BENJAMIN O.FORDHAM 743 economic and security concerns on intervention and a more complete account of where each fits into the causal process Testing the effects of economic and security concerns in this way requires solu- tions to two research design problems.First,data on incidents in which Ameri- can intervention was a possibility must be gathered.Because arguments about the role of economic and security concerns apply to many classes of events,I will examine data on both civil wars and international crises.Second,economic and security concerns must be operationalized in realistic ways.I will use data on the behavior of rival states,alliance commitments,and American exports.The remainder of this section will explain these research design choices. Opportunities for Intervention:Civil Wars and International Crises Testing the influence of economic and security concerns on intervention requires data on incidents where intervention was a possibility.In principle, intervention requires no triggering event.States can and do use force without provocation.For research purposes,however,it is useful to begin with a class of events that has sometimes precipitated intervention.Arguments about economic and security interests can then be tested by estimating the influence of observa- ble indicators of these concerns on the probability of intervention within this class of events.I will consider two such classes of events here:civil wars and inter- national crises. Civil wars are a useful setting in which to consider the determinants of inter- vention.Given the potential costs involved,decision makers cannot undertake these interventions lightly.On the other hand,if policy makers believe that con- ditions in the affected states are important,then civil wars are difficult to ignore. Data on civil wars are relatively abundant,but information on intervention in these conflicts is not.Policy makers may decide the stakes in a particular conflict are important,yet use economic or military aid rather than committing their armed forces.The central question here concerns what makes particular civil wars sufficiently important to draw serious American attention,even if they do not involve direct military action.Thus it makes sense to look at intervention more broadly than just the use of force.Unfortunately,military action is the only type of intervention coded in many existing data sets.Patrick Regan's (2000, 2002)data set on civil wars is an exception,containing information about several forms of intervention.Including the provision of economic and military aid,the United States intervened in 38 of the 150 civil wars included in Regan's data, which cover the 1944-99 period. Although interstate conflicts differ from civil wars in many important respects, scholars have made many of the same arguments about economic and security concerns motivating intervention in both.Examining intervention in interna- tional crises complements the analysis of civil war intervention,ensuring that the patterns identified are not unique to that class of conflict events.The Inter- national Crisis Behavior (ICB)project has identified 435 international crises 1A third research design issue concerns the possibility of selection bias in the samples of civil wars and intema- tional crises considered here.Selection bias would be an important issue if American trade or alliance commit- ments influenced the probability that a crisis or civil war broke out in the first place.For example,if credible American alliance commitments always deterred potential challengers from threatening American allies,or per- suaded potential rebels that their cause was hopeless,then crises and civil wars would only be observed when Ameri- can alliance commitments were not credible.The sample of observed crises and civil wars would contain only cases where the American alliance commitment was shaky,perhaps leading to the conclusion that alliance commitments were not strongly associated with subsequent American intervention on behalf of its endangered allies.Although this is an important problem in principle,censored probit selection models designed to handle it turned up little evidence that it made an important difference in practice.The rho parameter was not significantly different from zero,and coefficient estimates were nearly identical to those presented in the next section.A more thorough pre- sentation of these models can be found in the online appendix posted with the replication data.economic and security concerns on intervention and a more complete account of where each fits into the causal process. Testing the effects of economic and security concerns in this way requires solu￾tions to two research design problems. First, data on incidents in which Ameri￾can intervention was a possibility must be gathered. Because arguments about the role of economic and security concerns apply to many classes of events, I will examine data on both civil wars and international crises. Second, economic and security concerns must be operationalized in realistic ways. I will use data on the behavior of rival states, alliance commitments, and American exports. The remainder of this section will explain these research design choices.1 Opportunities for Intervention: Civil Wars and International Crises Testing the influence of economic and security concerns on intervention requires data on incidents where intervention was a possibility. In principle, intervention requires no triggering event. States can and do use force without provocation. For research purposes, however, it is useful to begin with a class of events that has sometimes precipitated intervention. Arguments about economic and security interests can then be tested by estimating the influence of observa￾ble indicators of these concerns on the probability of intervention within this class of events. I will consider two such classes of events here: civil wars and inter￾national crises. Civil wars are a useful setting in which to consider the determinants of inter￾vention. Given the potential costs involved, decision makers cannot undertake these interventions lightly. On the other hand, if policy makers believe that con￾ditions in the affected states are important, then civil wars are difficult to ignore. Data on civil wars are relatively abundant, but information on intervention in these conflicts is not. Policy makers may decide the stakes in a particular conflict are important, yet use economic or military aid rather than committing their armed forces. The central question here concerns what makes particular civil wars sufficiently important to draw serious American attention, even if they do not involve direct military action. Thus it makes sense to look at intervention more broadly than just the use of force. Unfortunately, military action is the only type of intervention coded in many existing data sets. Patrick Regan’s (2000, 2002) data set on civil wars is an exception, containing information about several forms of intervention. Including the provision of economic and military aid, the United States intervened in 38 of the 150 civil wars included in Regan’s data, which cover the 1944–99 period. Although interstate conflicts differ from civil wars in many important respects, scholars have made many of the same arguments about economic and security concerns motivating intervention in both. Examining intervention in interna￾tional crises complements the analysis of civil war intervention, ensuring that the patterns identified are not unique to that class of conflict events. The Inter￾national Crisis Behavior (ICB) project has identified 435 international crises 1 A third research design issue concerns the possibility of selection bias in the samples of civil wars and interna￾tional crises considered here. Selection bias would be an important issue if American trade or alliance commit￾ments influenced the probability that a crisis or civil war broke out in the first place. For example, if credible American alliance commitments always deterred potential challengers from threatening American allies, or per￾suaded potential rebels that their cause was hopeless, then crises and civil wars would only be observed when Ameri￾can alliance commitments were not credible. The sample of observed crises and civil wars would contain only cases where the American alliance commitment was shaky, perhaps leading to the conclusion that alliance commitments were not strongly associated with subsequent American intervention on behalf of its endangered allies. Although this is an important problem in principle, censored probit selection models designed to handle it turned up little evidence that it made an important difference in practice. The rho parameter was not significantly different from zero, and coefficient estimates were nearly identical to those presented in the next section. A more thorough pre￾sentation of these models can be found in the online appendix posted with the replication data. Benjamin O. Fordham 743