Operationalization and Methodology
Sample and Selection
Rivalry data from Klein, Goertz, and Diehl’s (2006) rivalry data-set is used as the sample for this study. The data used in this study begins with rivalry observations in 1946 and ends in 2001. The level of analysis is dyad-year. There are observations from 216 different rivalries and the data-set contains one observation per year, per rivalry from the life of the rivalry and has a total of 3,432 observations.
Dependent Variables
Rivalry termination. The dependent variable in this model is rivalry termination. Rivalry termination dates were drawn from Klein, Goertz ,and Diehl’s (2006) data-set on rivalries. Rivalries are defined as terminating in this dataset 10-15 years after the last militarized interstate dispute. Rivalry termination is coded as (1) for the year in which a rivalry ends and (0) in all other cases with the exception of a rivalry that continues on after 2001 when the dataset ends, in which case it is coded as (1) for 2001.
Independent Variables
Average rivalry severity. The severity of a rivalry is determined by utilizing the severity score index, which measures the severity of each conflict that occurs between two rivals, developed by Klein, Goertz, and Diehl (2006). The severity score takes into account the number of conflicts between rivals, the number of casualties in a conflict, as well as the level of hostility between the parties involved. The severity score is used as a way to measure the amount of loss incurred because of the rivalry. As the rivalry becomes severe, the amount of perceived loss should increase accordingly. From these scores the average severity is calculated by taking the severity of all previous conflicts in a rivalry and dividing it by the total number of conflicts to date. In this sample, the average severity score ranges from (0) to (361).
Initial rivalry severity. Initial rivalry severity is derived from the average severity variable. The initial severity of a rivalry is calculated as the average severity of the interactions between two rivals after the first five years of a rivalry.
Control Variables
Regime change. The Polity Index, drawn from the Polity IV data-set, measures different facets of executive recruitment, constraints on the executive, regulation of political participation, and competitiveness of political participation to determine the level of democracy or autocracy within a state. Polity Index scores range from (-10), fully institutionalized autocratic regimes, to (+10) fully institutionalized democratic regimes. The polity change variable controls for significant changes in the political structure or regime of a state. A six point movement in the polity score of a state represents a move from consolidated democracy or authoritarianism to anocracy, from anocracy to democracy or authoritarianism, or from an authoritarian anocracy to a democratic anocracy, or vice versa. For cases when either party of a dyad has a change in polity score greater than six points, polity change is coded as (1). If neither state has a polity change of six or greater Polity Change is coded as (0).
Past polity change. In order to account for major polity changes in the past that could affect current decision making processes we utilize a second polity change variable, which measures major changes in the political structure in the past five years. A five-year period was selected because it is long enough to account for the effects of recent political changes that may affect decision-making. This period is short enough that it does not take into account changes in polity that no longer have major effects because the changes in behavior that resulted from those changes in polity are now institutionalized and have become the status quo. If a change of more than six points on the Polity Index occurs in a country in the past five years, polity change is coded as (1). Otherwise it is coded as (0).
Composite index of national capability ratio. The Composite Index of National Capability is drawn from the National Material Capabilities data-set developed by the Correlates of War Project. A country’s CINC score is derived from their iron and steel production, military expenditures, military personnel, energy consumption, population, and size of their urban population. The CINC Ratio variable controls for wide disparities in power of rival nations. The ratio is calculated by dividing the CINC score of one rival in the dyad by the CINC score of the second rival in the dyad.
Balance of power. Previous research shows that changes in the balance of power may have some effect on the termination of rivalries (Bennett, 1997). We use a dichotomous variable in an attempt to account for substantial changes in the balance of power between two states. We set a threshold of two percent in the difference between the capabilities of two rivals from the previous year as a significant change. A two percent threshold was set because a two percent change is sensitive enough to account for major changes in power, but only occurs in about one percent of cases and is therefore uncommon enough to be considered a departure from normal fluctuations in the balance of power. If the difference in capabilities between one state in a rivalry and the second state in a rivalry, as defined by their CINC score, changes two percent or more from the previous year we deemed this to be significant, and coded it as (1). If the change from the previous year was less than two percent, its power change was coded as (0).
Presence of outside threats. Previous research indicates that the presence of outside threats has an effect on the duration of rivalries (Bennett, 1997). The presence of an outside threat is defined as the involvement of either rival in a second rivalry in the current year. If either state in a rivalry is involved in a second rivalry in the current year, the presence of an outside threat is coded as (1); otherwise it is coded as (0).
Past outside threats. In addition to controlling for the presence of outside threats in the current year, we account for outside threats in the past that may influence current decision making processes. We created a variable that accounts for the presence of an outside threat in the past that could influence current decision-making processes. If either state in a rivalry was involved in second rivalry in the last five years, the past outside threat is coded as (1). Otherwise it is coded as (0).
Democracy. Democracies and authoritarian regimes operate in different manners. Democratic states are forced to deal with losses in different ways than authoritarian regimes, because in democratic states the electorate both holds political power and must directly bear the cost of conflict (Kant 1795). In contrast, rulers in authoritarian regimes are often removed from the direct costs of conflicts. Additionally, previous research regarding democratic peace states that democracies are less likely to go to war with one another (Oneal and Russett 1997). This makes the way democracies interact with one another different from the way democracies interact with non-democracies in situations involving conflict. To account for this difference in behavior, democracy is coded as (1) when both countries involved in a rivalry are classified as consolidated democracies. For example, they have a score equal to or greater than (5) on the Polity Index; otherwise democracy is coded as (0).
Time. Previous research indicates that the longer a rivalry continues, the greater the probability that it will terminate (Cioffi-Revilla 1998; Bennett 1998). We introduce a time variable that measures the amount of time, in months, that has elapsed since the beginning of a rivalry as a control for this relationship.
Method
Because the dependent variable is dichotomous, a logistic regression model was utilized to predict the probability of rivalry termination as the result of the initial severity of a rivalry. The logistic regression determines which, if any, of the variables used in the model are significant. Furthermore, because the coefficients output by the logistic regression model do not correspond to a direct linear change in the dependent variable as the result of changes in the independent variable, odds ratios are also utilized in order to provide easily interpretable and meaningful coefficients.