The initial theoretical model of the GEM project includes the hypothesis of the existence of a relation between the index of total early stage entrepreneurial activity and the development of the countries, usually represented by the countries GDP per capita (Reynolds et al., 2000).
In 1999, the first year of the project, as all participating countries (ten: United States, France, Italy, United Kingdom, Denmark, Germany, Japan, Canada, Finland, Israel) were developed ones, the simple relationship between a variable similar to the actual TEA but with a different name (SUBO99 = %total [2/99] new start-ups, owned businesses) and a proxy variable representing the 1999 GDP (GDPGLY99 = GDP change: average all 1998 quarters with data) showed a positive and significant simple regression line:
GDPGLY99 = -0.873 + 12.116(SUBO99)
The estimated linear simple regression coefficients were significant (0,074 and 0,035 ate 90% and at 95% confidence level respectively significations), and the model showed a goodness of fit of R2 = 44.7%
This preliminary results aimed GEM researchers that followed this line of investigation, trying to demonstrate and measure the positive impact of TEA figures on the GDP of the countries.
When underdeveloped countries began to participate in the GEM project, contradictory results appeared. Since then, they have tested different regression models and it has been admitted that the relationship between TEA and GDPPC is not linear and that the impact of TEA in GDPPC can have an indeterminate time gap (María Saíz, 2006). Researchers have seen that the relationship between these variables can be reversed and that the GDPPC can explain the TEA scope to some degree with around a 40% of goodness of fit. It also have been proved that the TEA corresponding to entrepreneurial initiatives created by necessity shows better explanation results than the TEA corresponding to entrepreneurial activities created by opportunity, because the first TEA has more variability among developed and less developed countries.
The most accepted model to the present is an U-shaped curve that is presented each year since 2005, in which the dependent variable is the TEA and the independent variable is the current GDPPC of each edition (Minitti, Acs, Bosma and al. 2005-2007)
The present paper discusses chronologically these models from the year 2000, the first in which the number of countries and their characteristics begin to have enough variability, to the last published edition, of the year 2007. Our research penetrates the classical problems that present the data involved in regression models and contributes to demonstrate that it is possible to improve the initial model if data receive previous treatment or if they are used other econometric techniques (weight estimation, multiple regression with GDPPCs of different years, and others) that solve some problems present in the data with respect to requirements of the regression model in general.