Mediating Effects of Learning Environment Adaptation and Self-Efficacy on College Students’ Self-Directed Learning

The mediation effect model is an analytical method used to explore how an independent variable (X) influences a dependent variable (Y) through one or more mediator variables (M). In this paper, there are two mediator variables: Self-efficacy (M1) and Learning environment adaptation (M2). Based on existing research, their mediating role model is illustrated in Figure 1 and Figure 2. The independent variable is Self concept (X), and the dependent variable is Self directed learning (Y).

Figure 1.

Mediation Effect Model

Figure 2.

Mediation Effect Model

A mediation effect model with two mediator variables can typically be tested through the following four regression equations. Equation 1 examines the effect of the independent variable X on the mediator variable M1; Equation 2 examines the effect of the independent variable X on the mediator variable M2; Equation 3 assesses the direct effect of the independent variable X on the dependent variable Y, as well as the effect of the mediator variable M on the dependent variable Y. Equation 4 evaluates the total effect of the independent variable X on the dependent variable Y.

The effect of the independent variable Self concept (X) on the mediator variable Self-efficacy (M1) can be expressed as: (1)M1=a1X+ε1$${M_1} = {a_1}X + {\varepsilon _1}$$

where a₁ represents the effect coefficient of the independent variable on mediator variable 1, and ε₁ is the error term.

The effect of the independent variable Self concept (X) on the mediator variable Learning environment adaptation (M2) can be expressed as: (2)M2=a2X+dM1+ε2$${M_2} = {a_2}X + d{M_1} + {\varepsilon _2}$$

where a₂ represents the effect coefficient of the independent variable on mediator variable 2, d represents the effect coefficient of mediator variable 1 on mediator variable 2, and ε₂ is the error term.

The indirect effect of the independent variable Self concept (X) on the dependent variable Self directed learning (Y) is: (3)Y=c1X+b1M1+b2M2+ε3$$Y = {c_1}X + {b_1}{M_1} + {b_2}{M_2} + {\varepsilon _3}$$

where c₁ represents the indirect effect coefficient of the independent variable on the dependent variable, b₁ represents the effect coefficient of mediator variable 1 on the dependent variable, and b₂ represents the effect coefficient of mediator variable 2 on the dependent variable, and ε₃ is the error term.

The direct effect of the independent variable Self concept (X) on the dependent variable Self directed learning (Y) is: (4)Y=c2X+ε4$$Y = {c_2}X + {\varepsilon _4}$$

where c₂ represents the direct effect coefficient of the independent variable on the dependent variable, and ε₄ is the error term.

Exploring mediation effects typically employs various methods to ensure the reliability and validity of the results. Common methods include regression analysis, structural equation modeling (SEM), and bootstrapping. Regression analysis is the most basic method; in regression analysis, if learning environment adaptation mediates the relationship between self concept and self directed learning, both a₂ and b₂ must be significant. If self-efficacy has a direct mediation effect between self concept and self directed learning, then both a₁ and b₁ must be significant, whereas d may not be significant but should be evaluable. If self-efficacy does not have a direct mediation effect between self concept and self directed learning, but has an indirect mediation effect, then a₁,d, and b₂ must be significant. Applying the bootstrapping method based on regression analysis can determine the mediation effect between variables, with each effect coefficient needing to satisfy specific equations, divided into the following situations:

If learning environment adaptation mediates the relationship between self concept and self directed learning, and self-efficacy has a direct mediation effect between self concept and self directed learning, the effect coefficients must satisfy the following equation: (5)c2=c1+(a1×b1+a2×b2)$${c_2} = {c_1} + ({a_1} \times {b_1} + {a_2} \times {b_2})$$

If learning environment adaptation mediates the relationship between self concept and self directed learning, and self-efficacy has an indirect mediation effect between self concept and self directed learning, the effect coefficients must satisfy the following equation: (6)c2=c1+(a1×d×b2+a2×b2)$${c_2} = {c_1} + ({a_1} \times d \times {b_2} + {a_2} \times {b_2})$$

The Self-Concept Questionnaire for college students utilizes the scale proposed by Zheng Yong, which is divided into two dimensions: intrinsic quality and external performance. The intrinsic quality dimension includes traits such as friendliness, family, integrity, academics, and aspirations, reflecting individuals’ self-perception regarding moral responsibility and academic capability. The external performance dimension encompasses aspects such as appearance, maturity, and communication, focusing on individuals’ self-evaluations in terms of appearance and interpersonal relationships. The questionnaire consists of 60 items, each rated on a 5-point scale. The overall test-retest reliability of the questionnaire is 0.86, with the test-retest reliability for each dimension ranging from 0.62 to 0.82, indicating good reliability and the ability to effectively assess college students’ self-concept structure and its influences. [16]

The Student Environment Adaptation Questionnaire employs the subscale of academic adaptation from the “Student Adaptation to College Questionnaire” developed by Baker, which includes 18 items, such as: ability to learn new material, adaptation to new learning environments, time management skills for studying, participation in classroom discussions, ability to complete assignments on time, comprehension of course content, engagement in learning tasks, coping with academic stress, tendency to seek academic help, ability to establish good relationships with professors, researching and preparing for exams, organizing study materials, active participation in group learning, ability to set academic goals, maintaining self-motivation, adapting to different teaching styles, ability to evaluate one’s academic performance, and self-confidence in academic achievements. Each option is rated on a 9-point scale. Internal consistency analysis of a sample of college freshmen yielded Cronbach’s α values ranging from 0.92 to 0.94 [17].

The sense of self-efficacy is measured using a self-efficacy scale developed by Li Xiaona. This scale covers three main areas: self-assessment of course learning ability (5 items), self-assessment of research activity capability (8 items), and self-assessment of social practice ability (3 items). The scale uses a 7-point Likert scale for scoring and has a high reliability coefficient of 0.943 [18].

The College Students’ Self-Directed Learning Scale is developed by Zhu Zude based on Zimmerman’s theory of self-directed learning, incorporating the practical learning situations of Chinese college students. Through factor analysis, two subscales—learning motivation and learning strategies—were derived, with each subscale containing 6 factors. Each item requires a response on a six-point scale. The overall internal consistency coefficients of the two subscales reached 0.8 and 0.9, indicating that the scale has good reliability and validity metrics and can effectively measure college students’ self-directed learning abilities [19].

This study collected data using an electronic questionnaire. After obtaining the consent of the school management, researchers distributed the questionnaire link to teachers in various colleges through online platforms, who then forwarded it to their students. Questionnaires that took too little time (<3 minutes) or too long (>20 minutes) to complete, as well as those with inconsistencies or missing more than three items, were excluded.

The sample population for this survey consisted of students from a higher education institution in Sichuan Province, with a total of 325 questionnaires collected. The survey revealed that 49.23% of the respondents were male, while females slightly outnumbered them at 50.77%, resulting in a nearly balanced gender ratio. In terms of educational level, undergraduates predominated, accounting for 74.46%, while graduate students made up only 25.54%. Regarding academic performance, 13.23% of students had scores above 76%, while the largest group, representing 34.46%, had scores ranging from 51% to 75%. Concerning marital status, the vast majority of respondents were unmarried, reaching as high as 98.15%. The distribution of majors was primarily in the natural sciences, accounting for 49.54%, while social sciences and arts majors accounted for 34.46% and 16.00%, respectively. Overall, the group consists mainly of unmarried, natural science majors with above-average academic performance at the undergraduate level, as shown in Table 1.

Normalization of the data was performed to standardize the range to 0-5 as shown in Figure 3. The results indicated that the average score of participants’ self-efficacy was 3.24 (SD±0.12), suggesting that most students exhibit a high level of recognition regarding their abilities and learning potential. This enhancement in self-efficacy is often closely associated with students’ motivation to learn and academic achievement, indicating that they possess greater confidence when facing academic challenges and can actively engage in learning activities.

Figure 3.

Mediation Effect Model

Further analysis revealed that the average score for self-directed learning was 3.06 (SD± 0.26), reflecting students’ competence in managing their own learning. The cultivation of self-directed learning abilities not only aids students in better controlling their learning processes but also enhances their problem-solving skills and effectiveness in autonomous learning. This increase in capability implies that students can set goals, plan learning strategies, and assess their progress throughout the learning process, thereby improving learning efficiency.

Conversely, the measurement results for learning environment adaptation showed an average score of 2.17 (SD± 0.36), indicating that despite many students performing well in self-efficacy and self-directed learning abilities, they still face certain challenges in adapting to different learning environments. This may be related to the diversity and complexity of learning environments, as students need to invest additional time and effort to adjust their learning strategies to accommodate the ever-changing learning contexts. Furthermore, the average score for self-concept was 2.39 (SD± 0.18), further demonstrating that students exhibit moderate levels of self-awareness and self-evaluation.

To begin with, it is essential to determine the central tendency and the degree of dispersion for the primary variables in question by calculating their mean values and standard deviations. Subsequently, we will delve into the pairwise relationships between these variables by employing the Pearson correlation coefficient, which is a statistical measure that evaluates the linear association between two continuous variables. The Pearson’s r coefficient assumes values that fall within the interval of -1 to +1, where a value of -1 signifies a perfect negative correlation, +1 denotes a perfect positive correlation, and a coefficient greater than 0.6 suggests a notably strong positive relationship between the variables in question. Mathematically represented, the Pearson correlation coefficient (r) can be calculated using the formula: (7)r=∑i=1n(Xi−X¯)(Yi−Y¯)∑i=1n(Xi−X¯)2∑i=1n(Yi−Y¯)2$$r = \frac{{\sum\limits_{i = 1}^n {({X_i} - \bar X)({Y_i} - \bar Y)} }}{{\sqrt {\sum\limits_{i = 1}^n {{{({X_i} - \bar X)}^2}} } \sqrt {\sum\limits_{i = 1}^n {{{({Y_i} - \bar Y)}^2}} } }}$$

Where X_i and Y_i are the individual sample points of the two variables, X¯$$\bar X$$ and Y¯$$\bar Y$$ are their respective means, and the summations extend over all the pairs of data points. The application of this formula facilitates the quantification of the strength and direction of the linear relationships under investigation.

As shown in Table 2, we observed a considerate degree of correlation between the core elements of the mediation model, including self-concept, self-efficacy, learning environment adaptation, and self-directed learning. Among them, the correlation between learning environment adaptation and the other three factors is relatively weak, and the correlation between learning environment adaptation and self-directed learning is the weakest.

The hypothesized model was estimated and tested. In Structural Equation Modeling (SEM), model fit indices are important indicators used to assess the degree of fit between the model and the observed data. Commonly used indices include the Comparative Fit Index (CFI), Non-Normed Fit Index (NNFI), Goodness-of-Fit Index (GFI), and Root Mean Square Error of Approximation (RMSEA). These indices help researchers determine to what extent the theoretical model established reflects the structure and relationships in the actual data. Below is a detailed explanation and calculation method for the four key fit indices.

CFI is the Comparative Fit Index, which evaluates the model’s fit by comparing the fitted model with a baseline model (usually the independence model, which assumes no relationships among the variables). The calculation formula for CFI is: (8)χ2=∑i=1n(Oi−Ei)2Ei$${\chi ^2} = \sum\limits_{i = 1}^n {\frac{{{{({O_i} - {E_i})}^2}}}{{{E_i}}}}$$ (9)dfmodel=k−p$$d{f_{{\text{model}}}} = k - p$$ (10)dfbaseline=k−1$$d{f_{{\text{baseline}}}} = k - 1$$ (11)CFI=1−max(χmodel2−dfmodel,0)max(χbaseline2−dfbaseline,0)$${\text{CFI}} = 1 - \frac{{{\text{max}}(\chi _{model}^2 - d{f_{model}},0)}}{{{\text{max}}(\chi _{baseline}^2 - d{f_{baseline}},0)}}$$

In this formula, O_i represents the observed frequency of the actual data; E_i represents the expected frequency predicted by the model; n is the total number of categories; k is the sample size; and p is the number of parameters estimated in the fitted model; χmodel2$$\chi _{model}^2$$ represents the chi-square value of the fitted model, and df_model represents its degrees of freedom. The baseline model’s chi-square value and degrees of freedom are denoted as χbaseline2$$\chi _{baseline}^2$$ and df_baseline, respectively. CFI values range from 0 to 1, with a value greater than 0.90 indicating a good fit, and values closer to 1 indicating better fit.

NNFI, also known as the Tucker-Lewis Index (TLI), is another index for evaluating model fit. Similar to CFI, NNFI also takes model complexity into account in its calculation. The calculation method is as follows: (12)NNFI=χbaseline2−χmodel2χbaseline2−dfbaseline$${\text{NNFI}} = \frac{{\chi _{baseline}^2 - \chi _{model}^2}}{{\chi _{baseline}^2 - d{f_{baseline}}}}$$

The value of NNFI typically ranges from 0 to 1, with values closer to 1 indicating better model fit. NNFI is particularly suited for comparing the fit of models of different complexities and provides additional evaluative information for model simplification.

GFI is the Goodness-of-Fit Index, which measures the model’s ability to explain the observed data. Its calculation formula is:

For the data matrix X with m rows and n columns, the calculation steps for the observed covariance matrix S are as follows: (13)X¯j=1m∑i=1mXij(j=1,2,…,n)$${\bar X_j} = \frac{1}{m}\sum\limits_{i = 1}^m {{X_{ij}}} \quad (j = 1,2, \ldots ,n)$$ (14)S=1m−1∑i=1m(Xi−X¯)(Xi−X¯)T$$S = \frac{1}{{m - 1}}\sum\limits_{i = 1}^m {({X_i} - \bar X)} {({X_i} - \bar X)^T}$$ (15)S=1m−1(X−X¯AT)(X−X¯AT)T$$S = \frac{1}{{m - 1}}(X - \bar X{{\rm A}^T}){(X - \bar X{{\rm A}^T})^T}$$

where A is a column vector of ones.

The covariance matrix fitted by the model is typically generated by structural equation modeling, using the following formula: (16)y=Λη+ ò$$y = {\rm{\Lambda }}\eta + \unicode{x00F2}$$ (17)S^=ΛΦΛT+Ψ$$\hat S = \Lambda \Phi {\Lambda ^T} + \Psi$$ (18)GFI=1−tr((Sobserved−Sfitted)2)tr(Sobserved2)$${\text{GFI}} = 1 - \frac{{{\text{tr}}({{({S_{observed}} - {S_{fitted}})}^2})}}{{{\text{tr}}(S_{observed}^2)}}$$ (19)D=Sobserved−Sfitted$$D = {S_{observed}} - {S_{fitted}}$$ (20)tr((D)2)=∑i=1n∑j=1nDij2$${\text{tr}}({(D)^2}) = \sum\limits_{i = 1}^n {\sum\limits_{j = 1}^n {D_{ij}^2} }$$ (21)tr(Sobserved2)=∑i=1nSii2$${\text{tr}}(S_{observed}^2) = \sum\limits_{i = 1}^n {S_{ii}^2}$$

In this formula: Λ is the factor loading matrix, which represents the relationships between observed variables and latent variables. Φ is the covariance matrix between latent variables. Ψ is the covariance matrix of measurement errors, which is usually a diagonal matrix.

S_observed represents the observed covariance matrix, while S_fitted represents the covariance matrix of the fitted model. GFI values typically range from 0 to 1, with values closer to 1 indicating better model fit. GFI is an intuitive and easily understandable fit index, suitable for preliminary assessments of model validity.

RMSEA is the Root Mean Square Error of Approximation, which is used to assess model fit and adjusts for model complexity. Its calculation method is as follows: (22)RMSEA=χmodel2−dfmodeln−1$${\text{RMSEA}} = \sqrt {\frac{{\chi _{model}^2 - d{f_{model}}}}{{n - 1}}}$$

where n represents the sample size. An RMSEA value less than 0.08 indicates a good fit, 0.07 to 0.08 indicates reasonable fit, 0.08 to 0.10 indicates poor fit, and values exceeding 0.10 indicate inadequate fit. RMSEA can also provide a confidence interval, further enhancing the evaluation of the model fit.The calculation methods for the upper β_up and lower β_low limits of the confidence interval are as follows: (23)βlow=χmodel2−dfn−1+1.962(df+1)n$${\beta _{low}} = \sqrt {\frac{{\chi _{{\text{model}}}^2 - df}}{{n - 1 + 1.96\sqrt {\frac{{2(df + 1)}}{n}} }}}$$ (24)βup=χmodel2−dfn−1−1.962(df+1)n$${\beta _{up}} = \sqrt {\frac{{\chi _{{\text{model}}}^2 - df}}{{n - 1 - 1.96\sqrt {\frac{{2(df + 1)}}{n}} }}}$$

Through these model fit indices, a comprehensive assessment of the fit of the structural equation model can be conducted. CFI and NNFI values should be close to 1 to indicate good model fit; GFI should be greater than 0.9 to ensure the model can adequately explain data variability; whereas RMSEA should be less than 0.08 to indicate that the model’s error is within an acceptable range.

Using SPSS 12.0 software, calculations for each index of M1. The model M1 has fit indices NNFI, CFI, and GFI all greater than 0.90, RMSEA less than 0.08, but further examination of the parameter estimates reveals that the path coefficients for “self-efficacy → autonomous learning” and “learning environment adaptability → self-efficacy” are both not significant. Therefore, a stepwise deletion method was used to modify the model. First, the insignificant path “self-efficacy → autonomous learning” was removed, yielding M1-2 as shown in Figure 4. The model fit improved, with RMSEA decreasing by 0.001, NNFI increasing by 0.006, and the chi-square increasing by 0.154.

Figure 4.

Mediation Effect Model M1-2

However, in the subsequent analysis of the M1-2 model, it was observed that the coefficients of the “learning environment adaptability → self-efficacy” path were still not statistically significant. This prompted further modifications, resulting in the deletion of the path and the formation of the M1-3 model, as shown in Figure 5. Although the adjustment process aims to simplify the model, the improvement in the fitting index of the M1-3 model is not significant. Specifically, although RMSEA remained consistent and did not show a significant decrease, it did not deteriorate either, indicating that despite removing the path, the model still maintained its overall adequacy.

Figure 5.

Mediation Effect Model M1-3

This iterative refinement process emphasizes the importance of critically evaluating model parameters to ensure that only important relationships are retained. By eliminating unimportant paths, Model M1-3 provides a more focused view of the relationships between remaining variables, thus gaining a clearer understanding of how learning environment adaptability and self-efficacy interact to influence self-directed learning. In these adjustment processes, the stability of the model fitting index reinforces the view that the improved model still has relevance and usefulness for further exploration of educational dynamics.

Considering that the order of deleting freely estimated parameters may affect the estimation results, the path “learning attribution → learning environment adaptability” was removed first based on M1, resulting in M1-4 as shown in Figure 6. The model fit did not deteriorate, but the coefficient for the path “learning environment adaptability → autonomous learning” was still not significant.

Figure 6.

Mediation Effect Model M1-4

After testing and modifying from M1 to M1-4, under the condition that all fit index values meet the requirements, the only model where all paths are significant is M1-3. However, Model MI-3 cannot reveal the mediating effect of self-efficacy. To verify whether self-efficacy really has no mediating effect, the following analysis will be conducted on Model M2.

Building upon the measurement model established, a similar methodological approach was employed to estimate and test model M2. Extensive calculations for each fit index were performed, yielding impressive results: the NNFI reached 0.995, CFI was recorded at 0.998, GFI stood at 0.957, and the RMSEA was calculated to be 0.052. Additionally, the ratio of chi-square to degrees of freedom was found to be less than 5, which is indicative of a well-fitting model according to conventional criteria. These findings collectively suggest that model M2 captures the underlying relationships among the variables effectively.

However, a deeper analysis of the parameter estimates revealed an area of concern: the path coefficient for “Learning environment adaptation → Self-directed learning” was not statistically significant. This insight raised important questions about the direct influence of learning environment adaptation on self-directed learning. Consequently, in pursuit of a more parsimonious and meaningful model, this path was eliminated, leading to the emergence of Model M2-2. Remarkably, upon reviewing the modified model, it was found that all remaining paths in M2-2 were significant, which further reinforced the robustness of the model.

With the establishment of model M2-2, a mediation correlation analysis was deemed necessary to explore the intricate relationships between learning environment adaptation, self-efficacy, self-concept, and self-directed learning. Utilizing SPSS 12.0, statistical analyses were conducted to assess these relationships rigorously. The results yielded standardized estimates between the variables, as documented in Table 2. These estimates provide crucial insights into how learning environment adaptation influences self-efficacy, which in turn affects both self-concept and self-directed learning.

As shown in the results of Table 2, the impact of self-concept on self-directed learning is significant (β=0.572, t=8.54***), indicating that the enhancement of self-concept significantly promotes students’ self-directed learning abilities. Moreover, the influence of self-concept on self-efficacy (β=0.483, t=4.21***) and learning environment adaptation (β=0.268, t=7.92***) is also significant, suggesting that self-concept plays an important role in bolstering students’ confidence and adaptability. In the path related to self-directed learning, the path coefficient for self-efficacy is 0.476 (t=5.78***), indicating its positive impact on learning. In the path concerning learning environment adaptation, both self-efficacy (β=0.312, t=9.11***) and learning environment adaptation (β=0.065, t=5.47***) demonstrate significance, further supporting their importance in the learning process. The interactions among the variables and the revised model can be detailed in Figure 7. The path coefficients of the four key variables satisfy Equation 6, proving that learning environment adaptation mediates the relationship between self-concept and self-directed learning, while self-efficacy has an indirect mediating effect between self-concept and self-directed learning.

Figure 7.

Mediation effect of model M2-2

Self-efficacy plays a crucial role in students’ learning processes, as it not only affects their motivation and emotional states but also directly relates to their learning outcomes. According to the research findings, self-efficacy (β=0.483) is influenced by self-concept and indirectly promotes self-directed learning through learning environment adaptation (β=0.312) (β=0.476). This indicates that self-efficacy can enhance students’ confidence when facing learning challenges, thereby improving their learning performance and self-management capabilities.

To effectively enhance students’ self-efficacy, comprehensive interventions can be implemented at multiple levels. Firstly, teachers should actively cultivate students’ self-concept by providing timely positive feedback and affirmation, helping students build confidence in their abilities. For example, teachers can focus on each student’s progress in class, promptly acknowledging their efforts and achievements, allowing them to experience the joy of success when achieving small goals. Such experiences of success will effectively boost their courage to face future challenges. Additionally, appropriate challenges and task settings are essential; allowing students to gain a sense of achievement while completing these tasks further enhances their self-efficacy. Secondly, the power of role models should be introduced. Teachers can share success stories or invite outstanding seniors to communicate with students, showcasing how others have overcome difficulties and achieved success. This modeling effect can motivate students to believe, “I can do it too,” thereby enhancing their self-efficacy. At the same time, schools should provide psychological counseling and support to help students cope with stress and anxiety during their learning journey. Through psychological counseling, students can learn coping strategies, enhance their psychological resilience, and further boost their confidence in their abilities. Moreover, offering a variety of extracurricular activities and practical opportunities is also an important avenue for enhancing students’ self-efficacy. Schools can organize various clubs, competitions, and volunteer activities, allowing students to develop their skills and capabilities through participation. In these activities, students can face real challenges and accumulate practical experience, thereby strengthening their self-efficacy. Furthermore, extracurricular activities can provide a platform for students to showcase themselves, helping them recognize their strengths in team collaboration and receive encouragement and support from peers. Such positive participation experiences will inspire students to continuously pursue improvement and enhance their confidence in facing future challenges.

The research results indicate that self-concept significantly influences learning environment adaptation (β=0.268), while learning environment adaptation positively promotes self-efficacy (β=0.312) and self-directed learning (β=0.476). Therefore, improving the ability to adapt to the learning environment can not only enhance students’ self-efficacy but also elevate their self-directed learning capabilities, laying a foundation for academic success.

To effectively improve students’ learning environment adaptation, schools should create a supportive and inclusive learning environment that encourages students to ask questions and actively participate. Offering adaptive training courses can also help students master skills such as time management, stress coping, and learning strategies to better adjust to the changing learning environment. Furthermore, promoting interaction and cooperative learning among students, as well as providing personalized learning support, are key factors. Interaction and cooperation can strengthen students’ social skills and collective problem-solving abilities, thereby enhancing adaptability. Personalized teaching by educators can provide tailored guidance based on individual differences, helping students better cope with learning challenges. Additionally, utilizing modern technological tools, such as online learning platforms, to provide flexible learning resources allows students to learn according to their personal needs and pace, further enhancing their adaptability. By implementing these suggestions, schools can effectively improve students’ learning environment adaptation, thereby promoting the development of self-efficacy and self-directed learning, aiding them in achieving better academic performance and fostering lifelong learning abilities.