Seminar

Diffusion and Influence in Networks

Date
Wed November 13th 2013, 12:30pm
Event Sponsor
Stanford Institute for Research in the Social Sciences & Stanford Sociology Department
Location
SCANCOR Conference Room
Diffusion and Influence in Networks

Sharique Hasan & Matt Jackson

This week Sharique Hasan and Matt Jackson join the Stanford Network Forum for a discussion about diffusion and influence in networks, and to discuss their recent empirical work on these topics in universities and villages in India.

Wednesday's forum will be held at our regular time of 12:30-2:00 PM in the SCANCOR Conference Room located on the first floor of CERAS. We have moved rooms a number of times this quarter, but if all goes well, all remaining events this quarter and next will be held in CERAS. A brief comment on format. We will welcome everyone at 12:30 and do a brief round of introductions following which Sharique and Matt will kick off the discussion by briefly framing their papers (5-10 minutes each). They may speak on what led them to their respective topics and questions, how they decided on the methodology or approach they chose, what conversations or debates the paper speaks to, and similarly, how they would situate the paper theoretically. While the papers are a good way to lead us into these topics, they are not necessarily intended to be the centerpiece of discussion. Once Sharique and Matt have introduced their papers, we will open up the Forum to questions and discussion. As we transition into the discussion portion of the meeting, participants and faculty panelists are encouraged to move beyond the specifics of the papers and engage broader theoretical questions raised by the papers.

The Diffusion of Microfinance

The Mechanics of Social Capital and Academic Performance in an Indian College

Questions for presenters

Questions for Matt Jackson

1. In light of Duncan Watt's recent presentation at the fall CSS Conference where he distinguishes broadcasting from viral diffusion, we wondered to what extent diffusion of knowledge about the microfinance program occurred "virally" from household to household, and to what extent "broadcasting" mechanisms might have played a role. In the context of this particular paper, "broadcasting" mechanisms could include leaders making formal or informal announcements about the microfinance program in public settings.

2. On the social science side, questions were raised about the following model assumptions: 1) households only consider the current state of the network when deciding whether or not to participate in microfinance, and 2) upon hearing about a microfinance opportunity, households must decide once and for all whether they will participate, prior to sharing that information with others, and they are unable to revisit their decision at later points in time. Fortunately Michael Leung was in attendance and was able to stem the tide of criticism by explaining that such stark assumptions were necessary in order to even estimate model parameters and standard errors. Nonetheless, these simplistic assumptions cause us to raise questions about the internal validity of the findings. Perhaps the broader and more constructive questions that many of us wonder about is precisely how wide the gap is between our questions and our methods, what are the greatest challenges to bridging these gaps, what have we accomplished to this point, and what are the prospects looking forward?

3. A theoretical question related to the collapsing of the dozen or so network measures collected into a single network tie. One of the amazing aspects of this data set is that multiple types of ties were measured, allowing future studies to theorize and test differences and relationships between different types of ties. We wondered if the authors believed that all types of ties measured are reasonable proxies for the diffusion of microfinance information. If some ties are better proxies than others, what subset of ties might the authors have selected to better model potential paths of information diffusion? Theorizing mechanisms of diffusion and constructing ties from a more discriminating subset of the relations measured could potentially increase the size and significance of observed effects.

4. What practical implications can we derive from these findings? Rather, is it appropriate to derive practical implications from these findings, given the highly stylized and unrealistic nature of the assumptions made. How should we understand the contributions of this paper?

5. When does a network perspective actually enlighten our understanding of a diffusion or influence story, and when is it more trouble than it's worth?

Questions for Sharique Hasan

1. How do overcome the issues introduced by regressing student test scores on roommate test scores? A question was raised about different scenarios that may yield the same roommate average. For example, a student may have two roommates, one who received 50 and one who received 100. Another student may have two roommates, both of whom received 75. The model would treat these two scenarios as identical; however we could imagine very different social environments in these two settings.

2. Roommate assignments partly construct the environment in which influence may be exerted. To what extent might informal residence hall and campus networks circumvent, supplant, and override the influence described in this paper? What are the actual mechanics of peer influence in this setting? What kind of data, methods, research design could offer a more convincing story?

3. Like diffusion processes, influence can flow in both directions of a relationship. Furthermore, social structures can both enable as well as constrain diffusion and influence. Can networks help us understand not only the paths through which information and norms diffuse, but also the barriers that they might pose to diffusion?

4. When does a network perspective actually enlighten our understanding of a diffusion or influence story, and when is it more trouble than it's worth?

 

Summary of Nov 13 Panel Discussion with Jackson & Hasan

Our faculty panel last week with Matt Jackson and Sharique Hasan began by responding to student questions raised about the theoretical framing of their papers and the mechanisms of diffusion and influence. Matt framed the theoretical contribution of his paper as exploring the process of diffusion while controlling for peer effects. His findings distinguish between the effect of simply hearing about a piece of information and acting on that information based on one's position in a social structure. In this particular instance, peer effects turned out not to be significant when deciding whether to participate in a micro finance program; as the benefits of micro-finance were self-evident to households, hearing that the program was coming to their village was sufficient information for households to decide whether or not to participate. For this reason, Matt believes that the model's assumption is not totally unrealistic, that upon hearing about a micro-finance program, households make a one-time decision whether to participate and do not revisit their decision based on neighbor's decisions. Had peer effects turned out to be important, this assumption may have been less defensible. Matt also defended his model's assumption that diffusion of knowledge about the micro-finance program traveled primarily by word of mouth, across dyadic ties among households, rather than through some broadcasting mechanism such as social media, town meetings, local news. Anecdotal evidence supported this as well as the fact that information infrastructure in the villages surveyed was quite limited.

In framing his paper, Sharique noted that both his and Matt's projects were driven by real-world questions, in his case, popular criticism of the quota systems used by elite Indian universities to advantage those from lower castes. Sharique acknowledged that his measures were only first-order approximations of influential relationships, but in a story opposite of Matt's, he believes that it was the strong ties developed among students as they helped each other study, and not the diffusion of norms or the sheer fact of caste differences, that best predicted student academic performance. There are several ways, Sharique offered, of observing people and measuring interaction: ethnography, linguistic and body language, surveys, etc. Any form of data collected may offer a different kind of perspective, but will also come with its own limitations. Matt added that within the paradigm of social science, it is easier to rule out possible explanations than explain things. Good design is important in order to do this well.

These two themes, good data and good design, resonated throughout the remainder of the discussion. Sharique distinguished between three domains of innovation: theoretical, methodological, and empirical (data). He noted that both papers are drawn from very unique data sets. This is true of all of the empirical papers we explored this quarter, including Dan McFarland's papers on speed dating and research collaboration to Srivastava and Banerji's study of collaborative networks in organizations. A lack of quality data and rich description impedes our ability to formulate good theories, he argued. Innovation may be required to capture the breadth of data our question requires; it is also required, he argued, to better get at the mechanisms of the questions we are interested in answering, or the explanations we are interested in ruling out.

Such examples of innovative data are Matt's network measures such as borrowing and lending kerosene. Analysis is difficult when the data comes before the questions, Matt observed. Good data is especially a challenge in network analysis where we are often missing large pieces of the network. The frontiers of network analysis, multiplex relationships and network dynamics over time, will also require not only theoretical innovation, but also innovation in research design and data collection.