Key factors underlying innovation in the context of female entrepreneurship: Opportunity perceptions, digitalisation, and economic policy

Structural equation modelling (SEM) is a statistical technique that allows researchers to theoretically establish a model containing relevant constructs (Guenther et al. 2023). SEM involves a model that represents the expected causal connections among various aspects of certain phenomena. SEM can facilitate the construction of models that relate latent variables to multiple observed indicators (Bollen et al. 2022). This approach subsequently involves estimating and evaluating the hypothesised relationships among the constructs included in the structural model. Since such constructs are not directly observable, they are assumed to be determined and estimated via a set of observable indicators on the basis of the measurement model (Hair et al. 2019; Davvetas et al. 2020). Measurement theory determines the choice of indicators and the directionality of the relationships between the indicators and the constructs included in the measurement model from a conceptual perspective. When the model is estimated, this approach generates proxies for the constructs on the basis of the available data on the basis of the observed indicators included in the measurement model as well as the specific mathematical operations of the method (Rigdon 2016; Sarstedt et al. 2016; Guenther et al. 2023).

The design of this research involves a sequential approach in which secondary data are used to analyse the relationships between women’s perceptions of opportunities and the innovation and internationalisation of women-led businesses. The structural relationships between latent and observed variables were modelled via the partial least squares method (PLS-SEM) with the assistance of SmartPLS version 4 software. The PLS-SEM method is very attractive to many researchers because it can facilitate the estimation of complex models that include many constructs, indicator variables, and structural paths without the imposition of distributional assumptions on the data. More importantly, PLS-SEM is a causal‒predictive approach to SEM that emphasises prediction in the estimation of statistical models, whose structures are designed to provide causal explanations (Wold 1982; Sarstedt et al. 2022).

The methodology employed in this study involves the use of PLS-SEM to estimate the relationships proposed in the theoretical model. PLS-SEM is a multivariate statistical technique that can be used to test complex relationships between latent variables and their indicators (Hair et al. 2021). The latent variables (see Table 1) included in the empirical analysis are theoretical concepts that are not directly observable but can be inferred from other observable variables or indicators. This approach is particularly useful in research on abstract concepts such as digitalisation, opportunity perceptions, and entrepreneurial skills as well as in efforts to estimate the reliability and validity of the indicators used in this context. In addition, the SEM approach offers several advantages, such as facilitating (i) the concurrent, systematic, and complete consideration of the relationships among multiple dependent and independent constructs; (ii) the combination of formative and reflexive variables; (iii) the calculation of measurement errors; and (iv) the refutation of a priori theory via relevant hypotheses and data (Hair et al. 2019; Henseler et al. 2009; Wong 2013). In the empirical analysis, the proposed model is estimated on the basis of the partial least squares (PLS) approach.

One of the main advantages of this methodology, as just explained, is that PLS-SEM offers solutions on the basis of small sample sizes in situations in which the relevant models include many constructs and a large number of items (Fornell and Bookstein 1982; Willaby et al. 2015). Technically, the PLS-SEM algorithm facilitates this task by calculating the relationships pertaining to the measurement and structural models separately rather than simultaneously. Specifically, the algorithm is used to calculate partial regression relationships in the measurement and structural models via ordinary least squares regressions separately (Sarstedt et al. 2016; Hair et al. 2019; Cepeda et al. 2024).

Therefore, in this empirical study, PLS-SEM was viewed as an appropriate approach since our analysis focuses on secondary data obtained from various international databases; these data are used to explore concepts that can be measured on the basis of several indicators. Consequently, Table 1 presents each of the indicators and the constructs to which they are assigned.

As indicated by the theoretical analysis, innovation and internationalisation are interconnected. Namely, internationalisation enhances innovation capacity by providing access to more and better resources, ideas, and know-how. Furthermore, internationalisation can facilitate the exploitation and appropriation associated with innovation, thereby reducing risks (namely, by mitigating the fluctuations associated with local or regional market cycles), generating economies of scale, facilitating responses to the desires and demands of foreign customers, and facilitating the exploitation of more markets (Kafouros et al. 2008; Nissan et al. 2012). Therefore, to measure the latent variable “women’s innovation”, two items drawn from reports issued by the Global Entrepreneurship Monitor (GEM) that measure the innovation and internationalisation of female entrepreneurship are used in this context (see Table 1).

Entrepreneurs try to introduce changes or differences with respect to their competitors with the aim of taking advantage of new, unexploited business opportunities and thus obtaining competitive advantages (Audretsch and Keilbach 2004; Petrakis et al. 2020); therefore, as the focus of the empirical analysis is on female entrepreneurship, only data concerning the percentage of women entrepreneurs are selected for this research. The term opportunity refers to both the existence of available market opportunities as well as people’s perceptions thereof. To investigate this factor, in the empirical analysis, the latent variable for women’s opportunity perceptions is introduced, which is measured by the indicator (OPW), and used. Since the data used in this study pertain to the pandemic and postpandemic periods, it is important to measure the business opportunities that emerged as a result of this health crisis. For this purpose, the indicator (POPW) is used.

Several items drawn from the NES questionnaire (GEM 2024) have been included in the empirical analysis. This questionnaire is used to collect experts’ views regarding how the dynamics of entrepreneurship can be linked to conditions that either enhance or hinder the creation of new firms. These environmental conditions directly influence the existence of entrepreneurial opportunities and shape entrepreneurs’ innovation. The aim of the empirical analysis is to identify the effects that some economic policy updates have had in the context of efforts to promote entrepreneurial activity.

Therefore, the following three indicators are used to form the latent variable “economic policy”: (1) Governmental programmes (GP) refer to the presence and quality of programmes that provide direct assistance to SMEs at all levels of government (i.e. national, regional, and municipal). (2) Governmental support and policies (GSP) refer to the public policies that support entrepreneurship. This item includes two components: entrepreneurship as an important economic issue as well as taxes or regulations that are either size-neutral or encourage new SMEs. (3) Taxes and bureaucracy (TXB) refer to the level of support that public policies provide for entrepreneurship, specifically with respect to whether taxes or regulations are size-neutral and whether they encourage the creation of new businesses and SMEs (GEM 2024).

The COVID-19 pandemic significantly affected entrepreneurs. The pandemic accelerated the adoption of digital technologies by businesses, especially new entrepreneurs. Businesses rapidly digitalised to survive, including by using e-commerce platforms, digital marketing, and online management tools. The health crisis generated new market demands, which promoted innovation and adaptation in terms of business models (Galindo-Martín et al 2023a,2024). Improved digital infrastructure was implemented to offer access to advanced technologies, such as artificial intelligence and data analytics, thereby enhancing operations and market strategies. Consumer behaviour shifted towards online shopping and digital services, thereby increasing the demand for innovative technological solutions (Nambisan et al. 2019; Si et al. 2023). Therefore, our analysis takes into account the influence of the adoption of new technologies that are used by both new entrepreneurs (DTTEA) and established entrepreneurs (DTEBO) to sell products or services.

To measure the digital environment, two dimensions of the composite Digital Economy and Society Index (DESI) are selected: human capital in terms of digital skills (HCDIG) and the digital technology integration component (IDTDIG).

Human capital in terms of digital skills (HCDIG) is divided into two main categories. First, basic digital skills refer to the ability to use digital technologies effectively to perform everyday tasks, including using office software, browsing the internet, and engaging in online communication. Second, ICT specialists, who are professionals who have advanced skills with regard to information and communication technologies, are capable of developing, implementing, and managing complex technological systems.

The digital technology integration component (IDTDIG) measures the extent to which digital technologies are integrated into businesses and society as a whole. This component evaluates aspects such as internet use, the adoption of advanced technologies, and the digitalisation of processes and services.

Finally, the latent variable “credit availability” is composed of two indicators. (1) Financing for entrepreneurs (FE) measures the ability of small and medium enterprises to access financial resources, equity, and debt, including grants and subsidies, and informal investors (IFINV) capture the proportion of adults between the ages of 18 and 64 years who have invested in another person’s new business within the past three years.

The sample includes 13 European countries (i.e. Croatia, Cyprus, France, Germany, Greece, Latvia, the Netherlands, Norway, Poland, the Slovak Republic, Slovenia, Spain, and Sweden) over the period between 2020 and 2022. The choice of this sample is based on two characteristics. First, all of these countries toile within Europe, and second, this approach ensure the availability of data in light of the indicators chosen to perform the analysis. However, the small sample is suitable for partial least squares estimation (Hair et al. 2021).

A graphical representation of the model under consideration is presented in Fig. 1. This model follows the theoretical framework outlined in the previous section. In SEM, latent variables are represented by circles, and observable variables (indicators) are represented by squares (Hair et al. 2021). The arrows between the circles represent the hypothesised relationships among the latent variables. In addition, the model is reflexive (see Fig. 1), as the chosen indicators represent the effects (or manifestations) of an underlying construct (Nunnally and Bernstein 1994), and the indicators are highly interchangeable and correlated; accordingly, they are reflexive (Diamantopoulos et al. 2008).

Bild vergrößern

The measurement model includes values that pertain to an item’s individual liability, internal consistency, and discriminant validity. The simple relationship between each item and the corresponding construct is measured in terms of Cronbach’s alpha coefficient. Composite reliability is used to indicate the reliability of the construct. Values above 0.7 indicated that the construct in question is reliable (Barclay et al. 1995; Nunnally and Bernstein 1994). Convergent validity (AVE) reflects the variance extracted from the indicators, including the common variance that is absorbed by the latent variable. An AVE value of at least 0.5 is accepted as a reliable measure of goodness of fit (Fornell 1982; Fornell and Larcker 1981).

On the other hand, the R² coefficient represents the combined effects of exogenous latent variables on the endogenous latent variable. Therefore, this coefficient represents the amount of variance in the endogenous construct that is explained by all the exogenous constructs to which that construct is linked. This coefficient is calculated in terms of the squared correlation between the actual and predictive values of a specific endogenous construct (Sarstedt et al. 2023). R² values range from 0 to 1, and a value of 0.2 is considered to be adequate for exploratory studies (Hair et al. 2021; Henseler and Sarstedt 2013).

Table 2 indicates that the reliability and validity of the models are adequate because the AVE values of all the latent variables are greater than 0.5. Moreover, the composite reliability values are greater than 0.7 in all cases. The R² coefficient of the endogenous variable women’s innovation exhibits a value higher than 0.369 (see Fig. 1).

Since PLS modelling does not offer significant measures, it is necessary to use nonparametric resampling techniques to validate the stability of the estimates thus obtained. Therefore, the bootstrapping technique is necessary to analyse the significance of the relationships among relevant variables. Figure 1 and Table 2 present the level of significance.

The cross-loadings measure the extent to which each indicator is influenced by more than one latent variable. The cross-loadings measure the correlations of each indicator with the other model (Henseler and Sarstedt 2013; Sarstedt et al. 2016). In this model, the latent variables are well designed because each indicator exhibits the highest cross-loading value on the construct to which it is assigned (see Table 3). Additionally, all indicators are significant (p ≤ 1%) at the 95% confidence level according to the bootstrapping procedure, with the exceptions of the indicators for OPW and WCP, which are significant (p ≤ 5%).

The Fornell–Larcker criterion is used to assess the discriminant validity of the indicators. This criterion ensures that each construct is distinct and unique from the others included in the model. Thus, the variance extracted value pertaining to each construct should be higher than the shared variance between any pair of constructs (Hair et al. 2021). This criterion indicates that the construct shares more variance with the indicators with which it is associated than it does with any other construct included in the model. Hence, in this case, the previous criterion would be met in the estimated model (see Table 4).

The results of the model estimation indicate that the latent variable “women’s opportunity perceptions” has positive effects on innovation and internationalisation among women entrepreneurs, with a path coefficient of 0.594, which is significant at p ≤ 1%. This finding allows us to answer Q1 and Q2 affirmatively. These results are in line with the claims of Ndikubwimana et al. (2020) and Van Stel et al. (2023). Additionally, with respect to the weights that form the latent variable “women’s opportunity perceptions”, the indicator that captures the entrepreneurial opportunities that emerged following the pandemic (POW) exhibits the highest weight, with a value of 0.901, which is significant at p ≤ 1%.

The existence of infrastructures pertaining to new digital technologies and human capital with training in the use of these new technologies promotes the use of new technologies to sell more products and services, with a coefficient of 0.268. In turn, this more intensive use of new technologies stimulates innovation and internationalisation among women entrepreneurs, with a path coefficient of 0.072, thus allowing us to confirm the existence of this positive correlation, albeit at low levels of significance.

On the other hand, economic policies that aim to promote entrepreneurial activity have a positive impact on women’s opportunity perceptions, with a path coefficient of 0.128. Economic policy measures, such as grants and subsidies, can provide an initial financial boost that can allow female entrepreneurs (Nakku et al. 2020) to implement their ideas and expand their businesses. In addition, government programmes can offer business training, advice, and access to networks; furthermore, they can simplify the regulatory environment (Audretsch et al. 2007; Castaño-Martínez 2020). By providing this type of support, governments can help female entrepreneurs overcome some of the obstacles that they face. Therefore, the results obtained in this context can answer Q4 affirmatively, albeit at low levels of significance.

Finally, a positive relationship is observed between access to funding and the latent variables “women opportunity perceptions”, with a path coefficient of 0.423 (significant at p ≤ 5%), and “innovation by women entrepreneurs”, with a path coefficient of 0.129, thus allowing us to answer question Q7 affirmatively. These findings allow us to confirm that expansionary monetary policies helped stimulate entrepreneurial activity among women and promoted their innovation and internationalisation. These results indicate that the availability of credit is essential during all phases of entrepreneurial activity, especially during the early stages of women’s entrepreneurial projects. As indicated in Fig. 1, the path coefficient between the latent variables “access to funding” and “women’s opportunity motivations” is higher. Similarly, the weight of informal investment is 0.864 and significant at p ≤ 1%; therefore, this funding is crucial because it provides vital access to capital, offers flexible conditions, and reduces risks by diversifying sources of funding. This factor is especially important during the early stages of entrepreneurship, as it can facilitate business growth and innovation. Similarly, the PLS estimation facilitates the calculation of specific indirect effects, i.e. access to funding —> women’s opportunity perceptions —> innovation by women entrepreneurs, with a coefficient of 0.251, which is significant at p ≤ 10%.