, with those of a control group (CG), assessed only at the first and last measurement points (MP).
• (4) Differences in initial achievement level and slope. We hypothesize that the LPA will reveal differences in children's initial achievement levels and developmental trajectories (slopes). This expectation is based on individual differences in vocabulary and grammar competencies reported for monolingual and multilingual children (Chilla, 2020; Kauschke, 2023; Ulrich, 2017).

3 Methods

3.1 Design and participants of the cross-sectional and longitudinal study

3.1.1 Cross-sectional sample

The LPA for oral language was administered by student assistants and the developer in a cross-sectional sample of 87 preschool children aged 4–6 years in 2021. The original design aimed to complete all 11 parallel tests within 2 weeks. Due to COVID-19 related kindergarten closures, not all children were able to complete all 11 parallel tests resulting in between 74 and 80 responses per test. Children who started testing in September 2021 completed significantly more tests than those who started in November 2021, which is plausibly explained by the subsequent lockdown period. However, children with fewer vs. more completed tests did not differ on background variables (age, gender, language background) or initial LPA scores. As the parallel tests were systematically rotated across measurement points to minimize confounding effects of item repetition and measurement points, missingness did not affect any specific test, and we therefore do not expect to have introduced systematic bias into the homogeneity analyses.

Children were recruited from six kindergartens located in both rural and urban areas of southwest Germany. In the German context, children attending kindergarten have not yet started formal reading and writing instruction, therefore systematic influences of written language on oral language are expected to be limited in this age group.

3.1.2 Longitudinal sample

To evaluate the LPA's sensitivity to changes over time, a longitudinal sample was recruited. A total of 127 children aged 4–6 years from seven kindergartens in rural and urban areas of southwest Germany participated in this study. To account for potential training effects-a phenomenon where repeated testing can influence results (Wilbert and Linnemann, 2011)-the sample was divided into an Experimental Group (EG, N = 88) and a Control Group (CG, N = 39). Based on recommendations for estimating mean learning progress using LPA, at least six assessments are required with equal time intervals between them (Christ et al., 2013, in Gebhardt et al., 2021). Accordingly, the EG was assessed seven times from January to July 2022 at 4-week intervals, whereas the CG was assessed twice, in January and July 2023. Due to organizational constraints during the COVID-19 pandemic, the EG and CG thus represent two consecutively recruited cohorts rather than a single cohort with different testing schedules; recruitment procedures, inclusion criteria and testing protocols were kept as similar as possible. To avoid potential biases caused by administering parallel tests in a fixed sequence, the tests were presented in a randomized order across sessions. The 11 parallel tests were distributed across the seven measurement points.

To minimize examiner-related variance, all sentence repetition tasks were administered by student assistants who were trained and supervised by the developer. The training followed an administration manual specifying speech rate, intonation and pausing (e.g., neutral prosody). During the development phase, we additionally compared a live spoken version of the LPA with an audio version within the cross-sectional sample. No systematic differences in children's performance emerged between these two versions [t(848) = −0.130; p = 0.896]. Therefore, we chose the live spoken version for the longitudinal study in order to maintain a natural testing situation for the children.

To document potential age-related differences in how children responded to the task format, examiners also recorded how often children required an additional explanation of the task format, a repeated instruction or an explicit prompt to refocus attention. An analysis of variance showed no age-related differences in this parameter between 4-, 5- and 6-year-olds [F(2, 99) = 0.271, p = 0.763; post-hoc comparisons n.s.].

3.1.3 Inclusion criteria

The inclusion criteria for both samples were as follows: (1) monolingual and multilingual children, (2) aged 4–6 years, (3) currently attending kindergarten, and (4) parental consent. Verbal consent was obtained from the educational staff and the children themselves, while written consent was provided by their parents. Compliance with the current version of the Declaration of Helsinki was confirmed. (5) The samples were designed to include both children with age-appropriate language development and children with below-average language performance, as identified by the standardized assessments described in Section 3.2. Children with articulatory disorders were excluded if their speech production made it impossible to determine whether an incorrect word production was due to articulation problems or to grammatical difficulties. Children with mild but systematic pronunciation difficulties (e.g., sigmatismus interdentalis, substitution of /ʃ/ by /s/) were included. For the cross-sectional sample and the CG in the longitudinal sample, data collection was complete. In the EG of the longitudinal sample, approximately 4% of the data were missing due to children's absence from kindergarten on the day of testing. At the kindergarten level, missing data rates ranged from 1.4 to 6%. In the kindergarten with the highest rate, most missing values occurred at the final measurement point because some children were already on vacation. To explore the missing data mechanism, we compared children with and without missing values on available background variables (age, gender, language background) and on their initial LPA scores. No significant differences emerged, and the proportion of missing values did not systematically increase across measurement points. Although the missing data mechanism cannot be determined with certainty, these findings did not indicate systematic missingness related to children's initial language performance or available background characteristics. In addition, sensitivity analyses using the non-imputed data yielded the same overall pattern of results. Given the very low proportion of missing data and the robustness of the findings, we retained the regression-based imputation procedure within percentile groups for the LPA distribution as a parsimonious alternative to more complex techniques such as multiple imputation or full information maximum likelihood.

3.1.4 Demographic data and information about children's languages

Demographic data, including information on languages spoken and age of onset for German as a second language, were collected from parents. The demographic characteristics of all samples are presented in Table 1.

Table 1. Demographic data on the cross-sectional and longitudinal samples.

Parents also provided information about their children's age at onset of German as a second language. However, due to missing data on the age of onset, children were categorized as either monolingual or multilingual based on the language reported by their parents. Analyses based on age at onset yielded results comparable to those based on the monolingual/multilingual categorization.

In the EG, 8% of monolingual children scored below average (T-value < 35) on at least one of the vocabulary, grammar, or sentence repetition tests (see 3.2.1–3.2.3), indicating a DLD (Lüke et al., 2023). All tests were standardized assessments designed to evaluate language development and identify any delays or disorders. Additionally, 15% of monolingual children achieved a T-value between 35 and 40, which is classified as a below-average range (Lüke et al., 2023). Among multilingual children, 78% scored below average (T-value < 40), performing lower than monolingual children in German. However, DLD can only be diagnosed if a deficit exists in the first language (Lüke et al., 2023). Despite this, all children with below-average scores required language intervention in German as a second language.

3.2 Validation measures

3.2.1 Vocabulary

To test convergent validity, we used subtests from standardized language development assessments that were not specifically designed to monitor progress. Vocabulary skills were assessed using the picture-naming subtest of the SET 5–10 (Petermann, 2018). According to the manual, this subtest demonstrates a strong correlation with other vocabulary measures (r > 0.70). The maximum score is 40 and Cronbach's α = 0.90.

3.2.2 Grammar

Grammatical skills were assessed using the grammar subtest of the P-ITPA (Esser and Wyschkon, 2010). The subtest has a maximum score of 57 and high internal consistency, with Cronbach's α ranging from 0.93 to 0.94. The correlation with another grammatical measure is strong, with r > 0.65.

3.2.3 Alternative sentence repetition measure

For comparison with another sentence repetition task not specifically designed to assess learning progress, we used the corresponding subtest from the SET-K 3–5 (Grimm, 2015). This subtest has a maximum score of 119, and Cronbach's α ranges from 0.90 to 0.92 (based on the standardization sample). According to the manual, the prognostic validity of this subtest is demonstrated by a high correlation with a subsequently assessed test of dyslexia (r = −0.59) and in independent studies (Goldammer et al., 2010).

3.2.4 Working memory

As phonological working memory plays a particularly important role in language development (Baddeley et al., 1998), we assessed it using Mottier Syllables, a non-word repetition task (Mottier, 1951).

3.2.5 Divergent validity measures

For divergent validity, we used the number sequence and digit knowledge subtests from the Test of Basic Mathematical Competences (MBK 0; Krajewski, 2018). The MBK 0 demonstrates very high retest-reliability, ranging from r = 0.83 to r = 0.88 and it is strongly correlated with other mathematical measures (r = 0.62 to r = 76). Fluid intelligence was assessed using the matrices and picture concept subtests of the WPPSI (Petermann and Daseking, 2018). The reliability of these subtests ranges from r = 0.74 to r = 0.86, with correlations with previous versions of the test between r = 0.85 and r = 0.89.

As a proxy for socio-economic status, we used the “book question,” which was illustrated with pictures of bookshelves (Heppt et al., 2022).

4 Results

Data were statistically analyzed using SPSS (Version 29, IBM Corp, 2024). Both descriptive and inferential statistical methods were applied, and the specific methods are described at the beginning of each section.

4.1 Homogeneity of the parallel tests

To assess the homogeneity of the parallel tests, we used data from the cross-sectional sample. As shown in Table 2, the mean values (ranging from 71.12 to 76.64 correctly repeated words) and the standard deviations (ranging from 25.79 and 29.61) are similar across the tests. The difficulty level (P) of each parallel test is within the median range, with values between 0.56 and 0.59. The discriminatory power of the parallel tests as a total scale is excellent, with values ranging from 0.93 to 0.96. Furthermore, the parallel tests show very high internal consistency overall, with Cronbach's α = 0.99 and a high mean inter-item correlation value of 0.91, indicating a high degree of homogeneity.

Table 2. Parameters for analyzing the homogeneity of parallel tests.

PT, parallel test, P, difficulty of each parallel test, r_{it − korr}, adjusted discriminatory power, confirmatory factor loading assuming one latent factor.

One-dimensionality was examined using CFA, with the latent factor language competence explaining 92% of the variance in the model. The one-dimensional model showed overall acceptable fit, χ²(8) = 14.70, p = 0.07, CFI = 1.00, RMSEA = 0.098, SRMR = 0.003. CFI and SRMR indicated excellent fit, whereas RMSEA was elevated and should be interpreted in light of the relatively low degrees of freedom. Factor loadings were consistently high, ranging from 0.95 to 0.97 (see Table 2, Column 7).

4.2 Reliability and validity

To evaluate the reliability of the parallel tests, we used longitudinal data from the EG (N = 88) and analyzed correlations between neighboring measurement points (Walter, 2014). Overall, the correlations between neighboring measurement points (MP1/MP2, MP2/MP3, etc.) were high with r = 0.97 at p < 0.001 level. This reflects a consistent ranking across all measurements. The intraclass correlation coefficient (ICC) was r = 0.99, indicating a high degree of agreement between absolute values. Additionally, the correlations separately for different age groups were between r = 0.95 and 0.97, p < 0.001 (4-year-olds), r = 0.97 and 0.98, p < 0.001 (5-year-olds) and r = 0.96 and 0.99, p < 0.001 (6-year-olds). These results indicate high reliability across all age groups.

Convergent and divergent validity were assessed between the LPA for oral language and pre- and post-test validation measures (see Section 3.2) using correlations. Differences between correlations were examined using Fisher's Z-test, and the results were interpreted following Cohen (1988) conventions (see Table 3). Positive and strong correlations were observed between the LPA for oral language and tests for vocabulary, grammar, and the SET-K sentence repetition task, with values ranging from r = 0.81 to 0.94, indicating large effect sizes. A moderate correlation was found between the LPA and mathematical competences (r = 0.56). The correlations between LPA, phonological working memory (r = 0.30/0.34), and socio-economic status (r = 0.44) were moderate, whereas the correlation between LPA and intelligence was low (r = 0.27).

Table 3. Correlations between the LPA for oral language and convergent and discriminant validity measures.

R, Correlation according to Pearson or Spearman, ^* Level of significance of 0.05, ^** Level of significance of 0.01.

The correlations between the LPA for oral language and convergent measures (vocabulary, grammar, and SET-K sentence repetition task) were significantly higher than those between the LPA and divergent measures (intelligence and basic mathematical skills), as expected. For example, the correlation between LPA and vocabulary was significantly higher than that between LPA and intelligence (Fisher's Z = 7.218, p < 0.001), and the correlation between the LPA and grammar was significantly higher than that between LPA and mathematics (Fisher's Z = 4.261, p < 0.001).

4.3 Sensitivity to change

The results are described in several steps. First, descriptive statistics are presented. Second, repeated measures analyses are used to test for significant differences in the means of all measurement points at the group level (e.g., Gebhardt et al., 2021). Third, the standard error of measurement (SEM) is reported to evaluate whether observed changes exceeded measurement error. Fourth, a latent growth curve model is presented as an additional model-based analysis of systematic change over time and individual differences in growth. Fifth, the comparison between the EG and CG is used to examine potential test-repetition effects. Finally, developmental trajectories are examined by initial performance and language background to assess whether growth rates differed as a function of children's initial skill levels and number of languages spoken.

4.3.1 Descriptive statistics

Table 4 presents the means (M), standard deviations (SD) and standard error of measurement (SEM) of the longitudinal sample, illustrating children's language development over a 6-month period. In addition, Table 4 reports effect sizes (Cohen's d) and their 95% confidence intervals for the change from MP1+2 to MP6+7, providing information on the magnitude and precision of the observed changes. Effect sizes were calculated by comparing the means of the first two measurement points (MP1+2) with the final two (MP6+7), using the pooled standard deviation (Lenhard and Lenhard, 2022). The mean of multiple measurement points provides a more reliable estimate of achievement levels than a single-point measurement (Walter, 2014).

Table 4. Means, standard deviations and standard error of measurement and Cohens d effect sizes for all MP¹, separately by pre-test level and number of languages spoken.

¹MP, measurement points. ²Effect sizes were calculated based on the means scores at MP 1 and 2 and at MP 6 and 7 and the pooled standard deviation (Lenhard and Lenhard, 2022).

The mean values for the overall sample and the subgroups increased continuously from the first to the last measurement point, with minor fluctuations at individual points. Notably, children in the upper quartile (75–100) showed no notable development between MP1 and MP7 (M_*MP*1 = 113.31, SD = 9.45; M_*MP*7 = 113.90, SD = 10.19). Children in the 3rd quartile (50–75) displayed only minimal changes (M_*MP*1 = 88.61, SD = 7.20; M_*MP*7 = 90.24, SD = 12.49). At the final measurement point (MP7), all groups scored slightly lower than at MP6, likely due to external factors, such as hot weather on the testing day or the imminent start of the summer holidays. For monolingual and multilingual children, similar growth patterns emerged at different initial levels (monolingual: M_*MP*1 = 89.60, SD = 27.22; M_*MP*7 = 92.37, SD = 24.52; multilingual: M_*MP*1 = 31.57, SD = 27.77; M_*MP*7 = 33.71, SD = 24.07). The effect sizes (Cohen's d) indicated a moderate effect for the lower two quartiles, a small effect for multilingual children, and negligible effect sizes for other groups.

4.3.2 Results of ANOVA

Figure 1 shows the progress in language development based on the mean values per measurement point for the entire sample (left graph) and the lower three quartiles (right graph). To analyze changes over time for the entire sample, a repeated measures ANOVA with Greenhouse-Geisser correction revealed statistically significant differences across measurement points, F(4.70, 408.91) = 9.00, p < 0.001, with a moderate effect size (η² = 0.09). The paired t-test further indicated that the differences between MP1 and subsequent measurement points became statistically significant starting at MP5. Specifically, significant increases were observed between MP1 and MP5 [t(87) = 3.16, p = 0.001, d_Cohen = 0.34]; MP6 [t (87) = 4.05, p < 0.001, d_Cohen = 0.43], and MP7 [t(87) = 3.19, p < 0.001, d_Cohen = 0.34], indicating small effects.

When analyzing only the lower three quartiles, a repeated measures ANOVA also showed significant differences across time points, F(4.71, 310.92) = 9.22, p < 0.001, with a moderate effect size (η² = 0.12). Significant differences from MP1 emerged as early as MP3 [MP3-MP1 t(66) = 2.24, p = 0.01, d_Cohen = 0.27; MP4-MP1 t(66) = 2.10, p = 0.02, d_Cohen = 0.26; MP5-MP1 t(66) = 3.73, p < 0.001, d_Cohen = 0.20; MP6-MP1 t(66) = 4.45, p < 0.001, d_Cohen = 0.54; MP7-MP1 t(66) = 3.58, p < 0.001, d_Cohen = 0.44), with effect sizes ranging from small to moderate.

4.3.3 Standard error of measurement

For the whole group, the difference between MP1+2 and MP6+7 is 4.84 points, which is below the SEM value of 6.51/6.10. In contrast, for the quartile group 0–25, the difference between MP1+2 and MP6+7 is 8.64, exceeding the SEM values of 1.94/2.15. For the quartile group 25–50, the corresponding difference is 8.45, which is also above the SEM values of 1.88 /2.69. Moreover, in the quartile group 0–25, four of six differences between neighboring measurement points exceeded the respective SEM (difference MP1 vs. MP2 = 3.5 > SEM = 1.99/2.06; difference MP2 vs. MP3 = 2.18 > SEM: 2.06; MP3 vs. MP4: −0.05 < SEM: 2.06/1.84; MP4 vs. MP5: 2.55 > SEM: 1.84/1.9; MP5 vs. MP6: 2.36 > SEM: 1.9/2.11; MP6 vs. MP7: −0.32 < SEM: 2.11/2.36), whereas in the 25–50 and 50–75 quartiles this was the case for two differences, and in the 75–100 quartile for only one.

4.3.4 Latent growth curve model

As an additional model-based analysis, a parsimonious linear latent growth curve model was estimated to examine change across the seven measurement points (Figure 2). The intercept factor was specified with all loadings fixed to 1, whereas a linear growth trajectory was modeled by fixing the slope loadings from 0 to 6. The model fit the data well, χ²(29) = 33.50, p = 0.258, CFI = 0.997, TLI = 0.998, RMSEA = 0.042, SRMR = 0.004. The slope mean was positive and significant, b = 0.93, SE = 0.19, p < 0.001, corresponding to an average increase of approximately one correctly repeated word per measurement point. The significant slope variance, b = 1.85, SE = 0.49, p < 0.001, further indicated that children differed in their individual growth rates.

Nested model comparisons supported the relevance of both growth parameters. Constraining the slope mean to zero significantly worsened model fit, Δχ²(1) = 20.95, p < 0.001, indicating that the data were better described when systematic average change was included in the model. In a second restricted model the slope variance as well as the intercept–slope covariance was fixed to zero. This model also showed a significantly worse fit than the freely estimated model, Δχ²(2) = 38.04, p < 0.001, indicating that the data were better described when interindividual differences in growth rates were allowed.

Another notable finding was the negative intercept–slope correlation, r = −0.42, indicating that higher initial performance was associated with smaller subsequent gains. This pattern is consistent with the descriptive and subgroup analyses, which indicated stronger gains among children with lower initial performance and only small changes among children with high initial scores. Overall, the latent growth curve model and the nested model comparisons provided additional evidence that the measure is sensitive to both systematic average change and individual differences in change over time.

4.3.5 Control group comparison

To confirm that the observed learning progress was not merely due to multiple testing, an analysis of variance was conducted with group (experimental vs. control) pre-test performance as a factor. No significant differences between the EG and CG were found at MP6 [F(1, 125) = 0.473, p = 0.493, η² = 0.004], suggesting that more frequent testing in the EG (seven measurements) compared to the CG (two measurements) provided no evidence that repeated testing substantially influenced the results.

4.4 Developmental trajectories by initial performance and language background

Figure 3 visualizes the developmental trajectories in language skills, with separate trend lines for quartile groups based on initial performance levels and for groups categorized by the number of languages spoken. Trend lines were evaluated using regression analysis. The means, standard deviations, and effect sizes for these groups are summarized in Table 4. Results of ANOVA indicate significant differences in pre-test levels across quartile groups [F(3, 84) = 399.965, p < 0.001, η² = 0.935] and between language background groups [F(1, 86) = 70.676, p < 0.001, η² = 0.451].

Children in the bottom two quartiles begin at a lower initial language level and demonstrate growth rates of 1.5 and 1.7 words per measurement point, indicating steady progress from a lower baseline. In contrast, children in the top two quartiles start at a much higher initial level, with growth rates of only 0.2 or 0.3 words per measurement point, indicating only limited progress. Multilingual children begin from a significantly lower baseline compared to monolingual children; however, the growth per measurement point is relatively similar for both groups, ranging from 0.8 to 1.1 words per measurement point.

5 Discussion

Using cross-sectional data, this study demonstrates that the LPA for oral language is well-suited for assessing the language abilities of children aged 4–6 years. The results indicate that the newly developed LPA exhibits a one-dimensional structure, excellent homogeneity, and high reliability across parallel tests. Correlations with measures of vocabulary, grammar, and sentence repetition from a standardized language test confirm the LPA as a valid tool for assessing oral language abilities across different age groups (ages 4–6). Furthermore, the longitudinal data provide initial evidence of the tool's sensitivity to change, particularly for children with poor language abilities (e.g., those in the 0–25 quartile group). This is supported by the observed effect sizes over 6 months, the results of the repeated-measures ANOVA, and the interpretation of the SEM.

The one-dimensionality of the LPA was confirmed based on the CFA with the latent factor explaining 92% of the variance and overall acceptable fit indices (excellent CFI and SRMR, but an elevated RMSEA that should be interpreted in light of the relatively low degrees of freedom). Given the relatively small sample size for CFA and the very simple one-factor model, these fit indices should be interpreted with caution. In small samples and with models of low complexity, such values can occur even when the model is only an approximate representation of the data and may partly reflect overfitting or limited model complexity rather than a truly perfect model–data match. Although such near-perfect fit values are unusual, the high and homogeneous factor loadings (ranging from 0.95 to 0.97) appear plausible given the specific characteristics of the task. The sentences used as indicators were constructed based on identical principles, designed to require the same underlying linguistic competencies, and carefully calibrated to ensure comparable levels of difficulty. These shared properties likely account for the high degree of consistency observed in the data. As expected, the correlations between the LPA and convergent measures were significantly higher than those with divergent measures, further supporting the LPA as a global measure of language skills. The finding that a single latent factor accounted for most of the variance in sentence repetition performance is consistent with the view that sentence repetition primarily reflects a broad underlying language ability factor (Tomblin and Zhang, 2006). At the same time, this psychometric one-dimensionality does not imply that sentence repetition relies on a single linguistic subcomponent. Instead, it reflects an underlying language ability factor that draws on the coordinated contribution of various interacting systems, such as lexical, grammatical, phonological, and sensorimotor processes (Bates, 1992; Kuhl, 2010; Charollais et al., 2013). It also aligns with clinical studies showing that sentence repetition is a sensitive marker of developmental language disorders in both monolingual and bilingual populations (Conti-Ramsden et al., 2001; Stokes et al., 2006; Marinis and Armon-Lotem, 2015; Abed Ibrahim et al., 2018).

The low to moderate correlations between the LPA and phonological working memory (r = 0.30/0.34) are compatible with previous findings linking phonological working memory to language skills (Gathercole et al., 1997). They also indicate that sentence repetition performance is not independent of memory-related demands. This is consistent with the nature of sentence repetition tasks, which require children not only to process lexical, grammatical, and phonological information, but also to temporarily retain and reproduce verbal material. Nevertheless, the substantially higher correlations with vocabulary, grammar, and another sentence repetition test suggest that LPS scores more strongly reflect oral language abilities than phonological working memory. The LPA should therefore be interpreted as a global indicator of oral language competence that includes some phonological working memory demands, rather than as a working memory task per se.

The correlation between the LPA and early mathematical competencies was substantial (r = 0.56), but not unexpected given the well-documented associations between language and mathematics (e.g., Peng et al., 2020). In particular, the acquisition of number words, number sequences, and digit knowledge shows clear parallels to vocabulary learning and symbolic language acquisition (Ennemoser et al., 2024). The association may also partly reflect broader child-level factors, such as working memory, attention, or general cognitive development, that influence performance across domains.

Analyses of the longitudinal data provided evidence that the LPA is sensitive to change, although the interpretation depends on the criterion used. At the level of the total sample, changes across measurement points were statistically significant, indicating reliable group-level increases over time. However, the overall difference between MP1+2 and MP6+7 did not clearly exceed the SEM suggesting that this change should be interpreted cautiously when applying a more conservative criterion of change beyond measurement error. In contrast, for children in the lower quartiles (0–25 and 25–50), the differences between MP1+2 and MP6+7 clearly exceeded the respective SEM values, indicating meaningful change beyond measurement error. Moreover, in the 0–25 quartile, four out of six differences between adjacent measurement points exceeded the respective SEM, suggesting that the LPA can detect meaningful change even over relatively short intervals of 4 weeks. This interpretation was further supported by the latent growth curve model, which fitted the data well and indicated a significant positive average slope across the seven measurement points as well as significant variance in individual growth rates. Taken together, these findings suggest that the LPA is sensitive to systematic developmental change over the assessment period. This is especially relevant because continuous monitoring is needed to determine whether children with language difficulties are making sufficient progress or may require additional support.

At the same time, the findings indicate that sensitivity to change was not equally pronounced across the full performance range. In the upper two quartiles, only limited growth was observed. This pattern is consistent with the negative intercept–slope correlation in the latent growth curve model and indicates that children with higher initial scores showed less subsequent growth. Plateauing or even negative slopes are often reported in the outer percentiles of progress-monitoring data (Jungjohann et al., 2021). Since language development typically continues beyond age six (e.g., Ulrich, 2017), this stagnation should not be interpreted as a developmental plateau. Instead, it more likely reflects limited measurement sensitivity of the current LPA in the upper performance range. Importantly, this does not necessarily indicate a pronounced ceiling effect in the strict sense, as the highest-performing quartile still scored, on average, clearly below the maximum possible score. Rather, the item set may not have been sufficiently challenging to capture further linguistic growth among children with advanced language skills. More complex sentences may be needed to ensure that performance in the upper range remains constrained by linguistic competence rather than primarily by more general task demands such as attention, sentence length, or phonological working memory. Future work should therefore develop and validate more challenging sentences (e.g., by increasing grammatical complexity or incorporating more advanced academic vocabulary) to extend the effective measurement range of the LPA.

The growth rates of monolingual and multilingual children were nearly identical, despite substantial differences in their initial performance levels. The comparable growth rates likely reflect an averaging effect across the range of language skills within each group, encompassing both high and low achievers. Nevertheless, the language skills of multilingual children in German as a second language remain inferior to those of monolingual children. This result aligns with previous studies on multilingual learning, which suggest that it can take 6–10 years for multilingual children to reach the language proficiency levels of monolingual children, depending on factors such as the age of onset of the second language and the quality of instruction in their first language (Collier and Thomas, 2017). These findings suggest that the LPA is a suitable tool for assessing the diverse oral language abilities of both monolingual and multilingual children for whom German is a first or second language. However, it cannot be used to make any statements about a multilingual child's overall language ability. This interpretation is consistent with international research showing that sentence repetition tasks can distinguish between children with and without language disorders in both monolingual and bilingual samples (e.g., Marinis and Armon-Lotem, 2015; Abed Ibrahim et al., 2018), and these tasks can capture developmental change over time when used to monitor the progress of dual language learners (Wood and Hoge, 2017).

Limitations and further directions

It should be noted that the reliability parameters reported in this study relate exclusively to the tests used within this sample and at the specific point of measurement (Rost, 2022). Nevertheless, we were able to replicate findings from a study conducted with children with DLD (Besca et al., 2023). Further evidence will be provided by a follow-up study, which has already been completed and includes approximately 100 primary school children. The results of this study are currently being analyzed.

The very high ICC of 0.99 indicates that children's relative positions on the LPA were highly stable across measurement points, i.e., rank ordering hardly changed over time. At the same time, such a high ICC may also reflect limited variability in individual change scores and a restricted range of abilities in this sample. Therefore, the reliability of change scores and potential ceiling effects at higher ability levels should be examined more closely in future studies with larger and more heterogeneous samples.

The sentences of the LPA for oral language are presented orally to create a communication situation that is as natural as possible for children. However, the variation in oral pronunciation between presenters could potentially introduce inconsistencies that might distort results. During the development phase of the LPA, we compared pre-recorded vs. live-spoken presentations and found no differences in children's performance. Despite this, recent findings on Mottier Syllables suggest that different presentation modes (e.g., digital vs. analog, fast vs. slow) may significantly affect repetition performance (Heeg et al., 2024). We are currently developing a tablet-based digital version of the oral language LPA that uses standardized audio recordings and an automatic speech recognition (ASR) module, to further standardize test administration und automate scoring. Pilot work on the use of automatic speech recognition in speech and language therapy suggests that AI-based tools may support more efficient and objectified scoring of oral language tasks in the future (Ehlert and Lüdtke, 2025). However, inter-rater reliability for the scoring was not formally assessed in the present study, so potential scorer variability remains an open question that should be addressed in future studies.

This study provides initial evidence of the LPA's sensitivity to change in language development. However, a key application of LPAs is to assess whether interventions yield the expected results, and this was not the aim of the present study. To date, this aspect has therefore not been investigated. The next step is to conduct an intervention study to evaluate whether the LPA can sensitively measure progress resulting from specific interventions. From a neurodevelopmental perspective, even subtle early constraints in the perceptual and sensorimotor systems that support language can have long-term consequences for language development and learning more generally (Bates, 1992; Kuhl, 2010; Charollais et al., 2013). Therefore, sensitive tools for monitoring short- and medium-term change are essential in order to detect whether children with such vulnerabilities actually benefit from specific interventions. Clinical intervention trials such as the LAMOPRESCO study on early language stimulation in preterm children (Charollais et al., 2024) demonstrate that brief, protocolized interventions can have measurable effects on different components of oral language in vulnerable populations. Our study complements this line of work by providing a tool that may be used to monitor such changes in everyday educational and therapeutic settings. A CG design will be employed, comparing the language skills of an EG with those of a CG, and testing for significant differences (Breitenbach, 2020). Additionally, intervention studies could assess whether the LPA can detect short-term progress after interventions in individual cases. This would be particularly important for determining whether the LPA is not only suitable for diagnosing language skills but also for evaluating intervention outcomes (Bishop and Thompson, 2023).

The sample sizes of both the cross-sectional and longitudinal studies, each with fewer than 100 participants, were sufficient to allow for initial reliable conclusions regarding the psychometric properties of the LPA. However, the participants were drawn exclusively from a single region in southern Germany, so generalization to other regions or educational contexts-and in particular to non-German-speaking children-is limited. In addition, children with severe articulatory disorders were excluded because their productions did not allow a clear distinction between articulation and grammatical errors; this reduces heterogeneity in the sample and limits the generalizability of the present findings to children without such disorders. It is therefore crucial to replicate the findings with a larger and more diverse sample of German-speaking children. Furthermore, additional background measures and demographic data should be collected to evaluate the representativeness of the sample. A large, representative sample is essential for developing a normative sample, which is required to convert raw scores into standardized scores. Standardized scores enable individual performance to be evaluated in relation to the normative population (Bishop and Thompson, 2023). They also provide a basis for deriving growth rates, which are necessary for setting realistic expectations for intervention measures (Förster et al., 2017).

Importantly, the CG was not included to compare the LPA group with an alternative treatment condition, but to control for potential test-repetition effects. Nevertheless, the consecutive-cohort design reduces the comparability of the EG and CG to some extent, and potential cohort effects cannot be ruled out.

The future practical application of the LPA in speech and language therapy lies in its potential to complement the evaluation of small-step goals in therapy. The aim is to determine whether progress achieved in small steps is also reflected in overall improvements in children's language abilities. This could provide evidence of measurable benefits from the intervention or indicate that the child has not yet sufficiently benefited, suggesting that therapy should be continued or intensified. On the one hand, this serves as an important basis for global therapy planning, and on the other hand, it provides justification for funding providers who finance therapy. In this sense, the LPA can be seen as a building block within broader data-based decision-making approaches and multi-tiered systems of support in early childhood education, in which regular progress monitoring is used to inform the intensity and focus of language support (e.g., Buysse and Peisner-Feinberg, 2010; Espin et al., 2012; Souvignier and Förster, 2025).

A further limitation concerns the treatment of missing data in the longitudinal sample. In the longitudinal sample, only about 4% of data in the EG were missing. Descriptive checks did not indicate systematic missingness, and sensitivity analyses using the non-imputed data yielded the same overall pattern of results. Missing values were imputed using a regression-based procedure within percentile groups of the LPA distribution. Given the very low proportion of missingness, the robustness of the findings, and our focus on group-level inferences, this parsimonious approach appears acceptable for the present study. However, more advanced methods, such as multiple imputation or full information maximum likelihood, should be considered in future studies particularly when higher rates of missing data are expected.

When comparing correlations using Fisher's Z, we did not apply formal corrections for multiple comparisons, as these analyses were primarily exploratory. Thus, individual p values should be interpreted with some caution, and emphasis should be placed on the overall pattern of convergent vs. divergent validity.

6 Conclusion

There is a significant research gap, as no established system currently exists for monitoring progress in oral language development, particularly in early childhood education. Monitoring the effectiveness of interventions is essential to identify early when interventions are not effective, enabling timely adjustments to ensure tailored support for each child (Deno, 2003). It is the responsibility of speech and language therapists to assess the long-term benefits of interventions for individual children. The development and testing of the LPA aimed to address this gap. From the perspective of data-based decision making and multi-tiered systems of support, brief, reliable and easy-to-administer progress measures in key domains are a central prerequisite for adapting instruction and intervention to children's needs (Deno, 1985; Buysse and Peisner-Feinberg, 2010; Espin et al., 2012; Souvignier and Förster, 2025). The present LPA provides such a tool for oral language in German kindergartens.

The findings suggest that the LPA is a reliable and valid tool for assessing oral language abilities in monolingual and multilingual children aged 4–6 years. The repeated-measures analyses, SEM comparisons, and latent growth curve model provide converging initial evidence of the LPA's sensitivity to change, particularly among children with lower initial language abilities. This study represents a first step in demonstrating the LPA's potential for documenting language development in children with poor language abilities. In sum, the LPA provides a psychometrically sound and practically feasible tool for monitoring oral language development in kindergartens. Future intervention studies will be crucial to determine whether it can also guide data-based decisions about the intensity and focus of language support for individual children.

Generative AI statement

The author(s) declared that Generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher's note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Footnotes

Edited by: Lucia Colombo, University of Padua, Italy

Reviewed by: Charollais Aude, Hôpital Charles-Nicolle, France

Neda Kianinezhad, Hakim Sabzevari University, Iran