In a sense, the study of comparative politics is the study of the role that context plays in structuring behavior. Institutional contexts, such as the nature of the electoral system or the existence of an independent judiciary, drive differences in electoral outcomes or human rights across nations. Individual-level contextual factors such as norms, culture, or ethnic or religious identity can be determinative in understanding when social movements will likely flourish and which cleavages will lead to political parties or ethnic strife. Variation in context leads to variation in political outcomes and behavior across both peoples and nations, and provides comparative politics with its richness.

Context also leads to dissent among scholars as we engage in internecine methodological warfare over the best way to understand the complexity introduced through the inclusion of contextual factors in our theories and data (Mahoney and Goertz Reference Mahoney and Goertz2006). Quantitative approaches tend to look at issues broadly but shallowly, effectively averaging out complexity to make generalizations across cases. Qualitative approaches tend to look narrowly but deeply, yielding a weaker argument for generalization to other cases but a stronger one for truly understanding the determinants of behavior in a particular case. Because practitioners of each methodological school require substantial training to be successful, an insufficient commonality tends to exist between them, leading to less than ideal communication.

Social networks and a relational approach to comparative politics promise a way to bridge the gap between scholars and employ insights derived from deep qualitative study in cross-national, quantitative analysis. This connection is made possible by the focus on a core and often overlooked contextual factor: the role that the structure of interactions between actors plays in understanding the behavior of individual actors, and, therefore, in understanding aggregate behavior. As this form of context is universal—humans are by nature social—the use of social networks minimizes the traveling problem (Sartori Reference Sartori1970). Networks may vary in importance by substantive topic and spatiotemporal setting, but their conceptualization is clear and constant: ties to one's parents are ties to one's parents, whatever the relevance of the ties. In this article, I first discuss the use of social networks in general terms, focusing on the way in which relational analysis links the qualitative and the quantitative, as well as the individual-level and the aggregate. I then offer a series of four examples applying the use of relational approaches to core problems in comparative politics.Footnote ¹

Let me begin this general discussion with an important question: how might the promise offered by a relational approach be fulfilled? The key lies in identifying the appropriate level of analysis and the quality and availability of data given this level of analysis. To see how this process might work, it is helpful to first consider individual-level analysis of behavior. The goal is to understand the determinants of some choice or action, be it for whom or for what policy to vote, whether to turn out to vote or to protest, when or how much to alter a political opinion, and so on. There is a broad and growing empirical literature identifying the roles that one's social connections play in altering one's beliefs and behavior across a variety of political activities (e.g., Huckfeldt and Sprague Reference Huckfeldt and Sprague1995; Leighley Reference Leighley1990; McAdam Reference McAdam1988; Petersen Reference Petersen2001; Sinclair Reference Sinclair2009). The causal pathways underlying the effect of network connections are varied and include direct influence (Friedkin and Johnsen Reference Friedkin and Johnsen1999); information transmission (Huckfeldt Reference Huckfeldt2001; Lohmann Reference Lohmann1994; McClurg Reference McClurg2006); norms of fairness (Gould Reference Gould1993); considerations of reputation (Gerber, Green, and Larimer Reference Gerber, Green and Larimer2008), credibility (Granovetter Reference Granovetter1978), or legitimacy (Opp and Gern Reference Opp and Gern1993); strategic complementarities (Marwell and Oliver Reference Marwell and Oliver1993); resource coordination (Verba, Schlozman, and Brady Reference Verba, Schlozman and Brady1995); and safety in numbers (Kuran Reference Kuran1991).

Although a few of these causal pathways (complementarities, information) particularly lend themselves to quantitative analysis (Jackson Reference Jackson2008), others (norms, legitimacy) require more case-specific knowledge for conceptualization and quantification. This knowledge can only come from a deep understanding of the sort of contextual factors that qualitative scholars have long identified as important. Thus, when attempting to map out the network connections among people in cases in which one can expect norms, culture, identity, or emotions to drive behavior both directly and indirectly via the behavior of others, qualitative scholarship is a necessity.

Once these connections are discerned, the tool of quantitative analysis becomes available. For example, one could include a simple measure of individual-level connectivity in cross-national regressions of comparative behavior. Given the strong dependence of behavior on network ties in the American context, even controlling for media effects (Mondak Reference Mondak1995), one would expect that connectivity would continue to be an important explanatory variable in the analysis of questions in comparative politics. This expectation assumes, of course, that networks have been constructed that are appropriate to the case in question. Casual discussion networks may be the appropriate data from which to draw the independent variable of connectivity when political opinion formation is the behavior in question, but the same networks may not be appropriate when considering an action such as protest, in which participants assume more personal risk. In such cases, the appropriate network to study may be strongly tied family and friends (McAdam Reference McAdam1988). Careful qualitative analysis is necessary to identify these important distinctions.

Beyond the individual level, attention to the properties of behavioral spread in a network can help translate these insights to the aggregate level. Scholars in this tradition detail how the degree of behavioral spread depends on such network characteristics as clustering, path length, connectivity, and the number of weak ties (e.g., Centola and Macy Reference Centola and Macy2007; Fowler Reference Fowler and Zuckerman2005; Gould Reference Gould1993; Marwell and Oliver Reference Marwell and Oliver1993; Rolfe Reference Rolfe2005; Siegel Reference Siegel2009). A mapped-out network of relations allows one to employ these theoretical insights to produce predictions about how a population of individuals within such a network will behave, given a particular distribution of individual incentives and a particular causal pathway driving changes in individual behavior. For example, a network with many weak ties (Granovetter Reference Granovetter1973) is likely to more effectively distribute information, allowing individuals to update policy beliefs, than to encourage a risky collective action that relies on individual attention to norms of fairness or legitimacy (Centola and Macy Reference Centola and Macy2007; Siegel Reference Siegel2009). We would therefore expect a network with many weak ties to have more of an effect in the aggregate on opinion formation than on risky mass participation. Given network data, one can test these predictions by looking at the substantive effect of the aforementioned network characteristics on aggregate behavior in regression analysis.

As can be seen, relational studies connect across several levels of analysis. Individual-level data on discussion partners or movement ties can be used both to make predictions on behavior at the individual level and to form a picture of network structure at the aggregate level. These structures are in turn predictors of aggregate-level behavior (Siegel Reference Siegel2009). That networks link levels of analysis implies their potential for addressing new questions. If one found, for example, that weak ties tend to be determinative in individual opinion, but that support for a political party is strongest in regions in which few weak ties are present, these findings would suggest that different causal mechanisms lie behind support for a political party, spurring further research.

A fair criticism of this approach is its data-intensiveness. This approach may translate well to industrialized democracies, but even years of careful study in areas in which security is more of an issue are unlikely to produce much more than group-level information (Petersen Reference Petersen2001). Without individual-level connectivity data, one cannot include connectivity as an individual-level variable in a cross-national regression.

Network analysis, however, is flexible enough to produce additional insights even when data are comparatively poor. At the individual level, it may still be possible to discern the network of elite interconnectivity, given structured interactions within the halls of power. These data can provide for the same analysis described previously, although only for an unrepresentative subset of the population. Given the influence of elites in these settings, however, the insight derived from such network analysis might still be sufficient to make broad predictions.

At the aggregate level, knowledge of every tie in the network is not necessary to determine the general type of the network. Siegel (Reference Siegel2009) offers four such network types: small world, opinion leader, village, and hierarchy. Differences in expected aggregate behavior can be discerned across these types when knowledge is available only of the types themselves. Even when detailed network data are sparse, qualitative data on network type—for example, do opinion leaders drive behavior, or is there more equality of influence across the network?—can be employed to produce predictions. Siegel (Reference Siegelforthcoming) illustrates this approach using Patel's (Reference Patel2005) ethnographic data to postdict the outcome of the January 2005 Iraqi legislative elections.

Thus, depending on data availability and the level of analysis of the research question, network analysis and a relational approach to political science can not only produce useful and novel insights into comparative politics, but also allow for productive collaborations between qualitative and quantitative scholars, taking advantage of the techniques and insights of each. The remainder of this article illustrates this potential in a series of examples drawn from core problems in comparative politics.