since many reforms were introduced January 1, which indicates that policy decisions were
taken the previous year. Furthermore, economic and political conditions, as well as gov-
ernment ideology, usually precede reform measures. Lagging our control variables makes
us reasonably certain that we can make a case for the causal nature of the findings.
Table Two shows the results of our tests of the standard economic models in the
taxation literature on the probability of flat tax adoption, with ideology and diffusion
included.
[TABLE TWO ABOUT HERE]
On their own, few of the economic variables pass the standard cutoffs for statistical
significance, though capital-account openness shows consistently strong and significant
improvements in the likelihood of a country’s adoption of the flat tax in a given year.
Again, including diffusion and politics directly into the mix, however, changes the story
somewhat. First, the model fit improves by including the political variables. Second,
both the brute number of champions of the flat tax and the spatial lags, along with
government ideology with respect to economics, are statistically significant.
The spatial-autoregressive coefficient, ρ, gives the impact of all other spatial units,
weighted by wi,j , on the outcome in country i; if significant, it signals the presence
of diffusion effects in the model, but it cannot be directly interpreted.31 We address
this problem and evaluate the spatial interconnectedness and diffusion in several ways.
31Franzese & Hays (2003, 2006) suggest the use of spatial multipliers to calculate short term and long
term spatial effects on the basis of the estimated OLS regression model. However, the calculation of such
a multiplier does not translate directly to the logit context — functional form in logit is different, non-
linear, and also, while the model with the continuous dependent variable includes the latter both on the
left and right-hand side of the equation (lagged dependent variable), which allows the construction of a
multiplier (see Franzese & Hays 2006, 5-11), in limited dependent variable model the dependent variable
is the probability of an event occurring, while the spatial lag is based on the actual event, which does
not translate into multiplier. At the moment, despite the advances in spatial econometrics in political
science (Franzese & Hays, 2003, 2006) and literature on binary-time-series-cross-section (BTSCS) that
emphasized the ways to address the serial dependence and cross-sectional heterogeneity in the data
(Beck, Katz, Tucker, Jackman, various years), there is no literature on spatial aspects of TSCS data with
a binary dependent variable.
23