chance that my variable captures effects other than domestic opposition. I also run a variety of robustness checks to ameliorate such problems as omitted-variable bias and measurement error. Finally, to better test whether a causal relationship exists, I evaluate my theory’s causal logic in a series of case studies in Chapters 4, 5, and 6. In all, the statistical results in this chapter constitute an important first step in identifying whether domestically unstable states are prone to FIRC.
In the sections ahead, I explain how I code the 133 cases of FIRC that appear in my data set (see Appendix 1).4 Then, I describe the structure of the data set and the model I use to test my theory’s primary prediction that domestic opposition in the target state leads to FIRC. Next, I discuss the various strengths and shortcomings of the most common approaches to measuring domestic opposition. I then explain how the two proxies I use avoid some of the pitfalls associated with these other approaches. I follow this with a discussion of the covariates in the model, including the controls used to test alternative arguments. The last section of this chapter presents my findings.
The Dependent Variable
I use a dichotomous dependent variable (RC) that is coded 1 for each year that an attempt at regime change occurred. To identify these occurrences, I drew from a variety of existing data sets on foreign intervention, imposed democratization, and FIRC.5 I also consulted historical sources to identify cases that fit my definition.6 In assessing the historical record, I looked for evidence that the policymakers of one state (1) chose to abandon negotiations with the leader or regime of a sovereign state, (2) implemented an explicit plan to depose that leader or regime, and (3) did not intend to annex, colonize, or otherwise permanently rule that state.7 I explain these components of my definition of FIRC more fully in the introduction. In this section, I explain coding decisions that affect how the data appear in the data set.
In contrast to many existing FIRC data sets, my data include instances of attempted FIRC. I include such cases because the circumstances that motivate FIRC should apply whether it succeeds or fails. I code only the first year of an attempt as subsequent years are not independent events. Including them could also bias the statistical results in favor of my argument. When states attempt to overthrow foreign leaders, the aid they provide the opposition typically strengthens it further. Even when their efforts at FIRC ultimately fail, if the intervention empowers the opposition for some period of time, then coding each year of an attempt would increase the chances of finding that higher levels of opposition lead to regime change.
My definition of FIRC also requires that the state seeking regime change implement a plan of action aimed at toppling the targeted leader or regime. When it takes several years for such a plan to come to fruition, I use the first year in which the imposing state takes either covert or overt action toward removing the targeted leader. For example, I code the year in which the Allies initiated their attempt to overthrow the governments of Germany and Japan as 1943. Although hostilities preceded this year, President Roosevelt declared the goal of unconditional surrender in 1943, thereby indicating that the Allies would fight until they achieved the victory necessary to impose regime change. The action undertaken by the imposing state must be more than merely symbolic or ad hoc.8 States, for example, frequently impose sanctions that they claim are aimed at regime change, though it is well understood that the sanctions are too mild to do more than satisfy a domestic demand for action. In examining such cases, I use evidence that sanctions were part of a larger campaign that included either overt or covert measures to topple the targeted leader.9 The Nixon administration’s sanctions on Chile, for example, were paired with covert operations explicitly designed to foment a coup against President Salvador Allende.
Although my definition excludes cases in which the foreign power annexes the target state or rules it as a colonial possession, I do include cases in which states restore the target’s sovereignty but later establish formal control. In 1882, for example, the United Kingdom deposed Ahmad Urabi Pasha al-Misri, commander in chief of the Egyptian army, who had challenged the khedive, Muḥammad Tawfīq Pasha, for power. Although the period following the khedive’s restoration is often referred to as a “veiled protectorate,” the United Kingdom did not formalize its control until 1914.10 Nevertheless, some cases fall short of the criteria for statehood as defined by the Correlates of War data set, which I use to construct several of the variables in the model.11 Because these cases are dropped when running the model, I list them separately in Table 15 of Appendix 1.
Finally, 24 of the 133 cases of FIRC I identify involve more than one intervening state. I include each state that played a major role in the operation if the primary goal of that state was regime change. For example, although I include the US invasion of North Korea, I exclude the South Korean joint invasion, because the South Korean government’s goal was to annex the North.12 I enter these multilateral FIRC events separately into the data, which brings the total number of FIRC events to 148. In the case of Germany following World War II, for example, the Soviet, American, and British decisions to impose regime change are counted as separate observations. Although these events are clearly related, each state’s decision to participate (or not) informs our understanding of when states pursue FIRC. To control for the nonindependence between multilateral cases, I use robust standard errors clustered on the target-state year.
The Data Structure and Model
I employ a logit model to test the probability of FIRC. The unit of analysis is the directed-dyad year, which means each observation contains a potential imposing state (Side 1) and target state (Side 2) for every year, starting in 1816.13 This produces an enormously large data set with many dyads for which the probability of conflict is almost zero. The large number of potentially irrelevant observations poses two problems. First, large data sets can be sensitive to very minor effects, which can overstate the importance of variables that have little explanatory power. Second, because my theory explains how states reconcile conflicts of interests, it assumes a population of dyads with conflicting interests. Testing the model on data that include many dyads with almost no potential for conflict introduces inefficiency and possible bias.
I correct for these problems by limiting the scope of the data in three ways. First, I test the model using politically relevant dyads, commonly used to exclude dyads that have almost no probability of conflict. Politically relevant dyads are dyads in which Side 1 or 2 is a major power or the two states are contiguous.14 Twelve FIRC events fall short of these criteria. To include these cases, I perform a robustness check using an expanded definition of politically relevant dyads, which includes dyads located in the same region rather than only ones that are contiguous.
Second, I test the model on interstate rivals. Given that rivals engage in repeated disputes, we know that they are likely to have conflicting interests. I use James Klein, Gary Goertz, and Paul Diehl’s rivalry data set, which defines rivals as states that have had at least three or more militarized interstate disputes (MIDs) concerning the same issue(s). Rivalries are coded as ending ten to fifteen years after the last dispute.15 The disadvantage of restricting the population of cases to rivals is that the data exclude dyads that reconcile their differences through negotiation without the use of force. These data also leave out FIRC events that involve nonrivals, such as the United States and Guatemala in 1954.16 Lastly, to account for these cases, I use a third approach in which I limit the population to dyads that have experienced at least one MID in their shared history. These dyads should have a higher probability of conflict than merely politically relevant ones, though they do not necessarily meet the criteria to be considered interstate rivals.17
Measuring Domestic Political Opposition
