The impact of entrepreneurship education on university students’ entrepreneurial skills: a family embeddedness perspective

A full description of the GUESSS project is available at the website www.​guesssurvey.​org. Several works based on the GUESSS project have already been published in entrepreneurship journals: see, for example, Bergmann et al. (2016); Minola, Donina, and Meoli (2016); and Sieger and Minola (2017).

The inclusion of the lagged values of the dependent variables raises concerns related to the potential autocorrelation in the error terms. The Cumby and Huizinga (1992) statistic is applied to test for autocorrelation of order 1 in the residuals under the null hypothesis of no autocorrelation. The statistic rejects the null hypothesis of entrepreneurial skills being serially uncorrelated (χ² = 169.062, p < 0.001) and shows that serial correlation between the dependent variable and its lagged value is correctly specified at degree 1.

VIF values above 5 were found only for the dummy variables bachelor and master because together they cover 96% of the sample and are thus highly correlated (− 0.90). By dropping the variable bachelor and keeping undergraduate students as reference category, VIFs remained below 2 with results substantially unchanged.

In support to this idea, the model was also run with robust standard errors, not clustered by university. In this model, the coefficient of EE was positive and statistically significant (β = 0.203, p < 0.10). Moreover, we run a model without university fixed effects, and it was found that the coefficients of EE in elective (β = 0.262, p < 0.05) and compulsory (β = 0.224, p < 0.10) courses were both positive and statistically significant. To gain further insights into the different effects produced by EE in elective and in compulsory EE, we also separately focused on compulsory EE vs. no EE and on elective EE vs. no EE: in the first case, entrepreneurial skills were regressed on EE in compulsory courses, leaving students with no EE as baseline and excluding from the sample students who received EE in elective courses; in the latter case, entrepreneurial skills were regressed on EE in elective courses, leaving students with no EE as baseline and excluding from the sample students who received EE in compulsory courses. In both cases, the university fixed effects were included. In these specifications, the coefficient of EE in elective (β = 0.358, p < 0.05) remains positive and significant, while the coefficient of compulsory EE remains positive albeit not significant.

Applying the Oster (2017) procedure (cf. Lyons and Zhang 2018), support is found to the need to control for potential endogeneity due to unobservables (correlated both to selection into EE and to outcomes of EE). Additionally, both the Wu-Hausman test and Durbin-Wu-Hausman statistics reject the null hypothesis that the EE indicator is not exogenous.

To measure variables at the local level that might affect students’ entrepreneurial learning, periods of multiple years have already been used in empirical models aimed at quantitatively assessing the impact of EE (cf. Hahn et al. 2017).

As described in Section 3.1, the complete longitudinal GUESSS 2013–2016 dataset records the answers from 1383 students from 21 countries, but the present analysis includes only questionnaires of those students who in the 2013 survey reported not having attended EE before. The sample also excludes students who attended EE in both elective and compulsory EE, students from countries whose number of complete responses was below 5, and respondents for whom it was not possible to build the variables of interest.