Topic territoriality and the cost of civility: examining the impact of IP address disclosure on Weibo

Abstract

This article introduces the concept of “topic territoriality,” a mechanism that governs participation in conversational spaces. When a discussion becomes prone to territorialization, individuals are more likely to claim topics (participating in discussions about topics they own as “stakeholders”) and defer (reducing participation in topics owned by others). They are also more likely to patrol topic boundaries (monitoring who is participating and confronting topic “intruders”). We document the operation of topic territoriality by analyzing 112,278 conversational turns on Weibo before and after a policy that reveals users’ broad geographic locations. We find that revealing these locations increased territorial behaviors, leading to more homogenous participation in conversations. Although the display of locations has improved the overall civility in language, the confrontations between stakeholders and intruders became more toxic. Our research emphasizes the impact of topic territoriality in online conversations and sheds light on the unintended consequences of social media policies.

Social media interactions, like those offline, are often based on group identities. Unlike in offline interactions, however, the fact that individuals on social media can easily move from conversation to conversation challenges the stability and homogeneity of conversational groups (Zhang et al., 2021). In offline space, individuals typically choose whom to communicate with before deciding what to say. In online space, by contrast, the topic can act as the magnet, pulling in participants who do not know whom they will meet in the conversation (Yang & Peng, 2020).

This fluidity in participation may suggest the suppression of group dynamics. If people are choosing what to talk about, instead of whom to talk with, they should be free from biases such as a preference for interacting with in-group members, or a hostility toward outgroups. But participation may still depend on identity, as topics themselves can also carry implicit group-oriented associations. This possibility raises new questions about how group dynamics influences participation in spaces.

In this study, we explore one such set of dynamics by introducing the concept of “topic territoriality.” A topic becomes a “territory” when it is perceived that members of one social group have a greater right to speak about it than members of other groups do. The territoriality of a topic is an emergent, collective property that arises from aggregation of simple communication behaviors performed by individual group members. These behaviors can, in turn, create a self-reproducing feedback loop in which the more people see that only one group speaks on a topic, the more that topic appears to be “their territory.”

We argue that topic territoriality emerges because individuals are inclined to assess identity-based differences in standing with respect to topics of conversation. This perceived standing may arise from the perception that members of certain groups, on specific topics, have a greater expertise on the topic, a greater vulnerability to the consequences of the conversation, or some combination of these. This perception creates a divide between topic stakeholders—those who are motivated to “claim” the topic—and topic intruders—those who do not or cannot claim it. These stakeholder and intruder statuses motivate individuals to alter and negotiate participation in conversational spaces similarly to the way they do so with physical spaces through behaviors we call claiming, deferring, and boundary patrolling. Stakeholders assert their authority by claiming ownership of the topic, while non-stakeholders defer, recognizing it is not their place to speak. Stakeholders patrol to prevent intrusions by non-stakeholders, ensuring their group’s control over the topic.

Topic territoriality is a theory of group interaction and is consistent with principles of inter- and intra-group interaction in online spaces, such as homophily (McPherson et al., 2001), Social Identity Theory (SIT; Tajfel & Turner, 1979), and the Social Identity model of Deindividuation Effects (SIDE; Reicher et al., 1995). However, distinct from these theories, it emphasizes a contextual variable—the topic of conversation—with which these interactions vary. Topic territoriality is also like the spiral of silence (Noelle-Neumann, 1974) or group polarization (Poole & Rosenthal, 1983), a theory of how individual-level perceptions can create an emergent, and self-reinforcing, collective phenomenon.

These features create a research challenge: identifying spaces where perceptions of identity are relevant and vary and where identity information is revealed, or hidden systematically, rather than by individual discretion. However, on April 28, 2022, Weibo, a prominent Chinese social media platform, enforced a policy mandating the disclosure of each user’s Internet Protocol (IP) address, revealing users’ geographic location (Baptista, 2022). By revealing each user’s IP address, all other users could now see the geographic location, at least to the level of the city/province in China, of everyone they interacted with. Geography was also a source of inter-group tension in China during COVID-19, as the pandemic and the Chinese government’s responses toward it shaped shared consequences (among group members) and differentiations in status (between groups) on Weibo, producing “stakeholder” status for different conversations. This mandate thus provides us with a unique opportunity to observe conversational behavior in two regimes—one in which relevant identity cues are sparse, which should weaken territorial behavior, and one in which relevant identity cues are ubiquitous, which should strengthen it.

In this study, we analyzed 112,278 conversational turns related to COVID-19 on Weibo 2 months before (N = 53,672) and 2 months after (N = 58,606) the policy’s implementation to measure who was participating and the level of toxicity in these interactions. Our findings suggest that interactions after the IP disclosure policy became more territorial. Individuals exhibited a stronger tendency to claim, defer, and boundary patrol. While the increased transparency of IP addresses did lead conversations to be less toxic, stakeholders’ confrontations with intruders became more toxic. Thus, by failing to account for territoriality, the policy inadvertently increases homogeneity in group-oriented conversations and potentially hinders supportive behaviors such as allyship.

We begin by reviewing the literature of human territoriality and extend it to online spaces. We then introduce the concept of topic territoriality and differentiate it from other established theories. Our empirical examination focuses on COVID-19 discussions on Weibo and the platform’s IP disclosure policy. Finally, we present our findings and discuss their implications.

Human territoriality in online spaces

Territoriality refers to behaviors demonstrating exclusive use or ownership of a physical space. It is exhibited by both humans and animals (Edney, 1974). For humans, a sense of territory is a psychological construct residing in the mind of the individual (Pierce et al., 2003). When individuals establish a psychological sense of ownership over an object or space, they are more likely to assert their position and exhibit active and reactive behaviors to claim and defend it (Altman, 1970; Goffman, 1971).

Territoriality is tied to group behavior. Physical territory is valuable to both animal and human groups because it provides benefits in terms of group security and group recognition (Ardrey, 1966). For humans, territory is also endemic to group interaction and identity. For example, in Sherif et al.’s (1954, p. 94) seminal experiment on group formation and conflict, in which boys were divided into groups at a summer camp, they found: “As the in-group structures of the Rattlers and Eagles became delineated, members formed attitudes toward objects and places of functional importance to them, appropriating these objects as ‘ours.’”

Despite this importance of group interaction, territoriality has received limited attention from communication researchers. However, territoriality can be easily extended to communication in online spaces. For example, particular spaces may be formally claimed by groups, such as Reddit forums that welcome only members of particular identities (Gaudette et al., 2021). Territorial behavior need not be so explicit. The concept of psychological ownership suggests people can act “as though” they own things even when this ownership is not formally recorded, and others may defer to them in this way (Kirk et al., 2018).

Exploring topic territoriality: mechanism and impact

Topic territoriality emerges from individual perceptions that lead to collective behavior reinforcing those perceptions. Similar to the spiral of silence (Noelle-Neumann, 1974), where the belief that a view is unpopular leads to its suppression and further reinforces that belief, topic territoriality encourages individual behaviors—claiming, deferring, and boundary patrolling—that further the appearance of ownership by one group. Thus, a topic can emerge as being owned by one group through conversation, even if the inclinations toward such a consensus were not strong when the conversation began.

Our argument starts with the notion that individuals perceive topics as “ownable,” similar to physical territories or objects. We posit that topics are likely perceived as ownable because people recognize that (a) expertise and/or (b) vulnerability should privilege some voices over others, though there may be other reasons as well. Those with expertise or vulnerability are perceived as entitled to speak, while those without these qualities are not.

The discussion of a topic can impact different groups unequally. When a topic concerns Group A’s identity, experience, or a crucial issue affecting them, Group A has a large “stake” in the discussion (Lewis, 2007). Due to their differential experience or vulnerability, these stakeholders have an incentive to exhibit territorial behavior, defending the topic from outsiders they perceive as intruders who might distort the conversation. Non-stakeholders, recognizing the risks of entering these “defended” spaces, should then tend toward caution.

Of course, the presence of these perceptions does not guarantee agreement or consensus. Much as physical territory can be disputed, different groups may both perceive that a topic should be “theirs.” There may also be controversies because some individuals do not think that anyone should “own” this topic. Nonetheless, there are reasons to suspect that, in many cases, perceptions will be consistent, leading a topic to emerge as a “territory.”

Differentiating topic territoriality from other theories

Although topic territoriality shares similarities with other theories of online interaction, it operates through different mechanisms and variables. One similar theory is homophily, in which individuals choose to interact and build relationships with people who share similarities with them (McPherson et al., 2001). While homophily and territoriality both lead individuals to interact more with in-group members online, the mechanisms, and thus some of their consequences, are distinct.

Homophily can mean a mechanism that leads people to seek out relationships with similar others (literally “same loving”). Homophily can also be an observed result of individuals forming connections based on connections to shared objects like topics, interests, or affinities. In this sense, topic territoriality can create observed homophily—a conversation among similar others—however, the mechanism is different. Specifically, topic territoriality suggests a more dynamic and sometimes contentious process.

First, territoriality can encourage heterophilous interactions. Stakeholders assert ownership over topics not only by speaking about it to other stakeholders, but also by seeking out and challenging perceived intruders (boundary patrolling). This potentially increases inter-group interactions, at least in the short term. Second, unlike homophily, territoriality can suppress all interactions, even among the like-minded. A conversation about a topic where everyone perceives they are an “intruder” can produce complete silence. This could occur in a group of otherwise like-minded individuals, that is, those who, by homophily, should interact.

Topic territoriality is informed by the principles of SIT (Tajfel & Turner, 1979), as it assumes that in-group members will have more favorable intentions toward one another than out-group members. Importantly, SIT accounts for the context-driven activation of different identities, recognizing that individuals may identify differently based on the situation. Thus, according to SIT, individuals may identify with different groups depending on the topic of conversation. In topic territoriality, the identification with the group is fixed, while the variation in behavior is based on an assessment of which group has ownership. It thus reflects a varying relationship between groups, not variation in an individual’s identity. In topic territoriality, the identification with the topic is fixed; it is the inter-group relationship that changes with the perception of ownership.

SIT also posits a consistent relationship between groups. Individuals favor in-groups and have more (though not necessarily substantial) conflict with out-groups. However, topic territoriality predicts that inter-group behavior can shift based on topic, even as identification and perceptions of in-groups and out-groups remain fixed. For example, group A may act like the “owner” of topic A, showing great hostility toward group B on this topic, but then deferring more politely and graciously to Group B on topic B. This idea that individuals will in some situations defer to outgroups on certain topics, privileging outgroups’ speech over their own, is not covered directly by the logic of SIT.

Topic territoriality is also related to but distinct from the SIDE model (Reicher et al., 1995). The SIDE model emphasizes how individual group-oriented behavior varies with an individual’s sense of identification with their group, with behavior being influenced by the degree of group identification. Territoriality highlights the variation in behavior based on specific topics for a given level of group identification. It underscores the significance of how identity is perceived by others in relation to a particular topic, rather than just the salience of identity to the individual. Essentially, those perceived as outsiders to a topic, regardless of their self-identification, are seen as intruders. Thus, the impact of territorial dynamics is influenced by how identity cues are perceived by others, focusing on the topic as a key driver of behavior, in contrast to the SIDE model, which focuses on the salience of identity to the individual.

The dynamics of topic territoriality: claiming, deferring and patrolling

Claiming ownership is a basic territorial behavior. One reason stakeholders can claim ownership of a topic is because they have a greater expertise related to that topic. In particular, their claiming of topics reflects a need to express their connection to specific issues (Miller & Effron, 2010; Sherf et al., 2017). Thus, individuals tend to believe they possess the legitimacy to act on a topic or an issue. An example is spaces discussing diseases and health concerns, such as cancer patient forums (Lee et al., 2019). In these spaces, it is recognized that current cancer patients and survivors should have a greater voice than non-patients. Individuals may begin their comments by stating their health status to establish their right to speak.

Stakeholders may also claim ownership of a topic due to their greater vulnerability in discussions related to that topic. Ideally, conversations where all participants’ information is accurately weighed and considered can lead to wiser outcomes (Sunstein, 2006). However, people tend to avoid sharing information that contradicts their beliefs or discussing topics that do not benefit their stance (Aruguete & Calvo, 2018; Green et al., 2020), and conversations can involve participants speaking without directly engaging with each other (Jiang et al., 2023). This creates a risk that other factors, such as power or convenience, over-rule careful and concerned reasoning, exposing the vulnerable. As Fraser (1990, p. 64) quotes Jane Mansbridge stating “subordinate groups sometimes cannot find the right voice or words to express their thoughts, and when they do, they discover they are not heard [or] heard to say ‘yes’ when what they have said is ‘no.’” If stakeholders do not speak up, the topic risks becoming “common property” in which everyone comments, even though the stakeholders bear the brunt of the consequences. This concern is heightened in online spaces, where conversations dominated by outgroups are more likely to go viral (Rathje et al., 2021). Thus, it is crucial for stakeholders to claim topic ownership to guide the discourse more effectively. A good example of vulnerability conferring stakeholder status is the #MeToo movement. Survivors of sexual assault claimed ownership of the narrative to ensure their voices were prioritized and accurately represented (Jackson et al., 2020). Similarly, in discussions of disability policies (Fox & Kim, 2004), people with disabilities are significantly more affected by these policies than others, so they should have a greater say in the discourse.

Territoriality is also recognized by non-stakeholders. Once they recognize that they are not part of the stakeholder group, people should defer topics to others due to lack of expertise or experience of consequences. Political science has observed the concept of “issue ownership” (Ansolabehere & Iyengar, 1994), which shows that the voting public often perceives candidates of particular identities to be better at handling a particular issue, regardless of the actual policies championed by that party. For instance, when voters view sexual harassment as a key issue, they often show a preference for female candidates (Ansolabehere & Iyengar, 1994).

Patrolling is another stakeholder territorial behavior concerned with the consequences of non-stakeholders “intruding” on the conversation. Stakeholders fear that if those outside their group enter the conversation, they will either pollute it with their weaker knowledge or push the conversation toward views or statements that are harmful to them. Therefore, stakeholders have an incentive to “patrol” and “enforce” the borders of their topic, making sure that members of their group can control the conversation.

With patrolling by topic stakeholders, people are also cautious about entering others’ domains (i.e., becoming topic intruders). Commenting on unfamiliar topics tied to out-groups often lacks credibility and may be seen as judgmental, causing unnecessary tension and conflict (Hornsey et al., 2007). Therefore, individuals must consider their motivations, respect identity-related norms, and choose constructive dialogue strategies when responding to topics owned by others (i.e., topic stakeholders).

Topic territoriality on COVID and IP disclosure on Weibo

COVID-19 is a topic where people are particularly concerned about its impact relative to their geographic location (Franch-Pardo et al., 2020). Due to their proximity, individuals should be more familiar with and possess a greater sense of ownership over COVID topics pertaining to their local areas than those in other regions.

Specifically in China, the pandemic and the government’s COVID policy turned regional groups into stakeholders. During the pandemic, there was broad national interest in how well COVID was being contained in each particular region, and speculation was commonplace about where COVID would spread and whether individuals in all regions were compliant with government policy (Burki, 2022; Guo et al., 2023). The virus itself also posed area-specific health risks. Thus, different regions might be targeted for criticism (e.g., if their COVID cases were rising) or stricter lockdowns (e.g., if the government believed these were needed). This dynamic became apparent as people kept track of COVID cases locally and across the country.

These concerns created a stakeholder status—residents of a given locale had a substantial interest in (a) limiting inaccurate information about their region spreading in online conversation; and (b) limiting any negative portrayals of their region from spreading. That is, these stakeholders had an incentive to “patrol” conversations about their region and make sure that those commenting were also part of the stakeholder group.

These factors should lead to a tendency for residents to claim topic ownership or in discussions of their region. In contrast, when the topic concerns regions outside their own, individuals may exhibit a weaker sense of ownership. In these instances, people often defer to the insights of those living in the affected regions (Huang et al., 2013; Lin & Margolin, 2014). Thus, we hypothesize that:

H1: People will act as topic stakeholders by a) posting more frequently and b) replying more frequently to COVID-related topics they own than expected by random chance.

H2: People will avoid acting as topic intruders by a) posting less frequently and b) replying less frequently to COVID-related topics owned by others than expected by random chance.

The recognition and enforcement of topic borders thus go beyond how individuals feel about their own relationship to the topic and extend to how others should or should not engage with it. However, in most social media settings, this ownership is known to the person speaking, but not to the public/audience. For example, a user in Beijing commenting on COVID-19 knows they are discussing their own community, yet this is not apparent to others unless explicitly stated (saying, for example “as a resident of Beijing…”). Thus, identifying “intruders” in such discussions is challenging without them revealing their outsider status.

A unique policy on Weibo, however, suddenly revealed stakeholder and intruder status related to regional groups. On April 28, 2022, Weibo introduced a controversial policy that required the disclosure of every user’s IP address, implemented without public consent. After the policy, all users could see where all other users were commenting from, suddenly revealing group identities related to geography to all users. Thus, now it was not only the case that users from Beijing would comment on COVID-19 in Beijing (asserting claims as stakeholders), but also that users could see whether individuals from outside Beijing were also commenting (i.e., topic intruders).

The policy thus facilitated border patrolling on the topic of COVID-19 in a particular location. It follows that, if topic territoriality is in operation, such patrolling will intensify. Namely, stakeholders can now see who the intruders are and so will confront them. When others’ IP addresses are revealed, individuals should become more vigilant and responsive to potential threats or challenges to their group’s norms and values (Pierce et al., 2003). In line with this reasoning, the entrance of topic intruders into stakeholders’ topics is expected to be less frequent. On this basis, we hypothesize the following:

H3a: After the policy requiring IP address disclosure, topic stakeholders are more likely to reply to topic intruders than they were previously.

H4a: After the policy requiring IP address disclosure, topic intruders are less likely to reply to topic stakeholders than they were previously.

As clear boundaries intensify confrontations, it also affects the manner in which topic stakeholders and intruders engage in confrontation. Specifically, topic stakeholders’ confrontation with topic intruders should be more toxic (i.e., impolite, discourteous, or irrational, and compelling others to withdraw from a conversation). As described earlier, once territories are made clear, stakeholders are motivated to patrol the boundaries and confront intruders. Consequently, these confrontations are expected to be more hostile compared to situations where users’ IP addresses remain hidden. In contrast, based on topic territoriality, when “topic intruders” engage in discussions owned by stakeholders, they tend to be less confrontational. Furthermore, decreased anonymity can prompt these intruders to use more respectful language, as they become aware of the real-world accountability for their actions (Rowe, 2015). This awareness can lead to a more mindful and considerate interaction. Thus, we predict that:

H3b: After the policy requiring IP address disclosure, topic stakeholders’ confrontations with topic intruders become more toxic than they were previously.

H4b: After the policy requiring IP address disclosure, topic intruders’ confrontations with topic stakeholders become less toxic than they were previously.

Civic consequences of the policy

General change in language toxicity

According to media reports, the policy of IP address disclosure was part of the regulators’ effort to increase civility online (Baptista, 2022), but territorial behaviors can complicate the matter. The disclosure of an individual’s IP address should, all else equal, promote civility. Reduced anonymity is found to reduce the prevalence of trolls that contribute to toxicity online (Rowe, 2015). In essence, reduced anonymity serves as a deterrent, dissuading individuals from participating in toxic behavior. Nevertheless, the disclosure of an individual’s IP address can potentially escalate toxic language, primarily among those willing to transgress boundaries. This increased ease of targeting out-groups facilitated by IP address disclosure may lead to a rise in toxic language. We thus ask the following:

RQ1: How does the policy requiring IP address disclosure influence the overall language toxicity of a) posts and b) replies on Weibo?

Topic homogeneity

Another civic value in online discussion is the sharing of information, particularly across boundaries. By making topic boundaries more prominent, an unintended consequence of this could be topic homogeneity. Topic homogeneity occurs when a higher percentage of participants within the same group are involved in the discussion, potentially limiting diversity of perspectives as well as support from outsiders, both of which are valuable. In intergroup settings, outgroup support or “allyship” is crucial for a group’s well-being, but homogeneous discourse can discourage it. For example, racial minorities rely on majority support to combat biases (Kam et al., 2022; Roden & Saleem, 2022). Territoriality, by promoting “staying home,” may lead to more bystanders than allies.

All else equal, increased topic homogeneity is a mathematical consequence of the combination of the patterns predicted in H3a and H4a. If, holding other effects constant, fewer intruders’ comment on topics (H4a), this will increase the ratio of stakeholder to intruder participation. Similarly, if, holding other effects constant, intruders’ comments are more likely to draw stakeholders’ replies (H3a), then this will also increase this ratio. However, this homogenization effect does not follow tautologically from H3a and H4a. It is possible that another aspect of the policy reduces the base rate of in-group participation, counteracting the effects predicted in H3a and H4a. We thus consider this a separate empirical hypothesis:

H5: After the policy requiring IP address disclosure, topics will have a greater proportion of stakeholders.

Figure 1 visually illustrates the underlying concepts of the hypotheses.

Method

We collected relevant Weibo data, including posts, replies, and IP address information, by simulating the browsing behavior of regular users. This was achieved through a Python script designed for web scraping, a method chosen due to Weibo’s restrictive and inefficient API, as highlighted by Yang & Peng (2020). Weibo is a microblogging platform like Twitter. The platform features short, real-time posts and replies. Users can follow other accounts and share updates within a specified character limit.

Our data collection focused on Weibo posts containing COVID-19-related keywords (specifically, “COVID” and its Chinese translation “新冠”) and references to 16 major cities in China that experienced COVID outbreaks during this period¹ (see “Full List of Cities and Keyword Combinations” in the supplementary materials).

To comprehensively study the policy impact, we collected data for two months both before and after the policy date (February 28, 2022 to June 28, 2022). We chose this timeframe as it provides a reasonable window to assess the policy’s effects. We also note that there was no significant social event that happened during this period other than COVID-19. Additionally, we conducted robustness checks using data from 1 month, 1.5 months before, and after the policy date, and found that our main results remained consistent (see “Robustness Check,” Supplementary Tables S3–S10 and Supplementary Figures S3–S6). Since we were interested in conversational turns, posts without replies were excluded. The final dataset had 112,278 conversational turns by 59,621 repliers toward 9,047 posts made by 6,972 posters.

Measurement

The IP address disclosure policy

On April 28, 2022, Weibo imposed a policy enabling the public to view users’ IP addresses, labeled with the city/province for locations within China and as country names for locations outside China (see Supplementary Figure S8 for a visual illustration). While the technical details of IP addresses may not be widely understood, their significance, particularly in revealing one’s location to the public, was acknowledged and discussed on Weibo as a trending topic.² Prior to its implementation, there were 3,892 posts (43.02%) and 53,672 comments (47.80%) on COVID. After the policy change, the IP addresses of users became visible. We treat the IP address revealed after the policy to be an indicator of the individual’s location before the policy. This may appear to be a strong assumption, however, an analysis of these addresses revealed significant stability. Specifically, we randomly sampled 100 users and found 96% of the users had their IP addresses align with their geolocation traces reflected in their manually added geo-tags or geolocation information indicated in their posts. This is consistent with the strict government rules restricting physical movement during this period (Yuan, 2022). Although users can change their IP address by using a virtual private network (VPN), the use of VPN is restricted by the government (Qiang, 2019). Finally, it is important to distinguish how users feel attached to a place from how they are perceived to be affiliated. That is, even if a user with a Shanghai IP address does not identify with Shanghai, the IP address grants them perceived “entitlement” to participate in the topic, at least in the eyes of other users.

Perceived poster–replier relationships

We categorized poster–replier relationships based on the locations mentioned in the posts and the IP addresses of the posters and repliers. The posters have 50 unique IP addresses, while the repliers have 95 unique IP addresses. Individuals whose IP addresses match the location of a topic are termed “topic stakeholders,” while those whose IP addresses do not match the location are termed “topic intruders.” Through our categorization process, we identified five distinct types of “poster-replier” relationships (see Table 1). Our primary focus in presenting results is on two relationships that are particularly relevant to topic territoriality: stakeholder–intruder (N = 7,925) and intruder–stakeholder (N = 22,135). The results of the remaining three relationships are included in the supplementary materials (Supplementary Tables S1–S2; Supplementary Figures S1–S2). It is possible that a post contains multiple locations that both the poster and the replier identify with, yet their locations are different. However, such instances are rare (N = 2,839 out of 112,278 or 2.529%) in our dataset. We conducted our analyses excluding these instances and found no substantial changes in the results (see “Robustness Check,” Supplementary Tables S13–S16 and Supplementary Figures S9–S10).

Toxicity and related attributes

We aimed to assess the overall language toxicity in Weibo content, not specific types (Guo et al., 2023), so we used the Google Perspective API.³ The API uses machine learning models to identify a range of abusive comments. We examined the primary attribute of “toxicity,” which encompasses comments that are rude, disrespectful, or unreasonable and have the potential to drive people away from a discussion. This last element is the essence of boundary monitoring. Additionally, we incorporated three other attributes—insult, threat, and identity attack. These were included because they not only strengthen the robustness of our findings but also provide the potential for additional insights into the nature of online discourse. The API assigns a probability score from 0 to 1 to indicate the likelihood that a reader perceives toxicity (replies: M = 0.087, SD = 0.110; posts: M = 0.123, SD = 0.120), insult (replies: M = 0.043, SD = 0.084; posts: M = 0.069, SD = 0.101), threat (replies: M = 0.019, SD = 0.053; posts: M = 0.022, SD = 0.043), and identity attacks (replies: M = 0.021, SD = 0.059; posts: M = 0.048, SD = 0.077). The accuracy of Perspective API on English tweets was validated by a previous study (Frimer et al., 2023), but we also independently validated its accuracy and found the API effectively predicts human reactions in the Chinese language context (see “Validity of Perspective API on Chinese Text” in the supplementary materials). Table 2 presents the examples of statements rated as low and high in toxicity by the API classifier.

Topic homogeneity

This is measured by the proportion of stakeholder participants in topics. The mean proportion of stakeholder participants across all posts, which varies between 0 and 1, is 0.429 (SD = 0.407) or 42.9%.

Control variables

To account for potential confounding factors, we included three control variables in our analysis. Specifically, we included the log-transformed number of comments (M = 4.145, SD = 1.944), the number of likes (M = 4.518, SD = 2.731), and the number of followers (M = 9.888, SD = 4.321) for each post initiator. These post-level variables are indicators of a post’s popularity and may influence interactions between posters and repliers, as well as their language styles.

Analytical procedures

The first two hypotheses are about users’ probabilities of posting and replying to their own or others’ topics. To test these hypotheses, we compared the observed probabilities with a random baseline. To establish the random baseline, we shuffled posts and authors randomly to establish a baseline probability for how often users posted specific topics. We perform permutations 1,000 times to create a distribution against which to compare our observed data. Then, we compared what we observed with what we would expect if users were randomly choosing topics to discuss and reply to. If the observed distribution appeared only rarely in the random pairings, we rejected the null hypothesis that the observed pattern was due to chance. This method has been used in other similar contexts (Koh et al., 2019; Park et al., 2023).

The remaining hypotheses and the research question concern the effects of the policy, so we compared the dependent variables after the policy with those before the policy. Specifically, since the “poster-replier relationship,” a five-level categorical variable, is the dependent variable in H3a and H4a, we used multinominal logistic regression. Linear regression was used for the remaining hypotheses and research question since there were no categorical variables with more than three levels in those analyses.

Results

Our first two hypotheses propose that individuals are more inclined to post and reply to topics owned by them (H1), while being less likely to do so for topics owned by others (H2). To test these hypotheses, we compared the observed probabilities to a random baseline. Our results indicate that individuals are more likely to post (M = 0.3316, 95% CI [0.3257, 0.3379]) and reply to (M = 0.3949, 95% CI [0.3932, 0.3966]) their owned topics (i.e., topic stakeholders), supporting H1. They are less likely to post (M = −0.0177, 95% CI [−0.018, −0.0174]) and reply to (M = −0.0218, 95% CI [−0.0219, −0.0217]) others’ topics (i.e., topic intruders), supporting H2.

We note that this statistically significant difference accounts for the tendency to select one’s “own” topic out of any other topic that could be commented on. However, because most topics available are not one’s own, the nominal rate of commenting on other topics is greater. For example, a user (located in Beijing) replied 3 times on their own topic, and 1 time each on 7 other topics. Thus, they showed a bias toward replying to their own topic—it was chosen significantly more often than any other specific topic. However, stakeholder replies still make up a minority of their posts (30%). Understanding this distinction is important for understanding the results for H3a and H4a.

H3a and H4a examine how the policy affects territorial behaviors in topics. H3a predicts increased responses from topic stakeholders to intruders, while H4a anticipates fewer responses from intruders to stakeholders. Table 3 reports the multinomial regression results that show the probability of “intruder-stakeholder” interactions compared with the reference group of “stakeholder-intruder” interactions (see Supplementary Table S1 for a full table with five categories). Specifically, after the policy “intruder-stakeholder” interaction decreased compared with “stakeholder-intruder” (B = −0.743, SE = 0.027, p < .001). The multinomial regression table is extremely difficult to interpret, in part because the coefficients refer to between-group differences (with stakeholder–intruder as the baselines). We thus simplified it by using the estimated marginal means of the main effects. In particular, the probability of “stakeholder-intruder” interactions increased from 0.043 (SE = 0.001, p < .001) to 0.055 (SE = 0.001, p < .001) after the policy; “intruder-stakeholder” interactions decreased from 0.341 (SE = 0.003, p <.001) to 0.211 (SE = 0.003, p < .001). These results support H3a and H4a. A visual illustration is presented in Figure 2 (see Supplementary Figure S1 for visual illustration of all “poster-replier” relationships).

Table 4 shows results for H3b and H4b which compares “stakeholder-intruder” interactions (i.e., stakeholder replies to a post made by an intruder), to “intruder-stakeholder” interactions (i.e., intruders reply to stakeholders). Compared with “stakeholder-intruder” interactions, the policy change had significant interaction effects with “intruder-stakeholder” interactions on all the language attributes (toxicity: B = −0.025, SE = 0.003. p < .001; insult: B = −0.012, SE = 0.002. p < .001; threat: B = −0.009, SE = 0.001. p < .001; identity attack: B = −0.017, SE = 0.002. p < .001). By calculating the estimated marginal means of the policy effects, we found that the policy change encouraged “boundary monitoring,” as it made “stakeholder-intruder” interactions more toxic (B = 0.012, SE = 0.003. p < .001), insulting (B = 0.008, SE = 0.002, p < .001), threatening (B = 0.003, SE = 0.001. p < .001), and containing more identity attacks (B = 0.010, SE = 0.001, p < .001), but had the opposite effects on “intruder-stakeholder” interactions (toxicity: B = −0.013, SE = 0.001, p < .001; insult: B = −0.004, SE = 0.001, p < .001; threating: B = −0.006, SE = 0.001, p < .001; identity attack: B = −0.007, SE = 0.001, p < .001). Therefore, both H3b and H4b are supported. Figure 3 provides a visual representation of these findings (see Supplementary Table S2 and Supplementary Figure S2 for results of all “poster-replier” relationships).

The research question aims to understand the general policy effects on toxicity related variables. Table 5 reports that the policy significantly decreased language attributes in toxicity (posts: B = −0.012, SE = 0.002, p < .001; replies: B = −0.007, SE = 0.001, p < .001), insult (posts: B = −0.008, SE = 0.002, p < .001; replies: B = −0.006, SE = 0.0005, p < .001), threat (posts: B = −0.005, SE = 0.001, p < .001; replies: B = −0.005, SE = 0.0003, p < .001), and identity attack (posts: B = −0.008, SE = 0.002, p < .001; replies: B = −0.005, SE = 0.0004, p < .001). Overall, the policy generally improved language civility on Weibo.

In accordance with H5, we anticipate observing topic homogeneity after the policy. The results presented in Table 6 indicate a significant increase in the proportion of participants who are topic stakeholders after the policy change (B = 0.142, SE = 0.008, p < .001). This supports H5.

Given the availability of timestamps for all posts and replies, we also investigated the response speed of topic stakeholders and intruders. We found that topic stakeholders respond more quickly to intruders, while intruders are slower to reply to stakeholders. This supports “topic territoriality” after the policy requiring IP address disclosure (see “Analysis of Policy Impact on Response Speed,” Supplementary Table S11 and Figure S7).

Discussion

The present study introduces the concept of “topic territoriality” as a set of mechanisms that influence how individuals manage participation in conversational spaces based on the presentation of topic and group identities. Through studying a unique policy on Weibo with a corpus of 112,278 conversational turns on Weibo, we provide empirical evidence that topics are treated like “territories” which are “claimed,” “deferred to,” and “patrolled” like physical spaces.

This study examines territorial behaviors in online spaces, extending theories of human territoriality to online communication where topics are central. While similar to concepts such as homophily (McPherson et al., 2001), SIT (Tajfel & Turner, 1979), and the SIDE model (Reicher et al., 1995), our concept of topic territoriality identifies the conversation topic as a key variable influencing interactions. Using COVID-related discussions on Weibo, we explore how the IP address disclosure policy affects participation. This mirrors physical territoriality, involving claiming space, showing deference, and patrolling boundaries. Such behaviors can amplify in-group perspectives, leading to topic homogeneity.

Identifying boundary conditions is crucial for new theoretical phenomena. Here are three key boundary conditions for topic territoriality. First, an identity must be perceived as having greater expertise or vulnerability. This occurs when experiential knowledge is valued and relevant, or when discussions significantly impact individuals with that identity. The second condition is the absence of a countervailing social norm of universal participation. For example, in previous centuries, it was held that landowners should have more civic rights because of their expertise and vulnerability, but eventually this view was defeated by the argument that everyone should have a voice (Lepore, 2018). Explicit moderation policies promoting universalist norms may thus reduce territoriality. Finally, for territoriality to emerge, identity must be salient or easily seen. If identities are obscured, territorial dynamics will be weaker, as intruders are less identifiable and stakeholders’ status is not easily perceived or trusted.

Our study on Weibo’s IP address policy and COVID-19 discussions highlights the broader implications of topic territoriality in intergroup communication. We found that people are more active in discussions where they are stakeholders, and revealing identity cues intensifies territorial behavior. Stakeholders become more vocal and vigilant, while intruders become less likely to respond and more deferential. This leads to assertive confrontation of intruders by stakeholders, effectively driving them away and resulting in topic homogeneity.

The concept of topic territoriality suggests that revealing group-associated information can impact participation by inducing stakeholder-intruder dynamics, especially in specific discussions. For instance, self-disclosing pronouns may affect gender-related discussions without influencing other topics. Thus, the revelation of identity cues can have unanticipated effects on communication, emphasizing the need to consider territoriality in communication research.

The fact that territoriality can be subtle and topic-specific may also be problematic in some cases. For example, minoritized groups may seek allies from the majority but may need to explicitly invite such support due to the territorial nature of the topic. Thus, profile pictures that appear to indicate racial or ethnic identity may inadvertently discourage outgroup “allies” from addressing a topic for fear that they will be confronted. To be clear, our point is not to label territoriality as inherently good or bad in any situation. Territoriality may be desirable to highlight specific voices in many cases. Our main point is that because topic territoriality is often hard to see, it may lead to consequences that conversational participants do not expect.

Our study also has key implications for online policies. While disclosing IP addresses can improve civility for some users, it also increases toxicity for others and reduces intergroup communication. Policies that violate user privacy are unlikely in democratic countries, but our findings show that targeting individual behavior can have unintended aggregate effects. This highlights the need for cautious policymaking in online communication (Mejia & Parker, 2021; Yu & Margolin, 2022).

Finally, a recent study (Guo et al., 2023) highlighted an increase in toxicity on Weibo following an IP disclosure policy, which appears to contrast with our findings of reduced toxicity. However, a closer examination reveals that the behaviors they saw increase—specifically, “location-based name-calling” targeting “foreign hostile forces”—involves hostile language from Chinese users towards perceived external adversaries. In other words, they observed increased toxicity in “stakeholders versus intruders,” exactly as we predict and observe as boundary patrolling in topic territoriality. A potential reason for the contrast between our findings of the general change in language toxicity may lie in the different measurements. Our measurement relies on the Perspective API to identify a range of abusive comments, not confined to location-specific insults. None of the Weibo content samples in our Table 2 or Supplementary Table S12 fall under the category of “location-based name-calling” as measured by Guo et al. (2023). Additionally, our study primarily focuses on discussions of COVID-19 in Chinese cities rather than “foreign hostile forces.” Therefore, both studies identify a similar pattern resulting from the Weibo policy; however, our analysis centers on the broader concept of topic territoriality.

Limitations and future directions

The current study has limitations that open avenues for future research. First, the strict censorship on Chinese social media may have affected the completeness of our data, possibly filtering out extremely toxic comments. However, our primary interest—the increase in territorial behaviors concerning COVID topics—runs counter to the government’s stated interest in civility and harmony (Baptista, 2022). Thus, we find it unlikely that this effect is a byproduct of government control.

Another challenge with observational data is explaining variance. We are not claiming that territoriality is a major factor in posting choices or toxicity, as shown by low R-squared values (0.041 or less). Instead, we highlight how a small change, like IP address disclosure, can have an unforeseen impact on participation. Given Weibo’s massive user base (586 million monthly active users in Q4 2022)⁴, even small changes can lead to significant aggregate effects. This is crucial considering the link between identity cues and intergroup interactions (Koh et al., 2019; Park et al., 2023), as online behavior can influence offline attitudes and emotions.

In this study, we posit that both expertise and vulnerability influence the perception of topic ownership. Expertise confers legitimacy and authority, while vulnerability emphasizes the need to protect one’s position and interests. These variables are crucial in determining stakeholder status. Future research can explore two key questions: (a) To what extent do expertise and vulnerability influence an individual’s sense of ownership over a topic? (b) To what extent do these factors affect the perception of another person’s ownership over the topic?

Finally, our study is also limited by the lack of a large dataset tracking the same individuals’ repeated behaviors over the two-month period around the policy change. This constrained our ability to provide stronger evidence on territorial topic dynamics. Future research could use experimental designs to more precisely investigate repeated territorial behaviors before and after an intervention.

Supplementary material

Supplementary material is available at Journal of Computer-Mediated Communication online.

Ethical considerations

We acknowledge the ethical implications of collecting and analyzing IP addresses on Weibo. Although the data are public, they can reveal identifiable information and require careful handling. Our dataset limits IP addresses to city or province levels. We used user IDs only for initial data aggregation and removed them during analysis. Sensitive information, including user IDs, will not be shared in replication data to ensure privacy and ethical standards.

Data availability

The data underlying this article will be shared on reasonable request to the corresponding author.

Funding

This work was supported by funding from the National Science Foundation (Award Number: 2106476).

Conflicts of interest: The authors declare that there is no conflict of interest.

Notes

References