Introduction and Contribution

Democracies face a host of ongoing challenges, including the rise of elected autocrats, income inequality, mainstreamed forms of xenophobic nationalism, and political apathy. All of these challenges pose threats to democratic participation — elected autocrats restrict it, inequality makes it easier for oligarchs to sway election outcomes, and xenophobia can discourage cultural outsiders from voting.

Apart from practical threats to democratic participation, an established intellectual tradition has viewed participation with deep skepticism. In this view, democracy is good simply because it ensures peaceful transfers of power and protects individual rights. Collective decisions, however, cannot be meaningfully viewed as representing the “will of the people” owing to manipulation, apathy, and the high costs of acquiring political knowledge. Is this “realist” vision the best one can hope for in democratic life?

In “Can deliberation have lasting effects?,” James Fishkin, Valentin Bolotnyy, Joshua Lerner, Alice Siu, and Norman Bradburn show how a three-day deliberative experiment in late 2019 had large and long-term effects on turnout and voting behavior. Those most likely to exhibit these civic behaviors nearly a year later had come to follow politics more closely and see their political opinions as valuable. At the same time, the experiment’s effects on participants’ policy views were significant only in the short term — most deliberators eventually reverted to their previously held policy positions.

That three days of deliberation had such lasting civic effects suggests that efforts to create more inclusive forms of democratic participation are both possible and scalable. Moreover, it suggests that academic skepticism about democratic participation is not an argument against citizens’ capacities for reasonable decision-making; rather, it is an argument against our imperfect contexts of participation. Deliberative experiments may offer hope for improving these contexts.

The Deliberative Experiment and Its Effects on Policy Views

In September 2019 — one week prior to the experiment — a treatment group (i.e., those who would deliberate) of 523 registered voters from around the US and a control group of 844 voters were surveyed on their political attitudes. Members of the treatment group then deliberated on five issue domains (the economy, environment, immigration, health care, foreign policy) in small groups and on 47 policy proposals (e.g., redistributing wealth in some way). 26 of these proposals were characterized by extreme partisan polarization, meaning significant numbers of those who identified as Democrats or Republicans held the most extreme views. After the deliberations ended, both the treatment and control groups were surveyed. Then, three subsequent surveys were conducted in late 2020.

Among the treatment group, deliberation produced significant, short-term depolarization on 20 of the 26 (polarized) policy proposals. In other words, the averages for participants who identified with each party moved closer together (though not necessarily toward the center). These changes were large, sometimes 40 percentage points, as in the case of Republicans abandoning extreme positions on immigration. Meanwhile, the control group’s policy positions changed hardly at all — pointing to the key role played by deliberation. Within the five issue areas, averages among deliberators shifted leftward on all but the economy.
 

Image

Figure 1. Policy-Based Score (PBS) Changes over Time
Note: Policy-based score (PBS) is constructed for each individual based on responses to 26 questions identified as the most polarizing. The upper chart shows the participant group, and the lower chart shows the control group. T1 is the survey wave prior to the deliberations, T2 is right after the deliberations, and T3 is 10 months after, in July 2020.
 

By late 2020, however, the treatment group’s policy positions mostly reverted to their pre-deliberation levels. The differences between these two points in time were still significant compared with the control group, yet relatively small in absolute terms. These policy reversions are perhaps unsurprising: deliberators returned to an environment of heightened polarization and aggressive campaigning during the 2020 election cycle. (To be sure, and from the standpoint of finding solutions to collective problems, policy reversion is not especially concerning — the aim of deliberation is to bring citizens together to reason and compromise, which the experiment accomplished.)

A Civic Awakening?

The lack of long-term policy effects suggests that three-day deliberations may be limited in their ability to create a more encompassing, participatory society. However, the treatment group demonstrated large and persistent changes in their intention to vote (i.e., turnout) and their candidate of choice. Among the control group, Joe Biden was favored over Donald Trump by about four percentage points — very close to Biden’s actual margin in the popular vote. Among the treatment group, however, Biden was favored by 28 percentage points. The gaps in turnout were similarly large. (Note that these are intentions, not reports of actual decisions. However, Tables 6 and 7 in the article show similar effects for recollected votes after the election.)
 

Image

Table 2. Voting Intention for Participant and Control Groups, Time 4
 

These civic outcomes are especially surprising because (a) voting behavior is thought to be stable and deeply rooted in one’s psychology and social context, and (b) experimental efforts to increase turnout have been most successful when undertaken shortly before elections, as opposed to one whole year prior. The effects were most pronounced among political moderates and those without college degrees — perhaps pointing to the educative effects of deliberation.
 

Image

Figure 5. Effects on Vote Intention Captured by Predictive Modeling, by Education
Note: Middle are those participants who have Policy-Based Scores between 3 and 5 (inclusive) at Time 1. Non-middle participants are all other participants. Positive prediction error shows that, on average, participants were more likely to vote for Biden than predicted by the model. Vote intention data are collected at Time 4, in October, 2020. Full calibrated model used to construct this figure can be found in the APSR Dataverse.
 

Why did deliberation produce only short-term policy effects but long-term effects on voting behavior? The authors posit that deliberation caused an “awakening of civic capacities.” They reason that deliberation was a transformative experience in terms of stimulating political engagement and a sense of efficacy. And indeed, the treatment group was, in the long term, more likely than the control group to follow the 2020 election campaign, believe their political opinions mattered, and acquire general information about American politics. (The latter is measured in terms of knowing which party controlled the House and Senate.)
 

Image

Figure 9. Having “Political Opinions Worth Listening to”
Note: Policy-based score is constructed for each individual based on responses to 26 questions identified as the most polarizing. Responses to the question “How strongly would you disagree or agree with the following statement?”[I have opinions about politics that are worth listening to.] were collected at T1 (just before deliberations), T2 (just after), and T3 (10 months later, July 2020).
 

The authors close by discussing efforts to scale up civic engagement, such as the Stanford Online Deliberation Platform. In all, “Can deliberation have lasting effects?” provides a rigorous case for the value of deliberation in strengthening democratic participation.

*Brief prepared by Adam Fefer.

Introduction and Contribution:

Gossip — sharing information, both positive and negative, about absent third parties — plays a major role in social life. Two friends spread a rumor about their neighbor’s infidelity, an employee’s hard work is praised at a manager's meeting, and so on. 

From an evolutionary standpoint, however, the origins and proliferation of gossip are somewhat puzzling. Not only is it costly — acquiring personal information, sharing it with others, perhaps risking ostracism for being a yenta — but the evolutionary benefits are not clear. Indeed, gossiping may be entertaining, but it seems like a stretch to think it could improve human adaptation, let alone survival.

In “Explaining the evolution of gossip,” Xinyue Pan, Vincent Hsiao, Dana S. Naub, and Michele J. Gelfand conduct a series of computer simulations to model the gossiping process and shed light on the benefits it generates for both society and gossipers. The authors posit two such benefits that help explain its evolution: First, gossip spreads information about others’ reputations, helping people identify those who will (not) cooperate with them. Second, gossip reduces selfishness, as those who otherwise would not cooperate will choose to do so with gossipers in order to manage their reputations. These are called the “reputation dissemination” and “selfishness deterrence” functions, respectively.

Readers come away with a sense of the power of agent-based modeling to solidify our intuitions about complex social processes. Indeed, gossip involves dynamics of communication, cooperation, geographical proximity, as well as both personal and collective gain. Models help simplify these dynamics. In addition, the authors illustrate how an ostensibly non-strategic activity has very important strategic dynamics and implications. 

The Simulation Setup:

Agents in the simulation develop a strategy based on two decisions: whether or not to gossip and how exactly to cooperate with other agents. The former is a binary choice, whereas the latter permits six choices. Of these six, three are especially important: The “exploitive” agent is one who only chooses to cooperate if they believe their partner cannot easily be taken advantage of in the game. In other words, exploiters condition their choice on the other agent’s reputation. Next, the “virtuous” agent cooperates if it believes the other agent will and otherwise defects. Finally, the “opportunist” agent will only cooperate with gossipers and never with non-gossipers.
 

Image

Figure 2b. Plot (B) illustrates an agent’s action as a function of their own strategy and their belief about the interaction partner’s strategy.

The other three cooperation strategies are as follows: “unconditional defectors” and “unconditional cooperators” are, as the names imply, insensitive to information about others’ reputations. Meanwhile, “reverse opportunists” only cooperate with non-gossipers and never with gossipers. Perhaps unsurprisingly, the authors will show that agents tend to discard these three strategies over time. If gossip is, in fact, beneficial from an evolutionary standpoint, then both cooperative and gossiping strategies must increase over time.

To model the gossiping process requires that agents respond to new information and update their strategies accordingly. This will help them maximize their success and avoid being exploited by others. Information can thus be obtained in two ways: either agents observe each other’s behavior during the game and use it to infer their strategy, or information is disseminated to them via gossip.

The simulation proceeds in three steps. First, agents play a “cooperation game,” whereby each player simultaneously decides whether to cooperate or defect. (The rules are as follows for two agents, A and B: A incurs a cost of 1 for cooperating while B’s benefit is 3, and vice versa. A thus gains by cooperating, but is tempted to defect to gain 3 while B pays a cost of 1, and vice-versa.) The second step involves the dissemination of gossip. Finally, strategies are updated based on the first and second steps. 

Simulation Results:

The authors show that over time, the vast majority of agents (90%) choose to gossip and to cooperate (78%). The three most common cooperation strategies are “exploitive” (57%), “opportunistic” (18%), and “virtuous.” By contrast, none of the other three cooperation strategies is chosen more than 5% of the time.
 

Image

Figure 3. The evolutionary trajectories of different strategies and behaviors. The lines are average trajectories from 30 simulation runs with all the six cooperation strategies under the default parameter choice. The shadows show the SEs of the average trajectories. Plot (A) illustrates the evolution of gossipers. Plot (B) illustrates the evolution of different cooperation strategies. Plot (C) illustrates the evolution of cooperation. 

At first glance, the high prevalence of cooperation and exploitive/opportunistic strategies may appear puzzling. However, the growth of gossip helps guard against exploitation and opportunism. In other words, it remains strategically rational to exploit when feasible, but doing so simply becomes less feasible over time. 

One surprising finding is that gossipers and opportunists actually need each other. Opportunists cooperate with gossipers to protect their reputations, and in doing so give gossipers a real survival advantage over non-gossipers. As co-author Michele J. Gelfand said in a recent interview, “Opportunistic agents kind of get a bad rap. They’re seen as kind of sneaky…But in fact, they’re actually helping a lot in the population.”

The final aspect of the simulation involves showing that both functions of gossip, (1) disseminating reputations and (2) deterring selfishness, are jointly necessary to explain its evolution. To do this, the authors begin by analyzing solely function (1). Yet, this provides no direct benefit to gossipers: they spend time and energy disseminating information that helps others, but without deterring opportunists, who are not necessarily more likely to cooperate with them. In this situation, gossiping strategies will not proliferate. This is why function (2) must be part of the explanation. Gossipers directly benefit when opportunistic agents must manage their reputations by cooperating with them. In this way, they can deter the prospect of defection, which non-gossipers still face.
 

Image

Figure 4. Results of Steps 1 and 2. Each condition is the average of 30 simulation runs. The value is calculated as the average value from the 4,000th to the 5,000th iterations of each simulation run. The error bars show the SEs. On the Left side of each plot are the results from Step 1; on the Right are the results from Step 2. Plot (A) shows that gossipers evolve only when both reputation dissemination and selfishness deterrence functions exist (i.e., with-gossip and with-rep-manage). Plots (B and C) show that the existence of gossipers (yellow) increases reputation accessibility and cooperation. Plot (D) shows that opportunists evolve only when both reputation dissemination and selfishness deterrence functions exist. Plots (E and F) show that the existence of gossipers increases the proportion of reputation-sensitive agents.

In all, “Explaining the evolution of gossip” is an insightful exercise in “abductive reasoning.” No historical account could ever hope to conclusively show how gossip actually developed. Instead, the simulation shows readers how gossip could possibly have developed in order to provide us with better information and encourage cooperation.

*Brief prepared by Adam Fefer.