Social tipping is a way of exploiting cascading behaviours in a population. Everything depends on convincing few people to change their behaviours from the status quo to an alternative behaviour. Once a critical mass of people changes behaviour (the tipping point), conformity and coordination incentives kick in. In response to an initial push of the critical mass, people not initially pushed to change their behaviours may follow suit. For example, if 40 % of a group change behaviour, a cascading effect may lead to the other 60% changing simultaneously or one by one.
Our knowledge of when and why tipping may or may not work is limited, despite the widespread appeal of tipping as a new form of policy instrument. We - a group of researchers from Switzerland and the United States - investigated this question in an online experiment conducted with participants from the United States. We collected data during the controversial 2020 US presidential election, marked by outgroup aversion between Democrats and Republicans. The sample of participants consisted of groups of participants who were strong partisans, that is, groups only consisting of Republicans ( Democrats) disliking the outgroup candidate Biden (Trump) and liking the ingroup candidate Trump (Biden).
Participants were matched and randomly rematched each round into pairs for up to 45 rounds. Each round, they played a coordination game: everyone faced the same two choices. If the two players in a pair selected the same matching choice, they received a substantial payoff, but if they selected mismatching choices, they received only very little. Of course, the trick is that participants could not communicate with each other, nor did they know the choices of their current "counterparts". They also did not know the partisan affiliation of their group before the game started. The coordination payoffs thus favoured coordinating without pointing participants to a specific option.
The experimental procedure was identical across treatments—first, participants were to establish a typical behaviour, a "status quo norm", by repeatedly playing the coordination game. This works quite well – over time, participants figure out one choice to coordinate on and keep going. Then after several rounds of repeated interaction, a tipping point intervention changed the incentives for a randomly selected subset of the group. To be on the safe side, we chose a strong tipping intervention, changing the incentives of 50% of the group so they would choose the opposite of the status quo behaviour. This meant that 50% of the group suddenly had powerful payoffs for virtually choosing the contrary of what the group had nicely played along with, while the other 50% had the same payoffs. The only difference between treatments was a minimal change in how the choice options were labelled. In the treatment condition, we relabeled choice options with images designed to activate partisan political identities. These labels had no material consequences. In the control options, labels had no political meaning with @ and # symbols.
How did we choose the partisan labels? We hired an artist and created several versions of different partisan labels. We had lots of discussions, read literature on the role of elites concerning "affective" polarization, and finally pre-tested these images with independent raters (who were 50% Republican and 50% Democrat). In the end we chose the two candidates in winning / losing poses as a way to trigger political identities. You can see the labels used in the following figure.
|Neutral labels (control)||Identity labels (treatment)|
We had at least one group of 12 people at the same time online, for about one hour or more. It is a tricky business to keep people online and focused for a longer time, especially when it's several people simultaneously. We had some measures we used – one of them was to have a chat window where a researcher was permanently present during the study to answer clarification questions. Second, we had a mechanism in place that our groups could survive with ten group members, thus being able to cope with the loss of up to two participants. In total, we ended up with 68 groups after removing groups terminated due to people dropping out before the intervention was triggered.
What did we find? We observed clear patterns driven by identities across control and treatment. Some groups converged on choosing @ as a status quo norm in neutral groups, while others converged on #. In the treatment condition, however, practically all Republican groups quickly converged on the ingroup "triumphant Trump" image, while virtually all Democrat groups convergence on the ingroup "triumphant Biden" image. To better understand the outcome of tipping point interventions, we use "spillovers" as a normalized measure of how expected the alternative behaviour is after the intervention. Spillovers quantify the outcome net the direct intervention effort. Positive spillovers arise when more individuals change their behaviour net of the direct intervention on the 50%. Spillovers were large and highly significant positive in our neutral treatment. In contrast, our identity treatment produced a large and highly significant spillover reduction relative to this benchmark. Indeed, spillovers were not significantly different from zero in our identity treatment. The figure below shows the main results.
Our results show that social tipping provided a powerful route to behaviour change in the neutral treatment. But the identity treatment, tipping, was highly unreliable, if not to say it did not work. This difference has negative consequences for participant payoffs in the identity treatment. With an intervention incentivizing an alternative that ran counter to their pre-existing identities, players were collectively unable to respond, and they accumulated substantial opportunity costs as a result.
Our experiment shows how simply identity cues can impede tipping dynamics post-intervention. This way of easily breaking a tipping point intervention should not be underestimated. Our findings rely on a simple but visible connection between otherwise neutral, almost technical choice options and symbolic markers. One may consider running an intervention to separate identity markers and choices before the intervention. For example, a recent CNN advertisement stressed wearing a mask's "neutral" nature. However, more research is needed in this regard. Political identity polarization often creates incentives for politicians and cultural entrepreneurs to create the connection between neutral choices and symbolic markers of group affiliations. Intergroup competition and its connection to political and social identities might have far-reaching implications for the feasibility of using indirect mechanisms of large-scale behaviour and norm change.
Read the full paper here: https://rdcu.be/cV7sJ
Our study was generously supported by the Swiss National Science Foundation.