Private-Public Opinion Discrepancy

In many Western societies there are rising concerns about increasing polarization in public debate. However, statistics on private attitudes paint a different picture: the average attitudes in societies are more moderate and remain rather stable over time. The present paper presents an agent-based model of how such discrepancies between public opinion and private attitudes develop at the scale of micro-societies. Based on social psychological theorizing, the model distinguishes between two types of agents: a) those seeking to gain or maintain a good reputation and status, and b) those seeking to promote group harmony by reaching consensus. We characterized these different types of agents by different decision rules for either voicing their opinion or remaining silent, based on the behavior of their proximal network. Results of the model simulations show that even when the private attitudes of the agents are held constant, publicly expressed opinions can oscillate and (depending on the reputational concerns of individual actors) situations can occur in which minorities as well as majorities are silenced. We conclude that the macro-level consequences of micro-level decisions to either voice an opinion or remain silent provide a foundation for better understanding how public opinions are shaped. Moreover, we discuss the conditions under which public opinion could be considered a good representation of private attitudes in a society.

1. Can individuals express opinions that aren't consistent with their privately held attitudes, instead of (selfishly) remain silent?
Thank you for your interesting suggestions. We think doing all of this (introducing lying, noice, and other typologies) might yield interesting novel insights. But for the moment we can not yet see a coherent theoretical viewpoint which would justify and necessitate us to take these next steps ourselves. We therefore invite other researchers to use our code to extend the model in these or other ways. We therefore included the following text at the end of paragraph 4.4: The present model assumes that agents either give their opinion or remain silent. Of  3. Authors state that "when there are many reputation seekers, it can happen that the public majority is actually the minority in terms of private attitudes." Can this situation be overturned and the public opinion will change to the private one?

4.
Authors have stated they would like to run experiments to test their model, but have they found empirical evidence from other authors that support the model?
Of course there are many studies which show that group dynamics, needs for inclusion or desire for social approval influence people in the opinions they express (e.g., the classic Asch paradigm). But we do not know of any research that examines the relationship between the two motives central to this study and the combination of private attitudes and publicly expressed opinions in the longer run: the dynamics shown in this paper have no empirical parallel that we know of. Interestingly, we do not believe that the decision to remain silent has itself been investigated in empirical research much. Including this option may lead to quite interesting results.
5. Have the authors considered setting different topologies for the networks the agents interact in? If so, could they have a better impact on "common benefits"?
6. Have authors considered other types of agents (instead of just C and R) with different dynamics or the fact that Reputation seekers (Consensus seekers respectively) could change their behavior?
Yes, we have considered other agents (or rather motives) as mentioned in the introduction at the start of paragraph 1.3, lines 80-83. But in the present paper we rely on the very familiar prosocial -proself (or group -individual) distinction. We have not tested any other agents yet.
With respect to changing behavior: see Reviewer 2, point 4.
7. Is the code for the model available?
Yes, we now include the simulation code in the supplementary material.
Reviewer #2: 1. The authors propose two motivations for agents to either remain silent or to voice their attitude. At times, the authors seem to suggest that their model provides insights into these motivations (e.g., line 141-147). However, the obtained results are rather descriptive and informative with regard to cyclical patterns of public opinion fluctuation and the discrepancy between private and public opinion on the macro level. Behavior that may be determined by striving for consensus or reputation on the other hand is determined by the model characteristics. In fact, I wonder whether the proposed pattern of behavior (Table 1) could not be explained by other motivations as well. Either way, while setting the stage and interpreting the findings it would be helpful to clearly distinguish between assumptions and potential explanations and warranted conclusions.
The reviewer's point includes two questions: a. Whether the proposed pattern of behavior could be explained by other motivations.
b. Whether we can clearly distinguish between assumptions and explanations and conclusions.
To address this point, we carefully read the manuscript again, to check whether our language reflected the difference between assumptions and conclusions. Where this was unclear, we made explicit when we were talking about assumptions...

For instance, in the introduction:
" We assume that those who individualistically seek to acquire a good reputation (reputation seekers) do so by airing their views when those views are both distinctive and likely to be supported---in other words when they can count on the support of a minority but they do not simply restate that which everyone says. With respect to pro-socially motivated people who seek to maintain group unity (consensus seekers), we assume that they will do so by airing their views when these are aligned with what the majority publicly expresses." (lines 141-147) And in the discussion: 2. I highly appreciate the choice of method that the authors employed. As it is less commonly used within social psychological research -the field in which the study is theoretically rooted -it would be helpful to see a short justification for this choice.
Thank you for emphasizing this point. We agree and included a short justification at the start of paragraph 2: "The use of ABM provides unique advantages for the study of social phenomena as it enables the linking of distinct levels of analysis. Indeed, by using properties of single agents to predict emerging behaviors in their immediate social networks one can explain phenomena at a societal level (Vallacher, 2017). By observing how the model behaves over time, the ABM approach allows us to check the consistency of our theoretical assumptions, and to examine results for unexpected and emergent properties (Goldstone, 2005)." (lines 171-181) 3. The authors decided to vary the size of the system and the relative numbers of reputation and consensus seekers; however, no rationale explaining this decision is provided: From the theoretical background, it is unclear why these two factors were selected to vary.
There is a methodological and theoretical rationale for varying these factors. From a methodological point of view, it is a logical first step in testing the model. For complex systems research it's common practice to study how systems stabilize and behave by varying the size of the system and the other control parameters. This allows us to validate the model's stability and to better understand its properties. Concerning the theoretical rationale for varying these factors, we now explain this in two sentences at the end of paragraph 1.3: "Because societies diverge in the extent to which each of these two motives is encouraged, and thus, likely to occur (e.g., Hofstede, 1980), we are interested in seeing how varying the relative number of reputation seekers and consensus seekers affects dynamics of public opinion. Besides this, we vary the size of the system, to get an insight in the stability of these dynamics when systems get larger." (lines 147-151) 4. With regard to future research, the authors could add some thoughts regarding potential downstream consequences on personally held attitudes for which the model currently does not assume any changes (line 167).
Thank you for this suggestion. Indeed, under specific conditions, attitudes can change as a result of being discussed (see for example the literature on persuasion). In fact, the current research provides clear support for a future research agenda that examines the dynamics between private attitudes and expressed opinions over time. We now make this explicit in paragraph 4.3: "This disconnect between public discourse and private attitudes is most striking in research which relates news content (i.e., one supposedly influential form of public discourse) to immigration attitudes. In line with our assumptions, this shows that immigration attitudes are only marginally related to news discourse overall, but that attitudes do change to the degree that news portrays individual immigrants: here, exposure to immigrants tends to improve attitudes (Boomgaarden, 2009). This implies that it is not just important to research why publicly expressed opinions may fluctuate but also why private attitudes may fluctuate. Studying both in tandem may shed light on their relationship." (lines 644 -652) 5. I could not find any information about the software used for the modelling and whether a script is available to allow replication.
6. Minor comments: In the abstract, it should be "rising concerns" and "however, statistics on private attitudes paint a different picture". Moreover, I recommend substituting the word "parrot" (e.g., line 108) with a less evaluative word.
Thank you for pointing this out. We made the changes you suggested.