Nonprobability Sampling for Conflict Communication Research
Sampling in conflict research poses particular challenges. At the interpersonal level, including dyadic and small-group interactions, the level of intimacy between interactants may range from low (e.g., strangers) to moderate (e.g., coworkers) to high (e.g., friends) to very high (e.g., romantic partners and family members).
The predominant sampling strategy for studies at the psychological and interactional levels is nonprobability sampling, which is often a convenience sample of undergraduates. So what’s new? Researchers implicitly rely on the notion that, although the mean levels of the variables under investigation may not represent the theoretical population’s mean levels, the covariances among the variables are not biased and are, therefore, informative about the process. The discussion and references cited above regarding external validity are relevant here.To study conflict, we must assume that it is possible to sample not only people but also situations that vary in the magnitude or type of conflict and that the sample of conflicts is reasonable in terms of its range on a variety of variables (e.g., intensity, duration, and theoretical cause of the conflict). This assumption is not met if the conflicts that we may need to know most about are inaccessible or have limited accessibility. For example, because of business secrets, we may not be able to sample representative negotiations between employers and employees; because of security issues, we may not be able to sample representative hostage negotiations. In these cases, we face a selection bias (Berk, 1983): Our conclusions are likely to be, at best, incomplete, or at worst, biased and misleading. For example, suppose that the only hostage negotiation transcripts that we have are those made available to us by negotiators for the government. Suppose further that this subset of transcripts is available because the government agents used significantly fewer threats—or threats of a lower magnitude—in these cases than in the other transcripts, which are unavailable to us.
Finally, suppose that as level of threat severity increases, hostage takers from ethnic group A become more confrontational, whereas hostage takers from ethnic group B become less confrontational. In other words, threat level and ethnicity interact in predicting confrontational behavior. Given that we have only low-threat cases, our conclusions will be wrong about the two ethnic groups. Although the sample is biased, it is more important to note that the analysis is likely to be inaccurate or incomplete.Another example based on hostage negotiation research illustrates a different problem. Suppose we have transcripts from all or from a representative (even if not random) sample of such negotiations; we thereby avoid a selection bias. Thus, these transcripts, although difficult to obtain, provide very useful information about high-intensity conflicts. To complete the picture and create appropriate theoretical models of conflict processes that include processes that vary in intensity, we need our sample to contain a range of conflict intensities. Thus, we would need transcripts of low- to high-intensity conflict that fall within the domain being investigated, but it may be hard to imagine low-intensity conflicts when we are considering hostage negotiations. Thus, even a good sample may not allow some aspects of a theory of conflict processes to be adequately investigated.
If a theory of conflict is to be developed, we may require samples that represent the universe of conflict situations, even if the sampling of such situations is a convenience sample. The interplay of these demands means that we need a theory that explains across the range of conflict situations, and we need a representative sample of situations to generate the theory. In the absence of such a theory, we can look to theories in neighboring domains to consider this sampling problem. For example, Foa and Foa (1974) described six types of resource exchanges: exchanges over love, status, information, money, services, and goods. We may identify conflict situations based on the resources primarily involved, thus generating a resource-exchange universe from which to sample conflict situations. Similarly, Marwell and Hage (1970) organized role relationships by differentiating them in terms of their intimacy, visibility, and regulation. If we examine how individuals in different roles create, resolve, escalate, maintain, or define conflict, we can use the three role relationship dimensions (intimacy, visibility, and regulation) to generate a 2 ? 2 ? 2 typology of role relationships from which we can purposively sample (see also Seeman, 1997). Thus, even in the absence of a strong theory, we may improve our sampling strategy so that it is defined by a theoretical universe and thereby enhances our ability to make theoretical discoveries.