Abstract
Entrepreneurial self-efficacy (ESE) seems to play a key role in the entrepreneurial career choice. The main goal of this study is to examine the reliability and validity of the ESE Scale proposed by McGee, Peterson, Mueller, and Sequeira both in Italy and Portugal. In particular, this study aims to strengthen and extend previous evidence of the Italian version of the ESE Scale and to assess its psychometric properties in Portugal. Furthermore, it aims at investigating the differences between two groups of participants both in Italy and Portugal: entrepreneurs and students. Construct, convergent, and discriminant validity of the ESE Scale were assessed through confirmatory factorial analysis and multigroups confirmatory factorial analysis using structural equation modeling. Configural, full metric, and partial scalar invariance were achieved. Moreover, correlational analysis, reliability analysis, and one-way analysis of variance were conducted. The findings support the use of the ESE Scale in Italy and Portugal for research and practical purposes. Limitation and suggestion for further research are also discussed.
Keywords
Recent contribution by Crook, Shook, Morris, and Madden (2010) highlights the imperative need to undertake an approach to entrepreneurship research aimed to discourage less-than-optimal-practices. These authors stressed that findings and implications from extant research are good and improving but are not yet what they could be, mainly due to weak and inaccurate measurement. One of the main issues raised is related to the adaptability and validity of the scales of measure adopted in entrepreneurship research. In particular, researchers in this field should use a rigorous approach to the examination of the measures’ reliability and validity in order to avoid inconsistent or contrasting results due to the weakness of the instruments. The main goal of this study is to examine thoroughly one of the most used variables in entrepreneurship research: entrepreneurial self-efficacy (ESE).
ESE is a construct that specifically measures a person’s belief in his or her ability to successfully launch an entrepreneurial venture (McGee, Peterson, Mueller, & Sequeira, 2009). Although several factors influence the decision to be involved in entrepreneurial activities, ESE plays a key role to move from intention to implementation (Barbosa, Gerhardt, & Kickul, 2007; Boyd & Vozikis, 1994; Zhao, Seibert, & Hills, 2005). Moreover, several studies highlight a strong relationship between ESE and entrepreneurial performance (e.g., Baum & Locke, 2004; Forbes, 2005; Hmieleski & Corbett, 2008). A number of scales measuring ESE have been proposed in the literature (e.g., Chen, Greene, & Crick, 1998; McGee et al., 2009). However, all of them were developed in English language countries and mainly tested with English-speaking samples. Actually, to our knowledge, just few previous studies attempted to adapt an ESE Scale into a different language (e.g., Luthans & Ibrayeva, 2006; Mueller & Dato-on, 2013; Spagnoli, Caetano, & Santos, 2015). Essentially, Luthans and Ibrayeva (2006) adapted the general self-efficacy scale proposed by Sherer et al. (1982) in the Kazakhstan and Kyrgyzstan entrepreneurial context, whereas Mueller and Dato-on (2013) and Spagnoli, Caetano, and Santos (2015) adapted the specific ESE Scale developed by McGee, Peterson, Mueller, and Sequeira (2009), respectively, in Spain and Italy. However, except of Spagnoli et al. (2015), these studies failed in not taking a rigorous adaptation and validation process approach. As entrepreneurship research worldwide has been growing, researchers working with populations in non-English speaking countries require more than the translation and adaptation of an established English language measure. As ESE is such an important and frequently used measure in entrepreneurship, the validation of ESE measures in non-English speaking countries is relevant and contributes to the rigorousness of research. Moreover, in line with the call by Crook et al. (2010), beyond a rigorous research design including specific attention on constructs measurement reliability and validity, a cross-cultural perspective is also strongly recommended in order to strengthen the evidence attained in this field.
Guided by this preoccupation, the present study aimed at the adaptation and validation of the ESE Scale developed by McGee et al. (2009) in Italy and Portugal. These are two European countries that have been facing an increase in their entrepreneurial activities during the last years. The early-stage entrepreneurial activity (TEA) index—the percentage of adults (18–64 years old) in the population who are involved in either nascent or new firms—has been increasing both in Italy and Portugal in the last 3 years. Particularly, TEA in Italy was 3.4% in 2013, 4.4% in 2014, and 4.9% in 2015. In Portugal, TEA was 8.3% in 2013, 10.0% in 2014, and 9.5% in 2015 (Kelley, Singer, & Herrington, 2016). Furthermore, there is a boom of entrepreneurship-related initiatives in both countries, aiming to create entrepreneurial awareness and the pursuit of business opportunities. In such a dynamic environment regarding entrepreneurial activities, the need for evidence-based practices and validated tools and instruments is of utmost importance. In particular, the ESE Scale developed by McGee et al. (2009) is based on the main important stages for developing a venture creation. Therefore, it appears to be very suitable for assessing entrepreneurial personal resources at the beginning and at the end of entrepreneurial training sessions to support an individual’s awareness of their own entrepreneurial competences. Thus, it could be a useful career-counseling tool for promoting and supporting entrepreneurial development.
In sum, this study intends to contribute to the measurement of ESE in three ways. First, it strengthens and extends the results of the first adaptation of the ESE Scale into the Italian context. Second, it reports the first examination of the ESE Scale in Portugal. Third, it allows a simultaneous analysis of the psychometric properties of the ESE Scale in these two European countries. The role of ESE in entrepreneurial intentions and performance and an overview of the existing ESE measures are discussed below.
ESE as a Key Antecedent of Entrepreneurial Intention and Performance
The process framework of entrepreneurship identifies that the individual entrepreneur is a main element that contributes to the process (Kuratko, Morris, & Schindehutte, 2015) and is characterized by specific stages involving purposive and intentional actions (e.g., Bird, 1988; Chen et al., 1998; Krueger, 2000). There are several individual factors that account for the personal choice of conceptualization and implementation of a new venture process. Individual factors such as a need for achievement, internal locus of control, and risk-taking propensity (Brockhaus & Horowitz, 1986); personality traits such as the Big Five traits (Brandstätter, 2011; Zhao & Seibert, 2006); cognitive mechanisms such as counterfactual thinking (Baron, 2000), overconfidence (Forbes, 2005; Simon, Houghton, & Aquino, 2000), alertness (Gaglio & Katz, 2001), and optimism (Hmieleski & Baron, 2009); emotional intelligence (Cross & Travaglione, 2003); achievement motivation (Collins, Hanges, & Locke, 2004); generalized self-efficacy, innovativeness, stress tolerance, need for autonomy, and proactive personality (Rauch & Frese, 2007); positive affect (Baron, 2008; Baron, Hmieleski, & Henry, 2012); and entrepreneurial potential (Santos, Caetano, & Curral, 2014) among others were found to be related with the entrepreneurial behavior.
However, Krueger (2000) showed that the direct effect of these antecedents of different types of entrepreneurial activity is less significant, where intentionality serves as an important mediating variable between the act of starting a business venture and the antecedents. In other words, and in accordance with the Ajzen’s (1987, 1991) theory of predicted behavior, intentions predict behavior, while, in turn, certain specific attitudes predict intention (Kim & Hunter, 1993). This theoretical framework identifies three attitudinal antecedents of intentions. Two of them refer to the perceived desirability to perform a specific behavior: the personal attitudes toward outcomes of the behavior and the perceived social norms. The third, perceived behavioral control, refers to the behavioral control that reflects the perceived feasibility of performing the behavior and is, thus, related to the perception of situational competence (self-efficacy). Bandura (1986) states that self-efficacy is related with beginning and pursuing behavior under uncertainty to setting higher goals and reducing threat rigidity and learned helplessness.
Thus, an individual’s belief in his or her capacity to pursue a particular goal has been identified as crucial to several activities (Bandura, 1997) and entrepreneurial activity is no exception. Self-efficacy is important for entrepreneurs because they must be confident in their abilities to perform different and often unanticipated tasks in uncertain situations (Baum & Locke, 2004). Individuals characterized by high self-efficacy are likely to persist when problems arose and actively search out challenges and, by extension, challenging opportunities (Bandura, 1997). Self-efficacy has been identified as a key antecedent of entrepreneurial intentions (e.g., Boyd & Vozikis 1994; Carr & Sequeira, 2007; Mauer, Eckerle, & Brettel, 2013; Zhao et al., 2005), business venture launch and success (Chen et al., 1998), and business performance (Forbes, 2005; Hmieleski & Baron, 2008; Hmieleski & Corbett, 2008). In fact, among all the individual factors that might account for the entrepreneurship process, ESE is one of the most relevant and consistent variables in this research field. Accordingly, ESE is also one of the most used variables in the entrepreneurship research, and due to its relevance, different measurement attempts have been developed. Next section will illustrate a brief overview of the ESE measures.
ESE Measures
ESE has been defined as a multidimensional construct either for research or in practice, and thus it has been calling the attention of entrepreneurship researchers and career practitioners. Specifically, the multidimensional nature of ESE has important implications in its measurement. The multidimensionality of ESE has been defined in different and varied dimensions in accordance with the purposes of the behavior. There are three main ESE multidimensional measurement options developed based on three studies: Chen, Greene, and Crick (1998); DeNoble, Jung, and Ehrlich (1999); and McGee et al. (2009). Chen et al. (1998) defined the ESE multidimensionality according to the entrepreneurial roles and tasks that entrepreneurs have to perform: marketing, innovation, management, risk-taking, and financial control. Chen et al.’s (1998) ESE measurement was one of the most used scales in entrepreneurship research for many years. However, the ESE measurement of Chen et al. (1998) showed poor predictive validity and low discriminant power within dimensions (Fitzsimmons & Douglas, 2011). Consequently, as an attempt to find a new measurement option and overcome these weaknesses, DeNoble et al. (1999) developed a measure of ESE including six dimensions similarly anchored on the key tasks and roles performed by entrepreneurs: developing new product and market opportunities; building an innovative environment; initiating investor relationships; defining core purpose; coping with unexpected challenges; and developing critical human resources. DeNoble et al. (1999) measure was widely used on several entrepreneurship studies (e.g., Barbosa et al., 2007; Morris, Schindehutte, Richardson, & Allen, 2006).
More recently, and taking into account the four phases of the venture creation process, McGee et al. (2009) developed a multidimensional measure of ESE including the main stages of the entrepreneurial journey: searching, planning, marshaling, and implementing. McGee et al. (2009) measurement was especially interesting because it included both the construct conceptualization and its measurement. Furthermore, the four stages of the entrepreneurial activity that define the multidimensionality of McGee et al.’s (2009) ESE construct are based in the robust theory of the entrepreneurial process of Stevenson, Roberts, and Grousbeck (1985), which has been systematically supported in research and practice.
Building on these three central ESE measures (Chen et al., 1998; DeNoble, Jung, & Ehrlich, 1999; McGee et al., 2009), consequent attempts to develop ESE multidimensional measurements have also been carried out, working mainly on minor refinements, such as in wording of particular items (e.g., Karlsson & Moberg, 2013). Other further suggestions addressed theoretical issues, such as Drnovšek, Wincent, and Cardon (2010) who proposed a different conceptualization of ESE, using three dimensions: target of ESE (business start-up or business growth activities), goal benefits of ESE beliefs (task or outcome goal beliefs), and control beliefs of ESE beliefs (positive or negative control beliefs).
Despite the relevance and the highly used ESE measurement, there are relative few works that build on the existing scales and focus on the measurement and validation. Working further on the existing measurements, contributing to provide psychometric validation and demonstrating the reliability of entrepreneurship-related scales, is an important goal to the entrepreneurship scholars’ community. Guided by this preoccupation, we studied further the McGee et al.’s (2009) ESE measurement, in two different countries (Portugal and Italy), analyzing its psychometric properties and its validation.
In particular, this study aims to examine the construct, convergent, and discriminant validity of the ESE Scale (McGee et al., 2009) in the two European countries. A previous study conducted by Spagnoli et al. (2015) contributed to the examination of construct validity of the ESE Scale in Italy. However, convergent and discriminant validity were not explored. In the present study, construct validity will be examined by establishing the measurement invariance of the ESE Scale and by comparing the ESE scores in different groups of participants. In particular, two different groups of participants were involved: entrepreneurs and nonentrepreneurs. Analyzing entrepreneurs and nonentrepreneurs contributes to the face validity of ESE. More specifically, entrepreneurs are expected to have high ESE when compared to nonentrepreneurs, as they have already experience on entrepreneurial activities. Using entrepreneurs and nonentrepreneurs to analyze ESE allows to evidence that the measure is covering the concept that it is proposed to measure. As entrepreneurs have already developed their perception of their ESE, they are expected to have higher ESE’s scores than other nonentrepreneurs participants. Based on this reasoning, we put forward the following hypothesis:
Convergent validity will be explored by testing the relationship between ESE and a measure of attitude toward the enterprise (Athayde, 2009). The attitude toward the enterprise scale was developed in order to assess the characteristics more associated with entrepreneurship. It includes four main attitudes: leadership, creativity, achievement, and personal control. As convergent validity refers to measures that should be theoretically related to the construct under analysis, we decided to use the attitude toward the enterprise because previous results showed that the enterprise potential is associated with entrepreneurial performance (Athayde, 2009). Essentially, it is assumed that individuals who are higher in their attitude toward the enterprise would be more prone to engage on entrepreneurial activities and then will have higher ESE. Furthermore, previous evidence suggested that ESE and attitude toward enterprise (ATE) should be theoretically and empirically associated (Santos et al., 2014). In line with this reasoning, we expect a positive and statistically significant correlation between attitude toward the enterprise and ESE as an evidence for convergent validity. Thus, we put forward the following hypothesis:
Validity also requires testing that measures that should not be related are really not related one to each other. This is the discriminant validity of measures and tests if the construct is distinct from other constructs that should not be associated with. ESE should not be related to variables that are not also related to entrepreneurial behaviors and attitudes. One of the most consistent variables that are not associated with entrepreneurship is external locus of control (Mueller & Thomas, 2001) as opposed to internal locus of control. Previous research showed that there is no association between external locus of control and entrepreneurial potential measures (Santos et al., 2014). Based on this reasoning, discriminant validity will be assessed through the correlation between the composite measure of the ESE and the external locus of control. Therefore, we predict that there will be a nonstatistically significant relationship between ESE and external locus of control. Accordingly, Hypothesis 3 of this study is:
Thus, in order to test these hypotheses, we developed an empirical study measuring ESE, enterprise potential attitudes, and locus of control with an Italian and a Portuguese sample including both entrepreneurs and nonentrepreneurs.
Method
Participants
A total of 815 individuals from Portugal and Italy participated in this study. Two different groups of participants from both countries were involved, namely, 370 Italian students, 159 Portuguese students, 163 Italian entrepreneurs, and 123 Portuguese entrepreneurs.
The educational level of the Italian students was secondary school (43.6%), bachelor (47.8%), and postgraduate (8.6%). Most of the Italian students were women (65.9%), age ranging from 17 to 60 years old (M = 23.45; SD = 4.50).
The educational level of the Portuguese sample was bachelor (53.5%) and postgraduate (46.5%). The Portuguese sample included mainly men (65.4%), aged between 21 and 30 years old (M = 24.70; SD = 1.93). The Italian entrepreneurs were all involved in small business ventures, such as retail, artisanship, services, and construction. They were mainly men (52.1%), aged between 22 and 67 years old (M = 42.14; SD = 9.69). The Portuguese entrepreneurs were from the technologies and health informational technology (IT), smart cities and industrial tech, enterprise IT, and ocean economy. They were mainly men (81.3%), aged between 38 and 74 years old (M = 54.45; SD = 8.50).
It has to be noted that the inclusion of the entrepreneurs group in the study of ESE is not so frequent in previous research. However, it is important and relevant for one principal reason. Self-efficacy is a psychological construct based on the experiences made in specific fields, in this case, entrepreneurship. Thus, the inclusion of the entrepreneurs group will allow the comparison of the ESE’s scores between entrepreneurs and nonentrepreneurs in order to assess the construct validity of the scale. More details of the groups involved in the study are presented in Table 1.
Groups Demographics.
Measures
The McGee et al.’s (2009) ESE Scale comprises 19 items. The authors reported five dimensions underlying the ESE construct corresponding to the four typical phases of starting a new venture. The number of items in each of the five dimension and item examples are as follows: research and development, 3 items (e.g., “How much confidence do you have in your ability to identify the need for a new product or service?”); planning, 4 items (e.g., “How much confidence do you have in your ability to estimate customer demand for a new product or service?”); marshaling, 3 items (e.g., “How much confidence do you have in your ability to get others to identify and believe in your vision and plans for a new business?”); implementing human resource, 6 items (e.g., “How much confidence do you have in your ability to recruit and hire employees?”); and implementing financial resource, 3 items (e.g., “How much confidence do you have in your ability to read and interpret financial statements?”). Respondents were asked to indicate on a 5-point scale (1 = very little, 5 = a lot) how much confidence they had in their ability to engage in each of the 19 entrepreneurial tasks. The complete Italian and the Portuguese version of the scale are available from the corresponding author.
ATE
The ATE test was developed by Athayde (2009). It includes 18 items measuring four dimensions: leadership (6 items), creativity (4 items), achievement (4 items), and personal control (4 items). After checking the internal consistency of the measure, a composite measure of ATE was computed in order to assess the convergent validity of the ESE Scale.
External locus of control
External locus of control was measured by 3 items of the Levenson scale (1973). One example of the item is “Normally, I achieve what I want by chance.” In this study, the measure of external locus of control was used in order to examine the discriminant validity of the ESE Scale.
Both ATE items and external locus of control items were answered in a 5-point scale, ranging from totally disagree (1) to totally agree (5), regarding the extent to which the respondents agree with the items.
Procedure
To ensure equivalence of meaning for the items between the Italian, the Portuguese, and the English versions of the McGee et al.’s (2009) ESE Scale, the same rigorous translation process was used in Portuguese and Italian. This included forward and backward translation and pilot testing. The translation process began with the translation of the English version into Italian and Portuguese by a bilingual translator in English and Italian and in English and Portuguese. Then, another bilingual translator (a native English speaker) independently translated the ESE Scale back into English. The translators then compared the back translation to assess the item-by-item consistency. Before the application of the ESE Scale, a pilot test was conducted in order to gather feedback on the readability and content validity of the translated instrument. These instruments were applied to 12 individuals, 10 students, and 2 entrepreneurs, and no significant word changes were made.
After this pilot testing, data collection in Italy and Portugal started using the same techniques. The questionnaires were administered both online and by hand delivery and return. The online administration was applied through a dedicated website, which constitutes a data-gathering platform for an international research project. The advertising strategies used to involve the participants on the website were snowball technique, request to specific entrepreneurial associations, and connection of the research website to the university portal. The hand delivery and return administration was conducted through the snowball technique by involving graduated students on work and organizational psychology courses who voluntarily took part in the data collecting phase of the study. These students were trained by the research team and had to administer a limited number of questionnaires to entrepreneurs. Other entrepreneurs filled the questionnaire online on the website linked to a specific entrepreneurship association. The samples were recruited both during university entrepreneurship seminars and classes and through the university website. All participants were informed of the anonymity and confidentiality of the survey. Ten questionnaires where gender and the entrepreneurial status were missing were annulled.
Data Analysis
According to the measurement invariance literature in order to test the construct validity of the scale, three levels of invariance should be achieved: configural, metric, and scalar invariance (e.g., Byrne, 2004; Brown, 2006; Davidov, 2008). Nevertheless, a precondition for performing the assessment of the measurement invariance consists of testing the fit of the dimensional models. Thus, confirmatory factor analysis (CFA) was performed in order to examine the fit of the dimensional models of McGee et al. (2009) for the overall sample. According to McGee et al. (2009), three models were tested: unidimensional, three dimensions model, and the original McGee et al.’s (2009) five dimensions model. The indices of the model fit considered were the comparative fit index (CFI), the root mean square error of approximation (RMSEA), the normed χ2 (χ2/df), the Akaike information criterion (AIC), and the root mean squared residual (RMR). The CFI assesses the extent to which the tested model is superior to an alternative model in reproducing the observed covariance matrix (Bentler, 1990; McDonald & Marsh, 1990). The CFI index varies from 0 to 1 and a cutoff criterion of CFI > .90 is needed in order to ensure that misspecified models are not accepted (Hooper, Coughlan, & Mullen, 2008). The RMSEA introduces a correction for lack of parsimony since, all other things being equal, more complex models are penalized. A cutoff value close to .06 (Hu & Bentler, 1999) or a stringent upper limit of .08 (Steiger, 2007) seems to be the general consensus among the researchers in this area. The normed χ2, or the χ2 to degrees of freedom ratios (χ2/df), is a further version of the traditional χ2. The advantage of the normed χ2 is that it might be less sensitive to the sample size. Thus, as the whole sample of the present study is quite wide, the option to not use the χ2 in the analysis was adopted. A normed χ2 lower than 5 is a sign of good fit (Schumacker & Lomax, 2004). The AIC is a comparative measure of fit. According to Burnham and Anderson (2004), lower values indicate a better fit and so the model with the lowest AIC is the best fitting model. The RMR is an absolute measure of fit, and a value of zero indicates perfect fit. A value less than .08 is generally considered a good fit (Hu & Bentler, 1999).
Then, the model reporting the best fit was tested separately on each subgroup in the study and simultaneously through the multigroup confirmatory factorial analysis (MCFA) in order to assess the configural invariance of the model. Configural invariance is the first level of invariance to be examined, and it is achieved when the model holds on the different groups included in the analysis (Byrne, 2004). Subsequently, further measurement invariance analyses, such as metric and scalar invariance, can be conducted in order to test, respectively, if the factor loadings and the intercepts are the same in all the groups studied. Thus, the fit of a constrained model including all the factor loadings fixed is compared to the fit of the free-to-vary model in order to test full metric invariance. Following Chen (2007) and Cheung and Rensvold (2002), CFI and RMSEA were also used to test measurement invariance. An increase of RMSEA by .015 and a decrease of CFI by .01 can be used as cutoff points for accepting measurement invariance (Chen, 2007; Cheung & Rensvold, 2002). If the fit difference between the models does not fall into the threshold for accepting full metric invariance, partial metric invariance could still be explored (Byrne, Shavelson, & Muthén, 1989; Millsap & Meredith, 2007), leaving at least two factor loadings fixed in a construct or in a factor when a construct is composed by several factors. Once that at least partial metric invariance has been established, scalar invariance can be examined. Scalar invariance analysis allows understanding whether the scores from different groups have the same origin, that is, whether the intercept across the group is the same. Similarly to metric invariance, in order to achieve at least partial scalar invariance, the intercept noninvariance can be explored by relaxing constraints on the intercepts one by one (Millsap & Meredith, 2007; Byrne et al., 1989). A specific structural equation modeling software (AMOS 16) was used to run CFA, MCFA, and measurement invariance analysis.
Afterward, test of the ESE differences between entrepreneurs and nonentrepreneurs was conducted in order to further support construct validity both for Italy and Portugal. Finally, convergent validity was assessed by testing the correlation of a composite measure of the ESE with a measure of entrepreneurial attitude (Athayde, 2009). Discriminant validity was assessed through the test of correlation between the composite measures of ESE and external locus of control. Descriptive statistics, general data managing, reliability analysis, and correlational analysis were conducted through SPSS 20.
Results
Table 2 presents the intercorrelation among the five ESE dimensions. All the bivariate correlations are positive and statistically significant at p < .001. More specifically, although all the correlation coefficients are broadly <1 showing the absence of complete overlapping between the ESE dimensions, the correlation coefficients for marshaling, planning, and searching are relatively high ranging from .65 to .72. 1 This result is consistent with McGee et al. (2009) and justifies the test of the three-dimension model, where these three dimensions collapsed in one dimension. As Byrne (2004) suggests, the first step for testing measurement invariance is to examine the dimensionality of the scale on each subgroup in the study. However, as McGee et al. (2009) examined the dimensionality of the scale on three different models (five factor, three factor, and unidimensional), preliminarily we assessed the three models on the overall sample in order to compare the dimensionality of the scale with that of McGee et al. (2009) on a robust sample. Thus, following the same procedure of McGee et al. (2009), three models were tested on the overall sample including both the Italian and the Portuguese subsamples: the original five factors proposed by McGee et al. (2009)—Model 1; the three factors model resulting from collapsing searching, planning, and marshaling—Model 2; and the unidimensional model—Model 3. Table 3 shows the results for the CFA of the three models. Although they are not much higher than those of Model 2, the fit indices for Model 1 appear to show the best fit, whereas the fit indices for Model 3 show the weakest fit. In particular, Model 1 reported the lowest AIC, indicating that this may be the best factorial solution. Furthermore, a proper method to test if the difference between Model 1 and Model 2 was statistically significant was conducted by computing the χ2 difference between the two models calculating also its p value. Evidence of the prevalence of Model 1 was found (M1χ2 = 895.640, df = 142; M2χ2 = 1,036.558, df = 149; Δχ2 = 140.917 df = 7, p < .001). Thus, we could conclude that the best factorial solution of the scale was the five-factors model.
Intercorrelations Among the Five ESE Dimensions.
Note. IHR = implementing human resources; IFR = implementing financial resources; M = marshaling; P = planning; ESE = entrepreneurial self-efficacy.
*p < .001.
Fit Indices for the Factorial Solutions of the ESE Scale.
Note. CFI = comparative fit index; RMSEA = root mean square error of approximation; RMR = root mean squared residual; AIC = Akaike information criterion.
aIncluding the three errors of covariations.
Although all the factor loadings were statistically significant and ranged from .41 to .79, few modification indexes indicated the need to freely covariate three pairs of errors in order to achieve a better fit of the model. A close inspection of the items was conducted in order to evaluate the possibility of setting the three pairs of errors free to covariate. Essentially, two couples of items included the words “new product and service,” whereas the other couple of items regarded the “financial issues.” Since the high value of these modification indexes (>.40) and the meanings’ overlapping between the related items, according to Kline (1998), we decided to set the three pairs of errors free to covariate. Thus, the revised five-factors model (Model 1.2) showed a good fit (χ2 /df = 5.258; CFI = .91; RMSEA = .072; RMR = .069). Table 4 shows the factor loading for the revised five-factors model.
Factor Loading for the Revised Five-Factor Model.
Note. IHR = implementing human resources; M = marshaling, S = searching; P = planning.
Then, the five-factors model was tested separately in each of the four groups in the study and simultaneously through MCFA in order to further examining the dimensionality of the model and its configural invariance (Table 3). Results show that in some cases, the CFI values and the RMSEA, although they are very close, they are not above the cutoff point of .90 and .08, respectively. However, according to Marsh, Hau, and Grayson (2005), this goodness of fit criteria may be too restrictive when used with multifactor rating instruments as in the case of ESE. Thus, all in all, the results of the fit indices in the four subgroups might be acceptable. Similarly to the CFA results, the fit indices for the MCFA show adequate fit, except for CFI, which falls slightly under the threshold (CFIMG1 = .88). Additionally, the comparison between the fit indices of the present study and those of McGee et al. (2009) shows that the CFI values in the present study are lower than that of the McGee et al. (2009), which were .96. Conversely, the RMSEA value in this study is rather better of that of the McGee et al. (2009), which were .06. Because the RMSEA is intended to be “one of the most informative fit indices” (Diamantopoulos & Siguaw, 2000, p. 85), it is more appropriate in the confirmative context (Rigdon, 1996), and the cutoff point of .90 might be too restrictive when using multiple factors instruments (Marsh, Hau, & Grayson, 2005), we could affirm that the MCFA analysis reports evidence of acceptable fit of the data to the theoretical model proposed by McGee et al. (2009). Accordingly, configural invariance was achieved.
Table 5 presents the Cronbach’s αs of the five ESE dimensions in the overall sample, in the Portuguese and Italian sample, and, for comparison purposes, in the McGee et al.’s (2009) study. The Cronbach’s αs in the overall and in the two country samples of the present study are slightly lower than those reported in the McGee et al.’s (2009) study. Nevertheless, except for the internal consistency of marshaling in the Portuguese sample, all the values are above the common cutoff point of .70 (Nunnally & Bernstein, 1994). Thus, adequate reliability of the two versions of the ESE scale was attained.
Cronbach’s αs of the Five ESE Dimensions in the Overall Sample in the Italian and in the Portuguese Samples and in the McGee, Peterson, Mueller, and Sequeira’s (2009) Study.
Note. ESE = entrepreneurial self-efficacy.
Then, we conducted the measurement invariance test, in terms of metric and scalar invariance, in order to give further support to the construct validity of the Italian and Portuguese versions of the ESE Scale and, thus, to compare the ESE factors’ scores. Table 6 shows the differences between the two fit indices tested, CFI and RMSEA, both in the free-to-vary model (MInvariance1 configural invariance) and in three constrained models. Specifically, three constrained models were tested: the full metric invariance model, which includes all the factor loadings fixed (MInvariance2); the full scalar invariance model, where all the intercepts are fixed (MInvariance3); the partial scalar invariance, which comprises just two fixed intercepts in each ESE dimension (MInvariance4). The difference between the values of RMSEA and CFI in the free-to-vary model (MInvariance1) and in the full metric invariance model (MInvariance2) did not fall in the range for rejecting the invariance (Chen, 2007; Cheung & Rensvold, 2002). Therefore, full metric invariance was established.
Test for Measurement Invariance of the ESE Scale.
Note. ESE = entrepreneurial self-efficacy; CFI = comparative fit index; RMSEA = root mean square error of approximation.
After establishing metric invariance, the scalar invariance examination was carried out. First, the full scalar invariance model (MInvariance3) was compared to the free-to-vary model (MInvariance1). The difference between the values of RMSEA in the two models did not fall in the range for rejecting full scalar invariance, whereas the difference between the values of CFI fell in the range for rejecting full scalar invariance. Thus, in order to strengthen the scalar invariance results, the partial scalar invariance examination was also conducted. Following Byrne, Shavelson, and Muthén (1989), in this case just two intercepts’ values in each ESE factor were fixed (MInvariance 4—scalar partial invariance). As the difference between the RMSEA’s values continued to be in the range for accepting the invariance, and the difference between the CFI’s values were noticeably reduced, the partial scalar invariance could be established. Hence, the construct validity of the Italian and Portuguese versions of the scale was supported and the comparison of the ESE factors’ scores can now be conducted correctly.
A one-way analysis of variance (ANOVA) was conducted in order to compare the five ESE factors’ scores of the Portuguese students and the Portuguese entrepreneurs (Table 7), and other one-way ANOVA compared the scores of the Italian students and the Italian entrepreneurs (Table 8). The results of the Portuguese entrepreneurs were significantly higher than the Portuguese students in the five factors of the ESE Scale (FIHR = 15.618; FIFR = 4.273; FM = 7.267; FP = 9.887; FS = 53.092). Similarly, the results of the Italian entrepreneurs were significantly higher than the Italian students in the five factors of the ESE Scale (Table 7; FIHR = 8.504; FIFR = 58.712; FM = 12.008; FP = 47.609; FS = 35.183). Effect size, calculated for the composite measure of the ESE scale, was η2 = .07 for the Italian samples and η2 = .08 for the Portuguese samples, indicating that, respectively, 7% and 8% of the variance in ESE is due to the status of entrepreneurs or students. Overall, these results support Hypothesis 1 and the construct validity of the scale.
ANOVA of the Two Portuguese Subgroups on the Five Dimensions of the ESE Scale.
Note. ANOVA = analysis of variance; IHR = implementing human resources; IFR = implementing financial resources; M = marshaling; P = planning; S = searching.
*p < .05. **p < .001.
ANOVA of the Italian Subgroups on the Five Dimensions of the ESE Scale and on the Composite ESE Measure.
Note. IHR = implementing human resources; IFR = implementing financial resources; M = marshaling; P = planning; S = searching; ANOVA = analysis of variance.
**p < .001.
Finally, a correlational analysis was carried out in order to examine the convergent and discriminant validity of the Italian and Portuguese versions of the ESE Scale in the groups of the Italian and the Portuguese students. Specifically, the correlation between a composite measure of the ESE Scale and a measure of the ATE test (Athayde, 2009) was performed to examine the convergent validity, and the correlation between a composite measure of the ESE Scale and external locus of control was conducted to assess the discriminant validity. The results of these analyses showed a statistically significant and positive relationship between the ESE Scale and the ATE Scale both in the Italian (Pearson r = .51) and Portuguese (Pearson r = .46) groups. Thus, convergent validity was achieved, and Hypothesis 2 was supported.
Moreover, both in the Italian (Pearson r = .01) and Portuguese (Pearson r = .03) groups, the relationship between the composite measure of the ESE Scale and the external locus of control was very weak and not statistically significant. Accordingly, discriminant validity of the Italian and Portuguese versions of the scale was demonstrated. This result supported Hypothesis 2.
Discussion
This study contributed to the measurement of ESE, showing the robust psychometric characteristics of the McGee et al.’s (2009) ESE Scale for the Portuguese and the Italian versions. Our results strengthen and extend the validation of the scale in the Italian context. In particular, evidence of the present study supported further the construct validity of the ESE Scale and supported its convergent and discriminant validity in Italy. Configural invariance as well as full metric and partial scale invariance were established both for the Portuguese and Italian versions of the ESE Scale. The results of the CFA and the multigroup confirmatory analysis reflected acceptable fit of the Italian and the Portuguese data to the theoretical model.
Except for the CFI values, which were slightly lower than those of the McGee et al.’s (2009) study, all the fit indices tested for achieving configural invariance showed good fit. In particular, similarly to the previous study in Italy (Spagnoli et al., 2015), the RMSEA values are even better than those of McGee et al. (2009).
For comparison purposes, it is essential to note that the original test of the fit of the scale reported by McGee et al. (2009) comprised two more variables, that is, attitude toward venturing and the status of nascent entrepreneur or not. Thus, as RMSEA value is not sensitive to complexity and would never favor the simpler model, the fact that its values in the present study are better than that of McGee et al. (2009) reinforces the configural invariance test conducted. On the other hand, the CFI values could be improved by estimating some of the few residuals higher than .1 reported by the modification indices analysis. However, we decided to estimate just three couple of very high residuals (>.40) and to leave the others for model parsimony. Moreover, according to Marsh et al. (2005), the goodness of fit criteria of .90 may be too restrictive when used with multifactor rating instruments. Thus, a CFI value slightly under this cutoff in a multifactor rating instrument, such as ESE Scale, could still be accepted. Furthermore, another sensitive point of the evidence is the lowest α (.53) value on the marshaling dimension in the Portuguese version of the scale. However, as Field (2005) noted, questionnaires designed to measure “knowledge” and “intelligence” should have Cronbach’s α coefficients above the customary cutoff value of .70, however instruments designed to measure “attitudes” may have lower αs (<.70) and still have acceptable levels of reliability. Moreover, it is important to note that although coefficient α is arguably the most commonly used statistic for evaluating internal consistency of a scale, when evaluating a multidimensional psychological construct in a confirmatory context, the results could be underestimated (e.g., Rios & Wells, 2014). Notwithstanding, future studies using this scale in Portugal might consider a further revision of these specific items.
Nonetheless, full metric invariance and partial scalar invariance were established allowing to further examine and achieve the construct validity of the Italian and the Portuguese versions. A further examination of the construct validity of the ESE Scale was properly tested through the comparison between the ESE factors’ scores of entrepreneurs and nonentrepreneurs specifically in each country. Accordingly, although the effect size was small, both the Italian and Portuguese entrepreneurs’ ESE scores were significantly higher than those of the nonentrepreneurs for all the five ESE dimensions. Thus, further support to the construct validity of the ESE Scale in the Italian and the Portuguese context was attained.
Furthermore, the convergent and discriminant validity of the Italian and Portuguese versions of the ESE Scale were achieved through the correlational analysis between a composite measure of the ESE Scale and, respectively, a measure of ATE (Athayde, 2009) and a measure of external locus of control.
Research Limitations and Suggestions for Further Research
Despite the rigorousness of this study and the effort to contribute to the reliability and validity of entrepreneurship measures in non-English speaking countries, there are some limitations that need to be considered. First, the samples of the participants are not representative of the Italian and Portuguese students and entrepreneurs population. Future research should be aimed at recruiting participants according to a stratified sample design. Nevertheless, our study was motivated by the need to offer the community valid and reliable instruments to be used in entrepreneurship practice in Portugal and Italy. These are two European countries that have been experiencing an increase in their entrepreneurial activity indexes, and there is a need to develop evidence-based practice.
Second, support for the validation of the Italian and Portuguese version of the scale should also include in the future concurrent and predictive validity. For example, a longitudinal design research could be adopted in order to test the predictability of the ESE Scale on entrepreneurial activity success outcomes. Finally, the entrepreneurs group involved in the study in Italy comprises solely owners of small business ventures. Although, this reflects the real picture of the entrepreneurial phenomenon in a great part of Italy, the results could not be generalized to other ventures of different sizes. Similarly, the entrepreneur’s sample in Portugal was comprised by individuals operating in the high technology industry, which is far from covering all the types of entrepreneurs that exist in Portugal. Finally, results reported a small effect size of the one-way ANOVA conducted for examining the ESE differences between entrepreneurs and students both in Italy and Portugal. This could be due to the effect of some confounds such us some sociodemographic variables that could account for the variability within the groups. Analyzing the effect of some sociodemographic variables, such as, for example, gender, educational level, and age, in the two groups of entrepreneurs and students would be very interesting especially on a stratified sample. For example, previous study of Spagnoli et al. (2015), addressing the gender difference between entrepreneurs and students on all the five ESE factors in Italy, reported evidence of a gender gap solely in the students’ sample, whereas in the group of entrepreneurs, evidence of a difference was found only in the “implementation of financial resource” factor. The main results of Spagnoli et al. (2015) suggested that with the growing numbers of women entrepreneurs, specific attention should be paid to the perception of ESE both before starting (during training) and after creating a new venture (during follow-up). Thus, future studies both in Portugal and Italy might address the effect of sociodemographic variables on ESE using stratified samples.
Conclusion
In sum, the Italian and the Portuguese versions of the ESE Scale can be adopted for research and practical purposes. ESE plays a key role to move from intention to implementation of entrepreneurial activities (Barbosa et al., 2007; Boyd & Vozikis, 1994; Zhao et al., 2005) and to the entrepreneurial performance (e.g., Baum & Locke, 2004; Forbes, 2005; Hmieleski & Corbett, 2008). Building on Karlsson and Moberg (2013) suggestion, we propose that the ESE Scale can be useful for designing individual and group training programs aimed at enhancing ESE both in Italy and Portugal.
Footnotes
Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) received no financial support for the research, authorship, and/or publication of this article.
