Especially with the increasing devaluation of the other group, but also with the increasing conviction of their own position, the supporters of COVID restrictions accept restrictions regarding the others’ freedom of expression

Polarisation and Silencing others During the Covid-19 Pandemic in Germany: An Experimental Study Using Algorithmically Curated Online Environments. Tim Neumann, Ole Kelm, Marco Dohle. Javnost - The Public, Sep 10 2021. https://doi.org/10.1080/13183222.2021.1969621

Abstract: In 2020, societies debated the use of government restrictions on public life to stem the COVID-19 pandemic. Many of these debates took place online. The Internet enables people to come into contact with like-minded content. Algorithms based on collaborative filtering can contribute to this process and might lead to homogenous like-minded online environments that contribute to a polarisation of society. This article therefore examines the effects of (1) like-minded versus opposing online environments, which were (2) randomly versus algorithmically curated. Data from a between-subject experiment embedded in a two-wave panel survey of German citizens (n = 318) show that attitude polarisation as well as affective polarisation are largely independent of exposure to different online environments. Moreover, the results indicate that polarised attitudes of supporters and opponents of the COVID-19-related restrictions relate to varying degrees of beliefs in the importance of silencing people with opposing opinions: While supporters’ polarised attitudes are positively related to the belief in the importance of silencing others, opponents’ polarised attitudes are rather negatively related to such beliefs.

KEYWORDS: belief in the importance of silencing otherscollaborative filteringexperimentonline discussionpolarisationselective exposure

Discussion

One central aim of this article was to examine the effects of homogenous and algorithmically curated opinion environments on polarisation in the context of the COVID-19 pandemic. For this purpose, data from a two-wave panel survey conducted among the German population with an embedded between-subject experiment were analysed. Participants were assigned to different online environments, in which they were exposed to either (1) like-minded or (2) opposing arguments for restrictions to combat the COVID-19 pandemic. The selection of these arguments was either (1) random or (2) based on an algorithm with collaborative filtering.

The results show that the generated online environments did not influence attitude polarisation, i.e. they did not affect how certain the respondents are in terms of their respective attitudes toward the government actions. One reason for this could be that the COVID-19 pandemic is not the ideal context to test the influence of like-minded online environments and the role of algorithms as, for example, many people are most likely confronted with information on this topic outside online environments (Viehmann, Ziegele, and Quiring 2020). This potentially limits at least the short-term effects of arguments presented in online environments. Moreover, the non-existent effects regarding the selection mode must be treated with caution, as the manipulation check regarding the selection mode failed.

Another reason could be that the respondents’ attitudes toward the government actions were so pronounced that they could not even be strongly influenced by the presentation of convincing arguments. However, other explanations for these non-existent effects are possible. For example, it might not only be the content of an argument that is crucial to its persuasiveness. Instead, also in algorithmically curated online environments, how an argument is presented and whether, for example, it is presented factually, emotionally or embedded in a narrative might be (more) relevant. Even more obvious, who has authored and spread the argument might also be important. In line with the echo chamber hypothesis, other people echo one’s own opinion. Through repeated interaction, people can build a relationship with these like-minded others, giving their arguments more weight than the arguments of other people or, which seems to be relevant in the case at hand, than the arguments that are presented via an anonymous platform. In addition, people more often use the content of sources and people they trust (e.g. Strömbäck et al. 2020). As a result, and in line with the filter bubble hypothesis, algorithms with collaborative filtering are likely to select more often information from these sources or from people with whom the people in question more often interact. Thus, the impact of the authors of the arguments, which did not play a role in the presented experimental study, may be crucial.

The experimental results also show that the different online environments created in this study did not influence affective polarisation. Again, the reason could be that the respondents’ attitudes toward the other group were so pronounced that they could not be strongly influenced by the content presented via the online discussion platform. However, it might also be the case that group identification in the context of the COVID-19 pandemic is too low. Many people in Germany had an opinion on the government actions, but they might not have seen themselves as part of a specific group—maybe also because they recognised that (almost) the entire population was and is affected by the consequences. This makes affective polarisation as a consequence of arguments that were presented only once more unlikely. More likely, affective polarisation may occur as a consequence of observing behaviour that one disapproves of (e.g. ignoring restrictions on demonstrations). Another explanation is that the presented arguments were within the democratic spectrum of opinion and did not contain, for example, incivility, exaggeration or satirical elements. These elements may trigger affective polarisation more strongly than the direction of an argument (Druckman et al. 2019).

Another central aim of the study was to examine how attitude polarisation and affective polarisation are related to the BISO. The study shows that affective polarisation among the opponents of restrictions is rather negatively related to the BISO. In view of previous research results (Tsfati 2020), this may initially be surprising. However, it is conceivable that the opponents of the restrictions consider the restrictions as an unreasonable limitation of their fundamental rights. Since they perceive their fundamental rights as already being restricted, they might fear further restrictions of such rights. If they themselves silence other groups, this will lead to cognitive dissonance, which they want to avoid. Moreover, as there are no short-term political majorities for their demands, they are, compared to the supporters of restrictions, more dependent on a public debate that is as open and broad as possible.

For supporters of the COVID-19-related restrictions, however, the results were different: Especially with the increasing devaluation of the other group, but also with the increasing conviction of their own position, the supporters of restrictions want to prevent positions perceived as dangerous from being spread—for this reason, they also accept restrictions regarding the others’ freedom of expression. The differences between the supporters and the opponents of the restrictions can also be explained by the object of silencing: Tsfati and Dvir-Gvirsman’s (2018) understanding of the BISO focuses on interpersonal communication—whether face-to-face or in online environments. From the point of view of the opponents, there are reasons to support a broad public debate (which, of course, includes all forms of interpersonal communication), whereas from the point of view of the supporters of the restrictions, there are reasons to limit this public debate to acceptable contributions. If silencing had been applied to journalism or authorities, the results might have been different.

The study has limitations. Although government restrictions and incidence numbers in Germany changed little during the fieldwork, many people’s opinions toward the restrictions changed. The results of the control group of this study also illustrate this: In total, 21.9 per cent of the respondents assigned to the control group changed their opinion toward the restrictions from t1 to t2. Thus, the time gap between t1 and t2 may be responsible for the fact that arguments selected by the algorithm were not (more) convincing. This also became apparent in the manipulation check, which did not show a significant difference in the persuasiveness of the arguments that were randomly selected and those that were selected by the algorithm.

Second, the respondents were only briefly exposed to the stimulus. This may have reduced its effects. This might be especially true for the chosen topic, about which most people have probably received extensive information via many sources. Further studies should examine the effects of different online environments in other polarised contexts that are less in the media spotlight (e.g. genetically modified food).

Third, the respondents were not able to choose which online environment they entered (“forced exposure”), and forced and self-selective exposure might have different effects on polarisation (e.g. Arceneaux and Johnson 2013).

Fourth, as the results of the regressions are based on cross-sectional data, no causal claims can be made, which is why no mediation models were calculated (see Chan, Hu, and Mak 2020), even if the results suggest potential mediation effects. The results imply that attitude polarisation has an effect on the BISO and that this effect is mediated by affective polarisation. To test these assumptions, experimental studies should (1) manipulate attitude polarisation to analyse its potential effect on affective polarisation and the BISO and (2) manipulate affective polarisation to analyse its potential effect on the BISO.

Finally, we only focused on one country. Thus, it is unclear whether the results are generalisable to other contexts.

Despite these limitations, the present study contributes to current research on how like-minded and opposing online environments influence polarisation. Even though it cannot be ruled out that the non-existent findings are due to the experimental set-up, it is also possible that the results indicate that the process of how like-minded or opposing information influences polarisation is more complex than previously suspected. More studies are needed that focus on the potential effects of being exposed to these online environments. Experimental studies using algorithmically curated stimuli may be helpful to determine the effects. Furthermore, this study has shown that the BISO is a theoretical concept that is also relevant in more (national) contexts than those previously tested (e.g. Tsfati 2020). Silencing others and the belief in the importance of doing so may be important concepts for understanding polarised societies.