BENJAMIN O.FORDHAM 749 (1998).It assumes that only information about whether a dyad had an alliance in the preceding time period is relevant to the probability that an alliance forms in the current period.This is a dubious assumption here.I will adopt the solu- tion Beck et al.(2001)propose to avoid:The model will include a variable indi- cating the number of years that the dyad has had (not had)an alliance and three natural cubic splines in the model of alliance dissolution (onset).To the extent that there is time dependence,the splines and state duration variables should capture its effects. Table 2 presents the results of two sets of models assessing the effect of exports on American alliances.The first set includes only exports and the variables model- ing time dependence.The second set adds several variables that previous research suggests may influence alliance decisions(Lai and Reiter 2000).The results,which are essentially the same whether control variables are included or not,indicate that increasing exports made American alliance formation more likely.Holding other variables at their mean value,a state that received U.S.exports one standard devia- tion above the mean was roughly 10 times more likely to form an alliance with the United States in a given year than a state that received the mean volume of exports. The effect of exports on alliance durability was smaller and less certain.The exports variable had the expected sign,but narrowly missed the threshold for statis- tical significance.Even if one were to lower the threshold for statistical significance and reject the null hypothesis,the estimated effect of trade on alliance durability was much smaller than its effect on alliance formation.An alliance with a state that received exports one standard deviation below the mean was roughly twice as likely to end in a given year as an alliance with a state that received the mean level of exports. The effect of exports on the probability of alliance formation is large compared to the other variables included in the model.The number of militarized disputes involving the prospective American ally during the preceding 10 years provides a rough indicator of the level of threat it faces.A one standard deviation increase in this variable above its mean value would make an alliance roughly twice as likely to form.The United States was more likely to make alliance commitments to relatively weak states.A one standard deviation decline in the capabilities of the prospective alliance partner would roughly triple the probability of alliance formation.The tenfold increase in the probability of alliance formation associated with a one stan- dard deviation increase in the value of U.S.exports was larger than all of these effects.These results differ from those reported by Lai and Reiter(2000),who examined a much larger sample of states and found no evidence that trade influ- enced alliance formation.American alliance decisions may well have been more sensitive to trade than those of most other states. The potential endogeneity of trade to alliance patterns could still threaten these results if economic actors changed their behavior in anticipation of an alli- ance.This problem would exist only if alliances really did change American trad- ing patterns,however.The empirical findings supporting this claim rest on models regressing levels of trade on the presence or absence of an alliance.For the reasons just noted,these findings do not reveal the direction of causality. A more convincing test of the alliances-increase-trade hypothesis focuses on changes in bilateral trade flows,rather than on their level.If the security exter- nalities of trade lead states to encourage commerce with their allies,or at least to restrict it with their adversaries,then trade with allies should grow more 3 One variable that had larger effects than trade was U.S.military capabilities.A one standard deviation increase in this variable made alliance formation roughly 12 times more likely.This result underscores the importance of the shift in American policy after World War II.Before the war,the United States maintained a relatively small mili- tary force and rarely entered into alliances.Since then,the United States has maintained a very large military force and built a worldwide network of alliances.The relationship between American power and alliance formation is not a simple one.Both are endogenous to broader policy decisions made during the 1940s.(1998). It assumes that only information about whether a dyad had an alliance in the preceding time period is relevant to the probability that an alliance forms in the current period. This is a dubious assumption here. I will adopt the solu￾tion Beck et al. (2001) propose to avoid: The model will include a variable indi￾cating the number of years that the dyad has had (not had) an alliance and three natural cubic splines in the model of alliance dissolution (onset). To the extent that there is time dependence, the splines and state duration variables should capture its effects. Table 2 presents the results of two sets of models assessing the effect of exports on American alliances. The first set includes only exports and the variables model￾ing time dependence. The second set adds several variables that previous research suggests may influence alliance decisions (Lai and Reiter 2000). The results, which are essentially the same whether control variables are included or not, indicate that increasing exports made American alliance formation more likely. Holding other variables at their mean value, a state that received U.S. exports one standard devia￾tion above the mean was roughly 10 times more likely to form an alliance with the United States in a given year than a state that received the mean volume of exports. The effect of exports on alliance durability was smaller and less certain. The exports variable had the expected sign, but narrowly missed the threshold for statis￾tical significance. Even if one were to lower the threshold for statistical significance and reject the null hypothesis, the estimated effect of trade on alliance durability was much smaller than its effect on alliance formation. An alliance with a state that received exports one standard deviation below the mean was roughly twice as likely to end in a given year as an alliance with a state that received the mean level of exports. The effect of exports on the probability of alliance formation is large compared to the other variables included in the model. The number of militarized disputes involving the prospective American ally during the preceding 10 years provides a rough indicator of the level of threat it faces. A one standard deviation increase in this variable above its mean value would make an alliance roughly twice as likely to form. The United States was more likely to make alliance commitments to relatively weak states. A one standard deviation decline in the capabilities of the prospective alliance partner would roughly triple the probability of alliance formation. The tenfold increase in the probability of alliance formation associated with a one stan￾dard deviation increase in the value of U.S. exports was larger than all of these effects.3 These results differ from those reported by Lai and Reiter (2000), who examined a much larger sample of states and found no evidence that trade influ￾enced alliance formation. American alliance decisions may well have been more sensitive to trade than those of most other states. The potential endogeneity of trade to alliance patterns could still threaten these results if economic actors changed their behavior in anticipation of an alli￾ance. This problem would exist only if alliances really did change American trad￾ing patterns, however. The empirical findings supporting this claim rest on models regressing levels of trade on the presence or absence of an alliance. For the reasons just noted, these findings do not reveal the direction of causality. A more convincing test of the alliances-increase-trade hypothesis focuses on changes in bilateral trade flows, rather than on their level. If the security exter￾nalities of trade lead states to encourage commerce with their allies, or at least to restrict it with their adversaries, then trade with allies should grow more 3 One variable that had larger effects than trade was U.S. military capabilities. A one standard deviation increase in this variable made alliance formation roughly 12 times more likely. This result underscores the importance of the shift in American policy after World War II. Before the war, the United States maintained a relatively small mili￾tary force and rarely entered into alliances. Since then, the United States has maintained a very large military force and built a worldwide network of alliances. The relationship between American power and alliance formation is not a simple one. Both are endogenous to broader policy decisions made during the 1940s. Benjamin O. Fordham 749