the primacy of rational learning over cognitive heuristics; in the case of the latter, all
governments would adopt the reform regardless of their ideological bent. All four ways of
modeling diffusion also have impact: proximity to Western Europe has positive and sta-
tistically significant effects. The number of flat tax countries in a given year (Diffusion)
has the strongest effects, which are mitigated only slightly by including the spatial lags
in the model. On its own, the effects of FDI competition are not statistically different
from zero, but weighing the spatial lag by FDI competition has substantively less strong
but statistically significant effects, suggesting the pull of interdependence among those
neighboring flat-tax countries.
Next we control for flat-tax adoption through modeling the predictions on the litera-
ture on optimal levels of taxation. To correctly specify our model, following Swank and
Steinmo 2002, we control for overall productivity, (GDP per capita, Income) structural
unemployment as a share of GDP (Unemployment), and the demands placed on the pub-
lic sector through measuring both public sector debt (Public Debt) and the percent of
the population aged 65 and older (Percent Elderly). We control for international eco-
nomic pressures by including openness to the world economy, as measured by trade as a
share of GDP (Trade); we expect that greater openness to trade would lead to reduced
or simplified tax systems to boost competitiveness (Garrett, 1998). We include capital-
account openness to further take into account the pressures of mobile capital; a higher
degree of capital mobility should increase the probability of adopting a flat tax. More
industrialized countries will regard income tax collection as less pertinent than corporate
tax collection because of a larger share of the latter in their revenue structure, and they
could be more willing to risk experimenting with income tax policies; we include industry
as a percentage of GDP (Industry) to control for this. All control variables are lagged,
22