Presentation on theme: "Emergent Leadership in Collective Action: An Experimental Approach Helen Margetts*, Peter John †, Stéphane Reissfelder* *Oxford Internet Institute, University."— Presentation transcript:
Emergent Leadership in Collective Action: An Experimental Approach Helen Margetts*, Peter John †, Stéphane Reissfelder* *Oxford Internet Institute, University of Oxford †University of Manchester Political mobilization increasingly takes place on-line and there are various features of on-line mobilization that are reshaping the dynamics of collective action. First, internet platforms can provide real-time ‘social information’ about the participatory decisions of other people which may influence an individual’s likelihood of joining the mobilization. Second, they can vary the extent to which an individual’s decision to participate is made public or visible. Third, leadership is less costly and activists are likely to depend less on extensive resources and established networks. We hypothesise that in this changed environment, charismatic’ leadership is less important and what matters is the willingness of some individuals to ‘start’ mobilization, which is then fuelled by social information and visibility. We test this hypothesis in an experiment where subjects participate in a public goods game under treatments where social information and visibility are varied. Post-experiment personality tests enabled us to examine the personality characteristics of those subjects who habitually ‘lead’ rather than ‘follow’. The research addresses previous work on leadership and tipping points in collective action (Marwell and Oliver, 1993; Schelling, 2005). Introduction and Research Interest We implement one deviation from the classical public goods scenario, namely the step-level paradigm – whereby individual members are asked to contribute to the collective good at a cost and receive a higher return if the number of participants is higher than a determined point. Further, we mirror the asynchronous nature of online collective action by implementing a real-time sequential protocol of play (as one treatment condition), which can capture the different social information early vs. late potential contributors receive. We recruited 185 subjects to the OxLAB laboratory. After the experiment, subjects completed a questionnaire, which notably includes a short personality survey to determine subjects’ locus of control using the Rotter scale. At each round (n=28), subjects are shown a step-level public good scenario phrased as a request to fund a local initiative. Subjects are endowed with 10 tokens and are informed about the provision point (60 tokens) and the number of participants in their group (N = 10). If the provision point is met, a fixed bonus is redistributed amongst all participants. Some sessions were not fully attended, so smaller groups were formed, in which the provision point was adjusted to meet 60% of the maximum amount collectable. Subjects are paid for one round only, which is selected at random, a design which means that the most rational behaviour is to treat each as if it were the only round (Bardsley 2000). Groups of 10 are randomly allocated at each round, so that players never interact with the same exact same group. Experimental Design Three treatments are implemented in addition to a control condition. Control rounds show no information about the decision of other subjects (anonymity condition). The first treatment shows information about the total amount raised within the group and the number of contributors (social information condition). These rounds are limited in time to 50s. The second treatment consists a visual display of individual contributions on a projector screen (visibility condition). The last treatment is a combination of the social information and visibility conditions. Figure 2 (left) shows the distributions of individual contributions broken down by treatment conditions(across all scenarios). The distribution under the control treatment (light blue) has a tri-modal shape. A large proportion of subjects contribute either 0 or 10 tokens (although these are not necessarily their consistent strategy across rounds), and other provide the “fair share” (60% of their endowment) which correspond to the amount each should contribute for the provision point to be reached at equal costs. In the social information condition (surface plot in middle), a larger proportion of participants are inclined to free-ride on the contributions of others by giving 0 tokens. When contributions become individually visible (dark blue surface), the largest proportion of individuals give 6 tokens, likely due to an impulse to be seen as contributing their “fair share”. Figure 2 (right) shows the aggregate effects of treatment conditions on the likelihood that provision points are collectively met across scenarios. The reasons for these strong differences is explored at the individual and group levels in what follows. Treatment Conditions Determinants of Contribution Amount Susceptibility to Treatment and Personality Features Conclusion Fig.1: Screenshot from the experimental interface. Here, the social information treatment shows the progress of fundraising round and displays the total amount raised Table 2: OLS Predictors of treatment effect size. Standard errors are clustered by individuals. N= 3235 We seek to identify the factors that determine individuals’ choice of how much to contribute to each round and run a tobit regression (summarized in table 1) which tests personal attributes, group dynamics and scenario features as predictors for contribution amount. This analysis reveals that the visibility treatment triggers a strong impulse to contribute more. The impact of the social information treatment is captured by the group dynamics variables: the further from the provision point, the less individuals will be prepared to give. 'Rank' is the order at which participants make decisions, the variable hence shows that earlier contributors are more generous. Agreement with and importance of scenario will, predictably, increase the contribution amount. Person- level features (measured in post-questionnaire) are the most revealing: Social value orientation (cooperative and individualistic, as compared to reference category 'inconsistent') is highly indicative of contribution amount. Individuals who like to take financial risk are willing to give nearly 1 token more on average. Men give less than women. Finally, group size (varying from 5 to 10 due to differential attendance) mattered in that subjects within larger groups tended to contribute more. Table 1: OLS Predictors of contribution amount. Standard errors are clustered by individuals. N= 4550 Further, we measure the treatment effects of each condition, by comparing the contributions of each subject under those treatment and the contributions in the control round for each scenario (results summarized in Table 2). The visibility treatment had a strong positive effect (+0.96 tokens), whilst the social information condition was not unidirectional. The negative coefficient of the 'rank' variable suggests that the treatments have more influence on early movers. Agreement to the issue was associated with a positive treatment effect, whilst subjects who rated the scenarios to be important tended to give less under treatment than in control rounds. Subjects with both cooperative and individualist orientations are responsive to treatment cues. The magnitude of the effects is relative strong, yet they are in opposite directions. Individualists tend to give more under treatment (possibly encouraged by the initiative of others) whilst cooperative types tend to contribute less under treatment (e.g. disillusioned by the low amounts given by others). Individuals who have an external locus of control are more responsive to treatment conditions (as compared to internal types) and tend to contribute more as a consequence. Measurement of respondents’ tendency to adopt innovations were done using the joining year to Facebook. This value was correlated with the susceptibility to treatment effect - a significant coefficient in contrast to a self-reported measure. Group size is a strong positive predictor, in that individuals in larger groups are more responsive to treatment conditions. Figure 3 shows how personality features affect susceptibility to treatment. Investigating the dynamics of on-line mobilization, we found a strong effect of visibility on an individual’s likelihood of participation, replicating findings of Gerber et al (2008) for a collective action context. The effect of social information had no overall significance but rather introduced a strategic element (with both +ve and –ve effects). For combined treatment effects, the results suggest that people with a co-operative personality, an internal locus of control and a low tendency for innovation adoption are less susceptible to changes in the information environment and therefore more likely to participate in the early stages of mobilizations. Further research of this kind could shed light on the distribution of ‘k’ in Schelling’s mobilization curve and on the likelihood of the ‘tipping points’ that he predicted, which our finding that earlier contributors are more generous appears to contradict.