The evolution of global trade and impacts on countries’ carbon trade imbalances

Social Networks 46 (2016) 87–100

Contents lists available at ScienceDirect

Social Networks journal homepage: www.elsevier.com/locate/socnet

The evolution of global trade and impacts on countries’ carbon trade imbalances Christina Prell a,∗ , Kuishuang Feng b a b

Sociology Department, University of Maryland, College Park, MD, USA Geographical Sciences Department, University of Maryland, College Park, MD, USA

a r t i c l e

i n f o

Article history: Keywords: Co-evolution Global trade Embodied carbon trade inequalities Ecological unequal exchange Regionalization Multi-regional input–output analysis Stochastic actor-oriented models

a b s t r a c t We examine carbon trade imbalances among 172 countries over a 10-year period (2000–2010). A carbon trade imbalance refers to the extent to which the carbon emissions embodied in a country’s exports exceeds the emissions embodied in its imports. Although past research has considered how such imbalances coincide with variances in wealth and/or trade patterns, none have considered how such imbalances arise in the context of an evolving network structure. In this paper, we consider trade networks and carbon trade imbalances as co-evolving phenomena, and study these using an innovative combination of multi-regional input–output (MRIO) analysis with stochastic actor-oriented models (SAOMs). Our findings both challenge and support arguments made by ecological unequal exchange theory and the Gravity Model, and highlight the role emerging economies play in shaping network structure and the distribution of carbon trade balances overtime. © 2016 Elsevier B.V. All rights reserved.

1. Introduction The interconnectivity of international trade implies that activities and events in one part of the world have consequences, with winners and losers, around the globe. These consequences are numerous, and in this paper, we focus on the distribution of carbon dioxide associated with trade. Carbon dioxide (CO2 ) is considered one of the main pollutants responsible for global climate change, with far reaching impacts for human and environmental systems (IPCC, 2014). As such, it is important to develop frameworks in which the allocation of responsibility of CO2 can be understood in a global context. In this paper, we develop such a framework by reconfiguring CO2 distributions in terms of how CO2 becomes embodied in export and import flows. This is done through combining multi-regional input–output analysis (MRIO), an analytical technique used for studying the impacts of global trade on a number of environmental indicators,1 with

∗ Corresponding author. E-mail address: [email protected] (C. Prell). 1 This method has been applied to global environmental issues such as land use (e.g. Yu et al., 2014), water consumption (e.g. Feng et al., 2012), biodiversity (e.g. Lenzen et al., 2012), and CO2 emissions (e.g. Davis et al., 2011; Kagawa et al., 2015; Steinberger et al., 2012). http://dx.doi.org/10.1016/j.socnet.2016.03.001 0378-8733/© 2016 Elsevier B.V. All rights reserved.

stochastic actor-oriented models (SAOMs), statistical models used for studying how networks and attributes of actors within these networks change overtime (Snijders et al., 2010). In combining MRIO with SAOMs, we are better equipped to test a number of intuitive concepts about the interdependencies of international trade and their impacts on carbon imbalances of countries. As of this writing, we are unaware of any other study that combines such approaches to explore the theoretical ideas that we present here regarding economic trade and carbon trade embodied in trade flows. To better explain what we mean by carbon (CTI) trade imbalances (CTI), we offer Fig. 1 below: Fig. 1 shows CO2 produced within a country via manufacturing and consumption activities (referred to as production-based CO2 , and sometimes referred to as territorial CO2 ), CO2 embodied in the goods a country imports (referred to as CO2 embodied in imports), and finally, the CO2 embodied in the goods and services a country exports (referred to as CO2 embodied in exports). ‘Embodied emissions’ refers to all upstream or lifecycle emissions that are related to the production of a certain product. Together, these forms of CO2 comprise a country’s CO2 trade imbalance (CTI), such that a CTI of a given country represents: (i) the CO2 emissions in country A that occur during the production of good and services exported to the rest of the world, compared to, (ii) the CO2 emissions in other countries that occur during the production of goods

88

C. Prell, K. Feng / Social Networks 46 (2016) 87–100

Fig. 1. Calculating a country’s carbon trade imbalance.

and services imported by country A, such that CTI = CO2 exports/CO2 imports.2 As such, a country’s CTI is either a positive number, meaning the country is a net exporter of embodied CO2 (i.e. has a carbon export surplus) or negative, meaning the country is a net importer embodied CO2 (i.e. has a carbon export deficit). And these numbers represent the “imbalance” of CO2 embodied in a country’s trade.

more developed countries, this lack of wealth often translates into a loss of an economic ‘buffer’ that would potentially mitigate the harmful health impacts associated with pollution-intensive economic activities (e.g. Prell et al., 2015). Taken together, the negative environmental impacts associated with unequal trade to wealthier countries leads to our first hypothesis:

1.1. Ecological unequal exchange

H1. Countries with wealthier trading partners tend to become or remain being net exporters of carbon overtime.

Research suggests that less-developed countries tend to be net exporters of carbon, while wealthy, developed countries tend to be net importers of carbon (Davis and Caldeira, 2010; Davis et al., 2011; Hertwich and Peters, 2009; Peters, 2008; Peters and Hertwich, 2008; Steinberger et al., 2012). An argument given for this disparity is that less-developed countries are caught in an ‘ecologically unequal exchange’ (EUE) pattern with wealthy, developed countries (Hornborg, 1998, 2011; Jorgenson, 2011, 2012; Moran et al., 2013; Rice, 2007).3 Here, wealthy, developed countries are seen as dictating the terms of trade with poorer, less-developed ones, sending financial investment and/or high value added goods in exchange for lower value added goods and natural resources produced in, or extracted from these countries. Consequently, the populations in these less-developed nations tend to experience higher amounts of environmental costs. Environmental regulations in these countries tend to be less stringent than in developed ones (e.g. Copeland and Taylor, 2004), discouraging local agents from investing in cleaner, more efficient technologies, while simultaneously encouraging wealthier nations to externalize pollution-intensive manufacturing to these less-regulated regions (e.g. Roberts and Parks, 2007). In addition, as the residents in lessdeveloped countries are, on the whole, poorer than those found in

2 Embodied, embedded or virtual refers to all direct and indirect emissions or resources consumed throughout the entire (global) production (also commodity or supply) chain. 3 In the case of carbon, when carbon-intensive manufacturing is offshored, this process is referred to as ‘carbon leakage’ or the ‘pollution haven thesis’ (IPCC, 2014). More broadly, however, when developing countries experience environmental harm associated with producing/extracting goods for consumption in the wealthy, developed world, this process is referred to as ecological unequal exchange (Hornborg, 1998, 2011).

This past research (and our current hypothesis) portray CTIs as an outcome of international trade patterns and/or economic development, yet other research suggests processes that give rise to these unequal exchanges, and thus, suggest a means by which to perceive CTIs and trade patterns as co-evolving. The next section summarizes this research. 1.2. Trade tie formation and carbon trade imbalances A common perspective found in the trade literature is that of ‘comparative advantage’ (Porter, 1990; Ricardo, 1821). According to this perspective, economic agents in a given country strive to produce goods at lower costs within their home countries in order to be competitve globally, and this in turn enables economic agents to increase their sales worldwide, subsequently leading to increases in their number of global trade partners. In relation to CTIs, firms striving for a competitive advantage in the production of carbonintensive goods are more likely to be net exporters of carbon (i.e. maintain a positive CTI), and such carbon imbalances tend to occur in less-developed countries, where production costs are lower and environmental laws more permissive (Dick, 2010; Leonard, 1985; Roberts and Parks, 2007). By maintaining this competitive advantage for carbon-intensive products, a country is thus more likely to expand their reach into the global market, which leads to our second hypothesis: H2. Countries that are net exporters of carbon (i.e. have a positive CTI) will tend to increase their number of export ties overtime. Ecological unequal exchange theory expands ideas pertaining to comparative advantage by considering between-country economic and environmental disparities. Here, acquiring a ‘comparative advantage’ in pollution-intensive goods is a strategy

C. Prell, K. Feng / Social Networks 46 (2016) 87–100

pursued by economic agents in less-developed countries, who essentially degrade their environments to satisfy increasing consumer demand in wealthier countries (Copeland and Taylor, 2004; Grimes and Kentor, 2003; Leonard, 1985, 2006; Roberts, 1996; Roberts and Parks, 2007). As such, acquiring a comparative advantage, from an EUE perspective, is perceived and understood within a larger, historical context pertaining to ongoing inequalities (both economic and environmental) arising from unequal trade patterns. This leads to our third hypothesis: H3. Countries that are net carbon exporters will tend to increase their export ties to wealthier countries overtime. 1.2.1. The role of middle-income countries In discussing the inequalities associated with trade, the EUE perspective also considers the unique role played by middle-income countries. Here, middle-income countries are seen as playing a dual, ‘buffer’ role, in that they exploit lower income countries just as they, in turn, are exploited by wealthier nations (Arrighi and Drangel, 1986; Chase-Dunn, 1998; Clark, 2010; Mol, 2011). In the context of EUE, middle-income countries are often depicted as experiencing the highest levels of production-based emissions, as agents in these countries tend to have enough accumulated capital to invest in manufacturing, yet not enough for investing in cleaner, more efficient technologies (Roberts and Parks, 2007). Empirically, a number of studies support the idea that countries’ emissions and wealth are related according to an inverted U-shape curve (Burns et al., 1997; Grossman and Krueger, 1994; Jorgenson and Clark, 2012; Shafik and Bandyopadhyay, 1992), implying that middle-income countries experience the most degradation , while the wealthiest countries manage to either maintain or continue to expand their wealth without having the associated increases in environmental costs. With regards to CTIs and trade flows, studies similarly indicate that low- and middle-income countries tend to be net exporters of carbon, while high-income countries tend to be net importers of embodied carbon (e.g. Davis et al., 2011; Moran et al., 2013), and that the export ties from middle-income countries tend to increase more overtime than other income groups (e.g. Kim and Shin, 2002). Taken together, the theory and empirical research on the role of middle-income countries in unequal ecological exchanges suggest two additional hypotheses: H4a. Middle-income countries are more likely to become (or maintain being) net exporters of carbon. H4b. Middle-income countries that are net carbon exporters are more likely to increase their number of export ties overtime. 1.2.2. Regionalization and the Gravity Model Research on regionalization and the Gravity Model offer further insights into the potential drivers of trade tie formation. Regionalization refers to the tendency for international networks, based on trade and other kinds of international ties, to form within multinational regions of the world (e.g. Beckfield, 2010). The main goals of regionalization are to enable better cooperation among member countries, with economic (trade) interests being the main focus of such agreements (Kim and Shin, 2002). In the context of trade, establishing regional trade blocs is a common strategy for countries to maintain greater leverage with (or protection from) non-bloc, third parties (Bhagwati, 1993; Bhalla and Bhalla, 1997; Kim and Shin, 2002). Trade blocs that have emerged from regionalization processes include the Association of Southeast Asian Nations, established in 1967, the European Union, established in 1992, the Southern Cone Common Market (MERCO SUR), established in 1991, and the North American Free Trade Agreement, established in 1994. Empirically, studies have shown that global trade (Kali and Reyes,

89

2007; Kim and Shin, 2002) and other global relations (e.g. Beckfield, 2010) have become more regionalized overtime, even as global networks, in general, become more integrated as a whole (e.g. Iapadre and Tajoli, 2014; Schulz et al., 2001). The Gravity Model (GM) is a special operationalization of regionalization, which has developed independently within the physicals and trade literature. Here, the presence of trade from country i to j is a function of the economic size of countries (e.g. total GDP) and the geographical distance between them, such that senders of goods (i) are attracted to export their goods to larger economies (as proxied by that country’s total GDP), but this tendency decreases as geographical distance between i and j increases (Anderson, 2011; Tinbergen, 1962). Empirically, scholars have found evidence for the GM model in assessing the global structure of trade networks (e.g. ˜ and Fagiolo, 2011), trade network agreeBergstrand, 1989; Duenas ments (e.g. Baier and Bergstrand, 2007), and networks of foreign direct investment (Koskinen and Lomi, 2013). Taken together, the research on GM and regionalization suggest that economic size and geographical distance are important drivers of trade tie formation: H5a. Countries are more likely to form trade ties with larger economies. H5b. Yet this becomes less and less the case, the more geographically distant the two countries are. More recently, Koskinen and Lomi (2013) extended this line of research by considering how economic size and geographical distance predict tie formation, when controlling for a number of endogenous, network tendencies. Using exponential random graph models (ERGMs) as their modeling environment (Robins et al., 2007), Koskinen and Lomi (2013) found support for key GM arguments, while also uncovering some important, underlying structural features of trade networks, such as a propensity for countries to have positive feedbacks in their out-degrees and indegrees overtime, thus reflecting tendencies of countries to be either active and/or attractive investment nodes (Koskinen and Lomi, 2013, p. 544). This consideration for network endogeneity in predicting trade tie formation, moreover, has gained ground in other studies of global networks. For example, Manger et al. (2012) and Kinne (2014) both used stochastic actor-oriented models (SAOMs) to consider the evolution of trade agreements, controlling for key endogenous effects, and Manger and Pickup (2016) used SAOMs to explore the co-evolution of preferential trade agreement (PTA) ties among countries and countries’ level of democratization. A common finding across these examples was that international ties (be they trade-based or other) tended to subscribe to processes of triadic closure, more generally, and transitive closure in particular. Transitivity refers to a local process of tie formation, where given a tie from actor i to j, and from j to h, then there is a strong likelihood of a tie also forming from i to h. In every day parlance, transitivity refers to the scenario where ‘friends of my friends are my friends too.’ In the context of international trade, firms might be introduced to new partners through existing ones, or firms with common trade partners may be interested in the same markets (Matous and Yasuyuki, 2015). In addition to characterizing trade flows, transitivity has been found to characterize inter-firm networks (Uzzi, 1996; Uzzi and Lancaster, 2003) and economic networks more generally (Jackson and Rogers, 2005). H6. Countries will be more likely to form trade ties according to principles of transitivity. Taken together, Hypotheses 1–5 explain CTIs and trade networks as co-evolving phenomena. As such, trade ties and CTIs are being conceptualized as interlinked, interdependent processes,

90

C. Prell, K. Feng / Social Networks 46 (2016) 87–100

and thus require a modeling framework that can unravel how network processes affect CTI changes, and vice versa. The next section introduces such an approach. 2. Analytical approach To handle the co-evolution of trade ties and countries’ CTIs, we make use of the stochastic actor-oriented models (SAOM), developed by Snijders and colleagues (e.g. Snijders et al., 2010). Briefly, SAOMs comprise a stochastic modeling environment for analyzing longitudinal network data, and are designed to disentangle processes affecting network formation from the impacts of networks on actors’ (in this case countries’) attributes. As we are interested in exploring how countries’ CTIs co-evolve with trade networks, these models are well-suited for our aims. We make use of two models within SAOMs. The first handles changes in networks over time by taking into account the endogenous features of the network (e.g. the general tendency for actors to form reciprocal ties with one another, or to form ties leading to closed triads), as well as exogenous features involving actor attributes (e.g. the likelihood of a given country to form an export tie with a geographically distant other, or the tendency to form export ties with countries having higher GDPs). In this paper, we will refer to this as our ‘trade network change’ model. A second model handles the impacts of network patterns on actor attributes (in our case, countries’ CTI levels), for example, whether the presence of export ties lead to changes in a country’s CTI. In this paper, we refer to this second model as our ‘CTI change’ model. These two models are estimated simultaneously, so that changes in one set of processes (i.e. the trade network change model) can affect processes modeled by the second function (i.e. the CTI change model). In modeling changes in networks alongside changes in actor attributes, SAOMs manage to disentangle these two processes, while controlling for the endogenous tendencies that typically characterize network data (e.g. reciprocity). The estimation process involves a computer simulation in which the observed data serve as input. In our case, we used the Method of Moments option, as described by Ripley et al. (2015). Finally, we implemented SAOMs using version 1.1-232 of the Simulation Investigation for Empirical Network Analysis (SIENA) software program in R and followed procedures outlined for model fitting and convergence as specified in Ripley et al. (2015). 2.1. Data We have international trade data that were extracted from the EORA database, a multi-region input–output database that provides a time series of high resolution input–output (IO) tables with matching environmental accounts for 186 countries (Lenzen et al., 2012a,b). The MRIO tables from EORA contain trade flows, production, consumption and intermediate use of commodities and services for 26 sectors, both within and between 186 countries (see http://www.worldmrio.com for more details). The EORA MRIO tables were constructed based on six data sources: input–output (IO) tables from national statistical offices, IO compendia from Eurostat, IDE-JETRO, and OECD, the UN National Accounts Main Aggregates Database, the UN National Accounts Official Data, the UN Comtrade international trade database, the UN Servicetrade international trade database. Methodological detail on EORA MRIO construction can be found in Lenzen et al. (2013). As we were not able to find corresponding data for all our other variables of interest for all countries in the EORA MRIO dataset, we were limited to 172 of the 186 countries. The benefit of using the global MRIO data (combined with the CO2 data described below), is that we are able to calculate the amount of how CO2 that are used for the production of embodied

in imports and exports, along the entire global production chain. This accounting of emissions along supply chains is referred to as embodied, embedded or virtual emissions. As part of our intention here is to calculate countries CTIs, such detailed economic trade data are necessary for arriving at these figures. Finally, these trade data consist of three time periods, or ‘waves’ representing a 10-year period, these being years 2000, 2005, and 2010. Our CO2 emission data are at the economic sector level, and were also collected from the EORA database. The EORA database includes 35 types of environmental indictors covering air pollution, energy use, greenhouse gas emissions, water use, land occupation and many others. In the EORA database, all environmental data is systematically assigned to different sectors in different countries. In this study, we use their sectoral level CO2 emissions data for 26 sectors and 186 countries. Data on countries’ GDP and GDP per capita which were taken from the World Bank’s database (http:// www.worldbank.org). Finally, data on the geographic proximity of nations’ capitals was calculated based on the great circle distances between capital cities of the world (see Gleditsch and Ward, 2001 for similar data). 2.1.1. Data transformations For computing countries’ CTI levels, we did not transform these MRIO data in any way. However, to create matrices of international trade data, we transformed the EORA data in a few respects. First, for each wave of data, all sectors were summed to represent one, summed trade network for all 172 countries, for each wave (2000, 2005, 2010). Next, given that the SAOM environment can only handle dichotomous network data, we dichotomized each trade network, such that all values above the 95th percentile were counted as 1s and otherwise 0, resulting in a trade network of very ‘strong’ trade flows. For sensitivity analyses (presented in the Supplemental Material, and discussed in Section 4 of the article), we created trade networks with two different cut-off values: 90% (i.e. the upper 10th percentile of flows), and also 97.5% (i.e. the upper 2.5th percentile of flows) to reflect ‘moderately strong’ and ‘very strong’ flows, respectively. Finally, to handle skewness, we took the logarithmic form of GDP total and GDP per capita before analyses were done. 2.2. Computing CTI with multi-regional input–output analysis For computing the CTIs of countries, we used multiregional input–output (MRIO) analysis, an accounting procedure used on economic input–output (I–O) tables (our EORA dataset being an example of one such table). The interested but uninitiated reader is referred to a basic but comprehensive textbook on IO by Miller and Blair (2009). At its core is an accounting procedure relying on regional economic input–output (I–O) tables and inter-regional trade matrices, depicting the flows of money to and from each sector within and between the interlinked economies, thus revealing each sector’s entire supply chain. Although more technical details are offered in the SI, we offer a summary of our procedures here. First, we began with the MRIO technical coefficients matrix A, which contains all input–output relationships of the economy, and took the inverse (I − A), where I is a unit matrix. Intuitively, this inverse matrix (commonly referred to as the Leontief matrix) can be expressed as a sum of a power series, and each term in the series is a result of the recursive procedure happening when sectors need input from other sectors, which in turn need input from the next layer of sectors further upstream in the production chain, and so on, for an infinite number of turns. Next, we calculated the total input requirements to satisfy final demand (y) by multiplying the inverse matrix by final demand of a particular consumption item in a given country. Next, to calculate

C. Prell, K. Feng / Social Networks 46 (2016) 87–100

the emissions embodied in import of region s, we use the following calculation: imp

CO2

= k∼s (I − A)−1 y·s

(1)

imp CO2

is the total embodied emissions in import of region where s; k∼s is a vector of sectoral CO2 emission coefficients with zeros for the sectoral emission coefficients of region s; y·s is the final demand vector of region s. Eq. (2) is used to calculate the emissions embodied in export: exp

CO2

= ks (I − A)−1 ys·

(2)

exp CO2

where is the total embodied emissions in export of region s; ks is a vector of sectoral CO2 emission coefficients with the sectoral emission coefficients for region s but zeros for all other regions; ys· is a vector of total sectoral final demand of other regions but excluding the final demand of region s. Finally, to arrive at the CTI, we used the ratio representation (CO2 exports)/(CO2 imports), and took the natural log of this calculation, such that ln(CTI) = ln(CO2 exports/CO2 imports). Advantages of this measure include: (1) the size effect of countries’ population is automatically removed, (2) the logged transformation smooths out skewness in the resulting data vector, and finally, that (3) the ratio carries the same meaning as traditional net export measures, with the only difference being that “CO2 export/CO2 import > 1 means net exporter and “CO2 export/CO2 import < 1 indicates a country is a net importer of CO2 . 2.3. Network effects for testing Hypotheses 1–6 For Hypotheses 1 and 4a, we made use of the CTI change model, which models impacts to countries’ CTIs overtime. For Hypotheses 1, we specified the total alters’ GDP per capital effect on ego’s CTI. Here, a positive parameter implies that the wealth levels of alters to whom a focal country exports positively impacts that focal country’s CTI level. For Hypothesis 4a, i.e. to test whether middle-income countries were more likely to have positive carbon imbalances, we included three dummy variables for countries’ income classifications, in particular, upper middle-, lower middle-, and low-income countries (the remaining high income category is the reference variable, and hence, was not included in the model).4 These income categories, moreover, were taken from the World Bank’s Income Classifications, i.e. http://data.worldbank.org/news/2015-countryclassifications. For the remaining Hypotheses (2, 3, 4b, 5a, 5b, and 6) we specified effects in the trade network change model, which models impacts to trade network features overtime. For Hypothesis 2, we specified an CTI-ego effect, where a positive parameter is hypothesized, indicating the tendency of those countries that are net exporters of carbon to form more export (outgoing) ties. For Hypothesis 3, we specified an interaction effect called CTIego effect × GDP per capita-alter effect, where a resulting positive coefficient is hypothesized, indicating a tendency for net carbon exporting countries to form export ties to those with higher levels of wealth. For Hypothesis 4b, we specified the interaction terms consisting of Income ego × CTI ego, where Income refers to three dummy variables for countries’ income classification (in our case, upper middle-, lower middle-, and low-income countries). Here, a positive, significant effect is hypothesized for the interaction terms holding a middle-income dummy variable (either upper- or

4 Please note, Table SM4 in the Supplemental Material shows models in which Hypotheses 4a and 4b were tested by using (instead of dummy variables for income categories) by including a quadratic term for GDPpc. These results largely replicate those found here, in the main body text, but their interpretation is not as intuitive, and hence, we opted to not present them here.

91

lower-middle), indicating that middle-income countries that are also net exporters of carbon generate more export ties overtime. For Hypothesis 5a, we specified GDP total alter effect, where a positive parameter is hypothesized, which indicates countries tending to export to larger economies. For Hypothesis 5b, we created the interaction term GDP total alter × distance effect, where a negative parameter is hypothesized, indicating that countries’ tendency to export to larger economies decreases with geographical distance. For Hypothesis 6, we specified two endogenous network terms measuring transitive closure. These include the transitive triplets and transitive ties effects, where positive parameters are hypothesized for both, indicating tendencies for forming transitive triads.

2.4. Controls for trade tie formation and changes in CTI levels The interdependencies of network data require a number of controls. First, as Hypotheses 1, 3, and 4b involve interaction effects, we also included the primary terms for these interaction effects. Thus, for Hypothesis 1, we included the outdegree effect in our CTI change model, which measures the tendency of countries’ CTI levels to change in response to the number of export ties they hold (i.e. their level of outdegree centrality), and the control variable GDP per capita, to see the extent to which countries’ wealth impacts their CTI level(s). Past research has shown that countries more central to trade networks tend to experience more production-based carbon (Prew, 2010), yet they also tend to trigger more emissions along global supply chains through their consumption habits (Prell et al., 2014). For Hypothesis 3, we included the GDP pc alter as the primary term, where positive parameters indicate a tendency of countries to form export ties with wealthy others. For Hypothesis 4b, we included the distance effect as a primary term, where a negative parameter value indicates that countries are less likely to form trade ties with others at further distances. Other attribute-based controls we included were GDP total ego and GDP per capita ego effects, the CTI alter effect, and homophily effects for countries’ CTI, GDP total and GDP per capita. These additional controls were added, as research suggests that they can spuriously create hypothesized patterns (e.g. Koskinen and Lomi, 2013; Schaefer, 2013). In addition to attribute-based effects, we controlled for a number of endogenous network tendencies that affect tie formation in general, and which may also result in biased estimates of other specified effects if not included in the model (e.g. Mouw and Entwisle, 2006; Snijders et al., 2010). Our selection of these endogenous effects was aided by the use of goodness of fit tests (Lospinoso, 2012) found in the RSiena package, and explained in its manual (Ripley et al., 2015). These tests compare the average values of simulated, auxiliary statistics with values in the observed data, and if the distribution of these average scores corresponds closely to observed values, then the fit of the model is deemed good. Thus, through a process of trial and error using these GOF tests, we developed model specifications for endogenous network effects with the best fit. The endogenous effects we specified include: (i) reciprocity, which is the tendency for ties to be mutual, (ii) in- and out-degree popularity, where a positive parameter indicates the likelihood of country i to form a new export tie to some country j, as the number of ties held by j increases (i.e., sociologically referring to the “Matthew Effect”), and (iii) the 3-cycles effect, which indicates the tendency of actors to form triadic structures such that i → j, j → h, and h → i. Such a cycle effect is considered a tendency toward intransitivity, and is not very prominent in empirical networks (Holland and Leinhardt, 1970, 1971, 1972; Skvoretz, 1990), thus leading to the expectation of a negative coefficient. A full listing of these endogenous network effects can be found in Table 1.

92

C. Prell, K. Feng / Social Networks 46 (2016) 87–100

Table 1 All effects summarized. Endogenous network effects Outdegree effect Reciprocity Transitive ties Transitive triplets Number of actors at distance 2 3-cycles Dense triads

 j xij j xij xji j xij maxh (xih xhj ) j xij xih xhj

# {jxij = 0, maxh (xih xhj )} >0

 x x x j ij jh hi j,h

xij I

xij + xji + xih + xhi + xhj + xjh



≥ c,

where c is either 5 of 6.

Indegree popularity Indegree popularity (sqrt) Outdegree popularity Outdegree popularity (sqrt) Indegree activity Indegree activity (sqrt) Outdegree activity Outdegree activity (sqrt) Network formation effects covariate similarity covariate-alter covariate-ego Dyadic geographic distance

 x x+j j ij  j xij x+j x xj+ j ij  j

xij

xj+

xi+ x+i √ xi+ x+i 2 xi+

√ xi+ xi+

 j xij I{vni = vnj } xij vnj j  vni x  j ij j

Effects Impacting CTI

  j xij vj /x+i

Alter’s GDP per capita total

zi

Outdegree effect on CTI GDPpc and GDP total effects on CTI

zi j xij zi vni , where n = 1, 2

Rate and shape effects Rate (actor position) Rate (actor covariates)



xij wij − mean(wij )

net m , where m represents each time wave  h

˛h vhi

CTI linear shape

zi , where zi denotes the value of the dependent attribute of actor i

CTI quadratic shape

zi2 where zi denotes the value of the dependent attribute of actor i

2.5. Additional controls Additional, default controls built into SAOMs include the rate effect for both tie formation and changes to CTIs. For tie formation, the rate effect indicates the extent to which actors have opportunities to change their ties, and for CTIs, the CTI rate effect controls for the opportunities to change CTI values from one time wave to the next. The linear shape effect measures the overall tendency toward high or low CTI values; here, a negative parameter indicates that the majority of countries scored below the CTI mean, and a positive parameter indicates the opposite. The quadratic shape effect controls the effect of a country’s CTI value on itself, e.g. when the parameter is negative, this implies the tendency of the CTI value to decrease overtime, when the value was originally high. Conversely, when the coefficient is positive, this reflects the tendency for countries to score at the extreme ends of the scale for CTI values (Snijders et al., 2010). The effects outlined in Table 1 are brought into the SAOM environment, as explained in the next section.

3. Results We begin with descriptive statistics before showing the fuller model results. Tables 2 and 3 show descriptive statistics for the networks and country attributes.

We note that the Jaccard coefficient scores indicate very little change in the network between successive waves. Given that little change in the network would bias our models toward the null Hypothesis (no change in the network overtime), this decreases the likelihood for detecting significant tendencies in network dynamics, making our tests quite conservative. We also note that GDP total and GDP per capita are moderately correlated, and for this reason, we present Models (see Table 4 below) in which we separate out GDP total from GDP per capita. Next, Fig. 2 shows a digraph of the network in the first wave (2000) and third wave (2010). In Fig. 2, the size of nodes indicate outdegree centrality levels, with larger nodes indicating higher levels of centrality. We note that many of the large nodes represent ‘developed’ countries, e.g. France, Germany, Italy, Japan, United Kingdom and the US. China represents another large node, i.e. country with many export ties, and moreover, the outdegree centrality of China increases the most of all the other countries in our sample from year 2000 to 2010. This increase in outdegree centrality (number of export ties) most likely reflects that, in the time period of our study, China became a member of the World Trade Organization, as well as a member of a number of additional regional trade agreements (see Antkiewicz and Whalley, 2005) resulting in more trade partners globally. We also note that the color of nodes in Fig. 2 reflect ranging CTI levels, with black representing higher CTI levels (all levels greater than or equal to 0.051), gray representing CTI levels ranging between 0.05 and −0.045, and white representing CTI levels less than −0.045. As such, the black nodes represent ‘net exporters’ of carbon, the white nodes represent ‘net importers’ of carbon, and the gray nodes represent countries that more or less ‘break even’ in terms of their carbon trade imbalance (their CTI hovers around 0). Fig. 2 thus shows that many of the ‘developed’ countries are net importers of carbon, and that they share this CTI status with quite a large number of relatively isolated, less-developed countries, e.g. Cuba, Latvia, Madagascar, and Uganda. As poorer, less-developed countries tend to be less industrialized, whatever industrialized goods are consumed in these countries would likely be imported from more developed societies (e.g. Chase-Dunn, 1998). The gray nodes in Fig. 2 represent countries whose CTIs hover around zero, indicating that their carbon exports and imports are balanced. Here, we note that some of the developed countries that fall into this category in year 2000, e.g. Australia, Belgium, the Netherlands, and New Zealand, but by year 2010, most of these have become net importers of carbon, represented as white nodes in the 2010 digraph. With regards to net exporters of carbon, Fig. 2 shows China, India, Russia, Thailand, South Africa, and Canada as being the more prominent, central actors in this area of the network. These countries have fuel mixes that include a high use of coal (for China, India, and Russia) and/or they export carbon-intensive materials such as oil, as in the case of Canada (Brizga et al., 2013, 2014). We also see that a number of smaller nodes that are net exporters include many Eastern European countries, e.g. Romania, Belarus, and the Ukraine, where environmental standards tend to be lower and their carbon intensity of production tends to be higher (Brizga et al., 2013, 2014). Taken together, Fig. 2 offers some heuristic support for the idea that middle-income and/or emerging economies tend to be net exporters of carbon, while developed, wealthy countries tend to be net importers (Moran et al., 2013; Prell et al., 2014). With regards to edges, Fig. 2 displays reciprocal trade ties as darker edges, while lighter edges represent unilateral ties. We note that the larger (more central), ‘white’ nodes, representing developed countries, tend to all be reciprocally tied to one another, reflecting arguments stating that developed countries tend to form a cohesive, well-integrated ‘block’ via their trade tie patterns (e.g.

C. Prell, K. Feng / Social Networks 46 (2016) 87–100

93

Table 2 Trade network data for three time waves. Wave 1

Trade network

Wave 2

Wave 3

Ties

Density

Ties

Density

Ties

Density

1479

0.050

1478

0.050

1478

0.050

Note: The Jaccard coefficient ranged from 0.848 to 0.869 across both periods (between wave 1 and 2, and between wave 2 and 3). As this coefficient expresses the amount of change between two consecutive waves within a range from 0 to 1 (with 1 representing no change), our Jaccard values are quite high, indicating little change between waves.

Table 3 Country attributes (GDP per capita, GDP total, and carbon trade imbalance).

GDP per capita Carbon trade imbalance GDP total a

Na

Na missing

M

SD

GDPpc

CTI

500 504 494

16 12 22

8.09 4.56 8.71

(1.63) (1.04) (2.33)

0.22 0.56

0.33

Across three waves (years 2000, 2005, and 2010).

Clark, 2010; Mahutga, 2006). China also appears to be part of this cohesive block, although it does not share the same CTI profile of more developed countries (e.g. see Prell et al., 2014 who note and discuss a similar pattern). In contrast, the smaller nodes hold few to no ties with other countries, and very few of these are reciprocal. Taken together, Fig. 2 shows an interesting depiction of the distribution of CTIs and trade ties among countries in our sample. More developed countries tend to be highly central, net importers of carbon, and reciprocally tied with one another; yet they share (i) similar CTI levels with a high number of poorer, less-integrated countries, and (ii) a similar centrality level as China, a big net exporter of carbon. In addition, Fig. 2 offers support to arguments and past research suggesting that firms and other economic agents in middle-income, developing countries are more likely to acquire a comparative advantage in carbon-intensive products and/or become net exporters of carbon in efforts to further their economic development (e.g. Roberts and Parks, 2007). This appears especially true with regards to China: China maintains its high CTI level from 2000 to 2010, yet its number of export ties grows substantially, resulting in China becoming the largest ‘node’ in this network, in terms of outdegree centrality, while maintaining its status as a net carbon exporter. Moving on to our co-evolution model results, Tables 4 and 5 present the findings for our ‘trade network change’ and ‘CTI change’ models. In constructing these models, we first separated the GDP per capita and GDP total covariate effects from one another, as the two are rather strongly correlated (r = 0.56). In addition, for space reasons, we created a separate table (Table 5) for testing hypotheses dealing specifically with middle-income countries (Hypotheses 4a and 4b), which involved the use of dummy variables. Starting with Model 1, we first note that the rate effects indicate slightly more tie changes occurring in the first period than the second. The negative, significant coefficient for the outdegree effect indicates that countries tend to avoid forming too many export ties overtime. With regards to other default controls, the rate parameters in the CTI change model show countries tending to change their CTI levels slightly more in period two than in period one. In addition, both the linear and quadratic coefficient are negative and significant, with the later being slightly higher in value than the former. This implies that there is a downward drive for changing CTI levels. As these findings for the default controls remain largely the same across the remaining models (Models 1b–4), we will not comment on them further here. We now begin discussing results for our hypothesized effects. For Model 1, support is found for Hypotheses 1, 2 and 3. First, we find a positive, significant tendency of alters’ GDP per capita impacting countries’ CTI levels: countries’ tendency toward

becoming net exporters of carbon increase in response to having more export ties with wealthier alters, a finding that supports Hypothesis 1. We also see support for Hypothesis 2: countries with higher CTI levels tend to form more export ties overtime, as indicated by the significant, positive parameter for the CTI ego effect. Finally, we see support for Hypothesis 3: countries with higher CTI levels tend to form more export ties with wealthier alters, as indicated by the positive, significant parameter for the CTI ego × GDP per capita alter effect. The basic model for GDP total effects, found in Model 3, shows support for Hypotheses 5a and 5b: here we see that country i is more likely to export to larger economies, as indicated by the GDP total alter effect, and in addition, this tendency to connect to larger economies decreases as geographical distance increases, as indicated by the negative, significant parameter value for the GDP total alter × distance effect. Taken together, the two basic models (Model 1 and 3) suggest that the patterns posited in Hypotheses 1–3 and 5a–b are present in our data. However, once we turn to more complex models that control for underlying, related tendencies, we see that some of these patterns have changed. Starting with Model 2, we note that Hypothesis 1 is no longer supported: country i’s CTI level no longer appears to change in response to the GDP per capita of its alters, but rather, to i’s own GDP per capita. That is, i’s wealth positively predicts the likelihood of i’s CTI increasing. Or, in other words, as countries’ wealth levels increase, so does the likelihood for becoming a net exporter of carbon. To interpret this finding, we turn to Fig. 2, and note that four net exporting countries shown in Fig. 2 are BRICS countries (Brazil, Russia, India, China, and South Africa). Given that these economies are in similar stages of development (i.e. they are considered emerging economies), and that all are projected to be among the most dominant economies by 2050 (FinancialTimes 6 November, 2006), their economic ascendency may help explain this trend we are picking up in our data. For example, a comparison of changes in CTIs and GDP per capita scores between 2000 and 2010 shows that the BRICS countries (and other, middle-income, emerging economies shown in Fig. 2, such as many of the Eastern European countries) all experienced upward increases in GDP per capita, although few increased their CTI scores during that time period (please also see Table SM3 in the Supplemental Material, or SM). As such, increases in GDP per capita may be enabling these countries to maintain productivity of carbon-intensive products, and/or increase their productivity while maintaining the same CTI level. This interpretation is further strengthened when we look at the findings in Model 2 for Hypothesis 2. Here, we still see a positive, significant parameter for the CTI ego effect, suggesting countries

94

C. Prell, K. Feng / Social Networks 46 (2016) 87–100

Fig. 2. Digraphs of trade network in 2000 and 2010.

that are net exporters of carbon are expanding their number of export ties overtime, even when controlling for other network tendencies. Again, focusing on BRICS countries, we see that the biggest increases in outdegree scores occurred with these countries (again please also refer to the SM): China, a middle-income country with a quickly developing economy and high CTI, experienced an increase of 43 export ties between the time periods of 2000 and 2010, and such a pattern is similar for India (+13 ties) and Russia (+8). Thus, middle-income, developing countries that attain a comparative advantage with regards to carbon-intensive products (with

BRICS countries being foremost examples) increase their number of export ties overtime. Taken together, the results pertaining to Hypotheses 1 and 2 in Model 2 suggest that CTIs, trade ties and wealth appear to be interlinked, such that middle-income, developing countries (e.g. BRICS) first acquire a comparative advantage in carbonintensive commodities, which enables them to expand their reach in the global market (H2); then, through these increased exports, countries’ economies grow (i.e. their GDP per capita increases), and this increase in economic growth seems, in turn, to enable

C. Prell, K. Feng / Social Networks 46 (2016) 87–100

95

Table 4 All results for trade network and CTB change models. H#

H6 H6 H6

H2 H5a

H5b H3

H1

Trade changes

Model 1

TRADE rate (period 1) TRADE rate (period 2) Outdegree (density) Reciprocity Transitive triplets 3-cycles Transitive ties # Actors at distance 2 Indegree – popularity Indegree – popularity (sqrt) Outdegree – popularity Outdegree – popularity (sqrt) Indegree – activity (sqrt) Outdegree – activity Outdegree – activity (sqrt) Distance CTI alter CTI ego CTI similarity GDPtot alter GDPtot ego GDPtot similarity GDPpc alter GDPpc ego GDPpc similarity Int. GDPtot alter × dist Int. CTI ego × GDPpc alter

1.952* 1.773* −2.731*

(0.145) (0.135) (0.148)

1.688*

(0.249)

CTI changes

Model 1

Rate CTI (period 1) Rate CTI (period 2) CTI linear shape CTI quadratic shape CTI outdegree CTI: effect from GDPtot CTI: effect from GDPpc CTI: alter’s (TRADE) GDPpc total

0.465* 0.686* −1.012* −0.429*

(0.075) (0.102) (0.293) (0.143)

0.022*

(0.009)

0.084*

Model 2

(0.025)

2.539* 2.319* −7.714* 0.818* 0.100* −0.134* 1.047* −0.002 0.086 −0.093 −0.019 0.118 0.191 0.026* 0.226 −0.910* 0.314 0.802* 0.144

Model 3 (0.279) (0.190) (1.288) (0.303) (0.022) (0.049) (0.259) (0.008) (0.084) (1.351) (0.105) (1.471) (0.679) (0.013) (0.369) (0.118) (0.160) (0.239) (0.753)

−0.245* −0.491* 1.541*

(0.111) (0.083) (0.723)

−0.113*

(0.038)

Model 2 (0.080) (0.104) (0.347) (0.189) (0.027)

0.299* 0.004

(0.152) (0.025)

countries to maintain their comparative advantage – for example, through investments in technology and/or outsourcing that will spur greater carbon productivity at home – while not increasing (and in some cases decreasing) their CTI levels. Moving onto Hypothesis 3, the interaction term in Model 2, which tests this Hypothesis (CTI ego × GDPpc alter effect) now contains a significant, negative coefficient, implying that countries with high CTIS are now less likely to export to wealthy others. Such a pattern is contrary to what was found in Model 1, and to what we hypothesized. Again, focusing on BRICS countries may offer an interpretation: as these countries have the biggest increases in outdegree centralities between 2000 and 2010 (see the SM), the new trading partners they pick up tend not to be with wealthy, developed countries (to whom they were already tied in year 2000), but rather, with smaller, poorer economies. Such a strategy is not uncommon among emerging economies (Aulakh, 2006; Jain, 2006). Thus, BRICS and other developing countries may indeed export to wealthy countries, but in the time frame of our study, the majority of new export ties formed tend to be with less wealthy markets. Model 2 also includes transitivity related effects for testing Hypothesis 6: we find evidence that new strong trade ties tend to form according to principles of transitivity, that is, sectors in country i seek to form ties with sectors in country j when there is a third sector in country h shared in common. Thus, Hypothesis 6 is also supported. With regards to the other, remaining endogenous network and covariate effects included as controls in Model 2, we first note that none of the in- and out-degree activity or popularity effects are significant (although all were necessary for achieving a good GOF test score and distribution). However, all three GDP per capita effects

Model 5

(0.154) (0.136) (0.086)

2.860* 2.268* −7.423* 1.020 0.045* −0.122* 1.339* 0.005 0.146* −0.812 −0.131 1.638 −0.421 0.000 0.430* −0.928*

(0.231) (0.193) (0.774) (0.267) (0.025) (0.040) (0.334) (0.009) (0.073) (1.021) (0.092) (1.378) (0.709) (0.010) (0.191) (0.123)

0.243*

(0.022)

−0.492 0.649 0.652

(0.347) (0.588) (1.506)

−0.114*

(0.022)

Model 3

0.454* 0.682* −0.946* −0.494* 0.004

Model 4

2.035* 1.657* −2.662*

0.138*

Model 4

(0.044)

2.627* 2.355* −7.963* 0.851* 0.074* −0.139* 0.872* 0.008 0.103* −0.310 −0.072 0.827 −0.318 0.009 0.400 −1.058* 0.253* 0.875* 0.210 −0.135 0.853 0.291 −0.269* −0.654* 1.508* 0.057 −0.099*

(0.292) (0.187) (0.804) (0.229) (0.020) (0.033) (0.283) (0.010) (0.046) (0.698) (0.080) (1.253) (1.009) (0.014) (0.283) (0.164) (0.070) (0.414) (0.809) (0.335) (0.867) (1.081) (0.066) (0.252) (0.676) (0.047) (0.037)

Model 5 0.457* 0.696* −0.780* −0.558* −0.006 0.280 0.251* −0.009

(0.080) (0.108) (0.314) (0.212) (0.032) (0.177) (0.149) (0.027)

are significant: The negative coefficients for the GDPpc alter and GDPpc ego effects indicate that poorer countries are more likely to increase both their import and export ties overtime. Again, given our earlier comments regarding the high network activity of BRICS countries, in particular, and developing countries in general, seeing this tendency is not surprising. The positive coefficient for the GDP per capita similarity effect indicates that countries of similar wealth export to one another; when looking at Fig. 2, we see that wealthy countries tend to be (i) the largest nodes (implying high outdegree centrality) and (ii) reciprocally tied to one another. This tendency for developed, wealthy countries to form a cohesive block in global trade networks is not uncommon (e.g. Clark, 2010; Mahutga, 2006), and may help explain the finding we see here. Finally, neither the CTI alter- or similarity effects are significant, implying that CTI levels do not seem to play a role in either attracting incoming trade ties or in forming ties with others of similar CTI levels. Models 3 and 4 focus on the GDP total and Distance effects in the trade change model, as a means for testing Hypotheses 5a and 5b, both of which test regionalization and GM arguments. Starting with Model 3, which is the simpler model, we note that both Hypotheses 5a and 5b are supported: countries tend to form ties with larger economies (as indicated by the positive, significant coefficient for the GDP total alter effect), and this tendency tends to decrease with geographical distance (as indicated by the negative, significant coefficient for the GDP total alter × distance effect). Model 4 includes all network endogenous effects, GDP total effects as controls, and the distance effect as a main control for the GDP total alter × distance effect. In this more complex model, all the GDP total alter effect is insignificant, and the GDP total alter × distance effect has acquired a positive, significant parameter. As such, these new set of results

96

C. Prell, K. Feng / Social Networks 46 (2016) 87–100

indicate that Hypothesis 5a is thus no longer supported, i.e. that countries do not tend to export to larger economies. However, the distance effect has a negative, significant parameter, indicating that geographic distance does appear to play a role, i.e. that countries tend to export to nearer neighbors (Tobler, 1970). Finally, the interaction term GDP total x distance now has acquired a positive, significant coefficient, which is contrary to Model 5. As such, this suggests that now, controlling for additional network tendencies, countries are more likely to form export ties with larger economies at further distances. Said differently, there is something about the network patterns of these countries that, when controlled for, reveal a different kind of relationship between distance and economic size when predicting trade tie formation. One explanation may be that, similar to comments made earlier, middle-income, emerging economies are expanding their markets to even farther, more distant lands. As such, the ‘first law of geography’, which states that ‘everything is related to everything else, but near things are more related than distant things’ (Tobler, 1970, p. 236), does not apply to these kinds of countries. These fast growing, developing economies depend on increasing their exports for further developing their economies (Pao and Tsai, 2010). China, in particular, by its recent inclusion in the WTO, has used that opportunity to extend its reach world-wide, diversifying its exports and increasing its penetration into large, industrialized countries such as the US, UK, and Germany (Rumbaugh and Blancher, 2004). Yet similar increases in exports have been experienced in recent years by other emerging economies, where analysts argue that exporters’ strategy for increased economic rewards are (in part) linked to increasing their domestic economies-of-scale while simultaneously increasing their number of foreign markets, both near and far (Aulakh, 2006). This contrasts with developed economies, who already occupy a core, central position in the global trade network (e.g. Clark, 2010; Mahutga, 2006), and who tend to specialize in exporting more sophisticated, high-value commodities (Edwards and Lawrence, 2010). As such, these developed economies need not expand their markets in order to maintain their strategic standing in the global trade network. Thus, the network patterns we are seeing here – although counter to the trade narratives presented by GM and, to some extent, EUE theory – make sense for the time period in which we are examining, i.e. a time in which emerging economies are expanding their markets and growing in wealth. Our last model to discuss in Table 4 is Model 5, which includes both GDP per capita and GDP total effects. Although the two forms of GDP show some collinearity (r = 0.56), we note that past studies have included both terms in models to account for both wealth and size of an economy (e.g. Fratianni and Kang, 2006; Reinert, 2009). Comparing this model with Models 2 and 4 shows that only one major change has happened as a result of including both sets of covariate terms in the same model: the GDP total alter × distance effect has become insignificant, indicating that there is now no support for either Hypothesis 5a or 5b. As such, in this final model, we have no support for either Hypothesis pertaining to regionalization and GM arguments. Finally, we note that the endogenous network terms measuring for transitivity are significant across Models 2, 4, and 5, thus offering consistent support for Hypothesis 6. Thus far, our interpretation of our findings in Table 4 has heavily relied upon our intuitive understanding of the role that middle-income, developing countries play in the larger global trade network. Table 5 looks more closely at these middle-income countries: In Table 5, we test hypotheses considering changes to carbon trade imbalances, or CTIs (Hypothesis 4a) and changes to export trade ties (Hypothesis 4b). Beginning with Model 6, our simplest model, we see that all countries that are net exporters of carbon tend to form export ties overtime (again, confirming support for

H2), and that both lower middle and low income countries form significantly fewer ties than high-income countries, as indicated by the significant negative coefficient for these two Income ego effects. In contrast, upper middle-income countries do not differ in any significant way from high-income ones, as indicated by the nonsignificant parameter for the UpMid ego effect. In Model 7, we see interaction terms used to test whether middle-income countries that are also net exporters of carbon are more likely to increase their export ties overtime (Hypothesis 4b). We see support for this tendency, as indicated by the positive, significant coefficient for the CTI ego × UpMid ego effect. Here, being a net exporter of carbon has nearly double the effect on upper middle-income countries than high-income ones, in terms of predicting the formation of export ties. In contrast, the low- and lower middle-income countries appear to be no different from the high-income ones, as indicated by the insignificant coefficients for the CTI ego × LowMid ego and CTI ego ×Low egoeffects. Exporting carbon, in other words, has the same effect for all countries, except for the upper middle, where being a net carbon exporter seems to be a very strong predictor of increasing export ties overtime. Thus, Model 7 lends support for Hypothesis 4b. In Model 8, we include terms in the CTI change model to test Hypothesis 4a. Here, we find that lower income (i.e. the lower middle and low income) countries have a strong tendency to not be net exporters of carbon, or said differently, they tend to import more carbon relative to high-income countries. In addition, there is no significant result for upper middle-income countries. Hence, we conclude our data show no support for Hypothesis 4a. Such a finding, moreover, supports the linear relationship established earlier, in Table 4, between GDP per capita and CTI. We note, too, that high-income countries include big emitters such as Venezuela, Saudi Arabia, Canada and Russia, who, consequently, also have positive CTIs (see Fig. 2). In short, wealth and economic development do not appear, in our dataset, to transform all countries from a ‘net carbon exporter’ status to ‘net carbon importer’ one, as other research indicates (e.g. Moran et al., 2013). In addition, we note that the CTI ego × UpMid ego effect in Model 8, although still positive, is no longer significant. In Model 9 and 10, the coefficients for the interaction terms in the Trade change model have switched from positive to negative. In addition, the Income ego coefficients have changed: the UpMid ego effect has become positive and significant, and has actually increased in strength from Model 9 to Model 10. This is similarly true for lower middle-income countries, as indicated by the positive, significant coefficients for the LowMid ego effect in Models 9 and 10. Meanwhile, CTI egoeffect remains positive and significant across both models. Collectively, the models in Table 5 indicate that, once controlling for a number of endogenous, network tendencies, a different set of patterns arise between countries’ wealth, CTI and trade tie formation. High-income countries are the ones in our sample where having a positive CTI has the most impact in predicting increases in export ties overtime. For all the other income categories, being a net carbon exporter has basically the same effect. Agents in middle-income countries are, indeed, working hard to expand their number of trade partners (as indicated by the positive significant coefficients for the UpMid ego and LowMid ego effects), but such processes are not linked to whether these countries are net carbon exporters. Another way to interpret these findings in Table 5 is to say that Model 7 supports theory and past research suggesting that middle-income countries are more likely to build their economies via acquiring a comparative advantage in carbon intensive products for export elsewhere (Roberts and Parks, 2007). Yet additional models in Table 5 expand on this theory, suggesting that an underlying reason for seeing this pattern in Model 7 has to do with

C. Prell, K. Feng / Social Networks 46 (2016) 87–100

97

Table 5 Results for trade network and CTI change models. H#

H6 H6 H6

H2

H4b H4b H4b

H4a H4a H4a

Trade changes

Model 6

TRADE rate (period 1) TRADE rate (period 2) Outdegree (density) Reciprocity Transitive triplets 3-cycles Transitive ties # Actors at distance 2 Indegree – popularity Indegree – popularity (sqrt) Outdegree – popularity Outdegree – popularity (sqrt) Indegree – activity (sqrt) Outdegree – activity Outdegree – activity (sqrt) CTI alter CTI ego CTI similarity UpMid alter UpMid ego UpMid same LowMid alter LowMid ego LowMid same Low alter Low ego Low same CTI ego × UpMid ego CTI ego × LowMid ego CTI ego × Low ego

2.130* 1.848* −3.073*

(0.147) (0.145) (0.126)

Model 7 2.151* 1.847* −3.059*

(0.154) (0.146) (0.115)

2.161* 1.871* −3.090*

(0.157) (0.151) (0.128)

2.619* 2.270* −7.504* 1.365* 0.101* −0.199* 1.067* 0.018* 0.139* −0.958 −0.126 1.183 0.024 0.006 0.561

(0.282) (0.241) (0.665) (0.195) (0.021) (0.043) (0.274) (0.009) (0.069) (0.975) (0.069) (0.967) (0.560) (0.016) (0.400)

0.855*

(0.100)

0.846*

(0.114)

0.988*

(0.186)

0.992*

(0.345)

0.015

(0.204)

−0.355

(0.270)

−0.351

(0.327)

1.874*

(0.851)

−1.008*

(0.251)

−1.029*

(0.269)

−0.971*

(0.279)

1.596*

(0.932)

−4.346*

(0.573)

−4.349*

(0.553)

−4.359*

(0.569)

−0.089

(0.978)

0.875* 0.242 0.302

(0.407) (0.268) (0.281)

1.162 0.263 0.253

(0.650) (0.328) (0.314)

−0.763 −1.254 −1.338

(0.727) (0.804) (1.068)

CTI changes

Model 6

Model 7

Rate CTI (period 1) Rate CTI (period 2) CTI linear shape CTI quadratic shape CTI: from UpMid CTI: from LowMid CTI: from Low

particular network tendencies of these middle-income countries. Middle-income countries tend to build more trade ties, and to do so according to principles of transitivity. Given that this time period of our study involves critical historical events, such as China entering the WTO, and the expansion of trade agreements more generally among emerging economies, seeing these kinds of network patterns is not so surprising. What is perhaps an interesting twist, however, is that once we control for these patterns, then the impact of being a net carbon exporter is wiped away for middle-income countries’ tie formation, whereas this impact still remains for high-income ones. As we noted earlier, high-income countries include countries such as Saudi Arabia, which rely heavily on fuel exports. Thus, these kinds of wealthy, fossil-fuel intensive countries are most likely the ones driving the results we see here. 4. Discussion and conclusion The models presented in Tables 4 and 5 offer an interesting portrayal of the dynamics of trade and carbon trade imbalances. First, our analyses revealed a number of processes affecting trade tie formation, some of which run counter to the narratives presented by both regionalization and EUE perspectives. With regards to regionalization and the GM in particular, economic size does not appear to play a role in countries’ tendencies to form export ties, when additional country-level attributes are considered (namely, GDP per capita- and CTI effects). Including these additional explanatory tendencies seems to cloak the role of economic size in

Model 8

Model 9

Model 8 0.472* 0.658* −0.819* −0.465* −0.388 −1.238* −1.417*

Model 10

Model 9 (0.073) (0.096) (0.274) (0.197) (0.402) (0.601) (0.732)

0.471* 0.671* −0.818* −0.472* −0.390 −1.227* −1.429*

2.463* 2.177* −8.697* 1.548* 0.086* −0.179* 1.118* 0.021* 0.127 −0.886 −0.132 1.381 −0.437 0.004 0.959* 0.317* 1.248* −1.557 0.601* 2.582* 0.355* 1.061* 1.985* −0.363* 0.850 −0.384 −0.759 −1.148 −1.790* −1.973*

(0.228) (0.193) (0.906) (0.283) (0.023) (0.047) (0.335) (0.012) (0.078) (1.009) (0.086) (1.094) (0.660) (0.017) (0.554) (0.086) (0.574) (0.934) (0.238) (1.140) (0.145) (0.523) (1.171) (0.215) (3.493) (3.599) (3.133) (0.781) (0.994) (1.107)

Model 10 (0.080) (0.091) (0.279) (0.190) (0.404) (0.596) (0.712)

0.468* 0.668* −0.808* −0.464* −0.381 −1.200* −1.405*

(0.085) (0.111) (0.257) (0.183) (0.445) (0.600) (0.716)

predicting the likelihood of a country to form an export tie with another (larger) economy. Thus, one lesson here is that the patterns found in GM studies may arise from underlying, network mechanisms that previous research has overlooked. Chief among these include the tendency of countries with higher CTI values to form more export ties overtime, a point we will return to shortly. In contrast, when network endogenous terms are included, economic size interacts with geographical distance in an interesting, unpredicted way: rather than conform with the GM narrative, which suggests that countries will tend to avoid forming ties with larger economies at further distances, our results indicate that countries are more likely to export to larger economies at further distances. To explain this unexpected result, we looked carefully at the differences in outdegree scores among developed and middleincome and/or BRICS countries, noting that the later experienced larger increases in export ties (i.e. outdegree) between the first and third time period of our data. Drawing upon literature on emerging economies, we drew the conclusion that these countries are expanding their markets at a much faster rate than other countries, and such expansion often involves penetrating larger (and wealthier) markets, which may or may not be geographically close (Aulakh, 2006; Edwards and Lawrence, 2010; Jain, 2006; Rumbaugh and Blancher, 2004). Given that emerging economies tend to have less ties to begin with than developed ones (i.e. rich countries tend to be saturated with ties), expanding their number of trade partners, i.e. increasing their centrality in the time period studied here, makes sense.

98

C. Prell, K. Feng / Social Networks 46 (2016) 87–100

Another consistent finding across the models found in Table 4 pertains to the notion of comparative advantage. In line with this perspective, we argued that a high CTI would imply that a country has a comparative advantage in carbon-intensive products, and that such an advantage would subsequently lead to the formation of more export ties overtime (H2). This hypothesis received consistent support across models found in Table 4, i.e. countries appear to increase their number of export ties in response to having higher CTI values, as reflected in the positive, significant parameter for the CTI ego effect. In addition, extending this argument further, EUE theory suggests that poorer, less developed countries would be the ones more likely to acquire a high CTI, and moreover, that their subsequent export ties would target wealthier, developed economies. Our findings, however, are contrary to this expectation: countries with higher CTIs are more likely to export to poorer economies, as reflected in the negative, significant parameter found for the CTI ego × GDP per capita alter effect. Given the sharp increase in export ties among BRICS countries in particular (and other middle-income, developing countries to a lesser degree), we see these findings as possibly reflecting the tendency of economic actors within developing economies to expand their foreign markets as a means to stimulate growth (Aulakh, 2006), especially given that the carbonintensive goods produced in these countries tend to be lower in price than products exported from more developed economies (Edwards and Lawrence, 2010). In the context of this study’s time frame, such new, foreign markets translate into involving poorer economies. Finally, Table 5 tested two hypotheses dealing with middleincome countries. Simpler models in Table 5 offered support for the idea that those middle-income countries that are also net exporters of carbon are more likely to form export ties overtime (H4b). Yet controlling for network endogeneity, especially the tendency to form transitive triads overtime, revealed that being a net exporter of carbon was really only significant for highincome countries, in terms of predicting export tie formation. After noting that some of the net carbon exporting countries in our sample are also high-income ones, we discussed that an underlying reason for these findings potentially lay in the idea that middle-income countries network differently, and hence, controlling for these network patterns actually wiped away the kinds of links between trade, carbon, and wealth past research and theory suggest. While we were able to uncover a number of interesting processes affecting trade tie formation, our results regarding the tendencies affecting changes in countries’ CTIs were fewer in number. First, we were able to detect our hypothesized pattern that countries’ CTI levels were likely to increase in response to increased exports to wealthier economies (Model 1, Table 4). However, including additional effects in our model (Model 2, Table 4) quickly revealed that this pattern, while apparent in the data, was really a result of other processes. These included the positive impact of countries’ own wealth (GDP per capita) on their CTIs. In addition, as mentioned earlier, countries with high CTI levels tend to form more export ties to poorer countries. Bringing these two findings together, a likely scenario for explaining the result(s) in the CTI change model is that the economic growth experienced by developing countries came, in part, from the increase in export ties to poorer countries, and this economic growth, in turn, helped spur greater productivity in carbon-intensive goods for meeting foreign demands (e.g. Jain, 2006). Altogether, our study has shown, through a number of analytical steps, the intertwined nature of trade tie formation and countries’ CTI levels. One potential restriction of our study is that we rely on binary trade data composed of the upper fifth percentile of our valued trade matrix. Hence, we were limited to looking at only a small portion of the trade flows (i.e. strong ones) circulating

between countries. Given this limitation, we performed additional sensitivity analyses, the results of which are presented in the SM. These additional analyses were done on two different binary trade data matrices, which were dichotomized using cut-off values of the upper 10th and upper 2.5th percentile of trade flows from our original, valued trade matrix. As such, they represent ‘moderately strong’ and ‘very strong’ flows, relative to the ones presented in Table 4.5 Our new set of findings are found in Tables SM1 and SM2 in the Supplemental Material. Looking at these, we note that many of the findings found in these Tables are similar to those presented in Table 4. For example, the results for the main, hypothesized effects across both sets of models, when no controls are added (i.e. Models SM1a, SM2a, SM3a, and SM4a), mirror findings found in Models 1a and 2a in Table 4. When controls are added, however, both the moderately strong and very strong trade change models show the CTI ego × GDPpc alter effect as insignificant, as well as all GDP per capita effects (alter, ego, and similarity). Given that the simpler models (i.e. Models SM1a, SM2a, SM4a, and SM5a) still detect the hypothesized patterns, we conclude that changes in cut-off values for dichotomizing the trade matrix may result in further ‘cloaking’ of some of our hypothesized patterns. However, in spite of this limitation in having slightly different patterns emerge from trade data that has been dichotomized using slightly different cut-off values, the new models continue to show support for arguments made regarding transitivity, comparative advantage, and geographical distance as predictors of tie formation. Thus, although our analyses fail to take into account the full volume of trade passing between countries in our trade matrices, we feel confident that we have identified some real patterns shaping trade dynamics and carbon flows among these countries for the time-period of this study. This analytical framework has thus helped expand the view points on how global trade affects country-level outcomes by situating those same outcomes as potential drivers of trade formation, and enabled us to test these tendencies simultaneously. The result has been both a challenge and reinforcement on some commonly held narratives regarding economic globalization, carbon, and the patterns of trade. Acknowledgement This work was partially funded by the Deans Research Initiative, at the University of Maryland, USA. The authors would like to thank Klaus Hubacek, Mathew Mahutga, David Schaefer, Laixiang Sun, Reeve Vanneman, Christian Steglich, and Tom Snijders for helpful comments made to earlier versions of this paper. Appendix A. Supplementary Material Supplementary material related to this article can be found, in the online version, at http://dx.doi.org/10.1016/j.socnet.2016.03. 001. References Anderson, J.E., 2011. The gravity model. Annu. Rev. Econ. 3 (1), 133–160. Antkiewicz, A., Whalley, J., 2005. China’s new regional trade agreements. World Econ. 28 (10), 1539–1557. Arrighi, G., Drangel, J., 1986. The stratification of the world-economy: an exploration of the semiperipheral zone. Review 10 (1), 9–74. Aulakh, P.S., 2006. Global strategies of Brazilian firms in an era of economic liberalization. In: Jain, S.C. (Ed.), Emerging Economies and the Transformation of

5 Lower percentiles (e.g. upper 25th percentile) as cut-off values resulted in highly dense networks, which failed to reveal any meaningful patterns.

C. Prell, K. Feng / Social Networks 46 (2016) 87–100 International Business: Brazil, Russia, India and China (BRICs). Edward Elgar Publishing Limited, Northampton, MA, pp. 91–110. Baier, S.L., Bergstrand, J.H., 2007. Do free trade agreements actually increase members’ international trade? J. Int. Econ. 71 (1), 72–95. Beckfield, J., 2010. The social structure of the world polity. Am. J. Sociol. 115 (4), 1018–1068. Bergstrand, J.H., 1989. The generalized gravity equation, monopolistic competition, and the factor-proportions theory in international trade. Rev. Econ. Stat. 71 (1), 143–153. Bhagwati, J., 1993. Regionalism and multilateralism: an overview. In: Melo, J.D., Panagariya, A. (Eds.), New Dimensions in Regional Integration. Cambridge University Press, Cambridge, MA. Bhalla, A.S., Bhalla, P., 1997. Regional Blocs: Building Blocks or Stumbling Blocks. St. Martin’s Press, New York. Brizga, J., Feng, K., Hubacek, K., 2013. Drivers of CO2 emissions in the former Soviet Union: a country level IPAT analysis from 1990 to 2010 energy. Energy 59, 743–753. Brizga, J., Feng, K., Hubacek, K., 2014. Drivers of greenhouse gas emissions in the Baltic States: a structural decomposition analysis. Ecol. Econ. 98, 22–28. Burns, T.P., Davis, B.L., Kick, E.L., 1997. Position in the world-system and national emissions of greenhouse gases. J. World-Syst. Res. 3, 432–466. Chase-Dunn, C.K., 1998. Global Formation: Structures of the World-Economy, Updated Edition. Rowman & Littlefield Publishers, Lanham, MD. Clark, R., 2010. World-system mobility and economic growth, 1980–2000. Soc. Forces 88 (3), 1123–1151. Copeland, B.R., Taylor, S.M., 2004. Trade, growth, and the environment. J. Econ. Lit. 42 (1), 7–71. Davis, S.J., Caldeira, K., 2010. Consumption-based accounting of CO2 emissions. Proc. Natl. Acad. Sci. 107 (12), 5687–5692. Davis, S.J., Peters, G.P., Caldeira, K., 2011. The supply chain of CO2 emissions. Proc. Natl. Acad. Sci. 108 (45), 18554–18559. Dick, C., 2010. Do environmental characteristics influence foreign direct investment growth? A cross-national study, 1990–2000. Int. J. Comp. Sociol. 51, 192–210. ˜ Duenas, M., Fagiolo, G., 2011. Modeling the international-trade network: a gravity approach. arXiv:1112.2867v1. Edwards, L., Lawrence, R.Z., 2010. Do Developed and Developing Countries Compete Head to Head in High-Tech? NBER Working Paper No. 16105. In Working Paper Series. National Bureau of Economic Research. Feng, K., Siu, Y., Guan, D., Hubacek, K., 2012. Assessing regional virtual water flows and water footprints in the Yellow River Basin, China: a consumption based approach. Appl. Geogr. 32 (2), 691–701. FinancialTimes, 2006. Ask the expert: BRICs and investor strategy. In: Financial Times. Pearson, London, UK, 6 November. Fratianni, M., Kang, H., 2006. Heterogeneous distance – elasticities in trade gravity models. Econ. Lett. 90 (1), 68–71. Gleditsch, K.S., Ward, M.D., 2001. Measuring space: a minimum distance database. J. Peace Res. 38, 749–768. Grimes, P.E., Kentor, J., 2003. Exporting the greenhouse: foreign capital penetration and CO2 emissions 1980–1996. J. World-Syst. Res. 9 (2), 261–275. Grossman, G.M., Krueger, A.B., 1994. Environmental impacts of a North American free trade agreement. In: Garber, P. (Ed.), The US–Mexico FreeTrade Agreement. MIT Press, Cambridge, MA. Hertwich, E.G., Peters, G.P., 2009. Carbon footprint of nations: a global, trade-linked analysis. Environ. Sci. Technol. 43 (16), 6414–6420. Holland, P., Leinhardt, S., 1970. A method for detecting structure in sociometric data. Am. J. Sociol. 76, 492–513. Holland, P.W., Leinhardt, S., 1971. Transitivity in structural models of small groups. Small Group Res. 2, 107–124. Holland, P.W., Leinhardt, S., 1972. Some evidence on the transitivity of positive interpersonal sentiment. Am. J. Sociol. 72, 1205–1209. Hornborg, A., 1998. Towards an ecological theory of unequal exchange: articulating world system theory and ecological economics. Ecol. Econ. 25 (1), 127–136. Hornborg, A., 2011. Global Ecology and Unequal Exchange: Fetishism in a Zero-Sum World. Routledge, Abingdon, Oxon, New York, NY. Iapadre, P.L., Tajoli, L., 2014. Emerging countries and trade regionalization. A network analysis. J. Policy Model. 36, S89–S110. IPCC, 2014. In: World Meteorological Organization (Ed.), The Fifth Assessment Report (AR5). World Meteorological Organization, Geneva. Jackson, M.O., Rogers, B.W., 2005. The economics of small worlds. J. Eur. Econ. Assoc. 3 (2–3), 617–627. Jain, S.C. (Ed.), 2006. Emerging Economies and the Transformation of International Business: Brazil, Russia, India and China (BRICs). Edward Elgar, Northampton, MA, USA. Jorgenson, A.K., 2012. The sociology of ecologically unequal exchange and carbon dioxide emissions, 1960–2005. Soc. Sci. Res. 41 (2), 242–252. Jorgenson, A.K., 2011. Carbon dioxide emissions in Central and Eastern European Nations, 1992–2005: a test of ecologically unequal exchange theory. Hum. Ecol. Rev. 18, 105–114. Jorgenson, A.K., Clark, B.T., 2012. Are the economy and the environment decoupling? A comparative international study, 1960–2005. Am. J. Sociol. 118 (1), 1–44. Kali, R., Reyes, J., 2007. The architecture of globalization: a network approach to international economic integration. J. Int. Bus. Stud. 38 (4), 595–620. Kagawa, S., Suh, S., Hubacek, K., Wiedmann, T., Nansai, K., Minx, J., 2015. Identification of CO2 emission clusters within global supply-chain networks and the

99

implications for climate mitigation global environmental change. Glob. Environ. Change 35, 486–496. Kim, S., Shin, E.H., 2002. A longitudinal analysis of globalization and regionalization in international trade: a social network approach. Soc. Forces 81 (2), 445–471. Kinne, B.J., 2014. Dependent diplomacy: signaling, strategy, and prestige in the diplomatic network. Int. Stud. Q. 58 (2), 247–259. Koskinen, J., Lomi, A., 2013. The local structure of globalization. J. Stat. Phys. 151 (3–4), 523–548. Lenzen, M., Kanemoto, K., Moran, D., Geschke, A., 2012a. Mapping the structure of the world economy. Environ. Sci. Technol. 46 (15), 8374–8381. Lenzen, M., Moran, D., Kanemoto, K., Geschke, A., 2013. Buiding Eora: a global multiregional input–output database at high country and sector resolution. Econ. Syst. Res. 25, 20–49. Leonard, J.H., 1985. Divesting Nature’s Capital: The Political Economy of Environmental Abuse in the Third World. Holmes & Meier, New York. Leonard, J.H., 2006. Pollution and the Struggle for the World Product: Multinational Corporations, Environment, and International Comparative Advantage. Cambridge University Press, Cambridge. Lenzen, M., Moran, D., Kanemoto, K., Foran, B., Lobefaro, L., Geschke, A., 2012b. International trade drives biodiversity threats in developing nations. Nature 486, 109–112. Lospinoso, J.A., 2012. Statistical Models for Social Network Dynamics. University of Oxford, UK (PhD thesis). Mahutga, M., 2006. The persistence of structural inequality? A network analysis of international trade, 1965–2000. Soc. Forces 84, 1863–1889. Manger, M.S., Pickup, M.A., 2016. The coevolution of trade agreement networks and democracy. J. Confl. Resolut. 60 (1), 164–191. Manger, M.S., Pickup, M.A., Snijders, T.A.B., 2012. A hierarchy of preferences: a longitudinal network analysis approach to PTA formation. J. Confl. Resolut. 56 (5), 853–878. Matous, P., Yasuyuki, T., 2015. The effects of endogenous interdependencies on trade network formation across space among major Japanese firms. Netw. Sci., http:// dx.doi.org/10.1017/nws.2015.1037. Miller, R.E., Blair, P.D., 2009. Input–Output Analysis: Foundations and Extensions. Cambridge University Press, Cambridge, England; New York. Mol, A.P.J., 2011. China’s ascent and Africa’s environment. Global Environ. Chang. 21 (3), 785–794. Moran, D.D., Lenzen, M., Kanemoto, K., Geschke, A., 2013. Does ecologically unequal exchange occur? Ecol. Econ. 89 (0), 177–186. Mouw, T., Entwisle, B., 2006. Residential segregation and interracial friendship in schools. Am. J. Sociol. 112, 394–441. Pao, H.-T., Tsai, C.-M., 2010. CO2 emissions, energy consumption and economic growth in BRIC countries. Energy Policy 38 (12), 7850–7860. Peters, G., 2008. From production-based to consumption-based national emission inventories. Ecol. Econ. 65 (1), 13–23. Peters, G., Hertwich, E.G., 2008. Post-Kyoto greenhouse gas inventories: production versus consumption. Clim. Change 86 (1–2), 51–66. Porter, M.E., 1990. The Competitive Advantage of Nations. Free Press, New York. Prell, C., Feng, K., Sun, L., Geores, M., Hubacek, K., 2014. The economic gains and environmental losses of US consumption: a world-systems and input–output approach. Soc. Forces 93 (1), 405–428. Prell, C., Sun, L., Feng, K., Myroniuk, T.W., 2015. Inequalities in global trade: a crosscountry comparison of trade network position, economic wealth, pollution and mortality. PLOS ONE 10 (12), e0144453, http://dx.doi.org/10.1371/journal.pone. 0144453. Prew, P., 2010. World-economy centrality and carbon dioxide emissions: a new look at the position in the capitalist world-system and environmental pollution. J. World-Syst. Res. 16 (2), 162–191. Reinert, K.A., 2009. Gravity models. In: Reinert, K.A., Rajan, R.S., Glass, A.J., Davis, L.S. (Eds.), The Princeton Encyclopedia of the World Economy. Princeton University Press, Princeton. Ricardo, D., 1821. On the Principles of Political Economy and Taxation. J. Murray, London. Rice, J., 2007. Ecological unequal exchange: international trade and uneven utilization of environmental space in the world system. Soc. Forces 85 (3), 1369–1392. Ripley, R.M., Snijders, T.A., Boda, Z., Vörös, A., Preciado, P., 2015. Manual for RSiena. University of Oxford, Oxford. Roberts, T.J., 1996. Predicting participation in environmental treaties: a worldsystem analysis. Sociol. Inq. 66 (1), 38–57. Roberts, T.J., Parks, B.C., 2007. A Climate of Injustice: Global Inequality, North-South Politics, and Climate Policy. MIT Press, Cambridge, MA. Robins, G., Pattison, P., Kalish, Y., Lusher, D., 2007. An introduction to exponential random graph (p*) models for social networks. Soc. Netw. 29 (2), 173–191. Rumbaugh, T., Blancher, N., 2004. International Monetary Fund. International Trade and WTO Accession, China, Available at SSRN: http://ssrn.com.proxy-um. researchport.umd.edu/abstract=878859. Schaefer, D.R., 2013. Befriending substance-using peers during adolescence: an examination using dynamic network models and simulations. In: Network Perspectives on Criminology and Criminal Justice. Arizona State University. Schulz, M., Söderbaum, F., Öjendal, J., 2001. Regionalization in a Globalizing World: A Comparative Perspective on Forms, Actors, and Processes. Zed Books, London, New York. Shafik, N., Bandyopadhyay, S., 1992. Economic Growth and Environmental Quality: Time Series and Cross Section Evidence, Working paper. Washington, DC.

100

C. Prell, K. Feng / Social Networks 46 (2016) 87–100

Skvoretz, J., 1990. Biased net theory: approximations, simulations and observations. Soc. Netw. 12 (3), 217–238. Snijders, T., van de Bunt, G., Steglich, C., 2010. Introduction to stochastic actor-based models for network dynamics. Soc. Netw. 32 (1), 44–60. Steinberger, J.K., Roberts, J.T., Peters, G.P., Baiocchi, G., 2012. Pathways of human development and carbon emissions embodied in trade. Nat. Clim. Change 2 (2), 81–85. Tinbergen, J., 1962. Shaping the World Economy: Suggestions for an International Economic Policy. Twentieth Century Fund, New York.

Tobler, W., 1970. A computer movie simulating urban growth in the Detroit region. Econ. Geogr. 46 (2), 234–240. Uzzi, B., 1996. The Sources and consequences of embeddedness for the economic performance of organisations: the network effect. Am. Sociol. Rev. 61, 674–698. Uzzi, B., Lancaster, R., 2003. Relational embeddedness and learning: the case of bank loan managers and their clients. Manag. Sci. 49 (4), 383–399. Yu, Y., Feng, K., Hubacek, K., 2014. China’s unequal ecological exchange. Ecol. Indic. 47, 156–163.