The internationalization of small and medium-sized enterprises: the effect of family management, human capital and foreign ownership

For a review of the literature on entrepreneurial orientation and risk-taking in family businesses see Naldi et al. (2007).

See Piva et al. (2005) for an empirical analysis on the relationship between organizational/technological change and skills in Italy.

Made-in-Italy sectors correspond to the following NACE codes: 15, 17, 18, 20, 26.3, 26.7, 28, 29, 31.5, 33.4, 35.12, 35.4, 36.1, 36.2, 36.6.

1,311 firms completed this question: the variable assumes value 0 for the 266 non-exporting firms; 1 for 224 firms; 2 for 255; 3 for 203; 4 for 118; 5 for 109; 6 for 62; 7 for 32; 8 for 26 and 9 for just 16.

Some missing values justify the decision to use a dummy variable instead of a continuous one.

Applied econometrics literature considers correlation coefficients between regressors of less than 0.5 (in absolute value) as not being large enough to affect the estimates (see Belsely et al. 1980).

From a methodological point of view, the nature of this variable, which is a percentage going from 0 to 1, has raised critical issues regarding the estimation procedure (Wagner 2001). Studies on determinants of export performance generally use ordinary least squares (OLS), either a two-step or one-step model as estimation procedures. In a one-step approach all the observations (including the non-exporters) are used to estimate one equation modelling the export/sales ratio, while in a two-step approach two equations are estimated: one equation, based on the whole sample, models the decision to export or not; another equation, which is estimated with data only for exporters, models the share of exports over total sales. Following a two-step approach, the Heckman technique, which corrects the bias from nonrandom samples, has been implemented in our analysis to control for sample selection of exporting firms. The inverse Mills ratio from the probability of exporting step does not turn out to be significant even considering alternative models. This suggests that sample selectivity is not an issue in this sample. Another approach adopts the one-step model where export behaviour is analyzed using both exporting and non-exporting firms and is estimated using the Tobit procedure. As a robustness check we run both OLS and Tobit regressions using all the observations.