750 Power or Plenty？ TABLE 2.Transition Models of Trade and Alliances Alliance Alliance Alliance Onset Dissolution Alliance Onset Dissolution Log of exports in previous year 0.15(0.02)* 0.09(0.05) 0.30(0.05)-0.15(0.09) (millions of 2.000 dollars) Log of capital to capital distance from -0.03(0.02) 0.06(0.09) United States Log of MIDs in which United States was -0.05(0.10) -0.46(0.43) involved in preceding 10 years Log of MIDs in which other member of 0.13(0.07)0.01(0.12) dyad was involved in preceding 10 years Log of MIDs between United States and -0.39(0.22) 0.12(0.36) other member of dyad in preceding 10 years Log of COW capabilities index for 2.55(0.30)◆ 0.76(1.19) United States Log of COW capabilities index for -0.20(0.04) 0.30(0.12)* other member of dyad Polity score of other member of dyad 0.004(0.01) 0.00(0.02) Cold War era (1946-89) -0.002(0.15) 1.06(0.65) Previous alliance onsets -0.15(0.16) 0.06(0.22) Previous alliance collapses -0.41(0.21) -0.61(0.36) Constant -2.60(0.16)*-1.34(0.33) -1.25(0.75) 2.84(1.98) Pseudo R2 0.09 0.20 0.32 0.33 Observations 8,442 2.592 7,788 2,267 Note.Robust standard errors adjusted for clustering on the country are in parentheses.Asterisks indicate statistical significance at the p<.05 level.Each probit model also included variables indicating the number of years since the previous alliance,if any,and three natural cubic splines.Models of alliance formation included only observation in which there was no alliance in the preceding year.Models of alliance collapse included only observations in which there was an alliance in the preceding year.See Beck et al.(2001)for a detailed discussion of this approach. rapidly than trade with other states.Mansfield and Bronson(1997,102)present such a test in their examination of the influence of alliances and preferential trading arrangements on international trade,and find that this pattern only holds when alliance partners are also members of a preferential trading arrange- ment.However,the alliance variable they use,like the other components of their gravity model,is differenced.This is a reasonable specification,but it tests only whether trade tends to increase substantially in the year the alliance was formed. A more generous test would also consider the possibility that the change in trad- ing patterns happens more gradually,perhaps over the first few years of the alli- ance or even its entire duration. Table 3 presents the results of three differenced gravity models testing the effect of alliances on American bilateral trading patterns.The first employs a dif ferenced alliance variable like that which Mansfield and Bronson (1997)used. The second tests the hypothesis that the change in trade is greater on average over the entire life of the alliance.Because it is possible that the restructuring of trade occurs in the first few years of the alliance,the third model tests the hypothesis that the change in trade is higher during the first five years of the alli- ance.Statistically significant results in any of these three specifications would provide evidence of endogeneity.The need for GDP data to estimate the gravity model restricts this analysis to the 1950-2000 period.Gowa and Mansfield 4The expanded trade and GDP data assembled by Gleditsch (2002)were used to estimate these models because they cover a larger sample of states for the 1950-2000 period for which GDP data are available.Department of Commerce trade data were used elsewhere in the article because of their longer temporal coverage,which was more important when GDP was not required.rapidly than trade with other states. Mansfield and Bronson (1997, 102) present such a test in their examination of the influence of alliances and preferential trading arrangements on international trade, and find that this pattern only holds when alliance partners are also members of a preferential trading arrange￾ment. However, the alliance variable they use, like the other components of their gravity model, is differenced. This is a reasonable specification, but it tests only whether trade tends to increase substantially in the year the alliance was formed. A more generous test would also consider the possibility that the change in trad￾ing patterns happens more gradually, perhaps over the first few years of the alli￾ance or even its entire duration. Table 3 presents the results of three differenced gravity models testing the effect of alliances on American bilateral trading patterns. The first employs a dif￾ferenced alliance variable like that which Mansfield and Bronson (1997) used. The second tests the hypothesis that the change in trade is greater on average over the entire life of the alliance. Because it is possible that the restructuring of trade occurs in the first few years of the alliance, the third model tests the hypothesis that the change in trade is higher during the first five years of the alli￾ance. Statistically significant results in any of these three specifications would provide evidence of endogeneity. The need for GDP data to estimate the gravity model restricts this analysis to the 1950–2000 period.4 Gowa and Mansfield Table 2. Transition Models of Trade and Alliances Alliance Onset Alliance Dissolution Alliance Onset Alliance Dissolution Log of exports in previous year (millions of 2,000 dollars) 0.15 (0.02)* 0.09 (0.05) 0.30 (0.05)* )0.15 (0.09) Log of capital to capital distance from United States )0.03 (0.02) 0.06 (0.09) Log of MIDs in which United States was involved in preceding 10 years )0.05 (0.10) )0.46 (0.43) Log of MIDs in which other member of dyad was involved in preceding 10 years 0.13 (0.07)* 0.01 (0.12) Log of MIDs between United States and other member of dyad in preceding 10 years )0.39 (0.22) 0.12 (0.36) Log of COW capabilities index for United States 2.55 (0.30)* 0.76 (1.19) Log of COW capabilities index for other member of dyad )0.20 (0.04)* 0.30 (0.12)* Polity score of other member of dyad 0.004 (0.01) 0.00 (0.02) Cold War era (1946–89) )0.002 (0.15) 1.06 (0.65) Previous alliance onsets )0.15 (0.16) 0.06 (0.22) Previous alliance collapses )0.41 (0.21)* )0.61 (0.36) Constant )2.60 (0.16)* )1.34 (0.33)* )1.25 (0.75) 2.84 (1.98) Pseudo R2 0.09 0.20 0.32 0.33 Observations 8,442 2,592 7,788 2,267 Note. Robust standard errors adjusted for clustering on the country are in parentheses. Asterisks indicate statistical significance at the p < .05 level. Each probit model also included variables indicating the number of years since the previous alliance, if any, and three natural cubic splines. Models of alliance formation included only observation in which there was no alliance in the preceding year. Models of alliance collapse included only observations in which there was an alliance in the preceding year. See Beck et al. (2001) for a detailed discussion of this approach. 4 The expanded trade and GDP data assembled by Gleditsch (2002) were used to estimate these models because they cover a larger sample of states for the 1950–2000 period for which GDP data are available. Department of Commerce trade data were used elsewhere in the article because of their longer temporal coverage, which was more important when GDP was not required. 750 Power or Plenty?