Quantitative Analysis

\label{sect:quant}
In this section, I propose to test the hypotheses from Section \ref{sect:theory} using insurgency data from the Uppsala Conflict Data Program (UCDP)\cite{joakim_kreutz_how_2010} since the end of the Cold War and data I collected on insurgent maritime operations. Analyzing only post-Cold War insurgencies limited the number of conflicts studied, but reduced the likelihood of missing maritime conflicts unreported in available media. Since December 25th 1991, there have been 119 conflicts of which 104 were insurgent conflicts. Insurgent conflicts exclude coups (duration less than 30 days), civil wars where the country split into two organized parties, and terrorist groups without true ambition for territorial control. 104 is a low number of cases for multivariate regression, but is consistent with other cross-country studies of civil conflict.
To determine whether maritime insurgencies have greater duration than other insurgencies, I used a multivariate Cox proportional hazards regression with one observation per conflict. This regression indicates which factors are associated with increased or decreased conflict duration. Duration equals the time elapsed between a conflict’s start date and its end date or 1 January 2017.
I coded insurgencies as maritime if I found evidence they engaged in smuggling, piracy, or maritime terrorism (Appendix \ref{coding} contains coding rules). I did not attempt to capture the magnitude of the maritime operations. Developing a metric for insurgent maritime strength is beyond the scope of this investigation. Instead, I used a binary code to record whether or not they engaged in the behavior.
Because maritime insurgencies are located along navigable waterways, there may be factors other than the insurgency’s maritime campaign associated with duration. In my first model, I control for factors common to duration research: gross domestic production per capita (in logs), population, whether the country is a democracy, terrain ruggedness, ethnic fractionalization, whether the insurgency sought secession, and external support. In my subsequent models, I add controls for national defense spending per capita (in logs), whether neighboring countries were in conflict, and the median distance from the capital to insurgency attacks (in logs). Readers can find a detailed discussion of how each variable may effect duration and its source in Appendix \ref{controls}. Additionally, I pooled standard deviation by country to account for national omitted variable bias.

Descriptive Statistics

I consider 104 insurgent conflicts since the end of the Cold War, of which 37 are ongoing. I coded 19 insurgencies as maritime if they conducted maritime smuggling, piracy, or maritime terrorism. I also determined where maritime insurgent groups conducted operations: brown-water riverine, green-water littoral, or blue-water open-ocean. I coded groups as conducting these operations if I found two sources confirming the activity. For each variable, I compared the maritime and non-maritime explanatory variables with the Kolmogorov-Smirnov test if variables were continuous or the Wilcoxon test if the variables were discrete. I used the Kolmogorov-Smirnov test to determine whether there was a statistically significant difference between the cumulative distribution functions of maritime and non-maritime distribution functions. The Wilcoxon test served the same function, but tests discrete variables. I considered the two distributions significantly different if the p-value was .05 or less.