Assessing what counts: findings from a grade 1 mathematics national diagnostic assessment in Lebanese public schools

1 Introduction

Diagnostic assessments serve a strategic role in educational planning by identifying students’ prior knowledge, learning gaps, and misconceptions before instruction begins (Thompson, 2026). This approach allows for personalized learning and improves instructional efficiency. Research shows that adaptive diagnostic tools significantly enhance teachers’ ability to group students, customize instruction, and provide timely feedback. These tools enhance pedagogical decision-making and efficiency in teaching mathematics (Alfageh et al., 2024). Diagnostic assessment embedded within curricular frameworks has been shown to deliver more actionable, instructionally useful information compared to traditional assessments that do not differentiate among individual learning gaps (Fan et al., 2021). In early childhood education, diagnostic instruments for literacy and numeracy have been rigorously validated, offering reliable and valid measurement and informing early interventions (Fatimah et al., 2024). However, the usefulness of diagnostic assessment depends not only on the information produced but also on the validity of the interpretations and decisions made from assessment scores. Contemporary validity theory emphasizes that it is the proposed interpretations and uses of scores, rather than the test itself, that are subject to validation. The strength of these interpretations depends on the evidence supporting the inferences made from students’ responses to conclusions about what they know and can do (Kane, 2013). This consideration is particularly important when assessment results are used to identify learning gaps and inform educational interventions or policy decisions.

The Early Grade Mathematics Assessment (EGMA) has been widely utilized to measure foundational mathematical skills in early primary grades. Platas et al., 2016 discuss the development and intended uses of EGMA, emphasizing its role in informing policy and practice by providing reliable data on early mathematical knowledge. Similarly, Gove et al., 2017 present findings from early learning programs across six African countries, highlighting the effectiveness of large-scale assessments in evaluating and improving early mathematics education. The measurement of early numeracy requires attention to the multidimensional nature of foundational mathematical knowledge. Empirical work on early numeracy assessment has demonstrated the importance of measuring distinct but related skills, including counting, cardinality, number comparison, numeral identification, and early arithmetic reasoning, rather than treating early mathematics as a single undifferentiated construct (Purpura and Lonigan, 2015). More recent work has likewise emphasized the importance of establishing empirically supported benchmarks for different levels of early mathematical competence and examining the reliability of the resulting measures (Menacho et al., 2024). Gender disparities in early mathematics education have also been a subject of concern globally. Fonseca et al., 2023 conducted a multi-country analysis of gender differences in early grade reading and mathematics outcomes, revealing that while girls often outperform boys in reading, boys tend to have an advantage in mathematics. In the Lebanese context, Younes et al., 2023 explored how language and home factors influence students’ mathematics performance, concluding that socio-economic indicators, particularly home and family educational assets, significantly affect outcomes. National-level assessments can play a pivotal role in improving equity in early years’ numeracy education. Nortvedt et al., 2020 examined how a national-level assessment initiative in Norway may have improved equity in early years numeracy education, suggesting that such assessments can help identify and address disparities in educational outcomes. In addition, the integration of technology in early mathematics education has shown promise in enhancing learning outcomes. Pitchford et al., 2019 found that interactive apps could prevent gender discrepancies in early grade mathematics in a low-income country in Sub-Saharan Africa, indicating the potential of digital tools in bridging educational gaps.

At the system level, however, the interpretation of results from national or large-scale assessments requires consideration of assessment design and the scope of the claims that can reasonably be made from the resulting scores. Large-scale assessments necessarily sample selected domains and items from a broader curriculum or competency framework, meaning that observed scores represent evidence about a wider domain rather than complete measures of every component of that domain. The methodological literature on large-scale assessment also emphasizes that the assessment and sampling designs contribute to uncertainty in population-level estimates and should therefore be considered when drawing conclusions from such data (Cortes et al., 2025). The interpretation of proficiency levels also requires attention to the distinction between observed performance and performance standards. In educational assessment, cut scores are typically established through formal standard-setting procedures rather than representing inherently determined boundaries in the score distribution. Angoff and Bookmark are among the established approaches used to set performance standards, and empirical comparisons have demonstrated that alternative standard-setting procedures can produce different cut scores and classifications (Buckendahl et al., 2002). Thus, where assessment results are interpreted against a predefined proficiency threshold, the threshold should be understood within the assessment and policy framework through which it was established rather than assumed to represent a universally applicable boundary.

1.1 National diagnostic assessment (NDA) in Lebanon

In May 2025, the Lebanese Ministry of Education and Higher Education (MEHE), in collaboration with the Center for Educational Research and Development (CERD), launched a National Diagnostic Assessment (NDA) targeting all public-school students in primary education. This initiative marked a comprehensive assessment of foundational competencies in languages (Arabic and either English or French) and mathematics (Haj, 2025; MEHE, 2025a). The initiative adopted a “diagnosis before intervention” paradigm, paralleling global best practices in diagnostic assessment (Thompson, 2026).

1.1.1 Objective and rationale

The NDA was designed and administered to measure core competencies in literacy and numeracy, enabling the development of targeted intervention programs to address learning deficits in the 2025–2026 academic year. The assessment is a response to cumulative learning losses stemming from the COVID-19 pandemic, school closures, unprecedented economic crisis, teacher strikes, war, and overall instability in the country (Haj, 2025). Consistent with contemporary approaches to assessment validity, the interpretation of results should remain aligned with the claims supported by assessment design and available evidence (Kane, 2013). In particular, the assessment can provide evidence of observed achievement patterns and differences across domains and subgroups, while broader claims about students’ underlying competencies or the causes of observed differences require additional evidence.

1.1.2 Implementation process

The NDA was conducted in two stages. First, oral assessments were administered in Arabic and the first foreign language, focusing on reading and communication skills. Second, written assessments were conducted covering Arabic, the first foreign language, and mathematics. The process spanned three days and involved 871 public schools and adaptations were made for students with special needs through the Inclusive Education Unit (IEU). Preparation included collaboration with CERD, the Guidance and Counseling Directorate, IT units, and external specialists. Test questions were based on established scientific references and successful international models, adapted to Lebanon's context. Although the NDA is limited to the public sector only, the assessments were also available to private schools based on their request (Al-Markazia, 2025). The NDA instrument including its test blueprint, curriculum alignment, item development, translation protocols, and scoring rubrics, were developed and administered by MEHE and CERD. All relevant documentation is accessible via the repository link provided within the Data Availability Statement. The scale and structure of the NDA also place it within the broader context of large-scale educational assessment, in which practical constraints must be balanced against the need to obtain meaningful evidence across domains and populations. Accordingly, the present study treats the NDA results as evidence generated by a national assessment framework rather than as direct or exhaustive measures of all mathematical knowledge possessed by Grade 1 students.

1.1.3 Criticisms and counterarguments

The NDA received criticism concerning its logistical and administrative implementation. Regarding timing, MEHE justified administering the assessment during the final week of the school year, framing it as a year-end assessment intended to determine the knowledge and competencies students had acquired following completion of the instructional program. Thus, the NDA was positioned within the end-of-year assessment process while remaining diagnostic rather than a high-stakes examination determining students’ grades or progression (Haj, 2025). Nevertheless, from a measurement perspective, the low-stakes nature of the NDA constitutes an important consideration when interpreting student performance. In large-scale assessment research, test-taker motivation and engagement can influence observed performance, particularly when assessment results have limited direct consequences for examinees. This consideration may be particularly relevant when students undertake multi-item assessments over a relatively constrained administration period, as engagement may decline over the course of testing (Pools and Monseur, 2021). Moreover, the main concerns raised by teachers and school principals also involved systemic gaps in digital infrastructure, including unreliable internet access, inadequate computer laboratories, the absence of unified digital platforms, and weaknesses in data-management systems. These challenges may affect the capacity of educational systems to effectively implement and act upon diagnostic assessment results (Alfageh et al., 2024).

1.1.4 Grade 1 mathematics

The NDA encompassed a baseline understanding of early numeracy skills among Grade 1 students across public schools. Early mathematics proficiency is an important indicator of subsequent mathematical and academic development. Studies have shown that foundational skills in numeracy are strongly correlated with overall academic achievement and can influence long-term educational trajectories. For instance, research has demonstrated that early mathematical knowledge is predictive of future achievement and is teachable, underscoring the importance of early interventions (Platas et al., 2016). The assessment of such foundational competencies requires careful attention to the specific mathematical domains represented in the assessment. Research on early numeracy measurement has shown that early mathematical competence consists of multiple related skills and that assessments can provide more informative evidence when these components are examined separately as well as through an overall score (Purpura and Lonigan, 2015). Similarly, recent research on early mathematical assessment has emphasized the use of empirically supported benchmarks to distinguish levels of mathematical competence (Menacho et al., 2024). National assessments serve as valuable tools for identifying achievement gaps and informing educational policy.

Hence, this study analyzes the NDA data of Grade 1 mathematics to identify achievement patterns and equity issues related to gender and regional disparities. Examining performance for mathematical domains in the assessment provides a deeper understanding of how early numeracy competencies develop and where specific learning gaps emerge. Analyzing the three domains separately also recognizes that an overall mathematics score may conceal differences in students’ performance across specific areas of foundational mathematical knowledge. By situating these findings within the broader context of equity, the study highlights not only the overall trends in student performance but also the differentiated challenges faced across subgroups. Ultimately, this study aims to answer the following research questions: In all, this study generates evidence to guide targeted policy measures and instructional interventions aimed at strengthening early mathematics education in Lebanon.

• How do Grade 1 students in Lebanese public schools perform across the domains of Natural Numbers, Addition and Subtraction, and Geometry, and what patterns of proficiency and correlations emerge between these domains?
• To what extent are there differences in mathematics performance by gender and governorate, both at the domain and overall levels?

2 Methodology

2.1 Study context

In Lebanon, MEHE is the primary governmental authority responsible for setting national education policy, managing public education institutions, overseeing private schools, and coordinating with international organizations on education-related issues. In parallel, CERD is a semi-autonomous governmental institution under the authority of MEHE. The center is primarily tasked with curriculum development, educational research, and teacher training (CERD, 2020; World Bank, 2021). Within the Lebanese education system, public schools are completely managed by MEHE and primarily serve students from low-income backgrounds. They represent about 30% of the total student population in Lebanon (CERD, 2023).

At the Grade 1 level, mathematics instruction is structured around four main domains: arithmetic and algebra, geometry, and measurement. Within arithmetic and algebra, the curriculum emphasizes natural integers and the operations of addition and subtraction. The geometry domain introduces foundational concepts including location, recognition of solid and plane figures, and basic geometric transformations. Measurement is introduced through the concept of length, laying the groundwork for more advanced topics in later grades (CERD, 1997). For grade 1, mathematics is taught for five periods per week, with an estimated total of around 150 instructional periods per academic year.

2.2 Instrument

The assessment was a standardized instrument developed by MEHE in collaboration with CERD. It was designed to be completed in 50 min as a paper-based test. It included 30 questions covering three domains of foundational mathematics for Grade 1: natural numbers (12 questions), addition and subtraction (16 questions), and geometry (1 question). The final question addressed student comfort level, asking students to select a face (very comfortable to very uncomfortable) reflecting their experience after completing the assessment. The questions were a mix of multiple-choice, matching, and open-ended fill-in-the-blank formats. The natural numbers domain included six questions; each composed of two parts. The questions assess counting, number identification and writing, number sequence completion, and understanding of place value. Next, the addition and subtraction domain included six questions as well, with each having two or three parts. These questions cover basic addition and subtraction operations, missing number problems, and vertical calculation. Lastly, geometry was assessed through one matching exercise requiring students to identify and connect common two-dimensional shapes with their names (square, rectangle, triangle, and disc).

2.3 Data collection

The NDA was implemented in public schools under the supervision of MEHE. School principals coordinated the administration process, while Grade 1 teachers directly administered the assessment to students during regular school hours. To ensure consistency, teachers corrected the assessments according to a standardized scoring guide developed by MEHE. Results were then entered into a centralized electronic submission platform for compiling the national dataset. The NDA represents a census attempt targeting 871 public schools across all Lebanese governorates with a total of 19,640 students in Grade 1. Data integrity checks were performed by MEHE with the aim of verifying the accuracy of the correction and improving the quality of the final results.

2.4 Dataset and preprocessing

The datasets from the NDA were made publicly available by MEHE through its official website. The national database included anonymized school identifiers, anonymized student identifiers, gender, governorate, and question-level responses for each student. Each question was scored as correct (1) or incorrect (0), and a dash (–) response was used to indicate that the student did not attempt that question. In addition, the dataset contained responses to the final question on student comfort level, with answers ranging from A to E, as defined in the assessment instrument. Accordingly, a quantitative research approach was employed to analyze student performance on the Grade 1 mathematics NDA.

During preprocessing, data cleaning procedures were applied to improve the accuracy and reliability of the dataset. Cases with a dash (–) response across all questions of the assessment were identified. A total of 1,222 students in 53 schools across all Lebanese governorates did not attempt all the questions of the assessment. These students were excluded from the analysis, reducing the total sample size from 19,640 to 18,418 students across 818 schools, which led to a school participation rate of 93.9% and a student participation rate of 93.8%.

Student proficiency across domains was categorized according to the official NDA leveling guide established by MEHE and CERD (MEHE, 2025b). For natural numbers and addition and subtraction domains, students who score below 60% on the corresponding items are classified as Level 0, while those who score 60% or higher are classified as Level 1. The geometry domain is categorized binary-wise. Students who answer the 2D shape identification question incorrectly are classified as Level 1, and those who answer it correctly are classified as Level 1. Scores landing exactly at the 60% boundary are categorized as Level 1. This 60% cutoff represents an institutional diagnostic benchmark defined by MEHE and CERD for national monitoring, rather than an empirically derived psychometric standard-setting threshold.

2.5 Data analysis

All statistical analyses were conducted using IBM SPSS Statistics version 27 (IBM Corp., 2020). Descriptive and inferential statistics were used to examine student performance on the NDA and its associations with demographic and contextual variables. Descriptive statistics included frequencies, percentages, means, standard deviations, medians, and score distributions for the three mathematical domains and the total score. Frequencies and percentages were also calculated for domain-specific proficiency levels (Level 0 and Level 1) and student comfort levels.

For the initial analyses, Chi-square tests of independence examined associations between domain proficiency levels and gender and governorate. Independent-samples t-tests compared mean domain and total scores by gender, while one-way ANOVA examined differences across the seven governorates. A two-way ANOVA examined the main effects of gender and governorate and their interaction on domain and total scores. For governorate comparisons, homogeneity of variances was assessed using Levene's test. Because this assumption was violated for all variables (p < 0.001), Welch's test was used for the overall comparison and Games-Howell post-hoc tests were used for pairwise comparisons. Kruskal–Wallis H tests were used to compare total scores across student comfort levels, and Spearman's rank-order correlation assessed the association between comfort level and total score.

Effect sizes were reported alongside statistical significance. Cramer's V was used for Chi-square tests, Cohen's d for independent-samples t-tests, and eta-squared (η²) for ANOVA comparisons. Cramer's V values of 0.10, 0.30, and 0.50 and Cohen's d values of 0.20, 0.50, and 0.80 were interpreted as small, medium, and large effects, respectively. For η², values of 0.01, 0.06, and 0.14 were interpreted as small, medium, and large effects, respectively (Cohen, 1988). For the Kruskal–Wallis’ analysis, epsilon-squared was used as the effect-size measure. Pairwise comparisons following omnibus tests were adjusted using the Bonferroni method to account for multiple comparisons. Statistical significance was assessed at p < 0.05 and interpreted together with effect sizes and the magnitude of observed differences.

Because students were nested within schools, the potential non-independence of observations was explicitly addressed in the analysis. Students attending the same school may share instructional, organizational, socioeconomic, and contextual characteristics; ignoring this clustering can affect standard errors and statistical significance (Wright, 2019). The dataset comprised Grade 1 students from 818 public schools. School-level intraclass correlation coefficients (ICCs) were first estimated using unconditional linear mixed-effects models to quantify the proportion of variance in continuous achievement outcomes attributable to between-school differences. School-level descriptive statistics for mean total scores were also examined to characterize variation in achievement across schools.

Clustering-adjusted analyses were then conducted using mixed-effects models. For continuous outcomes, linear mixed-effects models (LMMs) were fitted using restricted maximum likelihood estimation, with gender and governorate as fixed effects and a random intercept for school. Gender × governorate interactions were examined within the same models. For binary domain proficiency levels, binomial generalized linear mixed-effects models (GLMMs) with a logit link were fitted, with gender and governorate as fixed effects and school as a random intercept. Pairwise comparisons of estimated marginal means for governorate effects in the linear mixed-effects models were adjusted using the Bonferroni method. These clustering-adjusted analyses were used to assess whether the substantive findings from the initial student-level analyses remained after accounting for the hierarchical structure of the data.

2.6 Ethical considerations

As this study relied exclusively on a publicly accessible secondary dataset officially published by MEHE, an independent Institutional Review Board (IRB) ethical protocol was not administered. Primary data collection was authorized and executed under national educational governance mandates. To uphold participant privacy, MEHE fully anonymized all student and school names prior to public dissemination. Crucially, as stated in its news article (MEHE, 2025a), MEHE explicitly released these raw, de-identified data into the public domain as an open invitation to researchers, academic institutions, and study centers to conduct independent secondary analyses aimed at informing evidence-based educational policies. Because this study was conducted strictly within the scope of this public research mandate using fully anonymized secondary data, formal IRB protocol review was deemed unnecessary under standard international research ethics principles governing open-access public data.

3 Results and analysis

3.1 Reliability of the assessment

The internal consistency of the dichotomously scored assessment subitems was evaluated using Cronbach's alpha, which is mathematically equivalent to the Kuder-Richardson 20 (KR-20) coefficient for dichotomous items. The overall 29-subitem assessment yielded a KR-20/α coefficient of 0.930. Domain-specific internal consistency was also high for natural numbers (KR-20/α = 0.853) and addition and subtraction (KR-20/α = 0.899). Geometry consisted of a single item and therefore did not permit estimation of internal consistency. Item difficulty was assessed using the proportion of students answering each subitem correctly, with frequencies and percentages of correct, incorrect, and unattempted responses reported at the individual-item level throughout the Results. Corrected item-total correlations ranged from 0.447 to 0.602 for the natural numbers domain and from 0.457 to 0.653 for the addition and subtraction domain, indicating positive associations between individual subitems and their respective domain scores.

Because several parent questions comprised multiple scored subitems, potential local dependency among subparts was also examined using phi-equivalent correlations. Correlations among subparts belonging to the same parent question ranged from 0.278 to 0.622 for natural numbers and from 0.278 to 0.727 for addition and subtraction. The relatively stronger correlations observed for some subparts were considered when interpreting internal consistency, given their shared parent-question structure.

An exploratory principal-axis factor analysis was conducted on the 28 subitems comprising the two multi-item domains. Sampling adequacy was high (KMO = 0.939), and Bartlett's test of sphericity was significant, χ²(378) = 200,322.361, p < 0.001. A two-factor solution accounted for 36.98% of the extracted variance, with the two factors substantially correlated (r = 0.641). The factor pattern was broadly consistent with the predefined multidomain structure. Thus, these findings present evidence of a correlated multidomain structure rather than strict unidimensionality. The geometry item was not included in the factor analysis because it constituted a single-item domain. Evidence of content and construct validity is grounded in the assessment blueprint and its alignment with the national mathematics curriculum established by MEHE and CERD.

3.2 Students and scores

The study included a total of 18,418 Grade 1 students from public schools across Lebanon. The sample represented a near-even gender distribution as reported in Table 1. Students were drawn from all seven Lebanese governorates, with the highest concentration of participants in the North (41.2%) and the lowest in Beirut (3.4%) as detailed in Table 2.

Table 1. Descriptive statistics of students by gender.

Table 2. Descriptive statistics of students by governorate.

Overall, students demonstrated a wide range of performance in the assessment. The total score, out of a maximum of 29, had a mean of 15.62. Performance also varied across the three mathematical domains, which will be examined in detail, as shown in Table 3.

Table 3. Descriptive statistics of total score and domain scores.

The percentile analysis of the total score (out of 29) for 18,418 students reveals substantial variation in performance, as presented in Figure 1. The 25th percentile was 10, meaning that one quarter of the students scored 10 or below, pointing to a significant group of low achievers. The median score was 16, indicating that half of the students achieved at or below this level, while the remaining half performed above it. The 75th percentile was 22, suggesting that the top quarter of students scored above 22, close to the maximum possible score of 29. This distribution shows that the middle 50% of students (interquartile range) scored between 10 and 22, a relatively wide span that reflects diverse levels of achievement. It is worth noting that 866 students (4.7%) scored zero on the assessment, which means their responses to all questions were incorrect.

From an equity perspective, the gap between the lowest and highest quartiles highlights disparities in learning outcomes, with some students advancing toward mastery while others remain at risk of falling behind. Such patterns underscore the need for targeted interventions to support low performing students while maintaining opportunities for high achievers to continue excelling.

3.3 Question and domain level performance analysis

3.3.1 Analysis of natural numbers domain questions

To evaluate foundational mathematical proficiency at the Grade 1 level, the NDA examined learners’ competence across procedural counting, visual comparison, number symbol recognition, forward number sequences, relational inequality, and base-ten place-value composition. Table 4 summarizes the descriptive statistics for the questions within the natural numbers domain. Learners demonstrated high proficiency in direct cardinal counting of small sets, with question 1a reaching the highest facility score across the domain. However, performance declined sharply when learners were required to map numerical quantities onto relational comparative terms (more, less, as many) in question 1b. This performance gap highlights a developmental divergence between basic procedural enumeration and abstract comparative quantification. Questions 2a and 2b evaluated translating verbal number names into two-digit numerals. Success rates declined from 2a (translating seventy-eight) to 2b (translating ninety-seven). The diagnostic distractors reveal a prevalent additive-concatenation error (writing 708 for seventy-eight or 907 for ninety-seven). Lower performance on question 2b suggests compounding difficulty associated with higher-decade numbers (90s).

Table 4. Descriptive statistics of questions under natural numbers domain.

Evaluation of forward verbal sequence completion revealed strong foundational skills when counting within familiar single-decade intervals as in questions 3a and 4a. Performance decreased moderately when tracking sequences in higher decades, such as the 30s range in question 3b. The alignment between missing-item placement across mid-sequence (question 4a) and sequence-terminal positions (questions 3a and 3b) suggests that students have established functional mental number-lines within lower numerical ranges. However, structural stability diminishes as numerical magnitude increases. Question 4b presented the highest difficulty across the entire natural numbers domain accompanied by the highest non-attempted rate (4.28%). When required to compare digits with reversed positions (78 … 87) using inequality symbols (<, >, =), over 56% responded incorrectly. This difficulty stems from two distinct cognitive hurdles. First, differentiating relative values when digits are inverted and second correctly interpreting and applying abstract relational operators (< vs. >). Comparing results across questions 5 and 6 provides clear evidence of the Concrete-Representational-Abstract (CRA) continuum. Questions 5b and 6a, supported by direct visual representations of base-ten blocks (4 ten-rods and 6 unit-cubes) yielded high success rates. Perceptual anchors allow learners to bypass abstract decomposition by relying on direct visual counting. Whereas performance dropped significantly in questions 5a and 6b, where visual scaffolding was removed or abstract unitizing was required such as recognizing that “10 ones is the same as 1 ten” or composing “6 tens and 4 ones” into the form “64” without pictorial aids.

3.3.2 Analysis of addition and subtraction domain questions

To assess computational fluency and operational reasoning at the Grade 1 level, the NDA evaluated performance across visual non-symbolic addition, base-ten scaffolding (adding 10), missing addend determination (algebraic initiation), basic addition facts crossing the ten-boundary, multidigit vertical addition algorithm performance, and basic subtraction facts. Table 5 summarizes the descriptive statistics for the questions within the addition and subtraction domain.

Table 5. Descriptive statistics of questions under addition and subtraction domain.

The results reflect a sharp cognitive gradient between visually supported binary addition and abstract symbolic multi-addend aggregation. Questions 7a, 7b, and 8a demonstrated high success rates which suggests that students effectively leveraged explicit iconographic representations (such as grouping objects or counting stars). In contrast, when presented with abstract symbolic three-addend addition in question 7c (6 + 5 + 4) performance declined markedly. Adding a decade (+10) using pictorial base-ten blocks yielded moderate performance in questions 8b and 8c. Despite the presence of physical block illustrations, nearly half of the cohort failed to execute the +10 operation efficiently. This difficulty directly aligns with the place-value weaknesses identified earlier in questions 5a and 6b. Learners who have not internalized the ten-unit block as a single countable unit struggle to apply the +10 pattern. This structural bottleneck is further demonstrated in basic addition facts crossing the ten-boundary (question 10), where facility rates fell across the board. Because these questions require decomposing an addend to construct a benchmark ten (7 + 9 = 7 + 3 + 6 = 16), the consistent error rate (around 56% - 57%) confirms that students rely heavily on counting-on using fingers rather than flexible mental strategies based on base-ten decomposition.

Question 9 evaluated missing-addend problems, which require operational flexibility and an early understanding of structural equality (for example, question 9a: 6 + ___ = 8). As structural complexity increases from simple unit distance to decade jumps, performance drops significantly, accompanied by high error rates (61.20% in 9c). The contrast between non-regrouping vertical addition (question 11a: 47 + 32) and regrouping vertical addition (question 11b: 28 + 45) reveals a strong procedural hurdle. Question 11b registered the highest failure rate across the domain (73.00% incorrect, 4.66% unattempted). In 11a, students could succeed by treating the tens and ones columns as independent single-digit additions without spatial interaction. However, when unit addition produces a value greater than 10 (8 + 5 = 13), students encounter procedural breakdown which reflects a gap in algorithmic execution and place-value integration. Basic single-digit subtraction in question 12 yielded consistent difficulty. Despite dealing with small single-digit quantities (<10), over 50% of students answered incorrectly or left the items unattempted. This performance gap compared to equivalent addition tasks (questions 7a and 7b) highlights the cognitive asymmetric difficulty between additive and subtractive processes in early childhood numeracy. While addition builds directly on forward counting strategies, subtraction requires backward sequence navigation or inverse-addition reasoning, both of which present greater cognitive load for early primary learners.

3.3.3 Analysis of performance on geometry: 2D shape nomenclature

To evaluate learners’ foundational spatial reasoning and geometric vocabulary, the NDA included a matching task (question 13) requiring students to link formal shape nomenclature (Square, Rectangle, Triangle, and Disc) to their corresponding two-dimensional geometric representations. Students had to match all four shapes correctly to receive the one-point credit on this question. Table 6 details the descriptive statistics.

Table 6. Descriptive statistics of the 2D shape-matching task (geometry).

The empirical performance indicates that slightly over half of the students struggle with basic 2D geometric recognition when formal vocabulary is introduced. According to the Van Hiele model of geometric thought (Van Hiele, 1986), early primary learners typically operate at Level 0 (Visualization), recognizing shapes holistically based on visual appearance rather than defined properties. Question 13 demands a transition toward Level 1 (Analysis) by requiring students to connect linguistic labels directly to spatial figures. Moreover, unlike simple multiple-choice formats, matching items require multi-step relational processing: The low unattempted rate (1.60%) confirms that students actively engaged with the task, but the high error rate (50.27%) demonstrates that visual familiarization does not translate into accurate nomenclature mapping without structured geometric instruction.

3.3.4 Analysis of natural numbers and addition and subtraction domain scores

The analysis of the assessment results reveals a clear performance disparity between the natural numbers and addition and subtraction domains. Tables 7, 8 present the scores in these domains.

Table 7. Descriptive statistics of natural numbers domain score.

Table 8. Descriptive statistics of addition and subtraction domain score.

For the natural numbers domain, while a significant number of students performed exceptionally well, with around 33% of students scoring 10, 11, or 12, there was also a marked group of low achievers, as evidenced by the 5.4% of students who scored zero. This suggests a wide range of foundational knowledge within the student population. In contrast, the addition and subtraction domain proved to be more challenging. Scores in this domain were more uniformly distributed, and the largest single percentage of students (8.9%) scored a zero, which is a significant indicator of widespread difficulty with these concepts. This data suggests that students, on average, struggled more with addition and subtraction than with natural numbers, particularly with more complex tasks like regrouping and finding missing addends. Since only one question involves a shape-matching task, the score on geometry is equivalent to that of question 13 as presented in Table 6.

3.3.5 Analysis of student performance levels by domain

Table 9 provides a breakdown of student performance levels by domain, using the leveling guide criteria. The results reveal significant differences in student mastery across the three domains.

Table 9. Descriptive statistics of students’ level by domain.

In the natural numbers domain, slightly more than half of the students (54.6%) achieved Level 1, meaning they scored 60% or higher on the questions in this domain. This indicates that around half of the students only demonstrated a solid understanding of natural number concepts. However, the other half were at Level 0, suggesting a substantial group struggles with these foundational skills. The addition and subtraction domain was the most challenging for students. A significant majority, 61.6% of students, were at Level 0, failing to achieve a score of 60% or more. Only 38.4% of students reached Level 1, indicating that less than half of the student population demonstrated a sufficient grasp of addition and subtraction skills. This is consistent with the analysis of Table 8, which showed a high percentage of students scoring very low in this domain. Finally, performance on the shape-matching task, representing geometry, was nearly evenly split with around half of the student population failing to correctly identify the basic shapes.

3.3.6 Correlations between scores

Pearson correlations were computed to examine the relationships among the three mathematical scores: natural numbers, addition and subtraction, and geometry. Correlations between each domain and the total assessment score were also reported descriptively. However, because the total score is composed of the three domain scores, these correlations represent part-whole associations and were not interpreted as independent evidence of relationships with overall performance. All correlations were positive and statistically significant (p < 0.001).

As presented in Table 10, the three domains were positively associated with one another. Natural Numbers showed a strong correlation with Addition and Subtraction (r = 0.726), while its associations with Geometry (r = 0.418) and the association between Addition and Subtraction and Geometry (r = 0.481) were moderate. These results indicate that performance across the mathematical domains was positively related, while the lower correlations involving Geometry suggest that performance in 2D shape nomenclature was less closely aligned with performance in the other two domains. The correlations between the individual domains and the total score were high, as expected, because each domain is a component of the total score. These values primarily reflect the mathematical overlap between the component scores and their aggregate and therefore should not be interpreted as independent evidence of the relative contribution of each domain to total performance.

Table 10. Correlation matrix between scores.

3.4 Comparative analyses across gender and governorates

3.4.1 Normality testing

Normality was assessed prior to inferential analyses. Given the large sample size (N = 18,418), tests such as Shapiro–Wilk and Kolmogorov–Smirnov were expected to indicate significant deviations even for trivial departures from normality. Inspection of Q–Q plots showed that the data closely followed the diagonal, with only minor deviations at the tails. These small departures are common in large samples and do not compromise the robustness of parametric tests.

3.4.2 School-level clustering

To assess the extent of between-school variation, null linear mixed-effects models containing only a random intercept for school were first estimated for the total score and the three domain scores. The resulting intraclass correlation coefficients (ICCs) are presented in Table 11.

Table 11. Intraclass correlation coefficients (ICC).

ICC = School-level variance/(School-level variance + Residual variance).

The ICCs indicate the proportion of total score variance associated with differences between schools; they do not imply that schools causally account for that variance. The ICCs indicated substantial clustering of student achievement within schools, supporting the use of school-level random effects in subsequent inferential analyses.

3.4.3 Associations of gender and governorate with student performance levels

The Chi-Square tests presented in Table 12 show a statistically significant association between student performance levels (Level 0 and Level 1) and both gender and governorate for all three domains.

Table 12. Results of chi-square test for gender and governorate.

The p-values for all these associations are less than 0.001, indicating statistically significant associations between performance level and both gender and governorate. However, despite the statistical significance, the effect sizes, as measured by Cramer's V, were small. For gender, Cramer's V values ranged from 0.059 to 0.094 across the domains, indicating small associations between gender and performance level. For governorate, Cramer's V values ranged from 0.116 to 0.154, also indicating small associations between school governorate and performance level. Thus, although gender and governorate were statistically associated with students’ performance levels, the magnitude of these associations was small.

To account for the clustering of students within schools, the three binary domain performance-level outcomes were reanalyzed using binomial logistic generalized linear mixed-effects models with a random intercept for school. The school random-intercept variance was statistically significant for all three domains, indicating meaningful between-school variation in the probability of attaining the defined domain performance level. Gender and governorate also remained statistically significant fixed effects across all three domains after accounting for school clustering as presented in Table 13.

Table 13. School-clustered generalized linear mixed-effects models for performance levels.

Relative to female students, male students had higher estimated odds of attaining the defined domain performance level in Natural Numbers [OR = 1.549, 95% CI (1.444, 1.662)], Addition and Subtraction [OR = 1.312, 95% CI (1.222, 1.409)], and Geometry [OR = 1.388, 95% CI (1.297, 1.485)], after accounting for school clustering. These results concern attainment of the defined performance-level threshold and should therefore be distinguished from the descriptive mean-score results, where female students had higher average scores.

Governorate remained significant in all three models. Relative to the South, Beirut showed higher odds of attaining the Natural Numbers performance level [OR = 4.818, 95% CI (2.671, 8.692)] and Addition and Subtraction performance level [OR = 4.370, 95% CI (2.383, 8.015)], while North also showed higher odds for Natural Numbers [OR = 2.070, 95% CI (1.519, 2.819)] and Addition and Subtraction [OR = 1.671, 95% CI (1.232, 2.267)]. Nabatieh showed lower odds than the South for Addition and Subtraction [OR = 0.565, 95% CI (0.383, 0.833)]. For Geometry, Beirut [OR = 4.077, 95% CI (2.390, 6.957)] and North [OR = 1.461, 95% CI (1.111, 1.920)] had higher odds relative to the South. These reference-category comparisons should be distinguished from the full set of pairwise governorate comparisons reported for the continuous outcomes below.

Overall, the clustering-adjusted analyses confirmed that the associations of gender and governorate with domain performance levels persisted after accounting for the nested structure of the data.

3.4.4 Gender differences in domain and total scores

Independent-samples t-tests were conducted to examine gender differences in student performance across the two domains (Natural Numbers, Addition and Subtraction), the 2D shape nomenclature (Geometry), and the total assessment score. Mean differences were calculated and negative values indicate higher average scores among female students.

As presented in Table 14, the t-tests indicate statistically significant differences between male and female students across all evaluated measures (p < 0.001). Across all measures, female students had higher mean scores than male students, as reflected by the negative mean differences, including a difference of 1.601 points in the total assessment score. Standardized effect sizes (Cohen's d, reported as positive absolute magnitudes) ranged from 0.146 to 0.212, indicating that the magnitude of the gender differences was small or below the conventional threshold for a small effect. Table 15 further presents the mean scores by gender. The average total score was 16.45 for female students compared with 14.85 for male students. Female students also had higher mean scores across the individual domains. These results indicate a statistically significant gender difference in performance, although the magnitude of the differences was limited.

Table 14. Results of t-test for gender.

Table 15. Means of scores by gender.

Gender differences in continuous achievement outcomes were subsequently reexamined using linear mixed-effects models with a random intercept for school. As shown in Table 16, gender remained statistically significant for the total score and all three domain scores after accounting for the clustering of students within schools.

Table 16. School-clustered linear mixed-effects models for gender.

After accounting for school-level clustering, female students continued to have higher model-estimated scores than male students across all four outcomes. Thus, the gender differences observed in the original analyses were not attributable solely to the non-independence of students within schools.

3.4.5 Governorate differences in domain and total scores

A one-way ANOVA was conducted to examine differences in student performance across the seven governorates. The homogeneity of variance assumption was assessed using Levene's test. Significant heterogeneity of variances was observed for all four outcomes (p < 0.001). Therefore, Welch's robust test was used for the omnibus comparisons, followed by Games-Howell post-hoc tests to identify pairwise differences while accounting for unequal variances. As presented in Table 17, statistically significant differences were observed across governorates for all four outcomes. However, the eta-squared values ranged from 0.013 to 0.032, indicating small effect sizes. Thus, although statistically significant differences in performance were observed across governorates, the proportion of variance associated with governorate was limited.

Table 17. Results of ANOVA for governorate.

Games-Howell post-hoc comparisons indicated that Beirut had significantly lower mean scores than all other governorates across all four outcomes (all p < 0.001). For the total score, Nabatieh had a significantly higher mean than each of the other governorates (all p < 0.001), while several comparisons among the remaining governorates were not statistically significant. In particular, Mount Lebanon did not differ significantly from Bekaa or South, and Bekaa did not differ significantly from Baalbeck-Hermel or South. Similar patterns were observed for Natural Numbers and Addition and Subtraction, for which most pairwise comparisons were statistically significant. In both domains, Beirut scored significantly lower than all other governorates, while Nabatieh scored significantly higher than all other governorates (all p < 0.05).

For Geometry, the pattern of pairwise differences was more limited. Beirut scored significantly lower than all other governorates, while North differed significantly from Mount Lebanon, Baalbeck-Hermel, Nabatieh, and South. Several other comparisons, particularly those involving Mount Lebanon, Bekaa, Baalbeck-Hermel, Nabatieh, and South, were not statistically significant.

Table 18 presents the descriptive mean scores by governorate. Beirut had the lowest mean scores across all four measures, including a mean total score of 10.44 out of 29. Nabatieh had the highest descriptive mean total score at 18.04, followed by South at 16.82. These descriptive differences should be interpreted alongside the Games-Howell results, as not all differences between the remaining governorates were statistically significant. Overall, the findings demonstrate statistically significant variation in student performance across governorates, while the small eta-squared values indicate that governorate accounted for a relatively modest proportion of the variance in scores.

Table 18. Means of scores by governorate.

To further characterize the extent to which achievement varied across schools, the mean total score was calculated separately for each of the 818 schools. The school-level distribution showed substantial variation in average achievement across the 818 schools. The mean school score (unweighted mean across the 818 schools) was 16.72 out of 29 (SD = 4.97), with a median of 16.70 and a range from 0.00 to 29.00. The middle 50% of schools had mean scores between 12.99 and 20.17 (IQR = 7.18). Together with the total-score ICC of 31.9%, these findings indicate substantial between-school heterogeneity in achievement.

Governorate differences in the total score and the three domain scores were re-examined using linear mixed-effects models with a random intercept for school. Governorate was treated as a fixed effect. As presented in Table 19, governorate remained statistically significant for all four outcomes after accounting for school clustering.

Table 19. School-clustered linear mixed-effects models for governorate.

The ICC values are based on the school and residual variance components from the corresponding governorate models. Each model included a random intercept for school. The significant omnibus effects indicate that differences among governorates remained evident after accounting for the clustering of students within schools. The ICC estimates from these models remained substantial, ranging from 20.1% for Geometry to 29.7% for the total score, indicating that meaningful between-school variation persisted after including governorate in the models.

Bonferroni-adjusted pairwise comparisons of estimated marginal means were then examined to identify the specific governorate differences that remained statistically significant. The significant comparisons are presented in Table 20. The pattern of pairwise comparisons was broadly consistent across the four outcomes, with Beirut differing significantly from the other governorates in most comparisons. Additional significant differences were observed among several other governorate pairs, particularly involving North, Nabatieh, and South. Importantly, these comparisons were estimated while accounting for clustering at the school level, rather than assuming that all student observations were independent. Taken together, the school-level descriptive analysis and the mixed-effects models indicate that achievement varied substantially across schools and that governorate differences persisted after accounting for this clustering.

Table 20. Significant Bonferroni-adjusted pairwise comparisons from school-clustered models.

Mean differences are based on model-estimated marginal means and are expressed as the score for the first governorate minus the score for the second. Negative values therefore indicate a lower estimated score for the first governorate. p-values are Bonferroni-adjusted.

3.4.6 Gender and governorate interactions

A two-way ANOVA was conducted to examine the main effects of gender and governorate and their interaction on students’ performance across the domains and the total score. The models included gender, governorate, and their interaction for the domains and the total score. Levene's tests indicated significant heterogeneity of error variances across the gender-by-governorate groups for all outcomes (all p < 0.001). Therefore, the results were interpreted with this assumption violation in mind. As presented in Table 21, although the gender differences were statistically significant, their effect sizes (partial eta-squared) were small. Governorate was also significantly associated with all four outcomes. The effect sizes were also small, indicating that governorate accounted for a relatively modest proportion of the variance in each outcome. In contrast, none of the Gender × Governorate interactions were statistically significant. The effect sizes (partial eta-squared) for the interaction were negligible. These results provide no statistical evidence that the difference in performance between male and female students varied significantly across governorates.

Table 21. Results of ANOVA for gender × governorate interactions.

The full models explained between 1.9% and 4.2% of the variance. These values indicate that the variables included in the factorial models accounted for a relatively modest proportion of the observed variation in student performance. Overall, the results indicate statistically significant main effects of both gender and governorate across the assessed outcomes, whereas their interaction was not statistically significant.

The gender-by-governorate interactions were reexamined using linear mixed-effects models with a random intercept for school. Gender, governorate, and their interaction were specified as fixed effects, while school was treated as a random effect. The clustering-adjusted interaction results are presented in Table 22. The corresponding school-clustered main effects of gender and governorate are reported previously in sections 3.4.4 and 3.4.5

Table 22. School-clustered gender×governorate interaction effects.

The gender-by-governorate interaction was statistically significant only for the Natural Numbers domain indicating that the magnitude of the gender difference varied across governorates. Model-estimated means for Natural Numbers showed that female students had higher estimated scores than male students in each governorate, although the size of the difference varied. The female-male differences ranged from approximately 0.41 points in Mount Lebanon to 0.98 points in Beirut, with intermediate differences across the remaining governorates. These estimated means describe the pattern associated with the significant interaction without introducing an additional series of unplanned pairwise significance tests. Overall, the school-clustered interaction analyses provided limited evidence of moderation by governorate: the interaction was evident for Natural Numbers but not for the total score, Addition and Subtraction, or Geometry.

3.5 Student comfort level

The assessment concludes with a final question on student self-perceptions by having them select a face reflecting their feeling after completing the assessment. Table 23 provides a descriptive analysis of how students felt after completing the assessment. The results show a strong positive trend in students’ comfort levels, with around three-quarters of students reporting that they felt either comfortable or very comfortable. Furthermore, descriptive patterns show that higher total assessment scores were observed among students in higher comfort categories. It is worth noting that in Table 23, the mean for each comfort level represents the average total score of all students within that response category.

Table 23. Descriptive statistics of student comfort level.

The 932 students (5.1%) who did not provide a response to the comfort-level question were retained in the descriptive statistics but treated as missing in the inferential analyses involving comfort level. A Kruskal–Wallis H test confirmed statistically significant differences in total scores across the five self-reported comfort groups, H(4) = 1,320.039, p < 0.001, with an epsilon-squared effect size of 0.075 (N = 17,486). Bonferroni-adjusted pairwise comparisons showed statistically significant differences between all comfort-level pairs except between the Uncomfortable and Neutral groups (adjusted p = 0.346). Consistent with these findings, a Spearman rank-order correlation demonstrated a statistically significant positive monotonic association between self-reported post-test comfort level and total assessment score (ρ = 0.253, p < 0.001, N = 17,486), indicating a weak positive association. Because the self-perception question was administered immediately following test completion, this association should be interpreted strictly as non-directional. While higher-performing students may experience greater ease and subsequently report higher comfort post-hoc, the observed association does not establish a directional or causal relationship between comfort and performance.

As an additional analysis, Chi-square tests of independence were conducted to examine whether comfort-level distributions differed by gender and governorate. Comfort level was significantly associated with gender, χ²(4) = 60.295, p < 0.001, although the effect size was very small (Cramer's V = 0.059, N = 17,486). Comfort level was also significantly associated with governorate, χ²(24) = 395.244, p < 0.001, with a small effect size (Cramer's V = 0.075, N = 17,486). Thus, while both associations were statistically significant, the small effect sizes indicate that gender and governorate were associated with relatively modest differences in the distribution of student comfort levels.

4 Discussion

The results from the Grade 1 mathematics NDA raise significant concerns of early numeracy in Lebanon's public-school sector. The average total score was 15.62 out of 29 (∼53.9%), which lies below commonly used proficiency thresholds and below MEHE's own 60% cutoff for Level-1 classification. Hence, this signals that Grade 1 students have not reached the minimum mastery expected of first-grade mathematics by the end of the school year.

Moreover, a marked proportion of students classified as Level 0 across domains. Nearly 45.4% of students were at Level 0 in natural numbers, 61.6% were at Level 0 in addition and subtraction, and 51.9% were at Level 0 in 2D shape identification (geometry). These domain-level prevalences reflect systematic weaknesses corroborated by question-level analyses presented in sections 3.3.1 through 3.3.3. Many students failed tasks probing place value, regrouping strategies, multi-digit addition, and missing-addend reasoning, while question 13 (geometry) also showed over half of students answering incorrectly, indicating important gaps in basic shape recognition and spatial reasoning. These results suggest that learning difficulties are both broad (affecting multiple foundational constructs) and deep (affecting essential building blocks like place value and simple operations), which are skills expected to be developed in the Grade 1 Lebanese mathematics curriculum (CERD, 1997).

Both unadjusted and clustering-adjusted analyses identified statistically significant differences in performance by gender and governorate, although the magnitude of these differences was generally modest. Across all three domains and the total assessment, female students had higher mean scores than male students (16.45 vs. 14.85 for the total score). This difference remained significant in the school-clustered linear mixed-effects models, with females scoring, on average, 1.472 points higher than males on the total assessment (p < 0.001). In contrast, the binomial logistic mixed-effects models showed higher estimated odds for male students of attaining the defined Level 1 performance threshold across all three domains (OR > 1). This apparent difference between mean scores and threshold attainment may reflect differences in the distribution of scores around the proficiency cutoff and should therefore be interpreted as indicating that mean performance and proficiency-level classification capture related but distinct aspects of achievement.

Across governorates, both the descriptive results and clustering-adjusted models indicated substantial differences in student performance. Beirut recorded the lowest mean total score (10.44 out of 29), whereas Nabatieh recorded the highest (18.04), a difference of approximately 7.6 points. Bonferroni-adjusted pairwise comparisons from the school-clustered linear mixed-effects models showed that Beirut scored significantly lower than each of the other governorates across all four outcomes. These findings indicate that low achievement is not distributed uniformly across the country but varies substantially across both schools and governorates. This pattern is consistent with broader structural pressures documented in Lebanon's general education system, including regional differences in operational capacity, teacher availability, and student-to-teacher ratios (Danageuzian, 2026).

The standard factorial ANOVA found no statistically significant Gender × Governorate interaction for any outcome. However, the school-clustered models identified a statistically significant interaction for the Natural Numbers domain, indicating that the magnitude of the gender difference varied across governorates. Model-estimated means showed that female students had higher Natural Numbers scores in every governorate, with the female–male difference ranging from approximately 0.41 points in Mount Lebanon to 0.98 points in Beirut. Because this interaction was not significant for Addition and Subtraction, Geometry, or the total score, the findings suggest that the variation in gender differences was specific to the Natural Numbers domain rather than a consistent pattern across overall mathematics performance.

Why does an apparently national problem arise? Lebanon's education system has experienced multiple, overlapping shocks that provide plausible contextual explanations for widespread learning deficits. First, prolonged COVID-19-related school disruptions contributed to learning losses. Second, the macroeconomic crisis severely reduced the real value of teacher salaries and contributed to strikes, attrition, and migration. Third, broader instability and the effects of the conflict in southern Lebanon further disrupted schooling and reduced instructional time. World Bank analyses document substantial learning losses associated with persistent disruptions in Lebanon's public schools since 2019 (Pushparatnam et al., 2024). Human-rights and policy monitoring reports have likewise documented reductions in effective teaching days and substantial teacher dissatisfaction (HRW, 2022; Khurma, 2023). In recent years, public schools have also adopted a four-day instructional week, rather than the usual five-day week, in response to pressures on the public sector (Nehme, 2023). Collectively, reduced instructional time and instability in the teaching workforce may have constrained curriculum coverage and instructional continuity, providing a plausible context for the low performance observed across governorates.

These external factors were not measured as student-level variables in the present dataset. Their broader structural context, however, is reflected in documented changes in Lebanon's education system over the past decade (Danageuzian, 2026). The national teaching workforce contracted by 2.5%, while public-school full-time teacher contracts declined by 27.1% and the number of educators aged 30 and below fell by 30.2%, amid the public-sector hiring freeze and severe salary erosion (Danageuzian, 2026). Structural pressures also varied across governorates, with increases in Student-to-School Ratios (SSR) and Student-to-Teacher Ratios (STR) particularly evident in the North, Bekaa, South, and Nabatieh. Nationally, the STR increased by 7.1% to 11.9, while private-school enrollment increased by 19.4%, reflecting shifts enrollment away from the public sector (Danageuzian, 2026).

These structural changes are consistent with broader international evidence that crises can amplify learning losses and educational inequalities, particularly when instructional disruption is prolonged and remedial provision is limited (Appels et al., 2024; Böhmer and Wills, 2025). Research on early numeracy similarly indicates that foundational skills are sensitive to limited instructional opportunities and differences in home numeracy support (Balala et al., 2021). In Lebanon, economic pressures, disrupted schooling, and variation in instructional provision may therefore compound existing challenges in foundational mathematics learning (Danageuzian, 2026; Younes et al., 2023).

5 Limitations

Several limitations should be considered when interpreting the findings. First, the NDA dataset covered only public schools and did not include students from private schools. Therefore, the findings should not be considered representative of all students in Lebanon and their generalizability beyond the public-school sector is limited. In addition, the dataset did not include socioeconomic or language-background variables. Consequently, the analysis could not examine whether differences in students’ socioeconomic circumstances or language of instruction were associated with observed mathematics performance. The cross-sectional nature of the data also limits interpretation to performance at the time of assessment and does not permit examination of changes in achievement over time or causal relationships. Second, assessment administration and scoring were conducted by school personnel, introducing the possibility of variation in administration or scoring procedures across schools. Also, unattempted responses were retained as a distinct category in the descriptive analyses but were treated as missing in analyses requiring complete item sets, resulting in reduced analytic sample sizes and the potential for exclusion-related bias. Finally, the geometry domain was represented by a single item, limiting the breadth with which this domain could be assessed and compared to the other domains. The low-stakes nature of the NDA also represents potential validity consideration. Because students were not subject to high-stakes consequences based on their individual performance; motivation, attention, or fatigue during the assessment may have varied, particularly given the number of items assessed. Such factors may have influenced observed scores independently of students’ underlying mathematical proficiency and should therefore be considered when interpreting the results.

6 Conclusion and policy implications

In conclusion, the combination of a mean score below expected proficiency, the large shares of Level-0 students especially in addition and subtraction, and the small effect sizes for gender and governorate indicate a foundational learning problem in Grade 1 mathematics within Lebanon's public schools. The patterns are consistent with broader evidence that economic and systemic crises acting through reduced instructional time, teacher attrition, and resource constraints challenge early learning. Policy and practice implications follow directly from the findings. Recommended priorities include:

• Immediate intervention. Conduct targeted remediation programs in Grade 1 (and Grade 2) that address place value, number bonds, regrouping strategies, and missing-addend reasoning. These programs should use short instructional modules with formative checks.
• Monitoring and evaluation. Implement rapid cycle monitoring to track the intervention's effectiveness and prioritize longitudinal follow-up to check whether early gains are sustained.
• Teacher support and retention. Stabilizing the teacher workforce through emergency salary support and incentives to retain qualified teachers, and professional development focused on early numeracy pedagogy is critical.
• Instructional time. Restoring policies that secure consistent full-week instruction in public schools are essential. Any continued practice of reduced teaching days is likely to perpetuate national learning loss.
• Targeted instructional support. Using question-level diagnostic analyses (as presented in sections 3.3.1 through 3.3.3) to design differentiated instructional pathways. Past evidence shows that diagnostic-to-instruction models yield more substantial learning gains than assessments used purely for reporting (Platas et al., 2016).

Generative AI statement

The author(s) declared that generative AI was used in the creation of this manuscript. During the preparation of this work, the author used ChatGPT to support the drafting and refinement of text, as well as to check clarity and consistency. After using this tool, the author reviewed and edited the content as needed and takes full responsibility for the content of the published article.

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: Gavin T. L. Brown, The University of Auckland, New Zealand

Reviewed by: I. Putu Ade Andre Payadnya, Universitas Mahasaraswati Denpasar, Indonesia

Mehdi Parsinia, University of Cologne, Germany