748 Power or Plenty? consideration of more than just its proximate causes.It must also include indirect effects that work through other independent variables in the model. The next section turns to this problem. Trade and Alliances The next step in the analysis is to disentangle the relationship between exports and alliances.As was discussed in the last section,this relationship could run in either or both directions,creating indirect effects on the probability of interven- tion.Because alliances strongly influence intervention decisions,exports may have an important indirect effect on intervention if they influence alliance for- mation,even though their direct effect is not large. The empirical association between trade and alliances is well established in the literature.What is less clear is whether this association stems from the influence of alliances on trade,or of trade on alliances.The endogeneity problem is espe- cially clear in the models used to test the hypothesis that alliances influence trade flows.This research employs a gravity model of the level of trade,adding a variable indicating the presence or absence of an alliance (e.g.,Gowa and Mans- field 1993,2004;Long 2003;Mansfield and Bronson 1997).The alliance variable in these models essentially compares the mean level of bilateral trade when the trading partners have an alliance to the mean level of trade in its absence.This relationship might exist either because alliances increase trade or because trade increases the probability of an alliance. One way to handle the endogeneity problem would be to identify exogenous instrumental variables for trade and alliances,then use these instruments in a simultaneous equation model of trade and alliances.Unfortunately,consider- ations that are related to trade but unrelated to alliance decisions,and vice versa, are difficult to find.There is empirical evidence that the variables commonly used in gravity models of trade also affect alliance decisions.The scarcity of gen- uinely exogenous instrumental variables would lead either to a weakly identified model,if the variables chosen were not really exogenous,or to inefficient esti- mates of the underlying relationships,if the instruments do not predict trade and alliances very well. An alternative approach to the endogeneity problem uses the temporal sequencing of trade and alliance formation to sort out their effects.Data on trade flows that occur in the absence of an alliance can be considered to see if high levels of trade make the formation of an alliance more likely.Data on trade occurring when an alliance is in place can be examined to determine whether high levels of trade makes alliances more durable.Finally,data on the annual change in trade can be used to test whether alliances alter trading patterns.This approach does not entirely eliminate the possibility of endogeneity,but it greatly restricts the conditions under which reverse causation is possible. The first step is to model the effect of trade on alliance formation and dissolu- tion separately,as is done in the transition model discussed by Beck et al. (2001).This model is similar to those employed by Przeworski et al.(2000)and Gleditsch and Ward (2005),estimating transition probabilities for the onset and the collapse of democratic rule.Like these researchers,I am interested in model- ing the probability that a particular kind of institution will be established or dis- solved.Like their data,those used here consist of discrete,binary,annual observations of whether the institution in question is present or absent.The transition model is essentially two probit models,one for the onset of an alliance when none was in place in the preceding year,and another for the collapse of the alliance when one existed in the preceding year. As Beck et al.(2001,8-9)point out,the transition model is a special case of the binary time-series cross-sectional model proposed by Beck,Katz,and Tuckerconsideration of more than just its proximate causes. It must also include indirect effects that work through other independent variables in the model. The next section turns to this problem. Trade and Alliances The next step in the analysis is to disentangle the relationship between exports and alliances. As was discussed in the last section, this relationship could run in either or both directions, creating indirect effects on the probability of interven￾tion. Because alliances strongly influence intervention decisions, exports may have an important indirect effect on intervention if they influence alliance for￾mation, even though their direct effect is not large. The empirical association between trade and alliances is well established in the literature. What is less clear is whether this association stems from the influence of alliances on trade, or of trade on alliances. The endogeneity problem is espe￾cially clear in the models used to test the hypothesis that alliances influence trade flows. This research employs a gravity model of the level of trade, adding a variable indicating the presence or absence of an alliance (e.g., Gowa and Mans- field 1993, 2004; Long 2003; Mansfield and Bronson 1997). The alliance variable in these models essentially compares the mean level of bilateral trade when the trading partners have an alliance to the mean level of trade in its absence. This relationship might exist either because alliances increase trade or because trade increases the probability of an alliance. One way to handle the endogeneity problem would be to identify exogenous instrumental variables for trade and alliances, then use these instruments in a simultaneous equation model of trade and alliances. Unfortunately, consider￾ations that are related to trade but unrelated to alliance decisions, and vice versa, are difficult to find. There is empirical evidence that the variables commonly used in gravity models of trade also affect alliance decisions. The scarcity of gen￾uinely exogenous instrumental variables would lead either to a weakly identified model, if the variables chosen were not really exogenous, or to inefficient esti￾mates of the underlying relationships, if the instruments do not predict trade and alliances very well. An alternative approach to the endogeneity problem uses the temporal sequencing of trade and alliance formation to sort out their effects. Data on trade flows that occur in the absence of an alliance can be considered to see if high levels of trade make the formation of an alliance more likely. Data on trade occurring when an alliance is in place can be examined to determine whether high levels of trade makes alliances more durable. Finally, data on the annual change in trade can be used to test whether alliances alter trading patterns. This approach does not entirely eliminate the possibility of endogeneity, but it greatly restricts the conditions under which reverse causation is possible. The first step is to model the effect of trade on alliance formation and dissolu￾tion separately, as is done in the transition model discussed by Beck et al. (2001). This model is similar to those employed by Przeworski et al. (2000) and Gleditsch and Ward (2005), estimating transition probabilities for the onset and the collapse of democratic rule. Like these researchers, I am interested in model￾ing the probability that a particular kind of institution will be established or dis￾solved. Like their data, those used here consist of discrete, binary, annual observations of whether the institution in question is present or absent. The transition model is essentially two probit models, one for the onset of an alliance when none was in place in the preceding year, and another for the collapse of the alliance when one existed in the preceding year. As Beck et al. (2001, 8–9) point out, the transition model is a special case of the binary time-series cross-sectional model proposed by Beck, Katz, and Tucker 748 Power or Plenty?