In the vast majority of cases, the explanatory variables are strongly
independent among each other. This is not true in five cases where
medium bilateral correlation coefficients are 0.5-0.6 involving the
correlation between pension and social security expenditure,
pensions and the distribution of human capital, taxes and social
expenditure, and FDI and Gini education. This may cause some
problems of multicollinearity and render the related parameters of
some of these variables non-significant. In more general terms,
however, the small bilateral collinearity among variables suggests
that there is no need to develop a structural multi-equation model –
as it might be suggested by economic theory because of the possible
(but not empirically verified) relations among regressors. Indeed,
one might surmise that the growth rate of GDP/c depends on the
international terms of trade, migrant remittances, or FDI, but the
related region-wide correlation coefficients between these pairs of
variables are only 0.26, 0.03 and -0.06.
3.B. Estimation procedure and regression results
The IDLA database is organized as a tri-dimensional matrix, with
18 countries on one axis, 18 years on the second, and the 12
dependent variables used in the analysis on the third. One may
wonder if the use of a panel of different countries may cause
heterogeneity in the data. Yet, the Breusch-Pagan test for data
poolability refuses, at the zero percent probability level, the null
hypothesis of heterogeneity of country data. As for the choice of
the best estimator, this kind of dataset demands that the procedure
chosen for the estimation of the determinants of income inequality
takes into account that each country is observed over several
periods. Such model takes therefore the following form:
it it uxy i it^
where y is the dependent variable (the Gini coefficient of the
distribution of gross income per capita), x is a vector of explanatory
variables (see above), the subscripts i and t represent respectively
the countries and the years of the panel, ui is the error term for each
country, it is a joint error term for countries and time periods, and
represent the parameters to be estimated. Given the nature of this