A Study on Self-efficacy of Undergraduates in Blended Learning in Chengdu, China--The Moderating Effects of Gender, Disciplines and Social Influences

: This study examines the factors influence learner' self-efficacy, such as perceived behavioral control and motivation, as well as moderating effects of gender, disciplines, and social influences. Blended learning reform in China was first stared in engineering majors and developed relatively mature, which became one of the models for other majors` reform. Blended learning has been a major goal of higher education curriculum reform in recent years, whereas the role students play in it is not discussed enough in China, including their subjective feelings and emotions, such as self-efficacy. Data was collected through an online questionnaire from a total of 500 undergraduates from Chengdu who have had blended learning experiences in the past year. Results indicate that both perceived behavioral control and motivation are significant predictors of self-efficacy. Gender plays a significant moderating role, while disciplines does not. When gender, disciplines, and social influence were considered together, influences from family and teachers and peers exerted different moderating effects, respectively.


Introduction
Following the nationwide application of online teaching in response to the COVIN-19 pandemic, China is now attempting to make blended learning the general form in higher education according to Xinhua News Agency (2022) [1] . It mentioned that by the end of February 2022, the number of online courses in China has exceeded 50,000, with nearly 800 million students taking courses and more than 300 million students studying in MOOCs. But while the rapid growth of online education resources and blended learning courses, these education reforms must be noticed that development at such an impressive speed are mainly driven by external forces, such as policy and investment support. And the subjective experience of learners, who are the subjects in education, deserves the attention and research of educators. This study investigated the key factors including perceived behavior control, motivation and self-efficacy and the effects of gender, disciplines and social influences moderating on the self-efficacy of undergraduates involved in blended learning.
In this study, blended learning is defined as a combination of traditional classroom learning and e-learning, which combine the interaction in face to face classroom and the technology advantages

Theoretical Background
Social Learning Theory and Theory of Planned Behaviour are the primarily theoretical framework of this study. They both focus on the interaction between individual learning and social influences.
Bandura proposed social learning theory and emphasized to examine education from the perspective of the joint functioning of cognition and behaviour   [11] . One core ideas of the theory is that psychological changes in learning do not occur in a vacuum and are mediated by cognitive processes. It means learning process could be influenced by educators, peers and individuals personal behavior and emotions in the learning situation.
TPB is one widely accepted behaviour theories that considers individual behaviour with social influences to explain the behavior of technology applications. The core constructs of TPB include perceived behavior control, self-efficacy and subjective norms (Zhang et al., 2019) [12] . Perceived behavior control is individual`s perception when practice or conduct certain behavior. It can facilitate or hinder an individual's behaviour.This study chose perceived control of behaviour as the independent variable, motivation as the mediating variable and self-efficacy as the dependent variable. And social influence, professional background and gender were as moderating factors.

Perceived Behavior Control
Perceived behavioral control (PBC) in blended learning is learner`s ability to control the factors that may impact on their learning behavior   [13] . PBC is an important element of the theory of planned behavior. Previous research in education has found PBC explains the extent of how available the resources and opportunities are to students (Leonard et al., 2004) [14] . Higher perceived behavior control leads to higher affective ratings and engagement tendencies for the system used (Martın & Jimenez, 2011) [15] .Thus, the hypotheses is formed as follows: H1: Perceived behavior control has significant positive impact on motivation H2: Perceived behavior control has significant positive impact on self-efficacy

Motivation
Motivation in blended learning is the willingness of learners to make an effort, expend energy and time to engage in blended learning (Tai, 2006) [16] . Learners`s motivation is proved to perform three main functions: motivating individuals to produce learning behaviour, directing them towards specific learning goals, and maintaining and regulating their behaviour (Leone et al., 2017 [17] ). In a study about students` information literacy conducted by Aharony and Gazit (2018) [18] found that motivation is a predictor for self-efficacy in online background.
H3: Motivation has significant positive impact on self-efficacy

Self-Efficacy
In this context, self-efficacy refers to individual's judgement of personal ability to engage in blended learning, to finish homework and achieve personal goals (Jungert et al., 2013) [ 19 ] . Self-efficacy could support students` confidence and concentration to solve problems and achieve academic goals (Schunk & Mullen, 2012) [20] . It enables individuals to take action when faced with challenges or to take responsibilities for high confidence required work, such as mentoring and helping others   [21] . It has been proved to be influenced not only by former experience and performance in using information systems, but also be able to impact willingness to use this technology in the future (Drumm, 2019) [22] .
However, self-efficacy is not a stable psychological element. It could be impacted by learner's background and other external factors then presented on learners` participants.
The moderating effect of gender varies across social contexts and technological situations in some extent, and it is valuable to test the gender moderation in different contexts (Venkatesh and Morris, 2000) [23] . Many studies have mentioned gender as an important factor when it comes to accept and continue to use information technology and electronic devices (Shao & Chen, 2020) [24] . Zander et al. (2020) [25] mention that girls generally have lower self-efficacy in mathematics learning and are less able to self-improve than boys. This may be one of the reasons why girls are still in the minority in STEM fields. Hä gg et al.'s (2022) [26] study on entrepreneurship education also mentioned that gender balance in the classroom as part of the educational environment and student socialization, could positively influence students' entrepreneurial behavior. It is partly because gender is closely related to social norms that shape students' perceptions and attitudes. Additionally, gender imbalance between students and teachers can also affect students' participation in classroom discussions (Opie et al., 2019) [27] . Thus, this study explores gender as one of the moderating factors to explore the development of self-efficacy in blended learning.
Another influencing factor is the student's academic background. Discipline differences in higher education have long been of interest to researchers, due to different disciplines would focus on different natures of knowledge, such as theoretical or practical (Foung & Chen, 2019) [28] . The impact of major on a student is subtle and multifaceted. A study on entrepreneurial intentions of engineering students mentioned that disciplines could moderate students' motivation to start their own entrepreneurship (Alam et al., 2020) [29] .
For blended learning, the impact of discipline difference has also been proved by previous research. Research of self-efficacy based on accounting students conducted by Mooi (2006) [30] found that, students with higher self-efficacy would to predict their learning outcome positively, and then make more efforts to it. But it did not imply that more motivated and higher self-efficacy could bring better the learning outcomes will be, the conservative self-efficacy students might benefit more. H4: Gender difference has significant moderation on factors impact on self-efficacy. H5: Major influence has significant moderation on factors impact on self-efficacy.

Social Influence
The social influence in blended learning refers to the pressure from important others make students think they need to participant. It belongs to social norms   [31] . Taylor et al. (2011) [32] mentioned peer influence might be stronger than other in universities in American. In a study of Chereau and Meschi (2021) [33] about entrepreneurial self-efficacy, they found that the entrepreneurial environment created by parents or families will constitute a different learning context and lead to difference in students' self-efficacy in entrepreneurship.
Based the literature and analysis before, the main hypothesis is: H6: Social influence has significant moderation on factors impact on self-efficacy. H6a: The teachers` influence (Social influence 1) has significant moderation on factors impact on self-efficacy.
H6b: The families` influence (Social influence 2) has significant moderation on factors impact on self-efficacy.
H6c: The peers influence (Social influence 3) has significant moderation on factors impact on self-efficacy

The Conceptual Framework of Study
In summary, the conceptual framework of this study is as follows Figure 1.

Research Methods
The questionnaire was composed of three parts, screening questions, demographic information and the main part "factors impact on self-efficacy in blended learning". The main body of the questionnaire containing 17 items to measure the conceptual framework with 5-point Likert scale.
The online questionnaire was distributed to undergraduates among five public universities in Chengdu and invited them to engaged in the survey voluntarily. The demographic information is in table 1. There was found that 53% of the students were majoring in social sciences (mainly study in languages, education and management) and 47% in engineering (mainly study in computer software engineering, mechanical engineering and energy and power engineering).

Data and Measurement Model Analysis
The normal distribution of each scale item and the Cronbach`s Alpha of the variables were examined. The results were as follows table 2.  [34] . The internal consistency of the entire scale was 0.851. In general, higher than 0.6 to indicate the scale is acceptable (Hair et al., 1989) [35] .The convergent and discriminant validity of the scale presented in the Table 3 as the values of AVE and CR. All variables have good convergent validity since the value of AVE is around 0.5 while that of CR is higher than 0.7. Table 4 shows that the discriminant validity is also acceptable.

Differential Analysis
In order to have an understanding of the demographic differences in the data, independent sample T-test and one way-ANOVA test were conducted by SPSS to analyse whether there was a significant difference between gender, majors and academic years.

Independent Samples T-test for Gender and Majors
Through independent sample t-tests, the study found that the standard deviation on motivation of females was larger than that of male. It indicated that female students' data were more discrete on the scale items of motivation. The variable perceived behavior control shown significant difference between male and female students according to Table 5.  Table 6, students study in social science and engineering, there was a significant difference in the motivation. The mean value of the motivation was 0.21 higher for engineering students than for social sciences students.

One Way-ANOVA Test for Academic Years
One-way ANOVA was chosen to test for differences between the grades. The results show that motivation and self-efficacy are significant differences by academic year. It can be seen in Table 7, the first grade students have more disparate views on blended learning and a larger gap in self-efficacy. Compared to other grades, first grade scored lower in perceived behavioral control and self-efficacy.

Measurement Model and SEM Analysis
Structural equation modeling is a method for building, estimating and testing causal relationship model (Henseler et al., 2014) [36] . Table 8 shows the metrics of model fit, all of which meet the criteria.

Mediating Effect Model Analysis
As showed in Table 9, the paths of the mediation model are all supported by the data,hypotheses 1 to 3 are proved. The second half of the mediation effect (MOT→SE) is significant at the 0.01 level and the other paths are significant at the 0.001 level. Perceived behavior control could significantly predict motivation (B=0.182, p<0.001) and self-efficacy (B=0.69, p<0.001), regression equations were formed. The mediating variable motivation also could positively impact on self-efficacy (B=0.19, p=0.004<0.01). From the standardized regression weight, the most powerful variables of the model is the independent variable, perceived behavior control (β=0.563, p<0.001).

Gender Moderating Effect Analysis
To explore whether there are differences in perceived behavior control and how it impact on self-efficacy and verify hypothesis H4 and H5, this study continue to investigate the moderating effect of gender. AMOS was used to model groupings for comparative analysis. Chi-square difference values of these models are checked. The model is stable. Illustrated in Figure 3, the regression weight value indicated that the mediating model is not supported by data of male. This is because the males presented insignificant in the first half of the mediation model (PBC→MOT), while in the second half (MOT→SE), the males' motivation negatively influenced self-efficacy at a significant level of 0.05.
Note: Values outside brackets are female, values in brackets are for males; ***p<0.001 The standardize regression coefficients are presented in Table 10 to compare the moderator effects. Firstly, males` perceived behaviour control (β=0.792, p<0.001) had a significantly higher contribution to self-efficacy than females (β=0.434, p<0.001). But for indirect effects, female students (β=0.285, p<0.001) showed a higher predictive power of motivation on self-efficacy, while male students (β=-0.137, p=0.015<0.05) showed a negative moderating effect of motivation on self-efficacy at a lower level of significance. The critical ratio of differences between parameters of measurement weights model, all values are greater than 1.96, indicating gender plays an important moderating role in the pathway. Hypotheses 5 to 8 are all supported by data.

Disciplines Moderating Effect Analysis
Amos was also used to analysis the moderating effects of disciplines. Comparing indicators of model fit result and chi-square test degrees of freedom could fine that the model is stable.
The biggest difference for disciplines between the regression equations constructed is the first half of the mediating effect (PBC→MOT), where perceived usefulness has no significant effect on motivation for engineering students, but this regression equation is significant for social science students. The results is presenting in Figure 4.
Note: Values outside brackets are engineering, values in brackets are for social science; ***p<0.001  Table 11, although the direct effects were close, students major in social science have higher impact on self-efficacy from perceived behavior control. The positive effect of perceived behavior control on motivation was stronger for engineering majors (β=0.239, p=0.003<0.01). Whereas the difference in the moderating effect of majors on the pathway of motivation on self-efficacy was smaller, both being significant at the 0.05 level, the standardized coefficient was slightly higher for engineering majors (β=0.159) than for social science students (β=0.149). However, the values of critical ratio of the difference between parameters are less than 1.96. These differences are not significant statistically.

Social Influence Moderating Effect Analysis
Model 59 of the PROCESSv4.1 program prepared by Hayes and the Bootstrap test (5000 times) were applied to analyse three kinds of social influence, the independent variable was perceived behaviour control, the mediator was motivation, the dependent variable was self-efficacy, the control variables were gender, major and academic year. Table 12, the interaction term between SI1 from the instructor and perceived behavioral control was significant at the 0.01 level and showed a negative moderating effect toward to motivation.   Table 13. When the influence from the teacher was low, its interaction term with perceived behavior control significantly attenuated the positive effect of perceived behavior control on motivation. The slope diagram is shown in Figure 5.

Family Influence Moderating Effect Analysis
SI2 measures the influence from family or relatives. According to the analysis in Table 14, the interaction term between SI2 and perceived behavior control was significant at the 0.001 level and showed enhanced moderation in the effect of perceived behavior control on self-efficacy.  Table 15 enhances the effect of students' perceived behavior control on self-efficacy only at high levels. The slope diagram is shown in Figure 6.   Figure 6: Slope diagram of the moderating effect of SI2 at high and low levels

Peers Influence Moderating Effect Analysis
Compared to the other two influences based on the data presented in Table 16, no significant moderating effect was found for the peer influence.

Moderating Effect of Peer Influence in Different undergraduates
Previous research implied that peer influence is consistently one of the strengths of blended learning. Blended learning provides a virtual learning communication environment that can enhance interaction between student and teacher and peers through email, discussion boards and virtual classrooms   [37] . And in Poon's (2012) [38] discussion of the definition of blended learning, it is mentioned that blended learning enables students to learn independently and engage in peer discussion at their own pace. Further exploration of non-significant peer influences was therefore undertaken to find whether there are moderating effects of similar size but opposite effects that make peer influence insignificant.

Moderating Effects of Peer influence and Gender Differences
The following results in Table 17 were the multilayer regression in groups tests the moderating effect of gender differences and peer influence. From the results, for male students' self-efficacy, the interaction term of peer influence and motivation played a significant positive effect (β=0.143, p=0.004<0.01), weakening the original negative relationship between motivation and self-efficacy (β=-0.177, p<0.001). In contrast, in the females' peer influence was not observed to be significant. 0.000 0.000 0.000 0.000

Moderating Effects of Peer Influence and Major Differences
When considering the major gap with peer influence, the SPSS multiple linear regression in Table 18 showed that for engineering students, the interaction term between peer influence and motivation enhanced the positive relationship between motivation and self-efficacy (β=0.123, p=0.036<0.05), whereas this moderating effect was not found on the data for social science students.

Moderating Effects of Peer Influence and Academic Year Differences
Combining academic year and peer influence reveals that for students in years 1 to 3, the moderating effect shown in Table 19 of peer influence did not present a statistically significant effect. For fourth-year undergraduates, the interaction term between peer influence and perceived behavioral control, on the other hand, was significant at the 0.01 level with a standardized coefficient of -0.303, weakening the positive effect of perceived behavioral control on self-efficacy. In summary, the results of the analysis of the hypotheses of this study are as follows.

Discussion
First, perceived behavior control is a strong predictor for self-efficacy and improving undergraduate's sense of control when they participate in blended learning could ensure the level of self-efficacy. Another finding of interest is the disparity between students of different academic years. The statistics show that undergraduates in the lower grades have relatively lower mean scores for both perceived behavioral control, motivation, and self-efficacy, with a greater degree of data dispersion.
Gender plays a significant moderating role in the model of perceived behavior control and motivation predicting self-efficacy. For female students, perceived behavior control has a positive effect on their motivation toward blended learning, while motivation are the most powerful contributor to self-efficacy in this model. This logic does not fully apply to male students, for whom perceived behavior control directly and positively influenced their self-efficacy but does not significantly influence motivations. Motivation, on the other hand, has a significant weakening effect on their self-efficacy. Looking back at previous research on self-efficacy, Karimi et al. (2014) [39] mentioned that self-efficacy can indeed be improved through education, but even at the same level of education, female's self-efficacy is lower. Pelegrini et al. (2021) [40] mentioned that the statistical differences of self-efficacy was existed and female were more concerned about their self-efficacy. When motivation is higher, female devote more attention and expectation to behavior, which may be one reason why female's motivation can have a positive effect on self-efficacy.
The moderating effect of social influences occurs primarily in motivation toward blended learning. However, teacher influence is significant interacting with perceived behavior control only when it is at low levels, and it would weaken the positive effect of perceived behavior control on attitudes. Influence from family, on the other hand, is significant only at high level and could promote positive effects of perceived behavior control on attitudes. This suggests that perceived behavior control` supports toward to students' motivation could be reduced when teachers do not provide enough influence, but when the influence from family is high enough, the path could be supported. A case study of Park (2015) [41] about drama teaching methods supports this. Asian students show the tendency to recognize the teacher's role in assessing their learning above their own. Siddiqui et al. (2019) [42] argued that teachers is one important participant in a successful learning. Teacher could provide a supporting learning environment to improve students positive and autonomous. Research of   [43] mentioned that compared with face to face classroom where teacher could communicate with students directly, it is disadvantaged for timid students in blended learning who do not like to engage in online activities regularly.
Another interesting finding is that the influence of peers is not significant in this model. For male students, peers influence could interact with motivation and significantly weaken the negative effect of motivation on self-efficacy. And for engineering students, peer influence can also have a positive effect on the same path. Similarly, peer influence did not have a significant effect on any of the 1st to 3rd year students, but it showed a significant negative effect on self-efficacy with the interaction term of perceived behavior control for the prospective graduates in their senior year.
Finally, there was no data support regarding the impact of disciplinary differences on students' self-efficacy in blended learning. This suggests that blended learning, and its accompanying instructional tools, are not discipline-specific, but instead can build on each other's good teaching models and curriculum, taking into account instructional design and content. This study does not demonstrate a statistically significant difference in self-efficacy and its impact factors for students from different majors in blended learning. From the previous studies, there are two possible reasons. The study by Foung and Chen (2019) [44] presented similar results. This may suggest that the classification of disciplines in the context of blended learning could be more detailed and take into account students' learning behavior and subjective experiences.
One possible reason is the discipline categories. Most of the participants involved in this study were from languages, management, computer software engineering, mechanical engineering. According to Biglan's (1973) [45] classification, engineering is an applied discipline corresponding to a pure discipline, and humanities is a soft science corresponding to a hard discipline. However, with the convergence of disciplines, some new majors are at the boundary of different disciplines, and also with the development of general education in universities, the "soft" majors like humanities have begun to include knowledge from many applied disciplines, such as statistics and big data, computational sociology, etc. This has narrowed the different nature gaps between disciplines.
The possible reason is the changes in assessment methods in blended learning. Blended learning often promotes and riches assessment and exam methods in HEIs. For example, the process of learning is more clear and quantitative, including the interaction with teacher and peers, and recording of assignments, these could help students from different majors to get a more comprehensive grades. Students from different majors receive a final score that combines multiple forms of assessment, rather than a grade determined simply by a final exam or essay, which narrows the gap between the final scores of students from different majors. Therefore, in the context of blended learning, students of different majors have obtained relatively complete and satisfactory learning results, which makes their sense of self-efficacy and related factors do not show a significant gap.

Implications
First, self-efficacy seems strongly linked to individual student characteristics, therefore the design of the blended learning curriculum, teaching methods and styles are deserving attention and adaptation based on the students` situation.
The curriculum should be designed to take into account the requirements and status of students at different stages. For undergraduates in the early grades, more group work and peer interaction can promote motivation and perceived behavior control and thus have a more positive impact on self-efficacy. For upper grade students, it is important to consider that as prospective graduates, they might have different plans for future. Peer activities should be selective and can be grouped according to the different needs of the participants. As suggested by Aharony and Gazit (2019) [46] , it is important for keeping high self-efficacy that individual is not exposed in enormous information. Because that would make people feel incapable to handle and lose confidence..
In terms of teaching methods and styles, lower grade students have lower motivation and perceived behavioral control in blended settings and need more highly supportive external environments to help them adapt and achieve high self-efficacy, including influences from teachers and families. For females, on the other hand, motivate teaching strategies could be applied to help them to maintain good attitudes toward blended learning and promote higher self-efficacy.

Limitation and Suggestions
There are certain limitations in this study. The sample for this study was focused on five universities in Chengdu, China, and given consideration of the level of economic development, higher education and information building in southwest China, the analysis results of data are geographically specific. It might limit the application and generalization. Secondly, the sample is only from public universities, private universities and vocational colleges are not included. In addition to the factors discussed in this study, blended learning could be influenced by other factors from the learner's background, such as the size of the country, and the timing of the implementation of blended learning, as mentioned by Poon (2014) [47] .