Entrepreneurship competence refers to the set of demonstrable features and skills that enable and improve efficiency or performance on a job ( Jafari-Sadeghi et al., 2020 ). Additionally, Al Mamun et al. (2019) define it as the ability to use resources to improve the performance of the firm. Moreover, Iglesias-Sánchez et al. (2019) see entrepreneurship competence as a comprehensive set of abilities, which involves both technical and soft skills, required for individuals to excel in entrepreneurial activities. Blokker et al. (2019) view entrepreneurship competence as a combination of knowledge, skills, attitudes, and personal attributes that enable individuals to effectively recognize, create, manage, and grow entrepreneurial ventures. Thus, entrepreneurial competence represents the overall capability to succeed in the complex and dynamic world of entrepreneurship ( Hodzic et al., 2015 ). González-López et al. (2021) conceptualize entrepreneurial competencies by using six competency areas (opportunity, relationship, conceptual, organizing, and strategic and commitment). This study adopted Jafari-Sadeghi et al.’s (2020) definition for its contemporaneousness and pertinence to the topic at hand.

Employability refers to the ability of an individual to achieve self-sufficiency in the labor market through sustainable employment ( Wei et al., 2023 ). Elsewhere, employability is defined as the ability to keep a current job or obtain a desired job ( Okafor & Enemuo, 2022 ) and an individual’s overall capacity and readiness to gain and maintain meaningful employment ( Kozlinska et al., 2020 ). It encompasses the skills, qualifications, knowledge, personal attributes, and attitudes that make a person valuable and attractive to potential employers ( Rahmat et al., 2012 ). Hosain et al. (2021) conceptualize graduates’ employability as employability skills, personality and academic performance. However, this study adopted Okafor and Enemuo’s (2022) definition because it is widely accepted by entrepreneurship researchers and it encompasses all aspects of future success and personal status.

Technical and Vocational Education and Training (TVET) is a term officiated in 1999 in Seoul, Republic of Korea at the World Congress, encompassing education and training activities ( Sohn et al., 2017 ). Mesuwini et al. (2023) added that TVET encompasses education, training and technical skills development relating to a wide range of occupational fields, production, services and livelihoods. Mesuwini et al.’s (2023) definition was adopted because it represents Tanzania’s tertiary training systems, graduates from which were the main focus of the current study.

The Human capital theory, propounded by Becker (1964) , states that formal education is essential to improve the productive capacity of a population. As human labor becomes a capital good, the productivity of human labor in terms of employability can be improved through formal education. Proponents of the theory viewed formal education, such as entrepreneurship education in terms of knowledge, training, and competence as a helpful resource to increase graduates’ employability ( Aboobaker, 2020 ; Jafari-Sadeghi et al., 2020 ). Employability depends on entrepreneurial knowledge, training and competence ( Aboobaker, 2020 ). Among the criticisms of human capital theory is its emphasis on the capitalization of human beings, removing the noble characteristics of labor ( Aboobaker, 2020 ). Although human capital theory primarily emphasizes the acquisition of individual skills and knowledge as a means of enhancing employability, it may overlook the broader societal and structural factors that influence employment outcomes ( Adom & Asare-Yeboa, 2016 ). Nonetheless, the theory has withstood those criticisms and it is relied upon by several researchers in explaining entrepreneurship knowledge, training, competence and employability ( Adom & Asare-Yeboa, 2016 ; Jafari-Sadeghi et al., 2020 ; Venesaar et al., 2022 ).

Self-efficacy reflects an individual’s sincere thoughts on whether they have the abilities perceived as important to task performance, as well as the belief that they will be able to effectively translate those skills into a chosen outcome ( Bandura & Wessels, 1997 ). It is the extent to which individuals are confident about their entrepreneurial skills to complete various tasks ( Adnyani & Suwandana, 2022 ). Furthermore, Ma et al. (2023) describe self-efficacy as individuals’ confidence in their ability to start a business venture successfully. Nonetheless, Doanh (2021) and Blokker et al. (2019) expound on entrepreneurial self-efficacy as individuals’ belief in their personal ability and capacity to successfully undertake entrepreneurial tasks, activities, and challenges. Yet, Schunk and DiBenedetto (2021) conceptualize self-efficacy as mastery experiences, vicarious experiences, social persuasion and physiological feedback. Schunk and DiBenedetto’s (2021) definition was adopted by this study as it contains the variables of interest. The idea of self-efficacy has been widely used in the literature on career theory to explain job preferences, perceived career options, and career-oriented behaviours ( Storme & Celik, 2018 ). Moreover, Bandura et al. (2001) listed self-efficacy as one of the many socio cognitive variables of children’s job goals, on which they found academic self-efficacy has the largest direct effect.

Previous studies on the link between entrepreneurship competence and employability in developed countries—Spain, the Netherlands, Australia and Portugal reported findings. For example, Hodzic et al. (2015) used a randomly selected sample of 113 unemployed adults in Spain, and a simple Chi-Square analysis technique reported a positive and significant relationship between their entrepreneurial competency and employability. Also in Spain, Iglesias-Sánchez et al. (2019) used a cluster-based sample of 329 public university students and the multiple regression analysis technique. Their findings showed a positive and significant relationship between entrepreneurship competence and employability. In the Netherlands, Blokker et al. (2019) used a conveniently selected sample of 704 young professionals from the informal sector. Applying SEM in data analysis, their study found that entrepreneurship competence and employability are positively and significantly related.

Conversely, studying the same link in Australia, Jackson (2014) used a convently determined sample of 1008 university undergraduate students. Utilizing the multiple regression analysis technique, the results showed that entrepreneurship competence and employability are positively but insignificantly related. Similar results are reported in Estonia and Latvia by Kozlinska et al. (2020) who also reported a positive but non-significant relationship between entrepreneurship competence and employability. Moreira’s study used a convenience sample of 2099 employees and utilized SEM for data analysis. Further studies by Ferreras-Garcia et al. (2021) . conducted in Spain revealed positive relationship between entrepreneurship competence and employability among higher education graduates.

Put together, the findings are based mainly on conveniently selected big samples, except Hodzic et al.’s (2015) which was small and randomly selected. Another exception is the use of one institution by Iglesias-Sánchez et al. (2019) . Although these conditions may question the generalizability of the findings, it is still observed that the findings are contradictory, calling for more empirical studies especially probing the conditions under which the relationship is strengthened or weakened. In such situations, Hayes (2017) recommends moderation analysis. Therefore, the first hypothesis of the present study was:

H₁: Entrepreneurship competence has a positive effect on the employability of TVET graduates in Tanzania.

Various research showed that entrepreneurship competence accelerates individual interest by influencing self-efficacy for employability ( Baiti et al., 2017 ; Doanh, 2021 ; Kusumojanto et al., 2021 ; Ma et al., 2023 ; Schunk & DiBenedetto, 2021 ; Wang et al., 2023 ; Wardana et al., 2020 ; Wilson et al., 2007 ).

Preceding studies suggest level self-efficacy is a crucial aspect in the study employability of graduates ( Ma & Bennett, 2024 ; Schunk & DiBenedetto, 2021 ). In the same line, previous studies, insisted that, high level of self-efficacy can enhance graduates’ confidence in their abilities, leading to more proactive engagement in job search activities, networking, and entrepreneurial endeavors which facilitates employability ( Doanh, 2021 ; Pollack & Lilly, 2008 ; Wang et al., 2023 ; Zhang et al., 2022 ).

While, Schunk & DiBenedetto (2021) argue that individuals with less self-efficacy level are unlikely to persevere and apply effort to overcome hurdles and enhancing their employability prospects. In this view and as encouraged by previous scholars, Wang et al. (2023) and Doanh (2021) claimed that the influence of entrepreneurship education on employability could either be moderated or mediated by different variables.

In the same note, series of studies have shown that self-efficacy is a positive modern actor of entrepreneurship studies ( Asimakopoulos et al., 2019 ; Wilde & Hsu, 2019 ). Furthermore, Wang et al. (2023) and Doanh (2021) , on their arguments, suggested that, the influence of entrepreneurship competence on employability can either be moderated by different variables. In line with the literature review made, and contradictory findings noted on the studies of entrepreneurship education and employability, this study therefore intend to assess the moderating effect on the self-efficacy on the relationship of entrepreneurship competence and employability of graduates; in order to strength the power of HCT in predicting employability of graduates. With regard to the reviewed literature, the study hypothesized that,

H₂: Self-efficacy positively moderates the relationship between entrepreneurship competence and employability of graduates from TVET institutions.

Control variables play a crucial role in research by ensuring the validity and reliability of the findings ( Kohler et al., 2024 ). Statistically controlling for the effects of variables ensures the accuracy of the estimated causal-effects models. Although the control variables (CVs) in a given study are not the variables of interest, they are used to rule out alternative interpretations of the study’s findings ( Shiau et al., 2024 ). Improperly handled CVs can confound the significance of the estimated relationships among the variables of interest. For example, any change in a CV, whether due to manipulation, social desirability bias, or other reasons, could distort the correlation between the independent and dependent variables ( Shiau et al., 2024 ). This study focused on entrepreneurship competence, one of the dimensions of entrepreneurship education, the other dimensions being entrepreneurship knowledge and entrepreneurship training. Hence, to control for the effect of the presence of these other submissions in the estimation of the effect of entrepreneurial competence on employability, the study used the two in the model as control variables. Firstly, to improve the interpretability of the findings, and secondly, to enhance the generalizability of the findings ( Ngatuni & Matoka, 2020 ). The conceptual model as shown in Figure 1 indicates the control variables applied in the study.

Entrepreneurship competence was measured by seven items, adapted from González-López et al. (2021) and Jafari-Sadeghi et al. (2020) . The measured items were; “I am able to identify customer’s need”; “I am capable of developing business opportunity”; I am able to treat new problems as opportunities”; “I am able to take reasonable business-related risks”; “I am capable of planning the operations of the business”; and I am able to determine strategic actions and overcoming obstacles.

Graduates’ employability was measured by seven items adapted from Hosain et al. (2021) and Matonya (2018) . The measured items were; “I have good communication skills (verbal and written)”; “I possess good problem-solving skills”; “I am able to treat new problems as opportunities”; “I am able to take reasonable business-related risks”; I am capable of planning the operations of the business”; and “I am able to determine strategic actions and overcoming obstacles”.

Self-efficacy was measured by six items adapted from Schunk and DiBenedetto (2021) and Wang et al. (2023) . The measured items were; “I have always been successful in doing well tasks related to business”; “I do well on even the most complex business problems”; “I admire peers who are professional in conducting business activities”; “I enjoy learning about new things by observing or hearing about others’ experiences”; I find it interesting to listen to others’ stories and learn from their mistakes; and “I am more likely to adopt new ideas or beliefs when they are endorsed by influential individuals.”

Entrepreneurship training was measured using nine items adapted from Glaub and Frese (2011) . The measured items were; “I received training that helped build my confidence” “I received training that motivated me to pursue my goals”; “I received training that provided me with practical strategies for overcoming obstacles and setbacks”; “I receive training that makes me employable”; “I received training on how to develop a proactive approach to problem-solving”; and “I received training on how to set and achieve entrepreneurial goals”. The other items were; “I received training on how to become more self-directed and take ownership of my entrepreneurial journey”; “I received training on how to manage my time effectively”; and “I received training on how to promote product and service”.

Entrepreneurship knowledge was measured by eleven items adapted from Karyaningsih et al. (2020) and Al Mamun et al. (2019) . The measured items were; “I know how to find the resources (e.g. financial) to set up a business”; “I have a good understanding of how to manage financial resources”; “I am aware of how to use technology to achieve business goals”; “I have good understanding on how to work on team” and “I am knowledgeable about how to identify and target potential customers”. Yet the other items were; “I am aware of how to create a marketing plan”; “I understand how to increase sales”; “I have sufficient knowledge in commercializing a business idea”; “I am aware of the legal requirements for registering a new business”; “I have knowledge of how to protect intellectual property rights for a new business”; and “I understand the importance of complying with tax laws and regulations”.

Respondents were asked to rate their degree of agreement to each of the measured item under each construct’s measurement scale on a five-point Likert-like scale ranging from 1 = strongly disagree to 5 = strongly agree. Demographic variables such as age, gender, level of education and the time taken to seek for employment were included for the purpose of describing the sample.

Structural equation modeling (SEM) technique with IBM’s Amos Statistics v. 23 was used for data analysis. SEM was preferred due to its ability to analyze and test complex relationships between observed and latent variables ( Glaub & Frese, 2011 ). First, the key assumptions underlying structural equation modeling which are normality, linearity, outliers, multicollinearity and homoscedasticity were all assessed. Normality was tested by generating and evaluating skewness and kurtosis statistics against the cut-offs of ±2 ( Al Mamun et al., 2019 ). The skewness and kurtosis ranged from −.084 to −.957 and from −.034 to .749, respectively all within the recommended threshold, suggesting satisfaction of the normality assumption. Pearson Product Moment Correlation analysis technique was used to assess two assumptions. The first assumption was linearity—the linear relationship between the dependent and independent variables. The linearity assumption is met if most of these correlation coefficients are higher than three (r > .3) and statistically significant ( Pallant, 2020 ). The second assumption was the no-multicollinearity assumption i.e., the absence of very high bivariate correlation between pairs of the independent variables. Field (2018) , Kline (2023) , and Tabachnick and Fidell (2019) suggest that a multicollinearity problem in the data would exist if the correlation between pairs of independent variables is higher than .9 (r > .9). Homoscedasticity was assessed using the scatter plot of regression standardized predicted residual versus the regression standardized predicted residual to ensure that the variance of error is constant across all levels of the independent variables. This is achieved if the residual displayed a random Patten with no systematic structure (no funnel shape). The residual displayed a random Patten with no systematic structure (no funnel shape), thus providing the evidence that homoscedasticity was attained.

Both the exploratory factor analysis (EFA) and confirmatory factor analysis (CFA) were employed to assess the quality of the model. In EFA, the Kaiser-Meyer-Olkin (KMO) measure of sampling adequacy and Bartlett’s Test (BT) were used evaluated against the recommended thresholds of above .6 and significance level of p < .001, respectively ( Field, 2018 ; Tabachnick & Fidell, 2019 ). In the CFA, the goodness of fit indices were evaluated against the recommended thresholds of CMIN/DF (χ²/df) ≤ 3, RMR ≤ .08, GFI ≥ .90, CFI ≥ .90, NFI ≥ .90, TLI ≥ .90, RFI ≥ .90, PCFI ≥ .50, and RMSEA ≤ .08 ( Hair Jr. et al., 2006 ; Hooper et al., 2008 ). Multi-Group analysis was used to test whether self-efficacy strengthens, buffers or reverses the relationship between entrepreneurship competencies and graduates’ employability. Before that, measurement invariance analysis was done as part of the CFA to establish whether the constructs’ measurements were the same across the self-efficacy levels/groups ( Collier, 2020 ).

A descriptive analysis (frequencies) was carried out to understand the samples characteristics. All 353 questionnaires were included in the analysis. The results shows that IAA contributed slightly more participating graduates (51.8 percent) than DIT. The majority of respondents were male (61.2 percent). While 49 percent were aged between 20 - 30 years followed by middle aged group (45.6 percent), 58.4 percent were holders of bachelor level of education. The graduates who were still seeking employment constituted 53.5 percent of the sample. Out of the 164 employed graduates, 64 percent indicted to have spent 4 - 5 years since graduation to secure employment, see Table 1 .

Source: Field data; 2024.

The Kaiser-Meyer-Olkin (KMO) was used to assess sampling adequacy using the threshold of KMO > .6 ( Field, 2018 ). Bartlett’s test (BT) was also used to determine whether the correlation matrix of the measurements is significantly different from zero (an identity matrix). A threshold of correlation > .7 (p < .001) was used, below which the sample would have been judged to be unfit for factor analysis ( Field, 2018 ). The results show KMO = .966, which is higher than the threshold of .6 ( Tabachnick & Fidell, 2019 ), indicating that the sampling adequacy was met. The Bartlett’s test of sphericity returned ${\text{χ}}_{\left(\text{703}\right)}^{\text{2}}$ = 12731.5, p < .001, indicating that the correlation matrix is statistically different from an identity matrix. Thus, the data qualified for exploratory factor analysis (EFA). The Principal Component Analysis Approach (PCA) was employed to carry out the EFA. The results show that the measurement items loaded to their designated factors (factor loadings > .5) ( Tabachnick & Fidell, 2019 ), leading to the predicted factor structure of five constructs. The validity and reliability of study’s constructs were assessed. The results show Cronbach’s Alpha coefficients ranging from .927 to .968, all above .9, showing excellent reliability of each construct’s measurement scale ( George & Mallery, 2022 ). Moreover, the Average Variance Extracted (AVE) values ranged from .505 to .784, all above .5 indicating that divergent validity of the construct was achieved, see Table 2 .

Source: Researcher, 2024.

The bivariate correlation coefficient was calculated to assess the linearity and no-multicollinearity assumptions. First, the results show correlation coefficients between the dependent variable and independent variables ranging from .420 to .704, all above .3 and statistically significant at p < .001 level, supporting the linearity assumption ( Pallant, 2020 ). Second, the results ( Table 3 ) also show correlation coefficients between pairs of the independent variables ranging from .275 to .571, all below the threshold of .9 ( Field, 2018 ; Kline, 2023 ; Tabachnick & Fidell, 2019 ), suggesting the absence of multicollinearity problem in the data, see Table 3 .

Notes: **. p < .001 (2-tailed). Source: Researcher, 2024.

Having passed the EFA, Confirmatory Factor analysis was performed in order to test whether the data approved the hypothesized measurement model based on the theory. The measurement models for individual constructs assessed to determine whether there was a relationship between the observed variables and their underlying latent construct. The results confirmed that the models fitted the data, confirming that the set indices observed variables grouped together to measure the underlying construct. Then the overall measurement model was constructed and evaluated.

The overall measurement model with all study construct was assessed. The results shows that the model fitted the data well with CMIN/DF (χ²/df) = 1.663, RMR = .034, GFI = .90, CFI = .991, NFI = .942 TLI = .990, RFI = .937, PCFI = .917, and RMSEA = .023; all meeting the recommended thresholds, see Figure 2 .

The results of the structural model show fit indices of CMIN/DF (χ²/df) = 1.198, RMR = .031, GFI = .914, CFI = .991, NFI = .947, TLI = .990, RFI = .942, PCFI = .909, and RMSEA = .024; all within the recommended thresholds. It was therefore concluded that the structural model was fit for hypothesis testing, see Figure 3 .

Regression estimates were obtained from the structural model to test the study’s hypotheses. The results show that entrepreneurship competencies were positively

and statistically related to employability (B = .541, p < .001). This indicates that holding entrepreneurship training and entrepreneurship knowledge constant, a unit increase in entrepreneurship competencies would lead to .541 increase in the graduates’ employability, supporting hypothesis one (H1), see Table 4 .

Note: ***p < .001. Source: Researcher, 2024.

Moderation analysis was performed to determine whether SEC strengthens, weakens or reverses the relationship between entrepreneurship competence and graduates’ employability. Since the self-efficacy construct was measured by a five-point Likert-like scale, mean scores were computed and used to categorize the construct into three groups. The groups were the low self-efficacy group if the mean score was less than 3, moderate self-efficacy group if the mean score was 3, and high self-efficacy if the mean score was greater than three. The results show that that 63.2 percent of the respondents were categorized into the high self-efficacy group while 34.3 percent were categorized into the low self-efficacy group. For statistical reasons, the moderate self-efficacy group with only 2.5 percent of the respondents was excluded in the subsequent analysis, see Table 5 .

Source: Researcher, 2024.

Measurement invariance testing, a requirement for the multi-group moderation analysis, was conducted to ensure that the measures of a given construct are consistent between the two self-efficacy groups. This step further ensures that measurement models yield a comparable representation of the same construct under various circumstances ( Hair Jr. et al., 2010 ). Xu and Tracey (2017) recommend the use of metric invariance to determine if the measures of a latent variable (i.e. the observed variables) varied between groups. Two tests were carried out to assess the measurement invariance—namely, configural invariance and metric invariance.

The fully unconstrained model was fitted to both low and high self-efficacy groups for entrepreneurial competence and graduates’ employability to assess the configural invariance. The results show a CMIN values of 114.988 for the unconstrained model which is statistically significant (p < .001), see Table 6 .

Source: Researcher, 2024.

In line with the results in Table 6 , the multi-group CFA generated the following indices: TLI = .926, CFI = .927, RMSEA = .044, NFI = 0,837, PCFI = .910 GFI = .784 and CMIN/df = 1.675, as shown in Figure 4 . These fit indices showed a satisfactory configural model fit. It was, therefore, concluded that the factor structure is a good fit for both low and high self-efficacy groups.

In assessing the moderating effect of SEC on the relationship between ECO and EMP, important tests including the configural invariance and the Chi-square difference tests were assessed. The finding revealed that, the models with low and high SEC attained the configural non-invariance as evidenced by the model fit indices as shown in Figure 4 where all model fit indices were within the prescribed cut-offs hence implying the best fit of data to the model across the two groups of SEC with the corresponding Chi-squared difference of 144.217 being statistically significant at p < .0001. Then, the comparative analysis of the unconstrained as well as the constrained model was conducted, thereby being followed by the path-by path analysis, see Table 7 .

The metric invariance test was carried out to assess whether the measured indicators are measuring the same thing across the two self-efficacy groups. This was achieved by comparing the factor loadings of the constrained model to those of the unconstrained model. The measurement weights should not be statistically significantly different across the two groups. The results revealed the Chi-square

Source: Field Data, 2024.

statistic of 15.685 being not statistically significant at a 5% level (p = .097). These results imply that the observed indicators measured the same thing across the two self-efficacy groups, see Table 8 .

Source: Researcher, 2024.

The presence of non-invariance further suggests that the effect of independent variables on the dependent variable might be different hence necessitating for the further moderation analysis called path-by-path analysis.

After constraining the ECO <--- EMP path, we performed a path analysis. The results showed that the computed Chi-squared values were (1897.974) which were even larger as compared to the threshold Chi-square value of 1039.58 at 5% as shown in Table 7 . The results indicate the presence of a statistically significant moderating effect of SEC on the relationship between ECO and EMP.