Negative Self-Concept: Cross-Country Evidence of Its Importance for Understanding Motivation and Academic Achievement

Abstract

Academic self-concept, the belief in one’s ability, is a key motivational construct in educational psychology and large-scale assessments. The construct is typically measured by instruments with positively (“I usually do well in science”) and negatively worded items (“I am just not good in science”). A single latent factor is often assumed. Here, we investigated this assumption using international large-scale assessment data across two age groups of children in fourth grade and adolescents in eighth grade (N = 296,320 students, 23 educational systems). We, instead, found strong evidence of the substantiveness of a negative self-concept factor derived from negatively worded items. Exploratory and confirmatory factor analyses uncovered negative self-concept as being distinct from positive self-concept. Furthermore, theory-driven modeling supported the internal/external (I/E) frame of reference model effect on negative self-concept: achievement has a stronger effect on eighth graders’ negative self-concept relative to fourth-grade children across many countries, especially for mathematics. Overall, understanding students’ negative appraisals and negative beliefs of their ability is an important theoretical and policy imperative.

1. Introduction

Self-concept is an individual’s comprehensive understanding and perception of themselves, encompassing their traits, abilities, attitudes, and values [1]. It is viewed as a hierarchical and multidimensional structure [2], which includes a global self-concept and further divides into academic and non-academic self-concepts [3,4,5,6]. Academic self-concept refers to a student’s self-perception of their capabilities in academic areas [7,8] and is subdivided into domain-specific self-concepts, indicating that individuals may hold distinct self-concepts in different academic fields [4,9]. Academic self-concept is a critical component of research on academic motivation, achievement, and development [10,11].

The formation of academic self-concept relies on an individual’s self-assessment of their capabilities in specific academic domains, which can be influenced by their past academic achievements and learning experiences [12]. Notable academic success can enhance a student’s sense of academic self-efficacy and vice versa; a positive academic self-concept significantly boosts a student’s motivation and performance. Therefore, academic self-concept is linked to academic achievement.

However, unlike the well-studied positive self-concept, the role of negative academic self-concept and its effects on academic achievement are still unclear. This study aims to fill this gap by exploring whether academic achievement similarly affects negative academic self-concept, the extent of this influence, and how the relationship changes across age groups. For this purpose, the study analyzes a recent international large-scale assessment, the 2019 Trends in International Mathematics and Science Study (TIMSS 2019), to examine the applicability of the aforementioned model across different educational systems. To address these research questions, we propose the following objectives:

Research Objective 1: To characterize the factor structure of academic self-concept. For this objective, we hypothesize that there is a two-factor structure, where negative self-concept is distinct from positive self-concept (H1). This objective involves conducting exploratory factor analysis (EFA) to determine the optimal factor number and confirmatory factor analysis (CFA) to validate whether students’ perceptions of their academic abilities are bifurcated into multiple domains. We will also assess the consistency of this structure across different grade levels and subjects.

Research Objective 2: To investigate the predictor(s) of academic self-concept. For this objective, we use the internal/external frame of reference model (I/E model) to examine the effects of academic achievement on the hypothesized dual factors of self-concept. We further determine how these predictors change across age groups. We hypothesize that within I/E model framework, academic achievement will significantly impact both positive and negative self-concepts (H2a). Furthermore, we expect that the impact of academic achievement on negative academic self-concept will be stronger in older students compared to younger students (H2b), indicating developmental changes.

1.1. The Relationship Between Self-Concept and Academic Performance

There are three prevailing perspectives on the relationship between academic self-concept and academic achievement. The first, known as the skill-development model, posits that academic achievement is a direct determinant of academic self-concept [6,13,14]. Within this framework, academic success shapes self-concept by influencing students’ evaluations of their academic abilities. The second perspective, the self-enhancement model [15], views academic self-concept as an independent motivator that drives academic achievement. In this model, a positive self-concept is considered a catalyst for academic success because it directly affects students’ engagement and motivation towards learning activities.

However, both the skill-development and self-enhancement models focus on a unidirectional relationship between self-concept and academic achievement. In contrast, the reciprocal-effects model proposes a more complex bidirectional relationship, where prior academic achievement influences self-concept, and a strengthened self-concept, in turn, fosters further academic achievement through a continuous interactive process [1,6,7,16]. Moreover, this model also accounts for the autoregressive effects of both achievement and self-concept [17], offering a deeper understanding of their mutual interaction.

1.2. The Internal/External Frame of Reference (I/E) Model

The relationship between academic self-concept and academic achievement is commonly examined through the lens of the I/E model. This model clarifies how comparisons—both internal and external—affect the development of academic self-concept [18]. Focusing on lateral comparisons between achievements and self-concepts across various domains, the I/E model delineates two principal types of comparisons: social and dimensional. Social comparison involves a student comparing their performance in a subject with the achievements of peers in the same subject. This type of external benchmarking typically results in a positive correlation between achievement and self-concept within that domain. For example, a student who excels in mathematics and notes that peers perform relatively poorly might develop a higher math self-concept. In contrast, dimensional comparison or internal comparison occurs when students assess their performance in one subject against their performance in another subject. This comparison may lead to contrast effects, where achievements in different domains negatively correlate with self-concepts [19]. For instance, a student might feel competent in mathematics compared to language arts, which could diminish their self-concept in language arts, even if their actual performance in language is commendable [20,21].

The I/E model explains that while academic achievements across different domains typically correlate positively, cross-domain self-concepts often exhibit near-zero or negative correlations [16]. Both internal and external comparisons contribute to shaping academic self-concept, with the interactions between these processes leading to stronger correlations among cross-domain achievements than among self-concepts across domains [19]. This model provides a robust framework for understanding the complex dynamics between academic achievement and self-concept, highlighting the role of comparative evaluations in shaping students’ academic identities.

Nevertheless, the I/E model assumes a consistent effect of comparisons across different cultural contexts, which may not hold true universally. In collectivist societies, which emphasize group harmony over individual competition, the reflected glory effect—where individuals’ self-concept is boosted by the success of their group—may counterbalance the negative impact of peer comparisons, leading to a lesser influence of peer achievement on self-concept [22]. Additionally, the I/E model’s focus on static comparisons may overlook the dynamic nature of self-concept development, which can fluctuate over time and across different stages of a student’s academic journey.

1.3. Extensions of the Internal/External Frame of Reference Model

The I/E model, initially focused on mathematics and language domains, has been validated for applicability across various academic disciplines through substantial research. Dimensional comparison theory (DCT) suggests that the similarity between subjects influences the formation of academic self-concept, with positive paths for matching domains and negative paths for non-matching domains [21,22]. For instance, high achievement in one subject positively affects the self-concept in a similar subject, demonstrating a matching effect (also called assimilation effect). Conversely, high achievement in one subject can negatively impact the self-concept in a dissimilar subject, leading to a non-matching effect (also called contrast effect) [22,23].

This model has been further tested in studies encompassing multiple subjects beyond mathematics and language, such as the sciences. A study included mathematics, language, biology, chemistry, and physics in their research, finding that achievements in similar domains (e.g., mathematics and physics) lead to assimilation effects, while achievements in dissimilar domains (e.g., mathematics and language) result in contrast effects, thus supporting the I/E model’s multi-disciplinary applicability [24]. Similarly, another study tested the cross-cultural universality of social and dimensional comparison processes among reading, mathematics, and science [25]. Their findings demonstrated that high achievements in mathematics and reading positively influence self-concepts in matched domains and that comparisons between science and reading tend to result in assimilation effects, while those between mathematics and science show contrast effects, with these patterns being consistent across a sample of students from 15 countries. Additionally, the adaptability of the I/E model has been confirmed in non-traditional subjects such as social studies [26] and physical education [27]. However, others [28] noted that matching domains do not always yield positive effects, and non-matching domains do not invariably lead to negative impacts, as near-zero or negative cross-disciplinary effects on academic self-concept can also be observed in highly similar subjects. This nuanced understanding underscores the complexity of interactions within the I/E model, affirming its broad relevance across various academic contexts.

Although the I/E model has been extensively studied, research on its application in STEM fields (science, technology, engineering, and mathematics) remains limited. Understanding how students in these areas form their academic self-concept is essential for fostering their long-term success, especially given the increasing focus on STEM education. Expanding research into these fields could clarify whether the contrast and assimilation effects found in other subjects also apply to STEM, providing a fuller understanding of academic self-concept development.

1.4. Negative Self-Concept and Academic Performance

The I/E model effectively demonstrates how students form self-concepts through social and dimensional comparisons of academic achievements. However, prevailing research predominantly focuses on enhancing student self-efficacy through positive feedback and accomplishments, overlooking the fact that not all students consistently perform at high levels. The formation of negative academic self-concepts in response to poor performance in certain subjects and the development effects of poor academic achievement impacting negative academic self-concepts in line with the I/E model warrant further investigation.

Limited studies have explored the link between academic performance and negative self-concepts. Research indicated that students who perform poorly in reading-related tasks are more likely to develop a negative reading self-concept compared to their better-performing peers [29]. Moreover, as students progress through grades, the gap in reading achievement between those with negative and positive reading self-concepts widens, suggesting that early poor performance can lead to a persistent negative view of one’s academic abilities. A positive self-concept is considered a protective factor against test anxiety, but for students with low academic self-concepts, poor grades predict increased exam anxiety, perpetuating a cycle of poor academic performance and heightened anxiety [30]. Additionally, students with a negative self-concept tend to avoid test-like tasks, require more time to regain optimal motivation, and exhibit less adaptive coping responses, which are associated with worsened emotional states [31]. A study on students transitioning to secondary school found that a positive math self-concept negatively predicts later poor math performance, with those whose self-concept becomes more negative facing a higher risk of secondary school retention [32]. In highly competitive educational environments, a negative academic self-concept can diminish students’ motivation to learn and adversely affect their emotional well-being and academic performance. Therefore, it is essential to delve deeper into the association between academic achievement and negative academic self-concept, examining how these elements relate to each other within the context of the I/E model.

1.5. The Developmental Relationship Between Self-Concept and Academic Achievement

Research further indicates that academic self-concept is developmental and malleable across different educational stages. A longitudinal study [33] involving cohorts of 5th and 10th-grade students over one year examined the developmental relationship between academic achievement and self-concept. The findings revealed that prior mathematics and Chinese achievements had a stronger impact on subsequent self-concept in these subjects for pre-adolescent children compared to adolescents. Conversely, for adolescents, prior self-concept in mathematics and Chinese had a stronger influence on subsequent academic achievement in these subjects than it did for pre-adolescent children. A cross-sequential study [34] spanning grades 1 through 12 revealed that as students advance in grade levels, the frequency of dimensional comparisons in self-evaluations significantly increases, leading to an increasingly differentiated academic self-concept over time. Younger students tend to rate themselves highly across all subjects, while older students demonstrate more specialization and distinction in their self-assessments, allowing them to identify their strengths and weaknesses across various disciplines more precisely. As students progress through their educational journey, both their mathematical and language self-concepts tend to decline significantly [9]. Increasingly aware of their abilities relative to their peers through social comparisons, students experience a heightened impact on their academic self-concept [35]. This awareness intensifies the link between academic achievements and attitudes toward learning [36]. Consequently, focusing on the evolving relationship between academic achievement and self-concept is crucial for understanding student development.

2. Data Source and Sample

We utilized the TIMSS 2019 dataset, which includes information on both science and mathematics achievements and self-concepts for fourth (G4) and eighth-grade (G8) students. TIMSS 2019 data demonstrate cross-cultural consistency in student attitudes towards science and mathematics [37]. Only countries with complete data for all variables and subjects for both grades were included. The final matched sample consisted of 296,320 students (139,129 at G4 and 157,191 at G8) from 23 countries, cities, and territories (henceforth, countries).

2.1. Variables

2.1.1. Science Achievement

Science achievement was measured using scale scores in general science from the TIMSS 2019, which ranges from 0 to 1000, with a mean of 500 and a standard deviation (SD) of 100. The assessment evaluates students’ knowledge across various domains including earth science, life science, physics, chemistry, and environmental science. Each student was assigned only one of the 14 different test booklets due to the extensive number of assessment items [38]. Utilizing Item Response Theory (IRT) scaling and latent regression, TIMSS generated five plausible values (PVs) for each student. The current study presents results from the first PV to assess each student’s science achievement. Very similar results were obtained for the other PVs.

2.1.2. Mathematics Achievement

Similarly, mathematics achievement was measured using the scale scores from TIMSS 2019, which covered algebra, geometry, data and probability, and numbers. Each student was assigned only one of the 14 different test booklets, and five PVs were also provided for their mathematics achievement [36,39]. This study uses the first PV to assess each student’s mathematics achievement. Very similar results were obtained for the other PVs.

2.1.3. Science Self-Concept

The measure of science self-concept was constructed based on student responses to items related to students’ perceptions of their performances, abilities, and emotional experiences in science. Items included positively worded statements such as “I usually do well in science”, and negatively worded “Science is harder for me than for many of my classmates” and “I am just not good at science”. Scores for these items were calculated using IRT and were collectively referred to as “students’ confidence in science” in the TIMSS 2019 [40]. The scores were based on students’ responses to a 4-point Likert scale, ranging from 1 (Agree a lot) to 4 (Disagree a lot). In the current study, the original measure for Grade 4 students comprised seven items (labeled A to G) with a mean Cronbach’s alpha of 0.788. For Grade 8 students, the measure included eight items (labeled A to H) with a mean Cronbach’s alpha of 0.847.

2.1.4. Mathematics Self-Concept

The measure of mathematics self-concept was constructed based on student responses to items related to general mathematics, referred to as “students’ confidence in mathematics” in TIMSS 2019. Items included positively worded statements such as “I usually do well in mathematics” and negatively worded statements such as “Mathematics is harder for me than for many of my classmates” and “I am just not good at mathematics”. Scores were based on students’ responses to a 4-point Likert scale, ranging from 1 (Agree a lot) to 4 (Disagree a lot). The original measure for both G4 and G8 students’ mathematics self-concept consists of nine items (labeled A to I). In this study, the mean Cronbach’s alpha is 0.824 for G4 students and 0.871 for G8 students.

2.2. Data Analysis

All individual data were school-mean subtracted first as we were mainly interested in student-level variance. To explore the structure of self-concept, we first conducted an EFA and parallel analysis (PA) to determine the optimal number of factors [41,42]. Subsequently, we performed a CFA to validate the factor structure’s adequacy. After identifying the factors of self-concept, we proceeded with structural equation modeling (SEM) to examine the skill development effects of academic achievement on the multi-factorial self-concept.

The data were randomly split into two halves, one half for the training set, used for EFA, and the other half for the validation set, used for CFA. We used the series mean to impute missing values to address missing data. All analyses were conducted using Mplus 7.4 [43], employing the robust maximum likelihood estimation method (MLR). Model fit was assessed using several criteria: the Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC) were used to evaluate model fit quality. The χ²/df ratio measured model fit adequacy. The Comparative Fit Index (CFI) and Tucker-Lewis Index (TLI) assessed improvements relative to an independent model. The Root Mean Square Error of Approximation (RMSEA) evaluated the extent of model error, while the Standardized Root Mean Square Residual (SRMR) measured the average size of the residuals [4,9].

To assess the strength of the differences, Cohen’s d was used to calculate effect sizes based on the means and standard deviations of the relevant groups. This provided a standardized measure of the differences in the relationship between academic achievement and self-concept across grade levels.

2.2.1. Exploratory Factor Analysis (EFA)

EFA with GEOMIN oblique rotation was conducted on the training sample to identify the underlying factors of science and mathematics self-concept. The number of factors to retain was determined based on eigenvalues greater than 1, PA, and the interpretability of the factors. The final factor structure was refined by including items with factor loadings of 0.30 or higher and excluding items with significant cross-loadings.

2.2.2. Confirmatory Factor Analysis (CFA)

CFA was subsequently conducted on the validation sample to validate the factor solution for science and mathematics self-concept derived from the EFA. We compared three models: a single-factor model (Model 1), a two-factor model (Model 2), and a single-factor model with method effects (Model 3). These are standard models with which to examine factor structure of constructs with positively and negatively worded items with potential underlying method effects [44] (Figure 1). The best solution was confirmed based on the interpretability of the factors and a combination of fit indices, including smaller values of AIC, BIC, χ²/df, RMSEA (<0.080), SRMR (<0.080), and larger values of CFI (≥0.90) and TLI (≥0.90).

2.2.3. Structural Equation Modeling (SEM) for Skill Development Effects on Positive and Negative Self-Concepts

Based on the factor results from the EFA and CFA, we conducted SEM on the data from G4 and G8 students from various countries to analyze the skill development effects of academic achievement on positive and negative self-concepts. Specifically, we examined the effects of science and mathematics self-concepts and their corresponding achievements. The model fit was evaluated using AIC, BIC, χ²/df, CFI (≥0.9), TLI (≥0.9), RMSEA (<0.080), and SRMR (<0.080) indices.

3. Result

3.1. EFA of Self-Concept

PA conducted on data from almost all countries suggested a two-factor structure for both G4 and G8 students’ science and mathematics self-concepts. Figure 2 shows the frequency of the optimal factors from the PA analysis for student samples from 23 countries. Based on item content, the factors for science self-concept were labeled “positive science self-concept” and “negative science self-concept”. For G4 students, the positive science self-concept factor included three items, while the negative science self-concept factor included four items. For G8 students, both positive and negative science self-concept factors consisted of four items each. Similarly, the factors for mathematics self-concept were labeled “positive mathematics self-concept” and “negative mathematics self-concept”. In both G4 and G8, the positive mathematics self-concept factor included four items, and the negative mathematics self-concept factor included five items. The exploratory factor analysis across all countries showed that items had high factor loadings (greater than 0.40) on both positive and negative self-concept factors for science and mathematics. This consistent structure was evident in nearly all country-specific datasets. Detailed factor loadings can be found in Supplementary Tables S1–S4.

3.2. CFA of Self-Concept

During EFA, there were significant cross-factor loadings in some countries for certain items in the G4 sample—specifically, “I usually do well in science” (positive science self-concept item A), “I usually do well in mathematics” (positive mathematics self-concept item A), and “Mathematics makes me nervous” (negative mathematics self-concept item E. These three items were excluded from the CFA analysis for the G4 sample. For the G4 students, the final CFA analysis included six items for science self-concept and seven items for mathematics self-concept.

CFA was conducted to validate three models of self-concept. Figure 3 shows an overview of the optimal models while Supplementary Tables S5–S8 provide detailed CFA fit statistics. Model 1 attempted to explain students’ self-concept using a single factor, but CFA results indicated poor model fit across most countries. This suggests that a single factor is insufficient to capture the complexity of students’ self-concept in the domains of science and mathematics. Model 2, which treated positive and negative self-concept as two distinct but correlated factors, demonstrated a better model fit for the vast majority of countries, particularly in terms of mathematics self-concept. This indicates that distinguishing between positive and negative self-concepts aids in providing a more precise representation of students’ self-perception. Model 3 extended the single-factor model by incorporating method effects. While some national samples supported Model 3, overall, Model 2 exhibited significantly superior fit compared to Models 1 and 3. Additionally, Model 2 more accurately captures the multidimensional nature of students’ self-concept in science and mathematics and fits a broader range of national samples. Distinguishing between positive and negative self-concepts provides a more comprehensive understanding of students’ self-perception, outperforming the simplified perspective of a single-factor model.

3.3. Skill Development Model via SEM

Following the skill development model, we modeled paths from science and mathematics achievements to both positive and negative self-concepts in these subjects. SEM analyses indicate that data from the majority of countries support the model of dual-factor self-concept influenced by academic achievement for both G4 and G8 students (Figure 4). Detailed SEM fit results are presented in

Table 1. Fourth Grade Skill Improvement Structural Equation Modeling.

	AIC	BIC	χ2/df	CFI	TLI	RMSEA	SRMR
United Arab Emirates	532,440.10	532,820.88	31.84	0.958	0.944	0.049	0.074
Australia	100,701.75	101,007.13	12.77	0.929	0.907	0.063	0.042
Bahrain	116,752.48	117,056.74	9.38	0.947	0.931	0.054	0.078
Chile	82,813.28	83,101.10	12.19	0.943	0.925	0.073	0.133
Ontario, Canada	58,988.25	59,271.68	11.61	0.917	0.891	0.074	0.058
England	53,251.05	53,528.35	11.05	0.907	0.878	0.077	0.056
Hong Kong SAR	47,548.68	47,819.11	6.10	0.946	0.929	0.059	0.042
Ireland	78,965.12	79,257.69	15.72	0.891	0.856	0.080	0.057
Iran, Islamic Rep. of	123,765.48	124,071.89	11.85	0.927	0.904	0.060	0.082
Italy	59,833.70	60,115.92	7.27	0.931	0.909	0.058	0.039
Japan	59,174.52	59,462.61	17.11	0.882	0.845	0.088	0.067
Korea, Rep. of	50,750.83	51,035.08	22.54	0.865	0.823	0.105	0.098
Kuwait	104,803.73	105,094.65	11.58	0.937	0.917	0.069	0.142
Norway	65,719.74	66,004.74	11.24	0.901	0.870	0.072	0.046
New Zealand	98,075.55	98,372.76	11.88	0.913	0.885	0.066	0.053
Oman	149,932.68	150,245.50	13.76	0.936	0.916	0.061	0.104
Qatar	106,104.04	106,400.37	9.44	0.943	0.925	0.059	0.089
Saudi Arabia	139,352.78	139,654.22	11.97	0.939	0.919	0.063	0.124
Singapore	94,354.57	94,660.77	13.19	0.944	0.926	0.064	0.043
Turkey	68,308.06	68,594.06	7.75	0.944	0.926	0.058	0.033
Chinese Taipei	56,235.17	56,517.71	10.67	0.931	0.909	0.072	0.050
USA	195,466.14	195,791.86	33.70	0.918	0.892	0.086	0.158
South Africa	283,965.63	284,306.84	42.26	0.918	0.892	0.083	0.185

AIC: Akaike Information Criterion; BIC: Bayesian Information Criterion; χ²/df: Chi-Square to Degrees of Freedom Ratio; CFI: Comparative Fit Index; TLI: Tucker-Lewis Index; RMSEA: Root Mean Square Error of Approximation; SRMR: Standardized Root Mean Square Residual; The bolded values in the table signify the models that meet the acceptance criteria.

and

Table 2. Eighth Grade Skill Improvement Structural Equation Modeling.

	AIC	BIC	χ2/df	CFI	TLI	RMSEA	SRMR
United Arab Emirates	530,636.02	531,097.23	22.33	0.958	0.949	0.044	0.062
Australia	188,438.30	188,842.66	16.95	0.951	0.941	0.059	0.045
Bahrain	135,446.28	135,821.72	9.72	0.938	0.926	0.055	0.053
Chile	75,532.66	75,887.29	8.25	0.936	0.923	0.059	0.045
Ontario, Canada	65,006.37	65,355.59	10.23	0.937	0.924	0.070	0.058
England	64,681.10	65,023.04	8.57	0.920	0.903	0.067	0.058
Hong Kong SAR	57,120.32	57,460.60	8.07	0.946	0.935	0.066	0.043
Ireland	84,454.50	84,809.19	6.78	0.960	0.952	0.053	0.034
Iran, Islamic Rep. of	122,684.66	123,062.85	8.92	0.944	0.932	0.051	0.048
Italy	67,348.35	67,695.13	11.56	0.920	0.904	0.076	0.087
Japan	75,330.91	75,690.43	22.23	0.872	0.846	0.098	0.080
Korea, Rep. of	60,972.11	61,322.72	15.80	0.912	0.894	0.088	0.084
Kuwait	116,290.66	116,651.96	9.65	0.919	0.902	0.061	0.088
Norway	76,691.51	77,048.43	14.43	0.922	0.905	0.079	0.074
New Zealand	108,678.22	109,056.43	13.60	0.921	0.921	0.065	0.058
Oman	176,418.57	176,804.39	11.00	0.949	0.939	0.054	0.109
Qatar	83,929.31	84,280.31	6.52	0.942	0.930	0.053	0.045
Saudi Arabia	177,033.92	177,408.87	19.06	0.857	0.827	0.080	0.099
Singapore	86,752.38	87,117.40	11.12	0.944	0.932	0.065	0.062
Turkey	78,790.85	79,144.90	8.62	0.939	0.926	0.061	0.053
Chinese Taipei	79,812.66	80,178.49	16.50	0.934	0.920	0.079	0.082
USA	212,585.98	212,987.78	19.19	0.952	0.942	0.065	0.129
South Africa	537,057.87	537,514.67	32.28	0.947	0.937	0.055	0.114

Same notes as Table 1.

.

Specifically, within domains, we consistently found that the path coefficients from academic achievement to positive academic self-concept were negative. This occurs because higher scores on self-concept measures indicate stronger disagreement with statements. Consequently, a negative path coefficient suggests that higher academic achievement is associated with a higher positive academic self-concept. Conversely, the path coefficients from academic achievement to negative academic self-concept were always positive, indicating that higher academic achievement corresponds to lower negative academic self-concept in both science and mathematics for G4 and G8 samples. In contrast, the majority of educational systems showed that in different domains, the path coefficients from academic achievement to positive self-concept were positive, and to negative self-concept were negative. This implies that higher achievement in science (mathematics) is associated with a lower positive mathematics (science) self-concept and a higher negative mathematics (science) self-concept, though these effects were relatively smaller. Detailed path coefficients for the impact of academic achievement on self-concept across different educational systems can be found in Supplementary Tables S9 and S10. Figure 4 presents the path coefficient overall means (s.e.). Overall, these results suggest that the skill development model applies to negative self-concept, as strongly as, if not more so than, positive self-concept, making negative self-concept an underappreciated substantive part of the skill development model.

3.4. Negative Self-Concept Difference Between G4 and G8

We next focused on negative self-concept as a new substantive factor, analyzing the impact of academic achievement on negative self-concept among G4 and G8 students. We found that in 7 countries, G8 students exhibited a stronger effect of science achievement on negative science self-concept compared to G4 students, with Cohen’s d ranging from 0.029 to 0.132 (mean = 0.090) (Figure 5A). A similar pattern was observed for mathematics achievement and negative mathematics self-concept in 12 countries, with Cohen’s d ranging from 0.050 to 0.195 (mean = 0.091) (Figure 5B). The main findings indicate that for within-domain comparisons from childhood at G4 to middle adolescence at G8, the link between academic performance and negative self-concept in science and mathematics becomes more pronounced in many countries.

In comparisons across domains, findings from 4 countries suggest that, compared to Grade 4 students, Grade 8 students exhibit a more pronounced impact of science achievement on negative mathematics self-concept, with Cohen’s d ranging from 0.062 to 0.125 (mean = 0.088) (Figure 5C). Additionally, findings from 3 countries indicate a stronger effect of mathematics achievement on negative science self-concept in Grade 8 students compared to Grade 4 students, with Cohen’s d ranging from 0.027 to 0.093 (mean = 0.068) (Figure 5D). For cross-domain comparisons, the influence of academic achievement on negative self-concept tends to remain relatively more stable across age groups. In conclusion, our findings suggest the developmental aspect of negative self-concepts.

4. Discussion

This study, using large sample data from multiple countries, explored through EFA and confirmed via CFA that negative and positive self-concepts coexist as separate substantive factors, thus confirming our hypothesis H1. Additionally, SEM demonstrated that the I/E model can be effectively extended to science and mathematics academic achievement and both positive and negative academic self-concepts, additionally supporting hypothesis H2a. Within-domain comparisons revealed that higher academic achievement is associated with lower negative self-concept scores. Conversely, across-domains comparisons indicated that higher achievement in one subject is linked to higher negative self-concept scores in another subject. Furthermore, our findings highlight developmental differences between G4 and G8 students. The impact of academic achievement on negative self-concept is stronger in G8 students compared to G4 students in numerous countries but not all, consistent with hypothesis H2b.

4.1. The Impact of Academic Achievement on Positive and Negative Self-Concepts

This study forms an emerging literature highlighting the construct of negative academic self-concept as distinct from its positive counterpart [45]. Furthermore, our study extends the I/E model to include comparisons of subjects of science and mathematics whereas previous studies have generally investigated STEM and non-STEM subjects for cross-domain effects. Our findings, based on SEM, show that data from most countries support the skill development model, which posits that academic achievement influences a dual-factor self-concept.

Our results corroborate previous research regarding the relationship between academic achievement and positive self-concept [16,18,21,22,46]. One key finding of this study is the impact of academic achievement on negative academic self-concept both within and across domains. Within the same domain, poor academic performance consistently increases negative academic self-concept. However, across different domains, poor academic achievement in one domain generally weakens negative academic self-concept in the other domain, although this effect is not as strong. This can be compared to the self-verification perspective [47], which posits that poor academic achievement in one domain leads to a decrease in negative self-concept in another domain, reflecting a tendency to maintain a consistent self-concept. A potential explanation could be that individuals engage more in internal comparisons regarding their academic achievement, leading to contrast effects on negative self-concept [19].

These findings underscore the significant influence of academic achievement on negative self-concept, a critical facet of academic self-concept deserving careful consideration. This contribution enriches our understanding of academic self-concept, thereby deepening our comprehension of students’ academic self-cognition.

4.2. Developmental Differences Between G4 and G8 Students

The developmental differences between G4 and G8 students are noteworthy. Unlike previous research that found a weaker influence of academic achievement on positive self-concept in adolescent students compared to pre-adolescent students [33], our analysis reveals that the impact of academic achievement on negative self-concept can be stronger in G8 students compared to G4 students in numerous countries. This finding points out the increasing importance of academic performance in shaping negative self-concepts as students progress through their educational journey. The varying influence of academic achievement across grade levels suggests that middle adolescence is a crucial period for the formation of negative self-concepts, especially in mathematics. From a developmental perspective, this heightened impact on G8 students can be attributed to several factors. As students grow older, they become more aware of their academic standing relative to their peers. This increased awareness can lead to more frequent and intense social and dimensional comparisons [35], which in turn can amplify the impact of academic achievements on their self-concept, particularly negative beliefs. In younger students (G4), self-concept is still relatively malleable [34] and influenced by a broader range of experiences and feedback, not solely academic performance. Theoretically, this finding aligns with the developmental trajectory of self-concept formation, which becomes more differentiated and domain specific as children grow older [14,34]. In G8, students’ self-concept is more likely to be influenced by academic success or failure due to the increased emphasis on academic achievement and its perceived importance for future educational and career opportunities. This effect may be even more pronounced in mathematics due to its objective nature, where clear-cut success or failure can intensify perceptions of ability. As students in middle adolescence increasingly view mathematics as a “gateway” subject, especially for STEM fields, those who struggle may experience heightened negative self-concept.

Most of the countries showing significant increases in negative self-concept were from Asia, including the Middle East. In many Asian countries, students are influenced by Confucian culture, which emphasizes academic achievement but is often associated with negative psychological effects [48]. As a result, students in these regions tend to have lower self-concepts regarding their abilities [49] compared to non-Asian countries. However, research indicates that the I/E model is systematically weaker in Middle Eastern Islamic countries. Students in these regions may receive less evaluative feedback and are less encouraged to critically assess their academic skills compared to their Western and East Asian peers [50]. Further research is necessary to resolve the inconsistencies between these findings and those observed in Middle Eastern Islamic countries.

4.3. Practical Strategies for Supporting Student Self-Concept Development

Research has shown that several factors play a significant role in shaping both academic achievement and self-concept. Teachers’ expectations are particularly influential—students tend to perform better and develop a stronger self-concept when their teachers hold high expectations for them [51,52,53]. Positive feedback from teachers also boosts their self-concept [54]. Additionally, when teachers provide autonomy support—encouraging students to take initiative [55], and emotional support by fostering a supportive classroom environment [56], they help nurture a more positive academic self-concept.

Recognizing that the impact of academic achievement on negative self-concept becomes stronger as students get older underscores the need for targeted support at critical stages. For adolescent students, schools should consider implementing programs that specifically address the increased sensitivity of negative self-concept to academic performance. In addition, broader interventions can further support self-concept development. This might include offering additional academic support [57], creating a positive and inclusive school environment, and teaching social-emotional skills to help students handle comparisons and negative self-evaluations in a healthy way.

At the classroom level, teacher feedback [58] that is targeted and task-specific is more effective in shaping students’ self-concept than generalized praise. Feedback should focus on specific actions students can take to improve, rather than vague affirmations like “good job”. To foster a positive academic self-concept, teachers should address three key areas: goals, progress, and next steps [59]. Strategies include offering process-oriented feedback that helps students refine their learning strategies, providing growth-oriented feedback that highlights effort and improvement, and encouraging self-assessment over peer comparisons to promote internal progress. Moreover, through growth mindset interventions, teachers can enhance students’ academic self-efficacy by fostering the belief that their intelligence and abilities can improve through effort and effective learning strategies [60,61], thereby promoting academic achievement [62] and improving self-concept.

This research aims to address the current gap in understanding negative academic self-concept, enhancing our understanding of how poor academic performance may be linked to negative perceptions of one’s academic abilities. By examining the relationship between achievement and negative self-concept, this study provides a deeper insight into the complexity of academic self-concept, enabling educators to more effectively design supportive educational environments.

4.4. Limitations and Future Research

Despite the contributions of this study, several limitations must be acknowledged. We focused on the skill development model to examine the effects of science and mathematics academic achievement on negative academic self-concept using cross-sectional data. However, there is literature supporting the reciprocal model, which suggests a bidirectional relationship between academic achievement and self-concept [1,6,7,14,16,28]. Future research can consider how negative academic self-concept influences academic achievement in a longitudinal design. Such studies would help consolidate the foundational understanding of negative academic self-concept and its broader implications. Moreover, our study primarily concentrated on science and mathematics. Future research could benefit from examining other academic domains [22] to determine if similar patterns exist. Additionally, longitudinal studies could provide more comprehensive insights into how negative self-concepts develop and change over time [6]. It would also be valuable to explore the effects of factors such as school climate [63,64], family characteristics [65] such as parental involvement [66], and students’ learning motivation [67,68,69] and strategies [70,71]. Investigating how these aspects influence academic performance and negative self-concept could provide a clearer understanding of the factors involved.

In conclusion, this study confirmed the coexistence of positive and negative self-concepts and demonstrated the applicability of the I/E model to science and mathematics academic achievement and to negative academic self-concept. Our findings revealed developmental differences between G4 and G8 students across various educational systems, highlighting the increasing impact of academic performance on self-concept as students mature. These results reinforce the importance of tailored educational interventions that address both positive and negative self-concepts to enhance student outcomes and psychological well-being.