Mathematics is regarded as a vital foundation for success in today’s modern society; however, it poses significant challenges for many students (Chu et al., 2017). In Ghana, as in numerous other African countries, mathematics is a core component of the educational curriculum at the basic school, senior high school (SHS), and colleges of education (CoE). It is a mandatory subject for all learners at these levels, serving as a crucial gateway for further academic pursuits in the nation (Ampofo, 2019). The low achievement in mathematics has been linked to various factors, including self-esteem, behavioral values, and anxiety (Phiri, 2019; Mwape, 2021).

Studies by Heinze et al. (2005) and Sauer (2012) have explored the relationship between mathematics achievement and interest, indicating that a positive interest in mathematics learning significantly contributes to academic performance. Building on this evidence, students’ interest in mathematics is increasingly understood as being influenced not only by achievement outcomes but also by underlying psychological and contextual factors that shape engagement and learning. Consequently, factors such as self-efficacy, mathematics anxiety, cultural and societal perceptions, and relevance of mathematics have gained attention as important determinants of students’ interest in mathematics. Self-efficacy is generally defined as a student’s belief that he or she is able to successfully accomplish a particular goal or result (Renner et al., 2008). According to research conducted by Arthur (2018), there is a positive correlation between academic achievement and self-efficacy, which presents self-efficacy as a predictor of academic performance. Mathematics anxiety is broadly understood as a state of fear that comes about in mathematical situations, especially when individuals perceive such situations as threatening their self-concept (Uusimaki and Kidman, 2004). Studies also indicated that mathematics anxiety is one of the reasons for students’ underachievement in mathematics. The core discomforts associated with mathematics are aversion, concern, and fear, and occasionally behavioral signs of tension, frustration, distress, helplessness, and mental disorganization (Ma and Xu, 2003).

Cultural beliefs and social norms have a tendency to establish expectations about who will excel in math and how mathematics is perceived in a society or community. For instance, mathematics is viewed in some cultures as being of very high value and worth, which motivates students to develop a good attitude and increased interest in the subject (Legewie and DiPrete, 2012). Cultural perceptions, including gender stereotypes and societal expectations, can affect students’ interest and participation in mathematics (Khalil et al., 2024). Relevance of mathematics refers to teachers’ ability to relate mathematical concepts to real-life applications and other subjects. When teachers effectively link mathematical ideas to phenomena in the world around them, students are more likely to develop a desire to learn mathematics (Retnawati, 2022). Studies have shown that students who recognize the usefulness of mathematics tend to perform better (Leyva et al., 2022). The relevance of mathematics to daily life and other curriculum areas profoundly affects the interest of students in the subject. Mathematical modeling is recognized to increase student attention, and its teaching and learning can benefit from the diverse educational views that mathematics education and its applications have brought about (Kaiser and Sriraman, 2006).

Integrating modeling into mathematics instruction can significantly improve students’ interest by connecting their real-life experiences with mathematical learning. This study seeks to contribute to this growing body of research by modeling factors that influence students’ interest in mathematics in colleges of education in the Ashanti region in Ghana.

The persistent issue of poor performance in mathematics among colleges of education students in Ghana and broader Africa remains a significant educational challenge. Despite various initiatives aimed at improving mathematics education, the disconnect between policy and student engagement persists, leading to a decline in interest and performance (Gray, 2014). Globally, there is a growing concern regarding students’ declining interest in mathematics, which has significant implications for their performance (OCED, 2019). In Africa, the situation mirrors global trends, with numerous studies highlighting the challenges faced by students in engaging with mathematics. Studies conducted in Zimbabwe indicated that factors such as teaching methods, societal perceptions of mathematics as a difficult subject, and curriculum relevance significantly influence students’ interest in mathematics (Mazana et al., 2019).

Investigating students’ interest in mathematics is one of the most critical areas that requires immediate attention in our quest to enhance mathematical performance in schools (Chen and Ennis, 2004). Several factors contribute to students’ lack of interest in mathematics, and even those who do express interest often do not perform as well as expected (Boadu and Boateng, 2024). While existing research has examined these factors individually, there remains a critical need to investigate their combined effects on student interest in mathematics in colleges of education. Without a clear understanding of these interrelated influences, efforts to enhance students’ interest in mathematics may be ineffective. The challenge of identifying how self-efficacy, mathematics anxiety, cultural and societal perceptions, and relevance of mathematics interact to influence students’ interest in mathematics remains underexplored in research. There is a pressing need to fill this gap by statistically modeling students’ interest in mathematics. Concerted efforts are required to devise strategies that enhance interest in mathematics in colleges of education in the Ashanti region in Ghana, as the issue of poor performance has become increasingly concerning. Therefore, it is essential to create a model that encompasses both global and localized factors that impact students’ interest in mathematics. This study aims to develop a model that builds on existing theories in the field to measure and explain the critical, often intangible constructs that determine students’ interest in mathematics in colleges of education in Ghana.

This study is underpinned by an integrated theoretical framework combining the foremost psychological, social, and educational theories. The dominant theories that inform this study are Social Cognitive Theory, Expectancy-Value Theory, and Sociocultural Theory. Bandura’s Social Cognitive Theory proposes that human behavior is a product of the interaction of personal factors, behavior, and the environment in a model of triadic reciprocity. Within the context of mathematics, self-efficacy is defined as a student’s belief in her or his capability to skillfully organize and execute the actions that are required in order to solve a problem and succeed in mathematics (Bandura, 2002). Social Cognitive Theory posits that students who have a high self-efficacy of mathematics are more willing to attempt difficult tasks, more persistent in the efforts they put in to overcome the challenges they encounter, and more passionate about the subject.

Their performance strengthens the positive self-efficacy belief and hence forms a motivational cycle. An extensive body of work confirms the strong positive mathematical correlation of self-efficacy with interest, engagement, and achievement. In a meta-analysis, Huang (2013) argues self-efficacy to be one of the strongest predictors of a learner’s performance in mathematics. In the same manner, Zimmerman (2000) showed that self-efficacy is associated with greater use of cognitive and self-regulatory strategies that sustain interest. This framework includes the emotional aspect mostly through the concept of mathematics anxiety, defined as the tendency to experience fear and apprehension that disrupt the ability to manipulate and solve numerical and other mathematical problems. Mathematics anxiety is posited to have a negative impact on interest via a dual process: it consumes the cognitive resources (working memory) crucial in problem-solving (Ashcraft and Krause, 2007), and it activates negative emotional responses, which result in avoidance behavior. This leads to the negative cycle in which anxiety manifests itself in poor performance, which in turn leads to diminished interest and elevated anxiety. Anxiety, interest, and achievement are known to be inversely correlated.

Wang et al. (2015) reported that mathematics anxiety is a negative predictor of students’ interest in pursuing a STEM career. Eccles and Wigfield’s Expectancy-Value Theory (2002) offers a comprehensive analytical framework for the concept of interest. The expectation of success (self-efficacy) and the perceived value of the activity are two factors that influence an individual’s decision, effort, and outcome, according to the Expectancy-Value Theory. The relevance of mathematics enhances the “value” part of the Expectancy Value Theory, which contains attainment value (the importance of doing well in mathematics for a person’s sense of self), utility value (the importance of math for future goals), and intrinsic value (the pleasure that comes from the subject). A student who perceives mathematics to be useful for their life and future will most likely assign to it high utility value and therefore, interest and motivation. Hulleman et al. (2010) demonstrated by experiments that students who were taught to appreciate the utility value of mathematics (for example, writing essays about the relevance of mathematics) showed increased interest and improved performance, which was initially lacking. According to Lev Vygotsky’s Sociocultural Theory Vygotsky’s (2012), the process of cognitive development, the formation of interests and attitudes in particular, hinges upon direct surroundings and cultural environment. Most of the time, mathematics is considered a difficult subject.

There is a mathematics anxiety stereotype; some people may consider themselves “bad” at math, and some people may think math is only for “brilliant and good” students. As an example, research done by Beilock et al. (2010) illustrated that mathematics anxiety among female teachers may result, by means of social learning, in an impeded performance of girls.

A conceptual framework is usually a diagrammatical depiction of variables to be researched (Boadu and Boateng, 2024). This is developed based on the relevant theory studied and gaps identified in the empirical study. The conceptual framework, which builds on previous studies in the field, outlines the key factors influencing students’ interest in mathematics from the structural equation model (SEM). The hypothesized model succinctly outlines that self-efficacy, mathematics anxiety, cultural and societal perceptions, and the relevance of mathematics all exert influence on students’ interest in mathematics. This conceptual framework depicted in Figure 1 was a custom-designed model by the researcher, tailored to address the specific problem at hand and align with the stated hypotheses of the current study. This model emerged from an in-depth review of existing literature, subsequently integrated to form a comprehensive composite.

This research is based on the positivist paradigm, a framework that supports the discovery of scientific facts through theory testing (Boadu and Boateng, 2024). The core of the positivist research paradigm is built on deductive reasoning, as it involves formulating theories and analyzing them through quantitative computations. Consistent with this view, the study used a descriptive survey design and quantitative methods. This design was chosen for its ability to efficiently collect data suitable for identifying and measuring relationships between variables (Boadu and Boateng, 2024). To ensure methodological transparency and replicability, the philosophical orientation (positivism), logical approach (deductive reasoning), and research design (quantitative descriptive survey) are explicitly stated, allowing future researchers to reproduce the study under similar theoretical and procedural conditions.

A research population encompasses all units or individuals sharing specific traits from which a sample is taken for study (Boadu and Boateng, 2024). The target population for this investigation includes the roughly 1,200 students attending colleges of education in Ghana’s Ashanti Region. This group is pertinent to the research objectives, as their firsthand experiences provide valuable insights into the core issues being examined. In research, a sample is a strategically chosen segment of the population intended to represent the whole (Boadu and Boateng, 2024). The validity of generalizing results back to the broader population depends heavily on this representativeness. Accordingly, a sample of 300 students was drawn from the target population. The population frame, geographical scope, and sample size are clearly defined to enable accurate replication in comparable institutional and regional contexts. The sample size was determined according to the method outlined by Boadu et al. (2023), which provides a systematic approach for calculating the optimal sample size. The formula is

Where n=requiredsamplesize = 300 N = total population size 1,200 e = margin of error (0.05) confidence level 95%

The explicit presentation of the sample size formula, parameters, and computation ensures transparency and allows exact replication of the sampling procedure.

Sampling techniques are the methods researchers use to select participants from a population, ensuring the collected data is valid and reliable (Boadu and Boateng, 2024). This study employed a combination of stratified and simple random sampling. Stratified sampling was first used to ensure adequate representation of key subgroups within the population. Following this, a simple random sampling technique was applied within each stratum to give every student an equal chance of being selected. The sequential application of stratified sampling followed by simple random sampling is clearly specified to allow exact replication of the participant selection process. The primary instrument for data collection was a structured, self-administered questionnaire. This tool was chosen for its efficiency in gathering quantitative data by allowing respondents to select from predefined options.

The questionnaire was organized into two sections: Section A collected demographic information (e.g., age, gender, class level, and type of basic school attended) using a nominal scale. Section B measured the study’s constructs using a five-point Likert scale, ranging from 1 (Strongly Disagree) to 5 (Strongly Agree). The questionnaire items were adapted from established instruments in previous studies (Arthur et al., 2017; Arthur, 2018; Boadu et al., 2023; Boadu and Boateng, 2024) to align with this study’s specific objectives and hypotheses. It measured one dependent variable, Students’ Interest in Mathematics (SIM), and four independent variables: Self-Efficacy (SEE), Mathematics Anxiety (MAA), Cultural and Societal Perceptions (CSP), and Relevance of Mathematics (REM), totaling 50 measurement items.

The source, construct mapping, and total number of measurement items are explicitly stated to support replication and content validity assessment. To ensure clarity and validity, a pilot study was conducted with 40 post-college school students. This group was selected for their comprehensive experience with mathematics teaching and learning, making them well-suited to provide feedback. The pilot aimed to identify and rectify any ambiguous or misleading items. Based on this feedback, the questionnaire was refined. The final instrument demonstrated strong internal consistency, with a Cronbach’s alpha coefficient of 0.76. The pilot testing procedure, sample size, purpose, refinement process, and reliability outcome are clearly documented to support methodological rigor and reproducibility. Subsequent statistical analysis of the main data revealed a significant positive correlation between the four independent variables and students’ interest in mathematics in colleges of education (r = 0.78, p < 0.005), a finding consistent with existing literature on active learning strategies. Formal data collection proceeded over 3 weeks. Prior to commencement, an introductory letter from the researcher’s department was sent to school authorities to secure permission and schedule sessions at mutually agreeable times, ensuring the process was organized and minimally disruptive. The duration, authorization process, and administration procedures of data collection are specified to ensure procedural transparency.

The researchers formally requested permission from the university to conduct the study in an official request. We confirm that informed consent was obtained from all participants involved in the study. The consent process was written and detailed the nature of the research, ensuring participants were aware of their rights, including confidentiality and the right to withdraw. For participants under the age of 18, we confirmed that parental/guardian consent was obtained prior to participation. The authors stated that the study was approved by the Akenten Appiah-Menka University of Skills Training and Entrepreneurial Development Institutional Ethics and Research Committee with the reference number AAMUSTED/IERC/2025/023. Written informed consent was obtained from heads of the departments and lecturers, as well as from students. The privacy and anonymity of the participants were respected. The inclusion of ethical approval details, consent procedures, and participant protection measures ensures ethical replicability and compliance with international research standards.

The demographic information depicted in Table 1 collected from participants provides a clear picture of the individuals involved in this study on factors influencing mathematics interest. The data reveals distinct patterns across gender, age, class level, and educational background. The study included 300 participants, with a notable majority being male.

Specifically, 210 male respondents accounted for 70% of the sample, while 90 female participants represented 30% of the total. This distribution indicates a significant gender imbalance within the participant group, which may influence the generalizability of findings across gender lines. The substantial male majority suggests that any gender-based analysis should be interpreted with caution, as female perspectives are underrepresented in this sample. Given the SEM modeling focus and the absence of population-level weighting benchmarks for colleges of education in the Ashanti Region, poststratification weighting was not applied. Participant ages showed a strong concentration in the young adult category. The largest group consisted of 240 students between 19 and 22 years old, representing 80% of the total sample. Younger participants aged 15–18 years accounted for 20 students (6.7%), while those aged 23–26 years comprised 30 participants (10%). The smallest age group included only 10 respondents (3.3%) who were 27 years or older.

This age distribution suggests that most participants are at a stage where they are making important educational and career decisions, particularly regarding mathematics. The distribution across academic levels showed a pronounced concentration in the second year of study. Exactly 200 participants (66.6%) were in their second year, forming a clear majority. First-year and third-year students were equally represented, with 50 participants each (16.7%, respectively). This distribution indicates that most respondents have some college experience but have not yet reached the final stage of their program, potentially influencing their perspectives on mathematics interest and application. The type of basic school attended revealed a substantial majority of participants coming from public educational institutions. Specifically, 230 respondents (76.7%) attended public schools, while 70 participants (23.3%) received their basic education from private institutions. This significant difference may reflect broader educational patterns in the region and could influence participants’ prior exposure to mathematics instruction and resources. The demographic characteristics are summarized in Table 1.

A descriptive analysis was conducted to examine the distribution of responses and assess the normality of the data collected through the questionnaire.

Normality testing is a statistical process used to evaluate if a dataset approximates a normal distribution, which is a symmetrical, bell-shaped curve commonly found in various natural and social phenomena (Boadu and Boateng, 2024). The findings from Table 2 showed that respondents rated all measurement items positively, with each obtaining a mean score above the midpoint of three. The calculated means and standard deviations for every construct fell within acceptable thresholds, indicating that the data met the assumptions of normality. As illustrated in Table 2, the distribution of scores across all survey questions did not significantly deviate from normality, supporting the use of subsequent parametric statistical methods.

A variable reduction method called exploratory factor analysis (EFA) is used to identify the latent constructs and underlying factors that determine a set of variable structures (Boadu and Boateng, 2024). Finding the factors that represent the measurement variables is the main goal of EFA. Factor analysis aims to clarify the connections between the complete collection of variables rather than forecast results (Boadu and Boateng, 2024). The EFA results are depicted in Table 3. The factor structure of the data was examined using SPSS (version 26) for this analysis, which also evaluated internal reliability and provided insights into the relationships between the variables. The researcher assessed how each observed variable loaded onto its matching latent variable during the EFA process. As indicated in Table 3, the Kaiser-Meyer-Olkin (KMO) measure of sample adequacy was found to be 0.785, over the acceptable threshold of 0.5. A Strong correlation between the components is indicated by this exemplary value (Arthur et al., 2022; Boadu and Boateng, 2024).

With a chi-square value of 2001.606 and 171 degrees of freedom, Bartlett’s test of sphericity also produced a significant result with a p-value of 0.000, suggesting that there are enough correlations to warrant factor analysis. Five factors were identified by the analysis; they were then taken out and rotated. With a determinant value of 0.001, these factors accounted for 64.204% of the total variation. A rotated component matrix using the varimax rotation technique is shown in Table 3. This method was used because it simplifies the factors and makes interpretation easier. In order to guarantee optimal model performance, items with weak factor loadings were methodically eliminated, with each deletion being followed by an evaluation of the fit indices. Items with factor loadings below 0.50 were considered weak and sequentially removed. After each deletion, model adequacy was reassessed using standard fit indices (KMO, Bartlett’s test, total variance explained, and rotated factor structure) to ensure construct stability and theoretical consistency.

Confirmatory Factor Analysis (CFA) was conducted using AMOS (IBM Corp, 2019, version 26) to validate the five latent constructs: Students’ Interest in Mathematics (SIM), Self-Efficacy (SEE), Cultural and Societal Perceptions (CSP), Relevance of Mathematics (REM), and Mathematics Anxiety (MAA). The results indicate that the measurement model demonstrates an excellent fit to the data: The Chi-square Minimum Discrepancy (CMIN = 172.631, DF = 138) measures the difference between observed and model-implied covariances, with smaller values indicating better fit.

The normed chi-square (CMIN/DF = 1.251) adjusts for model complexity, with values below 3.0 indicating good fit. The Comparative Fit Index (CFI = 0.982) and Tucker–Lewis Index (TLI = 0.977) compare the proposed model with a baseline model, with values above 0.90 indicating acceptable fit. The Goodness-of-Fit Index (GFI = 0.944) and Adjusted Goodness-of-Fit Index (AGFI = 0.923) reflect the proportion of variance explained by the model, with higher values indicating better fit. The Root Mean Square Residual (RMR = 0.049) shows the average difference between observed and predicted covariances, with lower values preferred. The Root Mean Square Error of Approximation (RMSEA = 0.029) assesses fit per degree of freedom, with values below 0.05 indicating close fit, and the p-close statistic (PClose = 0.998) tests whether RMSEA is ≤ 0.05, where non-significant values indicate close fit. Collectively, these indices confirm that the measurement model is well specified and suitable for structural analysis. All retained indicators loaded significantly on their respective latent constructs, with standardized factor loadings exceeding acceptable cutoffs.

Items with weak factor loadings (below 0.50) were systematically removed to improve construct validity. As a result, six items were removed across SIM, SEE, CSP, and REM, while seven items were removed from MAA. The final model retained four indicators each for SIM, SEE, CSP, and REM, and three indicators for MAA. The CFA results confirm convergent validity of the constructs, as each indicator demonstrated strong loadings on its intended factor, and the overall model exhibited minimal error variance. The validated five-factor measurement model therefore provides a robust foundation for subsequent structural equation modeling. The final CFA model is presented in Figure 2, and detailed factor loadings and fit indices are reported in Table 4.

Reliability concerns the consistency and stability of the results produced by the instrument over time and across different contexts. A reliable measure yields similar outcomes when reapplied under comparable conditions, indicating its dependability (Boadu and Boateng, 2024). Table 6 presents the reliability statistics for the study constructs using Cronbach’s alpha (CA) and composite reliability (CR). In this study, internal consistency reliability was assessed using Cronbach’s alpha (CA). This statistical test evaluates the extent to which items within a construct measure the same concept. A research instrument is generally considered reliable if results can be replicated using a similar approach (Boadu and Boateng, 2024). The reliability results are depicted in Table 6. Student interest in mathematics (SIM) achieved a CA value of 0.834, self-efficacy (SEE) reached a CA value of 0.810, mathematics anxiety (MAA) recorded a CA value of 0.757, cultural and societal perceptions (CSP) obtained a CA value of 0.831, and relevance of mathematics (REM) attained a CA value of 0.783.

According to the recommendations provided by Boadu et al. (2023), an instrument’s acceptability is determined by its Cronbach’s alpha score, which should be 0.7 or higher. The CA coefficients for all constructs exceeded the accepted threshold, confirming the instrument’s reliability and justifying its use in the study (Bannor et al., 2023). In addition to CA, composite reliability (CR) was computed to further assess the internal consistency of the constructs within the structural equation modeling framework. The CR values recorded were 0.789 for student interest in mathematics (SIM), 0.893 for self-efficacy (SEE), 0.789 for mathematics anxiety (MAA), 0.911 for cultural and societal perceptions (CSP), and 0.871 for relevance of mathematics (REM).

All composite reliability values exceeded the recommended minimum threshold of 0.70, indicating adequate construct reliability and confirming the consistency of the measurement items. The consistency of the measurements supports the credibility of the findings and strengthens confidence in the research outcomes. Both the reliability of individual constructs and the overall instrument were examined. The constructs, along with their corresponding Cronbach’s alpha values and number of items, are presented in Table 6. Together, the established validity and reliability form a solid foundation for drawing meaningful conclusions and making informed decisions based on the research results. The items used in this study were self-designed, achieving a minimum Cronbach’s alpha reliability coefficient of approximately 0.7, indicating that they are fairly reliable for self-constructed instruments (Arthur et al., 2022).

One technique for breaking down the covariations or correlations between two variables in a structural equation model is path analysis. The objective of this method is to measure the degree to which a theoretically supported causal relationship between the variables accounts for the covariance. By making it easier to examine relationships between different independent factors and dependent variables, path analysis supports preexisting hypotheses put forth by other academics. The direct effect analysis, which was carried out using Structural Equation Modeling (SEM) using Amos (version 26), is shown in Table 7. The structural model demonstrated a satisfactory fit to the data. The chi-square value (CMIN = 172.631) with 139 degrees of freedom yielded a CMIN/DF ratio of 1.251, which is well below the recommended upper threshold of 3.0, indicating good model parsimony.

The incremental fit indices further confirmed the adequacy of the model, with the Tucker–Lewis Index (TLI = 0.977) and Comparative Fit Index (CFI = 0.982) exceeding the recommended cutoff value of 0.90. In addition, the absolute fit indices showed favorable values, with a Root Mean Square Error of Approximation (RMSEA = 0.029) and Root Mean Square Residual (RMR = 0.049), both below the acceptable threshold of 0.08. The Probability of Close Fit (PClose = 0.998) further supports the conclusion that the model closely approximates the population covariance structure. Collectively, these indices indicate that the proposed structural model provides an adequate and reliable representation of the observed data. The standardized structural path estimates are presented in Figure 3.

The relationship between two variables that happens independently of any other variables in the model is referred to as a direct effect (Boadu and Boateng, 2024). Researchers can ascertain the degree to which one variable influences another by assessing direct impacts, which can be crucial for developing hypotheses and theories. Specific path coefficients are used in structural equation modeling (SEM) to describe direct effects. These coefficients shed light on the strength and importance of the connections. While a non-significant direct effect can suggest that the relationship is not as strong, a strong direct effect implies that the variables are closely related.