Small worlds or worlds apart? Using network theory to understand the research-practice gap

Among psychologists, Stanley Milgram is perhaps best known for his controversial 1963 study of obedience, which involved testing subjects’ willingness to follow an authority figure's orders to deliver painful electric shocks to another person. However, a few years later he began a series of experiments designed to explain the small world phenomenon, and which subsequently gave rise to the popular notion of six degrees of separation ( Milgram, 1967 ). The small world phenomenon refers to the experience of two strangers discovering in the course of conversation that they are linked by a series of mutual acquaintances (i.e. a friend of a friend), at which point, one of them remarks, “Gee, it's a small world.” Milgram framed this phenomenon as the “Small World Problem” because it seemed impossible that two strangers would be linked, given the size of the world's population (in 1967) and the fact that most individuals’ social circles are relatively closed (i.e. my friends are mostly friends with each other).

To investigate this phenomenon, he developed what has come to be known as the small world experiment. In its abstract form, the small world experiment asks a “starter” to deliver a message to a stranger, the “target,” by passing it only through personal acquaintances. For example, a starter might first pass the message to his neighbor, who passes it to her boss, who finally passes it to the target. Of particular interest is whether the message successfully reaches the target, and when it does, the number of intermediate contacts that participated in the chain. In this simple example, the starter and target, despite being strangers, were separated by just two intermediate acquaintances: the neighbor and the boss.

The first full-scale small world experiment asked a sample of 296 starters living in Nebraska and the Boston area to deliver a message to a stockbroker living in Sharon, Massachusetts, who served as the target ( Travers & Milgram, 1969). 2 2

A preliminary small world experiment asked a sample of starting persons living in Kansas to deliver a message to the wife of a divinity school student living in Cambridge, Massachusetts.

Milgram's initial small world experiment offers an experimental and conceptual framework for thinking about the research-practice gap. But, this experiment has been replicated using different starters and targets several times. In this section and in Table 1 , we briefly summarize the methods and findings of these subsequent small world experiments, drawing out three key lessons that can teach us about the research-practice gap. Our review focuses only on studies that replicated the key features of Milgram's small world experiment, but excludes studies that used elements of the method for other purposes (e.g., Bochner, Buker, and McLeod, 1976 link-tracing approach). For a comprehensive review of all small worlds research, see Schnettler (2009).

The “starter” and “target” columns in Table 1 identify the settings and populations involved in replications of Milgram's small world experiment, and highlight two key ways that these experiments have differed. First, some small world experiments draw starters and targets from the same population, like Killworth, McCarty, Bernard, and House's (2006) study of telephone survey interviewers, while others draw starters and targets come from distinctly different groups, like Lee's (1969) study of women searching for abortionists or Erickson and Kringas’ (1975) study of residents contacting politicians. The exchange of information between researchers and practitioners more closely resembles the latter case, because researchers and practitioners often represent distinct populations, each embedded in their own settings and with their own professional and personal goals ( Green et al., 2009).

Second, some small world experiments focus on information exchange in a specific closed setting, like Shotland's (1976) study of communication at Michigan State University, while others examine information exchange in society at large, like Dodds, Muhamad, and Watts (2003) large internet-based study. Both types of situations may be relevant to the research-practice gap because the researchers and practitioners who need to exchange information may both be located in the same closed setting (e.g., teachers and in-district research staff in a large urban school district), but may also be situated in the larger society (e.g., a community mental health professional in the U.S., and a mental health researcher at an institution in Spain). Focusing on this distinction, Killworth et al. (2006) contend that starters will more successfully reach targets in open systems than in closed systems. In closed systems, starters are likely to rely on their (possibly incorrect) understandings of the system's structure to reach targets. For example, a teacher seeking information from an in-district researcher may, based on her knowledge of the district's organizational chart, think that contacting the district's office of professional development is the appropriate next step, when in reality contacting the research office directly would be more useful. In open systems, starters “are more aware that they cannot comprehend the structure, and so use only attributes of their acquaintances in an attempt…[at] completing the chain” (p. 95). For example, when a community mental health professional seeks information, because there is no global organizational chart of mental health researchers to potentially mislead her initial search, she can instead focus on the attributes of the people she knows (e.g., “Is my colleague likely to know any mental health researchers?”).

Lesson 1: Spatial and social distances are related to the length and success of chains . The success and length columns of Table 1 show the percentage of chains initiated by starters that successfully reached the target, and the average number of intermediary brokers involved in these chains. Small world experiments frequently seek not only to identify these values, but also to understand why some chains are successful while others are not, or why some successful chains are shorter than others. The final column of the table identifies some of the key variables that explain these findings. In most cases, the explanatory variables capture various aspects of distance: chains are more likely successful and are shorter when the starter and target are separated by small spatial distance (i.e. they live nearby) and small social distance (e.g., same ethnicity, same race, same rank). However, a few other explanatory variables are noteworthy as well. Guiot (1970) attributed the high rate of success to his experimental design, in which the researcher himself contacted participants as the message was forwarded, rather than leaving the task of forwarding the message up to the participants themselves. Similarly, Dodds et al. (2003) attribute their low rate of success to attrition among the intermediaries; there was little incentive to participate, and thus most chains terminated early. Finally, Lundberg (1975) linked differing success rates to differing levels of bureaucratization in two settings: the utility company had a strict hierarchy and policy manual that hindered informal contacts, while the electronics company managers were “proud of the practice of using first names among all personnel” (p. 209).

Reduction of a research-practice gap would involve maximizing the likelihood that information will move from a researcher to a practitioner (or vice versa; i.e. success) and minimizing the number of brokers necessary to facilitate this transfer (i.e. length). Factors of spatial and social distance are likely influential in this context. That is, the research-practice gap is likely narrower when researchers and practitioners work in the same community, and when they share socio-demographic features. While this remains a hypothesis to be tested, Singh, Hansen, and Podolny's (2010) study of consulting firm employees (i.e. practitioners) seeking information from content area experts (i.e. researchers) offers more concrete insight into factors affecting the success and length of chains in this context. They found significant inequality among employees in their ability to contact experts and obtain information. Specifically, they found that newer employees were at a disadvantage, both because they “had lower cognitive awareness about who knows what” (p. 1433) and were more likely to turn for help to others who were similarly peripheral to the organization. This suggests that the research-practice gap may be less severe for those who have been researchers or practitioners for some time, but also highlights the potential for subtle inequalities in who might experience the disadvantages of a research-practice gap. For example, this finding might also suggest the research-practice gap could be more severe in areas that are dominated by researchers and practitioners who are newer to their fields.

Many of the small world experiments shown in Table 1 focus on starter-target pairs that require crossing social boundaries including race ( Korte & Milgram, 1970), organizational role (Shotland, 1976), and ethnicity (Weimann, 1983 ). In each case, chains were more often successful and were shorter when the starter and target came from the same group, than when the starter and target came from different groups. In this latter case, a successful chain requires crossing the social boundary from one group to another at some point. For example, a successful chain from a white starter to a black target in Korte and Milgram's (1970) experiment requires, at some point, that a white participant passed the message to a black participant. Investigating this boundary crossing requirement, Weimann (1983) found that “in mixed chains, where at least one cross-[group] link is forced, it is delayed by the participants until the stage when it is unavoidable, [but] once a gatekeeper is found, the chances for successful completion rise significantly” (p. 298). That is, effective communication between starters and targets from different groups hinges on that critical moment when the message jumps from one group to another. This phenomenon has a clear analog in the research-practice gap context because the researcher and practitioner communities often operate as different social groups. The successful transmission of information from a researcher to a practitioner requires that at some point the information leaves the researcher community, and enters the practitioner community.

Lesson 2: Social boundaries may lead to echo chambers or closed loops that hinder chain success . Some small world experiments have identified the risk of an “echo chamber” when a message fails to cross a social boundary. In their study of hospital workers’ attempting to locate a person capable of handling a patient complaint, Stevenson and Gilly (1991) expected “managers, as the exemplars of formal procedure, to forward folders directly to formally designated complaint handlers” (p. 921), but instead found that “managers relied more on… other managers rather than formal problem solvers, and tended to keep problems within their own professions” (p. 926). That is, once a patient complaint landed in the hands of a manager, it often simply circulated among a group of other managers without making progress toward someone who could address the complaint. In a different context, Killworth et al. (2006) found that starters could not reach their targets because the messages would get stuck in these types of closed loops in nearly 24% of all chains. This is also frequently observed in both researcher and practitioner communities: researchers regularly publish their findings in research journals that are available to and read by only other researchers, while practitioners seek advice from other practitioners. More generally, sending information to, or receiving information from, another member of one's own community serves to exacerbate the research-practice gap.

Lesson 3: The longer the chain, the lower the likelihood of success . The risk of attrition represents still another challenge to transferring information from a starter to a target via a chain of intermediary brokers. At each step of a message's journey toward the target, the current message holder must decide whether to pass the message along or not. As a result, long chains have a lower likelihood of success because they involve more opportunities for the chain to prematurely terminate. This also means that the observed chain lengths reported in Table 1 are systematic underestimates of the actual separation between the starters and targets because they include only successful, and thus likely shorter, chains. Some small world studies have attempted to use statistical models to determine the actual mean chain length in the absence of attrition. For example, while the completed chains observed by Singh et al. (2010) included 1.89 brokers on average, if chains had not prematurely terminated due to attrition, they would have been more than 25% longer and included an average of 2.37 brokers. Dodds et al. (2003) suggest that attrition may arise when participants lack incentives to help with the transfer of information, noting that “if individuals searching for remote targets do not have sufficient incentives to proceed, the small-world hypothesis will not appear to hold, but that even a slight increase in incentives can render social searches successful under broad conditions” (p. 828). This might have an important implication for the research-practice gap. If researchers do not have a clearly defined practitioner audience to serve as a target for their findings, or if practitioners do not have a clearly defined information need to serve as a target for their search, the gap is likely to persist because the ambiguity of the task of facilitating information exchange will deter all but the most dedicated brokers. That is, if brokers do not know what they are brokering, they are unlikely to help.

Milgram (1967) initially framed his small world experiment as a way to gain insight into the small world problem: how is it possible that, although the people we know mostly also know each other, which makes our social world feel small, we are nonetheless linked to the rest of the world by just a few intermediaries? For example, from the perspective of a white Angeleno in Korte and Milgram's (1970) study, the world is small and local: he knows mostly other white Angelenos, who live in the same area, visit the same shops and restaurants, and know each other. Likewise from the perspective of a black New Yorker in their study, the world is small and local: she knows mostly other black New Yorkers, who live in the same area, visit the same shops and restaurants, and know each other. However, despite their perceptions of living in small and bounded communities, the white Angeleno and black New Yorker are actually quite socially close, separated by just 6 mutual acquaintances. The many small world experiments confirmed this problem – the world seems small, but actually is big – but did not offer an answer to the apparent paradox.

To resolve the paradox, it is helpful to consider a couple different kinds of networks. At the one extreme, imagine a network in which people form closed communities within which everyone knows everyone else, but between which there is no communication. In the research-practice context, this is a worst-case scenario: researchers talk to other researchers, and practitioners talk to other practitioners, but they never talk to each other ( Fig. 2 A). In this kind of network structure, the world seems small to each individual, but it is also fragmented. At the other extreme, imagine a network in which people do not form clear communities, but rather know and talk to others randomly. In the research-practice context, this may seem ideal, with researchers and practitioners indiscriminately engaging with one another, but it does not allow researchers or practitioners to form their own communities, within which they might develop professional standards and cultivate collaborative relationships ( Fig. 2 C). In this kind of network structure, the world is very big because each person is just a few links from everyone else, but it is also disorganized.

Fig. 2

Small world networks.

Watts and Strogatz (1998) found that if a small number of random links are added to a network otherwise composed of closed communities, a unique new type of structure – a small world network – is formed. In a small world network, people still form relatively distinct communities that allow the world to seem small, but the few random links between these communities also mean the world is big and people are separated by just a few links. Interestingly, Watts and Strogatz (1998) discovered that many real world networks have this type of structure, and the structure of communication ties among researchers and practitioners is likely no exception. Researchers and practitioners each have relatively dense communities of their own, but just a few random linkages between these two communities dramatically reduce the gap between them ( Fig. 2 B). Within the communities, researchers and practitioners are able to form strong collaborative bonds and develop professional norms and standards, while between the communities they are able to efficiently share information through just a few intermediate brokers.

It is likely that the research-practice world is structured like a small world network, but this alone does not solve the research-practice gap. It is still necessary for researchers and practitioners to navigate this network as they attempt to push information out through dissemination efforts and pull information in through search efforts. That is, for example, although it might be possible for a practitioner to pull information in from a researcher via a (small) series of brokers, the practitioner does not necessarily know which brokers will be most effective for this purpose ( Kleinberg, 2000 ). When searching for information, a starter may or may not know the exact identity of the target, but they certainly do not know the global structure of the communication network. Instead, a starter has only local information about his or her own network, and thus must rely on search strategies intended to help find information efficiently. Among the simplest strategies is the high-degree search, which suggests that when a practitioner is looking for a piece of research, or a researcher is trying to get a piece of research to practitioners, he or she should seek the assistance of a broker who knows lots of people. Despite the intuitive appeal of such an approach, it is often not useful for finding the shortest chain toward a target ( Adamic & Adar, 2005 ), and thus is likely not an effective strategy for reducing the research-practice gap.

A second strategy that is widely used, but also nearly always ineffective, is a homophily search in which the starter turns for assistance to a broker who is like him or herself. This strategy is particularly common with respect to gender and race, with starters and subsequent brokers in the chain turning to others with the same demographic characteristics as themselves. But, it is also applied with respect to work roles and rank, where the risk of a research-practice gap is clear. Consistent with Lessons 1 and 2 reviewed above, a homophily search may lead to unsuccessful chains between researchers and practitioners, as well as echo chambers. If a practitioner seeking a piece of research turns to another practitioner, who then turns to another practitioner, the search for information spirals around within the practitioner community with little hope of reaching the target. The same is true for a researcher seeking to disseminate a piece of research, but who talks to another researcher, who in turn talks to another researcher.

Although the high-degree and homophily search strategies may not reduce, and may even exacerbate, the research-practice gap, they also hint at alternate strategies that may be more effective. First, rather than turning to a broker who knows lots of people, researchers and practitioners should seek assistance from brokers who know people they do not know. The feature of brokers that makes them critical in closing the gap is not how many people they know, but rather whom they know. Brokers that know different people, that is, whose social circles do not overlap with the information seeker, provide a bridge to other parts of the social world. Second, rather than turning to colleagues and coworkers, researchers and practitioners should seek assistance from brokers that are different from themselves . To be sure, this might be a particularly challenging search strategy to adopt because colleagues and coworkers are a natural source of information and advice, but they are also likely to be a source for redundant information that merely confirms the seeker's initial beliefs. Turning to different others for help in pushing or pulling information facilitates breaking out of one's own closed community and bridging into other communities, including the target's. These two strategies may help bridge social and spatial distances that limit successful searches (Lesson 1), reduce the potential for echo chambers (Lesson 2), and reduce the number of brokers needed to link researchers and practitioners (Lesson 3).

The fact that communications among researchers and practitioners likely take the form of a small world network structure means that there are relatively short chains of intermediate brokers through which researchers and practitioners can exchange information. Thus, the research-practice gap can, at least in principle, be overcome. Actually overcoming the gap requires finding these short chains, which can be aided by consciously eschewing some common strategies (e.g., degree, homophily) and consciously adopting alternate strategies for deciding where to seek assistance when trying to push information out or pull information in. These conclusions are informed by a wealth of research on the small world phenomenon that span many different contexts and more than five decades, but to date these phenomena and search strategies have not been studied specifically in a researcher-practitioner context. Thus, in addition to sketching a promising solution for the research-practice gap, these ideas also identify an important direction for future research. Future research on bridging the research-practice gap through a chain of intermediaries may also benefit from investigating the composition of those chains, exploring for example, the kinds of individuals or organizations that frequently appear in these chains and the types of brokerage in which they are involved (e.g. Gould & Fernandez, 1989; Neal, Neal, Kornbluh, Mills, & Lawlor, in press ).

The authors of this article declare no conflict of interest.

This research was supported by an Officer's Research Award (#182241) and Use of Research Evidence award (#183010) from the William T. Grant Foundation.