830 International Organization to capture the behavior of other host countries in the previous year,a chronology that makes more sense for the causal logic of diffusion as well.32 Note that the W matrix can represent not only geographic distances,but also economic,cultural, or political distances among countries.53 Our theory predicts interdependent decision making among host countries that compete for the same sources of global capital.Thus we need to determine the "competitive distance"between hosts.We create spatial weights that capture this distance in three ways.The first measures the degree to which host governments compete in the same foreign markets;that is,whether they have the same export trade relationships.54(All data sources and descriptive statistics are provided in the Data Appendix.)This is a useful indicator because trade competitors are also likely to be competitors for FDI and empirical studies show that the two are strongly correlated.We reason that countries that compete for export markets are structur- ally positioned to compete for the same sources of FDI as well.The second mea- sure records the degree to which nations export the same basket of goods.5 This measure captures the idea that investors choose between alternative locations for direct investment that they consider close substitutes with respect to the countries' traditional export products.For example,an automobile manufacturer might con- sider investing in countries that produce steel but will be unlikely to consider those whose leading export is cocoa.Our third measure captures the degree to which countries have similar educational and infrastructural resources.Assuming that potential foreign direct investors are concerned with a country's human assets as well as its technological and communications infrastructure,we reason that coun- tries with similar educational and infrastructural profiles will compete for the same 52.While spatial lags are common solutions to estimating the relational effects that we hypoth- esize,they do introduce a potential degree of endogeneity.Unless nondiffusion predictors are included in the model,spatial lags can absorb these effects when the domestic variables are correlated within the network.For this reason,some scholars have moved towards simultaneous equation modeling,in order to model the endogeneity.Recent Monte Carlo evidence reported in Franzese and Hays 2004 suggests that the costs associated with such models may outweigh their benefits in large samples.Our solution is to specify the nondiffusion components as completely as possible and to lag the spatial lag one year.Nonetheless,we recognize that effects from spatial lags may be slightly inflated. 53.See Elkins and Simmons 2005:and Beck.Gleditsch.and Beardsley 2006. 54.We use the IMF Direction of Trade data to produce an N by N by T matrix of correlations (between countries)across the countries'proportion of exports to each of the 182 partner countries. Two countries that export goods in the same proportions to 182 countries will have a score of 1;while those with entirely opposite relationships will have score of-1.For a similar approach,see Finger and Kreinin 1979.Network analysts often use this sort of measure to identify competitors;see Wasser- man and Faust 1994. 55.We calculate the distance between countries according to their export products,using informa- tion from the World Bank's World Development Indicators (WDI)that describes a country's export mix.These indicators tap the value of exports (in 1995 SUS)in sectors such as food,fuel,agricultural raw materials,ores and metals,and arms.We calculate the correlation between countries for each year across thirteen such indicators.The result is a measure,ranging from -I to 1,of the similarity between countries according to the products they export.to capture the behavior of other host countries in the previous year, a chronology that makes more sense for the causal logic of diffusion as well+ 52 Note that the W matrix can represent not only geographic distances, but also economic, cultural, or political distances among countries+ 53 Our theory predicts interdependent decision making among host countries that compete for the same sources of global capital+ Thus we need to determine the “competitive distance” between hosts+ We create spatial weights that capture this distance in three ways+ The first measures the degree to which host governments compete in the same foreign markets; that is, whether they have the same export trade relationships+ 54 ~All data sources and descriptive statistics are provided in the Data Appendix+! This is a useful indicator because trade competitors are also likely to be competitors for FDI and empirical studies show that the two are strongly correlated+ We reason that countries that compete for export markets are structur￾ally positioned to compete for the same sources of FDI as well+ The second mea￾sure records the degree to which nations export the same basket of goods+ 55 This measure captures the idea that investors choose between alternative locations for direct investment that they consider close substitutes with respect to the countries’ traditional export products+ For example, an automobile manufacturer might con￾sider investing in countries that produce steel but will be unlikely to consider those whose leading export is cocoa+ Our third measure captures the degree to which countries have similar educational and infrastructural resources+ Assuming that potential foreign direct investors are concerned with a country’s human assets as well as its technological and communications infrastructure, we reason that coun￾tries with similar educational and infrastructural profiles will compete for the same 52+ While spatial lags are common solutions to estimating the relational effects that we hypoth￾esize, they do introduce a potential degree of endogeneity+ Unless nondiffusion predictors are included in the model, spatial lags can absorb these effects when the domestic variables are correlated within the network+ For this reason, some scholars have moved towards simultaneous equation modeling, in order to model the endogeneity+ Recent Monte Carlo evidence reported in Franzese and Hays 2004 suggests that the costs associated with such models may outweigh their benefits in large samples+ Our solution is to specify the nondiffusion components as completely as possible and to lag the spatial lag one year+ Nonetheless, we recognize that effects from spatial lags may be slightly inflated+ 53+ See Elkins and Simmons 2005; and Beck, Gleditsch, and Beardsley 2006+ 54+ We use the IMF Direction of Trade data to produce an N by N by T matrix of correlations ~between countries! across the countries’ proportion of exports to each of the 182 partner countries+ Two countries that export goods in the same proportions to 182 countries will have a score of 1; while those with entirely opposite relationships will have score of 1+ For a similar approach, see Finger and Kreinin 1979+ Network analysts often use this sort of measure to identify competitors; see Wasser￾man and Faust 1994+ 55+ We calculate the distance between countries according to their export products, using informa￾tion from the World Bank’s World Development Indicators ~WDI! that describes a country’s export mix+ These indicators tap the value of exports ~in 1995 $US! in sectors such as food, fuel, agricultural raw materials, ores and metals, and arms+ We calculate the correlation between countries for each year across thirteen such indicators+ The result is a measure, ranging from 1 to 1, of the similarity between countries according to the products they export+ 830 International Organization