Mediation analysis of entrepreneurship development on the economic growth of women entrepreneurs in Nigeria

: The study used structural equation modelling (SEM) techniques to examine the mediation analysis of entrepreneurship development on the economic growth of women entrepreneurs in Nigeria. The study assumes the existence of three null hypotheses; entrepreneurship development does not mediate business performance, psychological, and knowledge of business practice indicators for economic growth. Data were collected from 500 randomly selected women entrepreneurs in Nigeria. Our results suggest that separating household and business income, maintaining a separate business account, stock-keeping, record-keeping among other were the knowledge of business practice indicators; openness to change, desire for self-independent among others were the psychological indicators, and reduced inactivity, increase in the size of the inventory among others were the business performance indicators that stimulate entrepreneurship development. The aforementioned catalyzed higher formalization, increases in capital investment among others as entrepreneurship development indices, which stimulate economic growth of contribution to personal and children school fare, reduction in financial dependence, and increase in affordable health care fee among others. We also established that entrepreneurship development mediates knowledge of business practice to economic growth.


Introduction
In a real sense, economic growth is a macro-economic concept reflecting the process of increasing the sizes of national economies (Alina-Petronela, 2012). Addison (2015) viewed economic growth as a powerful tool for reducing poverty and improving the quality of life.
Growth on its own is capable of generating a virtuous cycle of prosperity and opportunity. Adu et al. (2019); Ravallion (2007) contend that economic growth creates employment opportunity for improving incentives for women to invest in their children's education. This aspect of macroeconomic promotes the development of entrepreneurs. Ndulu et al. (2007) corroborate that economic growth should be measured from the point of physical capital (asset acquisition) and human capital (education and skills) development. Economic growth should be centred on developing strategies for poverty reduction tailored to sustainable economic development for women entrepreneurs. To drive this focus on economic growth, the capacity of women should be built to improve their management skills which are sort for by every entrepreneur. Standing on this economic growth mechanism, the economic capacity of the women should be empowered in form of women agency.
The term women agency means the act of giving women power and control over their own lives. This empowerment energizes the women to participate in social movements and the process of emancipation (Sharifah, 2015). Rajeshwari, 2015;Selvi and Bakialekshmi (2017) suggested that women agency or empowerment implies increasing the spiritual, political, social, educational, gender or economics strength of individuals and communities of women. Far back in 2009, Anju et al. contend that women empowerment affords the women the ability to make strategic life choices which they had been previously denied, seeing that this will help them to make and act on economic decisions (Golla et al., 2011). Selvi and Bakialeshmi (2017) noted that social background, culture and educational level plays an important role to getting the women empowered, this is because Rajeshwari (2015) contend that women empowerment will contribute to the development of the country's economic, social and political space. Empowering women will mean getting women involved in economic activities and these must be with a sensitive intention seen that women oftentimes are victims of gender discrimination, lack of equal opportunities in education, rape, abuse and torture, kidnap, financial constraints among others. These justify the need for the women to be organized and develop themselves into entrepreneurs. Women entrepreneurs by Anju (2015) connotes woman or groups of women who initiate, organize and run a business enterprise. While entrepreneurship as a process makes the women economically strong and freedom to take decisions since women entrepreneurs initiate, organize and operate the business enterprise. Entrepreneurship development helps to build a country's gross domestic product (GDP) as an important component of the demand side of job creation strategies in developing countries (Fox and Kaul, 2017) which Smriti (2020) contend that it is a key to economic development. Swetha et al. (2014) assert that entrepreneurship development was conceived by successive governments as a program of activities to enhance the knowledge, skill, behaviour and attitudes of individual and groups to assume the role of entrepreneurs who manages production resources for product developments. Magnus and Tino (2014) further opined that entrepreneurship has no specific definition noting that what was defined as entrepreneurship in some countries may not be viewed as entrepreneurship in another, they equally contend that entrepreneur is sometimes used to refer to anyone operating a private business regardless of size and activities. Far back in 1934, Schumpeter proposed that a true sense of entrepreneurship should be rested on firms that are innovative and growth-driven and should be capable to cause a ship in economic equilibrium. Later in 1942, Schumpeter added that the function of an entrepreneur is to revolutionize the pattern of production by exploiting an innovation or more generally an entire technological possibility for producing a new commodity or production of old ones in a new way by opening up a new source of supply for the product.
This clearly shows that most small businesses are not in true sense entrepreneurial as they do not bring innovation to the market. Most women in true sense are self-employed and are often mistaken as entrepreneurship, Magnus and Tino (2014) revealed that women in developing countries are more involved in farming, restaurant, child daycare, beauty salon among others, more attention should be paid to self-employed women hence it is capable of empowering them economically.
For quality entrepreneurship development in an economy, Zenobia (2018) contend that entrepreneurs should be made to undergo training which should be preceded with a grant, internship and mentorship which will help to empower the women whom Obianefo et al. (2019);Valerio et al. (2014) considered economically vulnerable. Corroboratively, Ayoade and Agwu (2015) stressed that governments should focus on encouraging entrepreneurship development due to its role in job creation, innovation, importance to large businesses and a dynamic economy. Considering the importance of the study, Zenobia (2018) pointed some indicators that signal entrepreneurship development in an economy whose outcome would bring about economic growth, these Zenobia's indicators are different from the income and profit indicator proposed by Cho and Honorat (2014). The Zenobia's (2018) indicators include business practice and knowledge (formalized record-keeping, separating household and business income, separate business account, improved marketing strategies, stock-keeping practices), business performance (income and profits, Sales, number of wage workers, size of inventory, business start-up, increased hours of work or increased employment, reduced inactivity, loans, savings, business survival, business growth ), and psychological indicators (women's agency or decision-making capacity, confidence, self-confidence and teamwork), with an intermediate indicator signalling entrepreneurship development (more start-ups, increases in investment, improved business knowledge/skills, improved agency over business decisions, higher formalization, improved business practices and performance, increased market access, and more employment) and outcome indicators that symbolizes economic growth. For the clarity of purpose, all these indicators are targeted at stimulating the economic growth of women. Thus, the study adopted a structural equation modelling approach with kin attention to mediation analysis for a better understanding of how the indicators interact with each other. Obianefo et al. (2020) assert that mediation is the introduction of an intermediate variable called a mediator that helps to explain how or why an exogenous variable(s) influences an outcome or endogenous variable(s). MacKinnon andFairchild (2009) andDouglas et al. (2013) noted that it is of great interest to identify the mechanisms through which an intervening variable (mediator) achieves its effect on the outcome variable(s). We, therefore, wish to conceptualize ( Figure 1  Subsequent sections of the article are organized as follows. The next section presents data and their description. Section 3 presents the methods while section 4 provides empirical results and discussion. Section 5 concludes.

Data
The study was carried out in Nigeria, Nigeria is an African country on the Gulf of Guinea with 36 States and 774 Local Government Areas (LGAs) and a Federal Capital Territory in Abuja. Nigeria is sub-divided into six Geopolitical zones (Southeast, South-South, Southwest, Northeast, Northwest and North-central) to aid planning. Nigeria is renowned in literature for commerce, adventure and dexterity. The United Nation's data estimated number of women in Nigeria as 102,590,998 been 49.4% of the total population. Nigeria is located on a latitude 9.0820°N and longitude 8.6753°E with a total land area of 923,768km 2 .
The study population comprised of all the women entrepreneurs involved in micro and small-scale enterprise in Nigeria. Five Geopolitical zones were purposively selected for security reasons from where two States were randomly selected for the study (Southeast; Anambra and Ebonyi, South-south: Akwa Ibom and Rivers, Northwest: Kaduna and Kano, North-central: Niger and Plateau, Southwest: Lagos and Ogun). A well-structured questionnaire and interview schedule was used to collect data from a cross-section of women entrepreneurs. Research assistants were engaged from each State and empowered with SurveyCTO data collect tool (Android data tool kit) through online training. Despite that the use of SurveyCTO complied with COVID-19 guidelines of reducing physical contact, it also helps to improve the quality of data which prevents falsification of information by the research assistants since the database is linked directly to the analyst SPSS package. The good thing about the SurveyCTO is that both the data and GPS coordinates of the respondents can be collected offline which is later uploaded over the internet to the researcher's database.
Finally, fifty women entrepreneurs were randomly sampled from each State to make a sample size of 500 women for the study.

Analytical framework
Schreiber (2008), Raykov (2005) and Byrne (2004) corroborates that structural equation modelling (SEM) is a multivariate statistical analysis technique that is used to analyze structural relationships between measured variables and a latent observation otherwise called construct.
This technique is a hybrid of factor analysis and multiple regression analysis. This method is often preferred by researchers due to its ability to estimates the multiple and interrelated (1) The error terms ( and ) are uncorrelated. These two structural equations are linked together to influence outcome simultaneously unlike two independent standard regression equation, while ( ) is the direct effect of the path from independent variables to the outcome variable through the mediator. Also, is the direct effect, and βxz + Yxy is the sum of the total effect (Clogg et al., 1992, Douglas et al., 2013and Obianefo et al., 2020. Imai et al. (2010) defined SEM in a reduced regression method without a mediator as: ( 3) Furthermore, the mediation effect of SEM was defined by Hair et al. (2013) in equation 4 as: (4) Where a, b, SEa, and SEb are the coefficient of exogenous (independent variable), coefficient of endogenous (dependent variable), standard error of the exogenous and standard error of the endogenous variable respectively. It was not out of place to ensure that the convergent validity was achieved where Fornell and David (1981) proposed composite reliability (CR > 0.7) and average variance extracted (AVE > 0.5) defined in equation 5, 6, 7 and 8 as: Where βi is the standardized regression weight of each variable on a construct, Var(εi) is the error variance, FL is the factor loading. To strengthen the mediation analysis, Fornell and David (1981) contend that the square correlation from the explanatory factor analysis (EFA) of all the construct must be less than the AVE.
SEM also allows for a modification of path until a good fit model is achieved. Due to its complexity, experts devised indices of "goodness of fit" or "approximate fit" using maximum likelihood estimation (MLE) to ensure that researchers come up with a model that meets the stated hypotheses. This goodness of fit indices should express the degree of approximation plus estimation discrepancy, and to provide an additional base for the acceptance or rejection of a 8 model. Almost all the goodness-of-fit indices (GFI) are based on Chi 2 (χ 2 ) and degree of freedom (df) as defined by Hu and Bentler (1998Bentler ( , 1999 in equation 9 as: (9) is the minimum value of the discrepancy function, is obtained by evaluating F with g from maximum likelihood estimation defined by Bollen (1989b) in equation 10: The second model fitness called Turker-Lewis coefficient or index (TLI) is defined by Betler and Bonett (1980) as: ( 11) where and d are the discrepancy and the degrees of freedom for the model being evaluated respectively, db is the discrepancy and the degrees of freedom for the baseline model. The typical range for TLI lies between zero and one, but it is not limited to that range, the value close to one indicates a very good fit. The third model indices we obeyed was comparative fit indexes (CFI) defined by Bentler (1990) as: ( 12) where is the discrepancy, NCP is the non-centrality estimate for the model being evaluated, , and db are the discrepancy, non-centrality and degree of freedom for the baseline model respectively. This CFI model is identical to McDonald and Marsh (1990) relative noncentrality index (RNI) defined as: The only distinguishing features of CFI and RNI is that Bentler (1990) contend that CFI is truncated to fall in the range zero to one. Thus, a CFI value close to 1 indicates a very good fit (10) model. Finally, we also bore in mind the root mean square error approximation (RMSEA) which has an indirect relationship with the residuals since it is based on Chi-square (χ 2 ), degree of freedom (df) and sample size (N). It is therefore defined by Hu and Bentler (1998Bentler ( , 1999 where the formula is expressed as: Several suggestions have been made regarding the critical cutoff values to determine the acceptance or rejection of a model, among which Hu and Bentler (1998Bentler ( , 1999  assertion that suggested convergent validity should not be less than 0.5. The study of Fornell and David (1981) examined the convergent validity using the composite reliability (CR) test and average variance extracted (AVE). The CR is a measure of internal consistency in scale items which Fornell and David (1981); Brunner and Süß (2005) contend that the benchmark for the establishment of the assumption is 0.7. Though, some scholars assert that the value is still debatable as others like Diamantopoulos and Siguaw (2000) suggested a benchmark of 0.6.
Furthermore, Tellis et al. (2009);Hair et al. (2006) noted that AVE is the average amount of variance in measured variables that a construct observation can explain. Fornell and David (1981); Brunner and Süß (2005) also suggested that an acceptable benchmark for the establishment of AVE is 0.5. Table 1

Discriminant validity
A default explanatory factor analysis (EFA) as shown in Figure 2 was estimated to calculate the discriminant validity of the SEM data, Engellant et al. (2016) contend that discriminant validity is the extent by which measures of different constructs diverge or minimally correlate with one another. Hair et al. (2006) noted that, for the assumption to be fully established, the AVE estimates should be higher than the squared correlation estimate. The above assertion corroborates the earlier opinion of Fornell and Larcker (1981) who contend that, for an acceptable discriminant validity test, any two constructs, the AVE for construct one and the AVE for construct two need to be larger than the shared variance (square of the correlation) between the two constructs. Table 2 represents the result of the discriminant validity test which shows that Knowledge of business practice indicator (KBPI) and business performance indicator (BPI); psychological indicator (PSYI) and business performance indicator (BPI); business performance indicator (BPI) and economic growth indicator (EGI); business performance indicator and entrepreneurship development indicator (EDI) are not highly correlated, therefore discriminant validity was fully established through the indicators. Also, KBPI and PSYI; KBPI and EDI are partially established therefore exhibit minimal correlation. Finally; the discriminant validity of KBPI and EGI; PSYI and EDI; EGI and EDI were not established showing a high correlation between the indicators. Thus, there is a need to watch out for the variables.
Down the Table is the model fit summary of the EFA, the five parameters used to judge the fitness of the model include the goodness of fit indices (GFI), comparative fit index (CFI), Tucker-Lewis index (TLI), normed fit index (NFI), and root mean square error of approximation (RMSEA) out of which Kenny (2012) opined that a GFI, CFI, NFI, and TLI closer to one shows a good model, while the RMSEA should be closer to zero. Since four (GFI, CFI, NFI, and TLI ) out of five parameters agree with Kenny (2012). The model is fit to accept the result of the discriminant validity.

Regression relationship between the construct and observed variables.
Table 3 and Figure 3 shows the result of the default SEM analysis done to determine the regression coefficient of the construct indicators and its measurement variables as well as the relationship between all the construct indicators. Down the table is a model fit summary showing a GFI (0.810), NFI (0.805), CFI (0.836) and TLI (0.814) close to 1 and an RMSEA (0.09) close to zero which was in agreement with Kenny (2012). This suggests that the model was a good fit model since they are within the appropriate threshold suggested by Hu andBentler (1998, 1999).
This regression weight represents the causal effect and relationship between the latent construct on the observed variables. Majority of the estimates in Table 3 Table 5 shows the result of the mediation analysis run to test for the stated hypotheses. For simplicity, we calculated the average all the measurement variables that made up each indicator(s) which enabled us to re-draw the path diagram (see Table 5 and figure 4). The factor loading and error variance were equally calculated and were later hand loaded into the path diagram (see Table 4).  983. 0.965,0.968,0.838,and 0.14. Four (GFI,NFI,CFI,and TLI) out of the five fit indices were in agreement with Hu and Bentler 91998, 1999) and Kenny (2012) except for RMSEA whose value is close to 1 against the benchmark of 0.06.

Mediation Analysis
From the table, the coefficient of knowledge of business practice indicator (KBPI) was positive and significant at 1% level of significance, this implies that a 1% increase in the KBPI will cause a 0.543 unit increase in the development of entrepreneurship among women in Nigeria. This was in agreement with the study of Cho and Honorat (2014); Zenobia's (2018) who identified knowledge of business practice as an agent of entrepreneurship development.
The coefficient of entrepreneurship development indicator (EDI) was positive and significant at 1% level of significance, this implies that a 1% unit increase in the women entrepreneurial ability will increase the unit of economic growth by 1.341. This finding was expected as the result was in agreement with the result of Cho and Honorat (2014) who contend that entrepreneurship development indicators acted as an intermediate variable for an outcome effect (economic growth).

4.5: Mediation Establishment
The result of

Conclusion
The study used structural equation modelling (SEM) techniques to examine the mediation analysis of entrepreneurship development on the economic growth of women entrepreneurs in Nigeria. Hypothetically, the study assumes the existence of three null hypotheses (entrepreneurship development does not mediate business performance indicators for economic growth, entrepreneurship development does not mediate psychological indicators for economic growth, and entrepreneurship development does not mediate knowledge of business practice indicators for economic growth) which we tried to uncover. Empirical reviews helped us to identify three entrepreneurship development indices as knowledge of business practice, business performance, and psychological indicators, whose target outcome is economic growth.
Data were collected from 500 randomly selected women entrepreneurs in Nigeria through an Android data collection tool known as SurveyCTO data collect, the choice of the data tool complied with Covid-19 protocols and the United Nation's guideline. Methodologically, we ran a convergent validity test of the reliability of variables that constitutes the measurement of each latent construct for internal consistency, the convergent validity test was in two parts of composite reliability (CR) and average variance extracted (AVE) where the rule of thumb by Fornell and David (1981); Brunner and Süß (2005) assumes a benchmark of 0.7 (CR) and 0.5 (AVE) respectively. This approach helped us to keep an eye on variables that are highly corrected with each other. We found that convergent validity was not established in economic growth indicator.
Discriminant validity was equally checked in-line with Hair et al. (2006) who contend that the AVE estimates should be higher than the squared correlation estimate between two constructs. A default explanatory factor analysis (EFA) was used to examine the discriminant validity where the study revealed that out of 10 discriminant test, only 3 was not established. In all these, we ensured that Kenny (2012); Hu and Bentler's (1998Bentler's ( , 1999 recommended values for the goodness of fit index (GFI), comparative fit index (CFI), normed fit index (NFI), Turkey-Lewis index (TLI) and root mean square error of approximation (RMSEA) were achieved to assure us more credible results for policy and novel contribution to literature.
Our results suggest that separating household and business income, maintaining a separate business account, stock-keeping, record-keeping, and use of ICT tools were the knowledge of business practice indicators that stimulate entrepreneurship development. Also, openness to change, desire for self-independent, team-work, and decision-making capacity were the psychological indicators that stimulate entrepreneurship development among women. Furthermore, reduced inactivity, increase in the size of the inventory, availability of market for the products, improved access to credit, and increased business savings were the business performance indicators that stimulate entrepreneurship development among women. Of all the entrepreneurship indices, our findings tend to corroborate Cho and Honorat (2014); Zenobia's (2018) indicator(s). Thus, our result calls for government and non-governmental agencies to concentrate effort in training women to become competent in such empirical areas. These above catalyzed higher formalization, improved agency over business decisions, increases in capital investment, and ownership of a business as entrepreneurship development indicators. Above all, the study revealed that these indicators under study helped the women to grow economically in the areas of contributing to personal and children school fare, reduction in financial dependence, increase in affordable health care fee, contributions to the family food budget, and increase in revenue or income.
We equally failed to reject the hypothesis one and two which states that entrepreneurship development does not mediate business performance indicators for economic growth, and entrepreneurship development does not mediate psychological indicators for economic growth. Lastly, we rejected the null hypothesis three (entrepreneurship development does not mediate knowledge of business practice indicators for economic growth) which was established at 1% level of significance. Thus, the study empirically suggests that entrepreneurship development mediates the knowledge of business practice indicator to economic growth.

Funding: This work received no funding from any party but solely sponsored by the researchers.
Availability of data and materials: The data for the study will only be made available on request.

Authors contributions:
CAO coded the SurveyCTO data collection tool, as well as analyzed the data. YI designed and developed the research instrument. LI reviewed and edited the work at each stage. ICE and GII conceptualized the study from introduction to analytical framework. IPO interpreted the result of the analysis.

Declaration of conflicting interests:
We declare that there was no conflicting interest.