Two Quebecois scholars, François P. Robert and Pierre Mongeau, have developed a valuable method for modeling the “socio-semantic network” formed when people discuss an issue.* I can envision this method used to assess deliberations, to give real-time feedback to moderators during conversations, and even to reveal patterns of discussion in fictional texts as a contribution to literary criticism.

The article is in French and it uses a lot of terminology from network analysis, so unless you already know how ideas like “directed degree centrality” are expressed in French, you may find it hard going. Since I worked my way through it, I will provide relatively extensive notes below. But here is the shorter version:

Individuals form a social network if they know one another. Each link is a relationship, such as an experience of having talked one-on-on together. Meanwhile, individuals form a “socio-semantic network” if they use the same or highly similar phrases in a conversation on a common theme. “Each word or phrase that two people use in common is a link.” They need not know each other to have a socio-semantic link.

The social relations among people and their socio-semantic networks are different but not necessarily independent. One could affect the other. For instance, we might find that people who are central in a social network disproportionately affect the ideas that are expressed throughout that network. Their prestige may give them influence. Or we might find that people who express ideas that are frequent in the network become more socially central: holding popular ideas may give them prestige.

Robert and Mongeau involved 95 Montreal residents in discussions of a public policy topic: college tuition. Some of these discussions occurred in a large group using formal procedures. (I imagine Robert’s Rules or some variation thereof.) Other discussions occurred in small groups that were less rule-guided.

Some participants knew each other before the experiment. The authors identified all the social links that existed before the deliberation and also found out who had talked to whom during the event. They could thus chart the changing social network of the participants. Meanwhile, the authors asked participants to write about the issue both before and after the deliberation, collected the writing produced during the discussions, and looked for similarities of phrases between pairs of participants. That allowed them to chart the changing socio-semantic network of the group.

They found: “The conventional method [a large group deliberation using formal rules] favors the emergence of a link between social centrality and socio-semantic centrality, while the alternative method [small group discussions] favors the emergence of a negative relationship between these measures.” Apparently, in large group discussions, people who are socially central—knowing many others or interacting with them one-on-one during the meeting—increasingly dominate the views expressed by all the participants. But “in the alternative deliberative format that uses discussions in small groups, the emergence of differences is promoted by providing a space for expressing views different from those of the socially central people.”

The authors draw a lesson for organizers of events. “In practical terms, these results suggest that it would be advantageous for a democratic organization to first use alternative methods so as to promote the expression of a diversity of views and then to continue the deliberation in a conventional manner (like that prescribed in codes of procedure) to develop consensus positions.”

Maybe–although the design of deliberative formats involves more criteria that these. I am more interested in the methodology, because I believe it could be developed and applied for other purposes. To me, the fact that opposite results emerged from different kinds of deliberation validates the method, especially since the authors have a plausible explanation for the patterns they found.

* François P. Robert and Pierre Mongeau, Caractéristiques sociosémantiques des méthodes conventionnelles et alternatives de deliberation, Revue internationale comminucation sociale et publique, no 12, Dec. 2014, pp. 101-120

A bit more detail on the instruments: Before the deliberation, individuals were asked, “Of the following people, with whom did you have [knowledge or contact] before January 28, 2012? ” They were also asked after the deliberation, “With whom [names listed] have you had an exchange, a confrontation, or a sharing of ideas (whether positive or negative, whether or linked or not to the experiment)?” “An extra open-ended question asked respondents to provide information to identify people whose names they did not remember; for example, ‘During the break, I spoke with the gentleman sitting in front of me’ or ‘I talked a lot with the young lady to my right at the first round of discussions.’”

And the findings: “There is a negative correlation (r = – 244; p <0.05) between socio-semantic centrality during the conventional activity [large group deliberation] and that observed in the alternative activity [small group]. This correlation indicates that the central people in the socio-semantic network when the method of deliberation is conventional do not tend to be central in the alternative method. Furthermore, socio-semantic centrality while using the alternative method is positively correlated with that observed after the experiment (r = 378, p <0.01). Thus, people using more words and similar expressions to all participants during the alternative method also tend to do so after the experiment.”

No correlation was found between socio-semantic centrality and “various social centrality measures …. No link can be established before the experiment between centrality in the social network of a person and centrality in the socio-semantic network.”

“In the conventional method [a large-group deliberation], only directed degree centrality in the pre-existing social network is significantly and weakly correlated with socio-semantic centrality (r = 223, p <0.05) (Table 5). Thus, the number of choices received (prestige) of a person in the pre-existing social network is linked, albeit weakly, with the person’s use of words and phrases used by all those present in the conventional method. When the correlations are based on social centralities observed in the circumstantial network centralities all social measures were significantly correlated with sociossemantic centrality in the conventional method. The link is moderate (see Table 6), because the correlations vary around a mean of 0.41 (p <0.01). These results confirm the presumed relationship between social centrality and socioesemantic centrality by our general assumptions (Hg1, Hg2) and the first specific hypothesis (HSA).

“In the alternative method, the correlations between sociosémantique centrality and three of the five social centraly measures are reversed … The correlations between sociosémantique centrality and betweenness centrality measures and proximity are no longer significant for measures of centrality in the circumstantial social network. Furthermore, the centrality of the close pre-existing social network is also weakly negatively related to socio-semantic centrality for the alternative method (r = – 213; p <0.05). …. The alternative method does not attenuate the relationship between sociosemantic centrality and social centrality; it appears to transform them insofar as the correlations are weaker in the alternative method, but reversed. …”

“Indeed, this correlation suggests that people with greater prestige in the network before the [large group] deliberation begins (that is to say, they arrive at the meeting with more  relational capital) have a slight tendency to use words and expressions similar to those used by all participants.

“Conversely, with the alternative method, the central people tend, but moderately, to use words and expressions different from those used by all participants. In other words, the greater the number of relationships a person has with other participants, the more the person’s vocabulary tends to be different, to a moderate degree. This can be explained either by a change of vocabulary on the part of people whose social centrality was already high even before the deliberation began, or else that new themes have become central. Since it seems unlikely that people have changed their vocabulary in such a short period, this suggests that other people have become central in the socio-semantic network during a deliberation using the alternative method. Indeed, the alternative method gives people more time and opportunity to hear a different perspective by using periods of small group discussion. In this situation, a person who brings a new perspective that is taken up and discussed in the group thus increasing his or her centrality in the socio-semantic network. Moreover, the proliferation of small groups not necessarily discussing the same topics explains the relatively weak correlation (r = – 24 on average; p <0.05).”

The post a method for mapping discussions as networks appeared first on Peter Levine.

Northeastern political scientist Nick Beauchamp has developed a remarkable, free tool (Plot Mapper) that identifies keywords in a stream of text and plots them on a two-dimensional plane. The words are placed automatically but meaningfully (using a form of principal components analysis), and a line traces the order in which the ideas unfolded over time.

Beauchamp’s analysis of the 2015 State of the Union was covered in the Washington Post. He demonstrated the neat arc of the president’s rhetoric:

I wanted to try the tool, so I looked around for another recent speech that might contrast with the SOTU. Sarah Palin’s Iowa speech has been widely panned on the right as well as the left, being called a “tragedy”–or at least “an interminable ramble.” I pasted the text into Plot Mapper and this is what I got:

I actually think the quality gap between these two texts is a little obscured by this form of analysis. “America,” “country,” and “people” seem to play similar roles in both speeches. We also see an actual organizing structure to Palin’s words: she moves from nationalism through right-populism (real people are conservative) to conclude with the local (veterans in Iowa). Finding such an arc may be giving the speaker a bit too much credit. Nevertheless, the tool has enormous potential for comparing discourse. I’d be especially interested in using it to analyze how people affect each others’ ideas when they deliberate.

The post diagramming Sarah Palin’s Iowa speech appeared first on Peter Levine.

In the Atlantic, Robert Wright describes research that my team at Tufts University conducted in partnership with him. He summarizes the research question:

Suppose you’re a conservative or a liberal, and you’re watching a debate, and the debater you consider your ideological opponent throws in a “to be sure” sentence—a sentence that qualifies his or her basic policy position, underscoring some point of agreement with your ideological ally. Will that make you more favorably disposed to the person—more likely to take their views seriously, less likely to demonize them?

This was the method:

The study involved some 1,600 people, about half of whom identified themselves as liberal and half as conservative. Everyone watched an excerpt from one of two debates: one between Tim Noah and Glenn Loury on whether the minimum wage should increase, and one between Sarah Posner and Michael Dougherty on whether the government should be able to mandate that employer-provided health insurance cover contraceptives.

The excerpt shown to each viewer was short, but it clearly conveyed which person supported which position—in other words, who was the conservative and who was the liberal (on the issue in question, at least). For half of the viewers, the clip also included, at the end, a segment in which the speaker on the other side of the ideological fence from them added a to-be-sure.

Our conclusions:

The researchers at Tufts found that “viewing a concession created a more positive reaction to the ideological opponent.” Viewers who saw their ideological opponent make a concession were less likely than viewers who saw no concession to call their ideological ally the more credible of the two or the more knowledgeable of the two. And they were less likely to say they liked the ally more than the opponent.

Finally, Wright draws out two implications:

First, a message to people on the left and right who opine in public: Don’t forget to throw in a to-be-sure sentence; it may sound like a “concession,” but it could wind up helping your cause, especially if your cause includes not seeing America consumed by bitter acrimony. And it’s especially advisable to do this when you’re in “enemy territory”—when liberals are on Fox News, when conservatives are on MSNBC.

Second, it would be nice if the formats that mediate our discourse made it practical to add a to-be-sure sentence. For example, a pet crusade of mine is to change the structure of Twitter in a way that, while maintaining the 140-character limit on tweets, would nonetheless make it easier to add a short elaboration.

The post viewing concessions dampens rancor appeared first on Peter Levine.