Balkans, and the Baltics). Second, we measure the spread of policy ideas by counting
the number of countries that have implemented the flat tax, for the regressions on flat
tax, and calculating the average tax levels for the region, for the robustness checks on
levels of individual taxation. We discuss these diffusion effects in more detail below.
These effects are estimated by the binary time-series cross-section model with temporal
controls (Beck et al., 1998; Beck et. al., 2001; Epstein et al., 2001). The model is thus a
duration-dependent logit (restricted Markov transition) model.
Finally, following Franzese and Hays (2006), we specify a spatio-temporal lag model
that includes weighted spatial lags to test whether policymakers observed and reacted to
tax policy change in neighboring countries. To assess whether countries blindly adopted
the policy or whether they did so on the basis of rationally evaluating the increasing
attractiveness of nearby flat-tax countries, we weigh each neighboring flat country by
FDI inflows. Thus, tax policy change in a country becomes a function of past flat-tax
adoption in a neighboring country, as well as FDI competition. Each spatio-temporal lag
model also includes other measures of diffusion, described above.29
The full spatio-temporal lag model to analyze the diffusion of flat tax policies through-
out Central and Eastern Europe is:
ex _ eX,p(αYi,t-i + ρWYi,t-i + γn Yt-1 + Xi,t-1β + ∈i,t
Pit = τ----:—τττττ----:—----τr^---
1 + exrp(αY i,t-i + ρWYi,t-i + γn Yt-ι + Xi,t-ιβ + e⅛,t
where Yi,t-1 is a one-period temporal lag of the dependent variable and α is the
temporal-autoregressive coefficient; W is an NxN spatial-weighting (standardized in-
29See Gleditsch & Ward, 2006 estimation of democratic diffusion using four different measures of the
latter).
19