The TPB, developed by Icek Ajzen in 1991, is a widely acknowledged psychological model used to explain and predict human actions and decision-making processes. According to the TPB, an individual’s behavioural intention serves as the strongest and most direct determinant of whether they will engage in a particular action. Ajzen (1991) expanded upon the earlier Theory of Reasoned Action by incorporating the concept of PBC, thereby enhancing the model’s ability to account for a wider variety of behaviours, such as entrepreneurial activities. Theory of Planned Behaviour offers a comprehensive and flexible framework for understanding EIs. Additionally, TPB proposes that intention is the strongest predictor of behaviour. The core components of TPB have been proven to strongly predict EIs across various cultural and educational contexts (Lihua 2022; Mfazi & Elliott 2022).

The application of TPB in entrepreneurship research, including among unemployed youth in South African townships, underscores its relevance in guiding interventions to foster entrepreneurial activities, thereby promoting economic empowerment and self-sufficiency in marginalised communities. These behavioural intentions are shaped by three fundamental components: an individual’s PA towards the behaviour, the perceived social pressure or SN and the PBC, which reflects one’s confidence in one’s ability to perform the behaviour.

Personal attitude denotes an individual’s overall favourable or unfavourable approach towards performing a particular behaviour. In entrepreneurship, this concept refers to the extent to which an individual holds favourable views of being an entrepreneur. Those who perceive entrepreneurship as a pathway to financial independence, personal fulfilment or social contribution are more likely to develop EIs (Krueger, Reilly & Carsrud 2000; Liñán & Chen 2009). Conversely, negative attitudes, such as fears of financial risk or business failure, can hinder the development of these intentions.

Prior South African studies have demonstrated that favourable attitudes towards entrepreneurship significantly enhance the likelihood of EIs (Fatoki 2014; Kolapo et al. 2023). Therefore, promoting positive entrepreneurial attitudes is critical in fostering intention. Drawing from these insights, we propose the following hypotheses:

Subjective norm relates to the perceived social pressures or influences from essential others regarding whether to engage in a specific behaviour (Azjen 1991). These ‘others’ could include family, friends and mentors. Within the entrepreneurial context, SNs represent the perceived social expectations or opinions of important individuals such as family, friends or mentors regarding one’s decision to pursue entrepreneurship (Liñán & Chen 2009). They can significantly impact EIs, especially in collectivist societies in which family and community approval are vital (Iakovleva, Kolvereid & Stephan 2011). For instance, individuals who feel supported to pursue entrepreneurship are more likely to form solid EIs. While some research indicates that social support strengthens EI (Arthur & Peprah 2024; Shinnar & Young 2008), others find social support to be a weak or non-significant predictor in under-resourced settings (Acuña-Duran et al. 2021; Iakovleva et al. 2011).

Accordingly, the following hypotheses are formulated:

Perceived behavioural control concerns an individual’s perception of how easy or difficult it is to perform a particular behaviour. It is one of TPB’s key components and incorporates two subdimensions: self-efficacy or belief in one’s own competence (Bandura 1977) and controllability or belief in control over external barriers (Ajzen 2002). While related to self-efficacy, PBC is broader in scope, encompassing both internal and external dimensions of control. Entrepreneurship encompasses assessing skills, knowledge and access to resources that are necessary to start and run a business (Krueger et al. 2000).

People who perceive themselves as having the necessary skills, abilities and resources to start a business are more inclined to develop intentions towards engaging in entrepreneurial activities (Liñán & Chen 2009). Therefore, it is argued that PBC influences actual behaviour. Fayolle and Gailly (2015) and Neneh (2019) state that people with high confidence in their abilities are likelier to act on their EIs, especially when the opportunities align with their perceived control. In township contexts, this construct is particularly important as a result of systemic barriers that may inhibit confidence and perceived feasibility. Therefore, the following hypotheses are set:

The TPB has been extensively utilised in entrepreneurship research to examine how PAs, SNs and PBC influence EIs in various settings. The theory has also been employed to analyse how these factors differ across cultural and contextual environments, offering insights into the diverse determinants of entrepreneurial behaviour. For example, Iakovleva et al. (2011) compared EIs between developing and developed countries. They found that the predictive strength of TPB components, particularly PBC, was more pronounced in developing economies. These findings underscore the importance of contextual factors, such as cultural norms and economic conditions, in shaping EIs. Entrepreneurship education has also been a vital area in which TPB has been applied to measure the effectiveness of interventions in enhancing EIs.

Fayolle and Gailly (2015) applied the TPB to examine the extent to which entrepreneurship education impacts students’ attitudes, perceived social norms and sense of behavioural control towards entrepreneurial activities. Their study demonstrated that educational programmes emphasising entrepreneurial skills and resources can significantly improve students’ EIs, especially by enhancing PBC. Additionally, the TPB has also been employed to investigate gender-based variations in EIs. Haus et al. (2013) observed that while women tended to report lower levels of EI compared to that of men, the underlying associations between the TPB constructs and EIs were largely consistent across both genders. Their research suggests that interventions to improve PBC and SNs could be particularly effective in encouraging women to pursue entrepreneurship.

Some scholars have expanded TPB to incorporate contextual variables to understand EIs in specific populations. Lüthje and Franke (2003) integrated environmental variables such as the availability of entrepreneurial resources, as well as the perceived feasibility of entrepreneurship, into the TPB model when studying EIs among engineering students. These extensions highlight the adaptability of TPB to different contexts, making it a valuable framework for studying entrepreneurship in diverse settings.

Theory of Planned Behaviour is a valuable theory for understanding EIs among unemployed youth in South African townships, where high unemployment rates and socio-economic challenges influence entrepreneurial aspirations (Mayekiso & Obioha 2022; Meyers-Mashamba 2021).

Applying the TPB framework helps identify critical areas for intervention. These include fostering positive attitudes towards entrepreneurship through success stories, reshaping social norms to support entrepreneurial activity and strengthening PBC by providing training and resources. Educational and community programmes can also leverage TPB components and promote entrepreneurship as a viable career option for young people with limited job opportunities (Dolan & Rajak 2018; Johri et al. 2024; Yan, Huang & Xiao 2023).

Theory of Planned Behaviour has been widely utilised in entrepreneurship studies to elucidate the psychological factors that precede and shape EIs (Liñán & Chen 2009; Neneh 2019). Additionally, the model is especially relevant in developing contexts where resource constraints and socio-cultural factors significantly influence entrepreneurial decisions (Iakovleva et al. 2011). Therefore, in township environments like Mdantsane, understanding how these constructs indicate associations or relationships is important for designing more effective interventions to support youth entrepreneurship.

The literature reviewed in the study is based on unemployment, educational barriers and entrepreneurship support in the Eastern Cape province.

A quantitative, cross-sectional research design was employed to test the TPB framework. This approach is appropriate for identifying patterns and testing theory-driven hypotheses with statistical rigour (Hair et al. 2014).

From a total population of 755 200, an estimated 265 475 individuals are unemployed youth, representing 35.1% of the population. The study targeted unemployed youth aged 18–35 years residing in Mdantsane Township. A non-probability purposive sampling method was used, appropriate for targeting specific, information-rich subpopulations (Pace 2021). A final sample of 248 respondents was deemed sufficient, based on structural equation modelling (SEM) sample size guidelines (Kline 2016). This non-probability method is justified due to the targeted nature of the population and study objectives.

The study employed measurement scales adapted from previously validated instruments: personal attitude (Liñán & Chen 2009), SN (Krueger et al. 2000), PBC (Fayolle & Gailly 2015) and EI (Neneh 2019). The reliability and validity of the constructs were verified through CFA. All variables were assessed using a five-point Likert scale ranging from 1 (strongly disagree) to 5 (strongly agree).

Data analysis was conducted by using the Statistical Package for the Social Sciences version 29 and Analysis of Moment Structures (AMOS) version 22. Confirmatory factor analysis was performed in AMOS to assess and validate the measurement model’s construct reliability and validity. Reliability was assessed by using Cronbach’s alpha, McDonald’s omega coefficients and the Jöreskog rho. A descriptive approach was used to describe the study’s demographic characteristics and the established main theoretical variables. A one-tailed test of the significance of the correlation (Pearson’s r) was used. Path analysis using structural equation modelling was used to determine the direct effects of the hypothesised frameworks. A bootstrapped maximum likelihood approach was used as an estimation approach for both the measurement and the structural models. To assess the significance of the beta estimates, 95% bias-corrected bootstrapped confidence intervals and the associated p-values were used.

Ethical clearance for this study was obtained from the Walter Sisulu University Research Ethics Committee (No. FEDREC15-06-23-3).

Descriptive statistics were generated to provide an overview of the respondents’ demographic profiles and to summarise the key study constructs, including PA, SN, PBC and EI. The demographic details of the study participants are presented in Appendix 1 Table 1–A1.

The measurement instrument underwent preliminary evaluation to assess its validity and reliability. Confirmatory factor analysis was performed to verify construct validity and to establish the measurement model. Additionally, reliability analysis was conducted to examine the internal consistency of the identified factors and ensure the robustness of the measurement scales. As the measurement tool was adopted by other authors, literature about the measurement instrument was used to guide the empirical factors to be loaded for the CFA. To examine the model fitness, we used some goodness-of-fit indices coupled with their respective criteria (Table 1).

Overall, the findings from the validity and reliability assessments of the measurement instrument derived from the CFA and reliability coefficients demonstrated strong psychometric properties. The most parsimonious measurement model was attained after removing items with low factor loadings. As in Table 2, all retained items exhibited factor loadings exceeding 0.60, confirming their statistical significance and construct relevance. The average variance extracted values for all constructs were above the recommended threshold of 0.50, indicating satisfactory convergent validity. Furthermore, reliability analysis revealed Cronbach’s alpha and McDonald’s omega coefficients greater than 0.80 across all constructs, signifying excellent internal consistency and measurement reliability.

Additionlly, the assessment of composite reliability (CR) using Jöreskog’s rho indicated values exceeding 0.80 for all significant constructs, confirming a high level of CR for the established measurement model. The finalised measurement model (see Figure 1) demonstrated substantial factor loadings, reflecting a strong and satisfactory alignment between the observed items and their corresponding latent factors.

An evaluation of the overall model fit indices further supported the adequacy of the measurement model. As presented in Table 3, the chi-square to degrees of freedom ratio (CMIN/df = 3.234) falls within the acceptable range of 3 to 5, indicating a reasonable fit. The standardised root mean square residual (SRMR = 0.041) is below the 0.05 threshold, suggesting a good model fit. Similarly, the root mean square error of approximation (RMSEA = 0.086; 90% CI [0.072, 0.100]) indicates an acceptable level of fit. In terms of incremental fit indices, the comparative fit index (CFI = 0.942) and Tucker-Lewis index (TLI = 0.926) both exceed the recommended minimum of 0.90, providing further evidence of an acceptable and well-fitting measurement model.

Table 4 provides a descriptive summary and correlation matrix for the principal theoretical constructs. The mean values (measured on a five-point Likert scale) indicate moderately high levels across the key study variables. A correlational analysis examined the linear relationships among the main constructs, serving as a preliminary step for subsequent inferential analyses aimed at testing the hypothesised model.

A one-tailed Pearson’s correlation test was employed to determine the statistical significance of the relationships. The interpretation of correlation strength followed the thresholds proposed by Cohen (1988), in which coefficients between 0.10 and 0.29 indicate a weak relationship, those between 0.30 and 0.49 represent a moderate relationship, and values above 0.50 denote a strong association.

The results revealed that EI demonstrated strong, positive and statistically significant correlations with SN (r = 0.755), PBC (r = 0.753) and PA (r = 0.780), all significant at p < 0.001. These findings suggest a substantial effect size, indicating that higher levels of EI are associated with corresponding increases in SN, PBC and PA.

Figure 2 illustrates the structural model developed to evaluate the hypothesised relationships outlined earlier. The goodness-of-fit indices indicated that the model achieved an acceptable and satisfactory fit to the observed data. Employing a maximum likelihood-based bootstrap estimation procedure, we computed the structural regression path coefficients to assess the direct effects among the latent variables within the model. The model fit summary shows that the model is a generally acceptable fit, thus RMSEA = 0.096, SRMR = 0.043 < 0.05, CMIN = 3.247 < 5, CFI = 0.928 > 0.90 and TLI = 0.909, which is greater than 0.90.

Table 5 presents the maximum likelihood estimates of the structural path coefficients, along with their corresponding bootstrapped 95% bias-corrected confidence intervals.

Hypothesis H1a: The analysis revealed that the direct effect of PA on EI was positive and statistically significant (β = 0.388; SE = 0.167; 95% CI [0.082, 0.719]; p = 0.023). This result supports H1a, indicating that a favourable PA significantly enhances EI. Consequently, the null hypothesis is rejected.

Hypothesis H2a: The regression path coefficient for the direct effect of SN on EI was positive but not statistically significant (β = 0.258; SE = 0.182; 95% CI [−0.060, 0.660]; p = 0.103). This finding does not empirically support H2a, suggesting that SN does not significantly and directly influence EI. Therefore, the null hypothesis (H2₀) is accepted.

Hypothesis H3a: The direct effect of PBC on EI was found to be positive and statistically significant (β = 0.364; SE = 0.141; 95% CI [0.115, 0.669]; p = 0.004). This outcome supports H3a, confirming that higher levels of PBC significantly contribute to stronger EIs.

The results revealed that PA had a positive and statistically significant direct influence on EI (β = 0.388; p = 0.023). This finding indicates that unemployed youth who perceive entrepreneurship favourably are more inclined to develop stronger intentions to establish their own ventures. This outcome supports the central proposition of the TPB, which asserts that a positive disposition towards a particular behaviour enhances the likelihood of intending to perform that behaviour (Ajzen 1991; Sibanda et al. 2020). The finding is also consistent with prior studies that identified PA as a crucial antecedent of EI (Liñán & Chen 2009; Urban 2010).

Similarly, Fatoki (2014) reported that South African students who regarded entrepreneurship as appealing and rewarding demonstrated higher EIs. This finding is significant in the context of unemployed township youth because it emphasises the need for interventions that cultivate positive attitudes towards entrepreneurship. These interventions could include entrepreneurship education, exposure to entrepreneurial role models and community programmes that showcase the benefits and opportunities of entrepreneurship.

Contrary to expectations, the SN did not have a statistically significant effect on EI (β = 0.258; p = 0.103), although the norm was positively correlated. While the social pressures from family, friends and community members may influence youth in other contexts, the lack of a significant relationship in this study suggests that external social influences are less pivotal in shaping EIs among unemployed youth in the township.

This outcome is not unexpected as previous research has reported inconsistent findings regarding the influence of SNs on EI. For instance, Iakovleva et al. (2011) observed that in several developing economies, SNs exhibited either weak or statistically insignificant effects on individuals’ intentions to engage in entrepreneurship. This observation may be a result of several factors, including the socio-economic realities of the township environment, where individuals may prioritise personal agency and control over external influences when considering entrepreneurship (Sibanda et al. 2020).

The lack of significant influence from SNs could also indicate that societal expectations in this context do not strongly promote or discourage entrepreneurship. As a result, youth may rely more on their personal beliefs and perceived ability to control entrepreneurial outcomes rather than social approval or disapproval. This finding highlights a potential gap in community support structures for entrepreneurship, suggesting the need for more targeted efforts to foster a culture of entrepreneurship within township communities (Makwembere et al. 2024).

The findings of this research add to the broader literature on EI by reaffirming the applicability of the TPB within a South African township context. The results highlight PA and PBC as key determinants of EI, whereas SN appears to play a comparatively weaker role in shaping entrepreneurial motivation among unemployed township youth.

The study provides context-specific insights into the applicability of TPB within a South African township, highlighting how socio-economic challenges may shape individuals’ perceptions of and responses to entrepreneurial opportunities. The significant role of PBC suggests that structural barriers, such as limited access to resources and skills, may be particularly salient in township environments, making self-efficacy and control crucial for EI.

The study was confined to one township and may have limited potential for generalisation to other contexts. Further research may extend the analysis into other townships and rural areas to explore what factors influence unemployment in the youths’ intention to be entrepreneurial. The study focused on a single township, limiting generalisability. Future research should consider multi-township or longitudinal approaches to assess change over time or comparative approaches across multiple provinces. The cross-sectional design precludes causal inference.