Controls

\label{controls}
In order to isolate the effect of the maritime campaign, I applied eleven control variables common to insurgency research.
Higher gross domestic product per capita (GDPPC) provide insurgent groups greater access to financial resources to fund their insurgencies. Fearon and Latin (2003) found an additional $1000 in GDPPC is associated with a 35% reduction in the onset of civil war. This variable is taken from the World Bank GDPPC data closest to the conflict’s initiation. I use logged GDPPC because the minimum and maximum score differ by three orders of magnitude.\cite{fearon_ethnicity_2003}
Insurgencies in countries with greater population recruit from a larger pool. A large population also makes it more challenging for the incumbent regime to isolate the insurgency. This variable comes from the World Bank’s population closest to the conflict’s beginning. I use logged population because the minimum and maximum scores differ by five orders of magnitude.\cite{_world_2016}
Countries with mountainous terrain are more likely to be weak states with insurgencies. This variable is the national variance in ruggedness drawn from the work of Andrew Shaver et al.\cite{shaver_andrew_david_b._carter_and_tsering_w._shawa._terrain_2016}
Countries with a large number of ethnic groups may have greater internal conflict as those groups compete. This variable is sourced from the replication data from Fearon and Latin (2003).\cite{fearon_ethnicity_2003}
Insurgencies with the goal of territorial seccession are associated with greater success. This variable is organic to the UCDP dataset.
Foreign support provides insurgencies greater resources and escalates the conflict. This variable is based on the research of Elbadawi and Sambanis, who an association between foreign support and increased duration.\cite{elbadawi_external_2000}\cite{sambanis_whats_2009} I based this variable on whether there was a second party on the insurgency side in the UCDP dataset.
Democratic nations are more likely to conclude conflicts by incorporating the demands of the insurgent group. This variable is drawn from the Polity project’s polity2 score. Countries with polity2 scores above seven at the beginning of the conflict are considered democracies.\cite{center_for_systemic_peace_and_societal-systems_research_inc_polity_2016}
Anocractic regimes are thought to lack the resources or political strength to overcome insurgency.\cite{fearon_ethnicity_2003}\cite{center_for_systemic_peace_and_societal-systems_research_inc_polity_2016}
Greater defense spending per capita indicates stronger regime capacity to combat insurgency. I developed this variable by taking the natural log of the average defense spending per person. This measure also proxies for naval capacity on the assumption that some portion of greater defense spending will be allocated to naval spending, especially if the country faced a maritime insurgency. Modern data for this variable comes from the Stockholm International Peace Research Institute, with figures before 1970 from the Inter-University Consortium for Political and Social Science Research’s Military Defense Expenditure Data.\cite{sipri_international_2009}\cite{gillespie_military_1984} I use logged average defense spending per person because the minimum and maximum score differ by four orders of magnitude.
Insurgencies cluster in different regions, sharing resources, sanctuary, and logistic networks. This control tackles the spatial autocorrelation of insurgency using a dummy variable of there was at least one insurgency in a neighboring country.
Insurgencies further from centers of power are more likely to persist because they less frequently threaten regime survival. I developed this measure by taking the natural log of the median distance from the national capital to UCDP geo-located insurgency events using geospatial packages for r statistical software. Buhaug et al. found that a similar measure significantly predicted insurgency duration.\cite{buhaug_geography_2009} I use logged distance because the minimum and maximum score differ by three orders of magnitude.