18
Analysis of the Second Hypothesis
In order to study regime change, the dependent variable in these regressions is a
binary variable that takes a value of “1” if there is a change from one year to the next in
the regime coding by Cheibub and Gandhi (2004). It takes a value of “0” if there is no
change.
I use a statistical model identical to the one underlying the tax revenue regressions
above, both so the results are comparable and also for underlying theoretical reasons. As
discussed above, an error correction model enables the study of both long-term and short-
term relationships between variables. Many theories of regime change imply that certain
independent variables could have both long-term and short-term effects. For example,
there are hypothesized effects on regime change of both the level of income in a country
(Lipset 1959) and economic growth and decline (Remmer 1991). For this reason, many
scholars (e.g. Przeworski, et al. 2000) include both level and growth of GDP per capita in
their regressions. However, few scholars do this with other variables they include in the
regressions.
A close reading of the resource curse literature, however, implies that a similar
approach should be taken to oil variables. It is often unclear whether scholars are talking
about the effects of levels or changes in oil revenue, but one can certainly generate
hypotheses about both based on the literature. For example, I read the following
hypothesis as one about the level of oil: “resource wealth retards democratization by
enabling governments to boost their funding for internal security” (Ross 2001, 328). And
I read this next one as being a hypothesis about the change in oil revenues: “during [oil]
booms politicians are likely to flood the domestic economy with revenues, spending
Convention