19
able to deter challengers through a strong and well-equipped security force (Fearon and Laitin
2003). GDP data are drawn from Gleditsch’s (2002) expanded GDP data, which has the best
temporal and geographic coverage; this measure is lagged to account for possible reverse-
causation. Secondly, a control for total population size (logged) is included. It may be the case
that countries with larger populations are more difficult to govern because state security forces
are stretched thin when they must monitor dissent among a large population. Because conflict
may affect population size (either through deaths or emigration) a once-lagged value is included.
Third, several studies have demonstrated that regime type is an important predictor of
civil conflict. It is argued that the most democratic and the most autocratic countries are least
susceptible to a violent challenge. Democracies encourage non-violent means of dissent while
the most authoritarian regimes can effectively deter any opposition. It is mixed regimes, or
“anocracies”, which are not fully-democratic but not extremely repressive that are most likely to
experience conflict (Hegre et al. 2001; Muller and Weede 1990). Therefore, the expectation is
that there should be a parabolic, inverted-U shaped, relationship between continuous measures of
democracy and conflict. Regime data for this study comes from the combined democracy-
autocracy score from the Polity IV data project (Marshall and Jaggers 2002). This widely-used
data is a 21-point scale ranging from most autocratic (-10) to the most democratic (+10)
regimes.
8
Because conflict is likely to affect regime type, I use once-lagged rather than
contemporaneous values for the Polity score. To test for the non-linear effect of democracy, I
include a squared Polity term as well.
Finally, it is argued that the ethnic composition of the country may be an important
predictor of conflict. However, there is little agreement as to what measure of ethnicity is most
8
Countries with special indeterminate codes ( -88, -77, -66) are assigned a value of zero, according to the standard
practice in the literature and the recommendation of the Polity project.
Convention