Using online negative emotions to predict risk-coping behaviors in the relocation of Beijing municipal government Authors: Qihui Xie, Hongyu Wu, Ruwen Zhang Published: 2024 ABSTRACT AbstractThis article explores the use of online negative emotions to predict public risk-coping behaviors during urban relocation. Through a literature review, the paper proposes hypotheses that anticipate advanced prediction of public risk-coping behaviors based on online negative emotions. The study’s empirical focus is on the relocation of the Beijing municipal government, using time series data for Granger causality analysis in EViews 10.0 software. Data on online negative emotions is sourced from Sina Weibo. After data cleaning, 1420 pieces of data related to the relocation policy of the Beijing Municipal Government within the period from June 9, 2015 to April 28, 2019 are retained. while risk-coping behaviors are measured through public information search behaviors and the incidence of violent crimes, the data coverage is also from June 9, 2015 to April 28, 2019. The results indicated that: (1) Online negative emotions regarding the relocation policy predict public risk-coping behaviors in advance. (2) Negative comments are more effective predictors than negative feelings; (3) Negative emotions about relocation policy formulation predict risk-coping behaviors better than those related to policy effectiveness and implementation; (4) Negative emotions from individuals better predict public risk-coping behaviors than those from institutions; (5) Negative emotions from key stakeholders better predict public risk-coping behaviors than those from non-key or marginal stakeholders. It is recommended that relevant departments establish a real-time monitoring system to track negative public opinions and emotions expressed online, adopt a stakeholder-centric approach to facilitate communication, and promote transparency and educational campaigns to address the challenges of urban relocation. In future studies, methods such as expanding the sample size and adding indicators will be used to address the limitations of potential bias in sample data.Keywords: Negative emotions online, Information search behavior, Violent behavior, Granger causality analysis, Relocation of Beijing municipal governmentSubject terms: Psychology and behaviour, Socioeconomic scenarios, Sustainability Introduction In the era of big data, the advent of online public opinions and the sentiments expressed by netizens have heightened the complexity of the urban relocation policy process. The surge in negative emotions observed online poses a challenge, potentially leading to offline public risk-coping behaviors. However, amid these challenges, it is crucial to recognize the positive impact of online public opinions and emotions. These digital expressions can serve as a mirror reflecting the genuine attitudes and coping mechanisms of the public during the urban relocation process. Moreover, they offer a unique opportunity to anticipate the occurrence of mass incidents. This study aims to validate the hypothesis that negative emotions online can predict the public’s risk-coping behaviors in the relocation process. If confirmed, it implies that real-time data on negative emotions can be monitored during urban relocation policy implementation, enabling timely and effective prediction of risk-coping behaviors. This approach serves as a supplementary method towards conventional means of policy risk assessment. The existing research substantiates the significant role of negative emotions in shaping public risk-coping behaviors1–3. Moreover, it establishes that negative emotions possess predictive capabilities regarding the public’s risk-coping behaviors4. However, the current body of research seldom addresses two critical questions: first, whether negative emotions expressed online can predict the public’s risk-coping behaviors in the context of urban relocation policy; second, exploring factors that influence the accuracy of this prediction. This study aims to address these two research questions. This paper posits hypotheses suggesting that negative emotions can predict the public’s risk-coping behaviors during the Beijing Municipal Government relocation. The accuracy of prediction is proposed to be influenced by types, issues, participants, and stakeholders, as derived from a comprehensive literature review. These hypotheses are then tested using the Granger Causality Analysis of time series data. The time series data on online negative emotions are sourced from Sina Weibo, while risk-coping behaviors are quantified through public information search behaviors and the incidence of violent events. Literature review and research hypotheses The prediction of risk-coping behaviors by negative emotions Risk-coping behaviors in urban relocation The displacement and resettlement resulting from urban relocation5,6have garnered increasing attention from scholars7. This phenomenon, deemed essential for enhancing sustainable urban development, concurrently introduces social risks to urban management8. The success of relocation plans hinges on the improved quality of life and economic well-being for the relocated public, along with the fulfillment of reasonable demands9. Successful implementation relies on cooperative efforts among all actors, stakeholders, and the public, fostering mutual understanding based on shared goals and visions10. This collaborative approach actively promotes the relocation process, attains expected policy outcomes, and contributes to urban development objectives11. However, if the living environment deteriorates post-relocation, resulting in short to medium-term livelihood stagnation and poverty, the public may engage in various risk-coping behaviors, impacting urban sustainable development and social stability. In situations of perceived risk, individuals undertake risk and response assessments, deciding on actions they deem appropriate12,13. Defined as conscious, purposeful, and flexible changes in emotions, cognition, behavior, and environment in response to stress and risk, risk-coping behaviors are conscious efforts to reduce the perceived risk of a threatening event in a given situation14,15. Although integral to individual stress management, risk-coping behaviors also hold significant value for effective social risk management. In this study’s research scenario, the coping behaviors adopted by the public to safeguard their interests introduce uncertain risks in the urban relocation process. Public risk-coping behaviors, as the focus of this paper, arise from dissatisfaction with public demands not being met. In the context of public risk response behaviors during urban relocation, representative actions include information-seeking behaviors and violent criminal behaviors. Information-seeking behaviors refer to actions taken by the public to understand relevant information, policies, and impacts of the relocation, which helps them better cope with the uncertainties and risks brought about by the relocation. On the other hand, violent criminal behaviors, which arise in some cases due to social unrest, dissatisfaction, or conflicts triggered by the relocation, represent extreme coping mechanisms adopted by some members of the public. Such behaviors are detrimental to both social stability and the smooth progress of the relocation. Lindell and Perry16identified information searching behavior as a manifestation of risk-coping behavior. In such contexts, individuals tend to engage in information search when perceiving a genuine threat and sensing an unacceptable level of personal risk. When presented with highly credible sources, the public may unquestionably comply with instructions, while ambiguity may prompt increased efforts in seeking and processing information17. In the Internet era, given the uncertainty surrounding relocation information and the perceived threat associated with it, the public frequently turns to online platforms to acquire relevant policy information—specifically, searching for relocation policies. Hu et al18.corroborated the correlation between the public’s information search behavior and their risk perception through a questionnaire survey. In essence, heightened information search activity may signal increased risk perception, correlating with a greater likelihood of resistance against the relocation. Violent conflict behavior represents another coping strategy that may introduce social risks during the relocation process. Dissatisfaction with relocation compensation and resettlement may lead to violent protests against the relocation, or large-scale conflicts between the involved parties. Some scholars argue that certain forms of relocation inherently involve violence. For instance, Ortega’s19research suggests that Manila, in its pursuit of becoming a modern and investment-friendly city, forcefully relocates informal settlers to the outskirts—an inherently violent form of suburbanization. Consequently, the public may resort to corresponding violent actions in response. Begega and Kohler20discussed protest behavior and the mobilization process in enterprise production relocation through case studies. Coffin et al21. examined relocation delays stemming from conflict behaviors in utility relocation processes. Furthermore, a decrease in poverty and quality of life resulting from relocation may contribute to an increase in violent crimes in the relocated area. Casas et al22. argue that areas of concentrated poverty and public housing face persistent issues with high rates of violent crime, while relocated residents may not necessarily experience less violence or improved safety in their new communities. Popkin et al23. contend that while most relocations reduce violent crimes, the relocation of medium and high-density communities may negatively impact domestic violence crimes. Negative emotions online and risk-coping behaviors Research on emotions originated in psychology, attributing emotions to physiological mechanisms such as the impact of the prefrontal cortex on the amygdala. Sociologists have also extensively studied emotions, recognizing their integration with social network relations, people, places, time, space, and events, giving rise to the concept of social emotions. It is believed that the formation of social emotions relies on individuals’ perception of society. Once formed, social sentiments influence social behavior, with social emotions driving fiscal, macroeconomic, and political behaviors, according to Prechter24. The internet’s global reach, breaking time and space constraints, significantly enhances individual social cognition levels, contributing to the formation of social emotions. On social media platforms, Internet users’ emotions, often magnified infinitely, shape and adjust netizens’ emotions and impact public social behavior25,26. It is posited that internet users’ emotions influence risk-coping behaviors, challenging the notion that social behaviors drive emotions27. When focusing on negative emotions, studies suggest their impact on risk-coping behaviors. For instance, Ramli et al1. found that controlling negative emotions during group gatherings reduces aggressive and dangerous behaviors. Lei and Yang2proposed that stronger negative emotions correlate with higher likelihoods of choosing high-risk coping behaviors. Hu et al3. quantitatively established a positive correlation between risk perception, negative emotions, and rumor-spreading behavior, with negative emotions playing mediating roles. It has been shown that negative emotions have a certain correlation with public risk coping behaviours, and that negative emotions increase the emergence of risk coping behaviours to a certain extent and increase the number of violent incidents in society. The research object chosen for this study is online negative emotions, and the collection of negative emotions is limited to the scope of the Internet, mainly because there are certain differences between online negative emotions and negative emotions. From the point of view of the scope of dissemination, online negative emotions can be quickly spread to a wider audience through the Internet and can cross geographical boundaries, which makes their influence wider. On the other hand, ordinary negative emotions are usually confined to face-to-face communication, which has a relatively small scope of influence. In terms of communication persistence, once posted, online negative sentiment may be permanently recorded unless it is deleted or hidden. This makes negative content highly persistent. In contrast, negative emotions tend to be temporary in real life and may fade over time. From the perspective of speaking anonymity, online negative emotions can sometimes be expressed anonymously or under a pseudonym, which may prompt some people who are afraid to speak out in real life to choose to speak out on the Internet and express their true emotions. Finally, from the perspective of information interactivity, online negative emotions can trigger chain reactions, such as follow-on comments and retweets by netizens, creating collective negative emotions. Whereas offline negative emotions may also resonate with others, but the spreading speed and scope of such resonance is relatively limited. Online negative emotions are to some extent more authentic and widespread than offline negative emotions. With the popularity of the Internet, many researchers choose to capture online comments and data for data analysis28,29. Therefore, this paper chooses to study the impact of online negative emotions on public risk response behavior. Despite evidence of negative emotions driving risk-coping behaviors, there is limited research exploring whether online negative emotions can predict public risk-coping behaviors in urban relocation processes. Building on prior studies, this paper hypothesizes that negative emotions online can effectively predict public risk-coping behaviors during urban relocation. H1: Negative emotions online can predict public risk-coping behaviors in the urban relocation process. Factors influencing the prediction effect According to the existing research, there is a certain correlation between negative emotions and risk coping30–32. This study further explores the differences in the predictive effects of various dimensions of negative emotions on risk coping behaviors. On one hand, we delve into the generation of the online text content containing negative emotions. The types of negative emotions are categorized based on the internal formation mechanisms of those online contents, while emotions related to different issues are classified according to the external problems that evoke these contents. On the other hand, we focus on the distinctions among the subjects that express online negative emotions. The negative emotions from different participants are categorized based on the organizational attributes of those subjects, whereas the negative emotions from different stakeholders are classified according to the interest attributes of those subjects. Based on the formation of contents containing negative emotions and the differences among subjects expressing negative emotions, this study identified four dimensions of the negative emotions, including the different types of negative emotions, the negative emotions about different issues, the negative emotions from different participants, and the negative emotions from different stakeholders. Different types of negative emotions Initially, emotional analysis methods were predominantly employed by computer and information science scholars, utilizing coarse-grained analysis focused solely on positive and negative attitudes. Watson and Tellegen33categorized self-reported emotions into positive and negative emotions, introducing the PANAS scale for emotional measurement. Subsequent research included intensity analysis based on this coarse-grained dichotomy. Russell et al34. developed a circular emotion model, classifying emotions according to pleasure and intensity, resulting in four categories: high-intensity pleasure, medium-intensity pleasure, high-intensity displeasure, and medium-intensity displeasure. As emotional analysis technology matured, scholars transitioned from a simple dichotomy to a more systematic, fine-grained analysis of public emotional information. Ekman35, for instance, divided emotions into six basic states: joy, sorrow, anger, fear, surprise, and disgust. Chinese text analysis also identifies various negative emotions. One approach is fine-grained, as seen in Liu’s36classification of negative emotions, including anxiety, sadness, anger, disgust, and fear. Another approach, based on coarse-grained dichotomy, involves differentiating negative emotions by degree. For instance, the HowNet Dictionary separates negative feeling words from negative comment words37,38. In comparison to fine-grained analysis, HowNet’s classification accentuates distinctions within negative emotions. This study employs HowNet’s classification to analyze the influence of different negative emotions on prediction effects. Negative feelings refer to the expression of one’s subjective feelings, such as sadness, anger, or disappointment, while negative comments refer to criticisms, accusations, or dissatisfaction expressed towards a certain event or phenomenon based on one’s personal values. Negative emotions belong to the realm of individual psychology, emphasizing the expression of personal feelings. These emotions may be private and subjective, and do not necessarily directly reflect or influence external social reality. Therefore, although the negative feelings expressed by individuals online may be genuine and intense, they may not necessarily immediately or significantly change the opinions or behaviors of the general public39. Negative comments fall under the scope of social psychology, emphasizing the expression of societal perspectives. Rational and well-grounded negative comments tend to be more easily accepted by the public, as they may be based on facts, logic, or shared values, rather than mere emotional venting40. Xie and Peng41 also confirmed that negative comments significantly impact risk perception, while negative feelings show no significant impact. Given the significant relationship between risk perception and risk-coping behaviors, research hypothesis 2 is as follows: H2: Negative comments online can more effectively predict public risk-coping behaviors than negative feelings online in the urban relocation process. Negative emotions on different issues Issue setting is a form of agenda setting42, wherein the media accentuates specific matters to influence the public’s attitudes or understanding. It is widely acknowledged that framing issues significantly impact public awareness and behavior43,44. Similarly, government policy making and synergies can influence and guide public behavior45. Therefore, the paper posits that negative emotions related to different issues will differ significantly in predicting risk-coping behaviors. Grounded in policy process theory, public policy undergoes stages such as formulation, implementation, evaluation, and termination46. Thus, issues within a public policy can be classified into three types: policy formulation, policy implementation, and policy effectiveness. The policy formulation process is more likely to draw public attention and discussion than other stages47. The outcome of the policy formulation will affect the policy implementation and the policy effectiveness48.Thus, the third hypothesis of this paper is presented as follows: H3: Negative emotions related to the issue of relocation policy formulation can more effectively predict public risk-coping behaviors than those related to issues of relocation policy effectiveness and relocation policy implementation. Negative emotions from different participants Online emotions serve as a reflection of the public’s mindset and behaviors. Numerous studies have confirmed that personal characteristics significantly influence behavioral judgments and public risk-coping behaviors49–51. Therefore, variations in the subjects of emotions may result in different risk-coping behaviors. Public opinion participants can be classified in various ways, such as categorizing them as opinion leaders, active users, or inactive users based on the enthusiasm of their online behaviors52, or as government, media, and the public based on their roles in online posting53. Other researchers divided public opinion participants into individuals, groups and communities based on the perspectives of different groups54. In contrast to negative emotions expressed by individual citizens, emotions emanating from government and media are subject to “gatekeepers” and may not directly mirror public opinion55. Thus, this paper proposes that emotions expressed by individuals are more effective in predicting risk-coping behaviors compared to those from government or media, leading to the formulation of the fourth hypothesis. H4: Negative emotions expressed by individuals are more effective in predicting public risk-coping behaviors than those expressed by institutions in the urban relocation process. Negative emotions from different stakeholders Stakeholders, initially an economic concept, refer to individuals or entities with claims to a company’s cash flow. In the realm of social science or management research, stakeholders encompass any entity affected by organizational decisions and actions in the external environment56. As per stakeholder theory, numerous risk factors influencing social stability often escalate into significant social conflicts not solely due to the actual magnitude of the risk, which can be challenging to control and resolve. Instead, these conflicts are largely rooted in varying risk perceptions among different stakeholders57. From a relevance perspective, stakeholders can be categorized into three groups: key stakeholders, non-key stakeholders, and marginal stakeholders58,59. Key stakeholders are essential resources that institutions rely on for survival and development; their support crucial for the institution’s continued existence and growth60. While the existence and development of institutions are intertwined with the support of non-key stakeholders, the degree of dependence on them for enterprise survival and development is notably lower than that of key stakeholders. Marginal stakeholders, on the other hand, are not indispensable for the institution’s existence and development. Numerous studies have explored the differences in risk perception among various stakeholders61. The majority of research findings indicate that the risk perception of key stakeholders surpasses that of other stakeholders62,63. As a result, risk-coping behaviors originating from key stakeholders are anticipated to be higher than those from other stakeholders. Therefore, this paper proposes Hypothesis 5. H5: Negative emotions expressed by key stakeholders are more effective in predicting public risk-coping behaviors than those expressed by non-key stakeholders or marginal stakeholders in the urban relocation process. Materials and methods Case study Beijing’s population had already exceeded 20 million in 2012, as shown in Fig. 1. If not controlled, overpopulation will lead to a series of megacity-related issues, such as traffic congestion, a decline in service levels, resource shortages, and the destruction of the ecological environment. In this context, both the central government and the Beijing government are contemplating the possibility of seeking a breakthrough. The relocation policy implemented by the Beijing Municipal Government is a public policy aimed at evacuating the population. Population Change in Beijing (Ten thousands). (Source: China National Bureau of Statistics). Figure 2 summarizes the policy process of the Beijing Government Relocation. Since the beginning of 2015, rumors about the relocation by the Beijing Municipal Government had been circulating among the public. Many people initially reacted with disbelief, but those who were well-informed began to purchase houses in the Tongzhou District of Beijing. By June 9 2015, Beijing’s stance on the implementation of the Beijing-Tianjin-Hebei Cooperative Development Plan was adopted, confirming the government’s policy of relocating to Tongzhou. On November 25, 2015, during the Eighth Plenary Session of the Eleventh Beijing Municipal Committee of the Communist Party of China, it was announced that all municipal administrative institutions in Beijing would collectively or partially move to the Tongzhou Administrative Vice-Center in 2017, with the relocation schedule being determined for the first time. Policy progress. Since December 2015, the negative effects of the policy have surfaced one after another. The most immediate consequence of the policy is the surge in housing prices in Tongzhou and its neighboring areas. The spike in house prices has led to sudden wealth for some individuals, while others have experienced a loss of income, resulting in social injustice and exacerbating social conflicts. For instance, on November 8–9, 2015, four consecutive intentional injury cases related to demolition and relocation took place in Xinzhuang Village, Yanjiao Town, Sanhe City, Hebei Province64. Although fraught with numerous risks, the relocation of the Beijing Municipal Government proceeded. By the end of December 2017, only a few departments had adhered to the relocation schedule outlined in November 2015, and the initial batch of ministries and commissions commenced the move to Tongzhou. On January 11, 2019, four groups comprising the Beijing Municipal Committee of the Communist Party of China, the Beijing Municipal People’s Government, the Beijing People’s Congress, and the Beijing Political Consultative Conference collectively relocated to Tongzhou, officially implementing the policy. Spanning nearly four years from its initial announcement to implementation, the policy had been widely discussed on the Internet since the emergence of rumors. Simultaneously, risk-related issues such as rising housing prices, education, and healthcare garnered public interest, with the repercussions of policy risks still ongoing. This paper selects this policy as a representative case of relocation policies to explore solutions for various challenges arising from China’s rapid economic development and investigate the relationship between online negative emotions and public risk perception. Based on the case analysis, the data for analysis was obtained from June 9, 2015 to April 28, 2019. The reason that we choose the period of data collection is: rumors about the relocation by the Beijing Municipal Government had been circulating among the public in June 9 and the Beijing Municipal Government relocation policy was officially introduced in July 2015, while the Beijing Municipal Government officially relocated into Tongzhou in January 2019 and the Beijing Municipal Government have settled down in Tongzhou in April 28. This study is concerned with whether there is any public opposition to this policy and negative emotions in cyberspace during this four-year period, which ultimately led to the policy not being implemented, so it is considered that the official relocation in 2019 means that the policy has already been implemented and has produced effects. This paper will investigate the public’s negative emotions on the internet during these four years. Variables of negative emotions Acquisition of analytical data Presently, Sina Weibo remains as the most popular and influential microblogging platform in China; thus, it was chosen as the sample source. Using keywords such as “Beijing government relocation,” “Beijing relocation to Tongzhou,” “Beijing government eastward relocation,” and “Tongzhou New Town,” searches were conducted on Sina Weibo from June 9, 2015 to April 28, 2019. Over 20,000 public posts about this policy were collected. Through the emotional judgment system developed by Boryou Technology (https://www.boryou.com/, accessed on August 12, 2022), 3,133 posts on Sina Weibo related to negative emotions were identified. After manual data cleaning, which involved removing information unrelated to the policy, eliminating duplicates, and excluding non-negative emotions, 1420 posts related to the relocation policy of the Beijing Municipal Government were retained. The time series data for the number of posts containing negative emotions were organized to align with the data on public risk-coping behaviors. The data used in this study were all open data and were obtained from the Internet. The data on online negative emotions and risk-coping behaviors are shown in the Figs. 3 and 4, containing a total of 1420 data between June 9,2015 and April 28,2019. To make the data presentation more scientific and aesthetically pleasing, this study has undergone data normalization. Online Negative Emotions data. Risk-coping Behaviors data(information searching and violent crime behaviors). Ethical approval and consent to participate All procedures performed in studies involving human participants followed the ethical standards of the institutional and national research committee along with the 1964 Helsinki declaration and its later amendments or comparable ethical standards. The study protocol was approved by the Ethics Committee of the school of Law and Humanities, China University of Mining and Technology (Beijing). All participants provided voluntary informed consent upon survey completion. Furthermore, our data underwent anonymization procedures to ensure participant privacy. Manual coding of variables Table 1shows the list of variable codes. To ensure the validity of the coding, two coders with backgrounds in linguistics and public administration were selected and trained to perform the coding independently. If the repetition rate between the two coders exceeds 90%, the confidence level of the results is considered acceptable65,66. In this study, the coding of the data was examined through the Holsti reliability test, and the overall reliability coefficients of the two coders were calculated to be between 0.90 and 1.0, which shows that the results of the reliability between the 2 coders are acceptable. Table 1Variables coding list.FactorsVariablesManual codingTypesA1:Negative commentsIf the content of the text is expressions of different views about the public policy, and the opposing attitude is more moderate, then code 1. If not, code 0A2:Negative feelingsIf the content of the text is expressions of strong dissatisfaction and strong opposition, then code 1. If not, code 0IssuesB1:Policy formulationIf the content of the text is to express opinions of policy making (such as agenda setting, policy plan, etc.), then code 1. If not, code 0B2:Policy implementationIf the content of the text is to express opinions of policy implementation (e.g. group incidents, rumors, etc.), then code 1. If not, code 0B3:Policy effectivenessIf the content of the text is to express opinions of the policy effect (for example the policy caused the rise of housing prices, the policy caused the problems of education, medical care, etc.), then code 1. If not, code 0ParticipantsC1: InstitutionsIf the publisher is media or government, the code is 1, if not 0C2: IndividualsIf the publisher is an ordinary person, the code is 1, if not 0StakeholdersD1: Key StakeholderIf the negative information is published in Beijing, the code is 1. If not, the code is 0D2:Non-Key stakeholderIf the negative information is published in Tianjin or Hebei, the code is 1. If not, the code is 0D3:Marginal stakeholdersIf the negative information is not published in Beijing, Tianjin or Hebei, the code is 1. If not, the code is 0 Variables coding list. Measurement of public risk-coping behaviors Based on the literature review, representative risk-coping behaviors in the relocation process include information-seeking behaviors and violent criminal behaviors. This paper assesses public risk-coping behaviors in the urban relocation process through an examination of these two dimensions. Information searching behaviors This paper employed the Baidu Index to gauge public information searching behaviors. The time series data capturing the public’s search volume related to the “Beijing government relocation” policy were obtained through the Baidu search index (http://index.baidu.com, accessed on June 1, 2022). Violent crime behaviors In previous studies, violent crimes have typically been derived from official statistics. The violent crimes investigated in this paper require time series statistical analysis in conjunction with online public opinion. Online data can be captured at specific intervals, even down to the minute or second, while offline official statistics are typically reported annually or quarterly, posing challenges for correlation studies. Therefore, this paper utilizes big data mining to acquire daily data on violent crimes. Davenport et al67. employed news reports to study public behaviors. Traditional newspaper media, unlike new media, tends to track crisis events, offering insights into the essence of these events. Hence, this paper initially identifies violent crime behaviors in the relocation area using the “newspaper database” from the China National Knowledge Infrastructure (CNKI). Keywords such as “Beijing Tongzhou,” “violence incident,” “mass incident,” “petition,” and “crime” were used in the search, and the data on behaviors were manually assessed and filtered. Moreover, official microblog accounts serve as extensions of government functions, and each piece of information released through them can be considered credible. Thus, government microblogs are selected as sources of data on violent crime behaviors. Given that the “Beijing Municipal Government Relocation” policy primarily impacted Beijing and Hebei, the official microblog accounts of the Beijing Municipal Public Security Department, “@ Ping An Beijing,” and the Hebei Provincial Public Security Department, “@ Hebei Public Security Network Spokesman,” were chosen to gather crime data. Keywords such as “violent events,” “mass events,” “petitions,” and “crimes” were used in the search, and manual judgment was applied to determine whether the behavior constituted a violent crime. By combining newspaper data and government microblog data, a time series dataset of violent crime behaviors was compiled. Granger causality analysis Granger causality analysis, pioneered by Clive W. J. Granger, the 2003 Nobel Prize winner in economics, has been widely employed by scholars to examine the predictive relationship between variables68,69. This test was initially developed to analyze Granger causality in economic variables. Granger defined Granger causality as dependence on the variance of the best least squares prediction using all the information at some point in the past. The original hypothesis of the Granger causality test is as follows: Hypothesis 0 Variable x cannot Granger cause variable y; Hypothesis 1 Variable x can Granger cause variable y. The calculation formula is presented in Formula 1. If the p-value is less than 0.05, it indicates that H1 is established; if the p-value is greater than 0.05, it means H0 is established. Granger causality analysis. This paper employs the Granger causality analysis method to examine the predictive impact of online negative emotions on public risk perception, through three steps: unit root test, VAR model construction, and establishment of the optimal lag date. Results The analysis was conducted using EViews 10.0 software. Data cleaning by using the Excel software. The negative emotions data section includes the variables A1-D3 in Table 1, and the data in the Public Risk-coping behaviors includes two parts, respectively, the public information searching behaviors and the number of violent incidents. The data of public information searching behaviors are derived from Baidu search index, the data of violent events is the sum of different violence reports in microblog news, newspaper news and website news in the same day. Verification of the prediction of public risk perception by negative emotions Unit root test Only stationary series can be used for the Granger test. The Unit Root Test was employed to assess whether the data represent stationary series. The time series data for negative emotions and risk perception underwent the Augmented Dickey-Fuller (ADF) test in EViews. Table 2 shows that at a 1% significance level, the variables of negative emotions, information searching behaviors (ISB), and violent crime behaviors (VCB) are stationary and meet the conditions for modeling. Consequently, an unconstrained VAR model was established. Table 2Augmented Dickey-Fuller test of negative emotions and risk-coping behaviors.t-StatisticProb.*Test critical values1% level5% level10% levelNegative emotions–24.921460.0000–3.964525–3.412980–3.128488ISB–8.2306610.0000–3.964534–3.412985–3.128490VCB–3.5996410.0059–3.434779–2.863383–2.567800 Augmented Dickey-Fuller test of negative emotions and risk-coping behaviors. VAR model construction The VAR model was estimated using the AR root estimation method to test the stationarity of the results. AR root estimation is based on the principle that if the reciprocal of all root modules of the VAR model is less than 1, indicating that all root modules are within the unit circle, then the model is stable. Conversely, if the reciprocal of all root modules of the VAR model is greater than 1, meaning that all root modules are outside the unit circle, then the model is unstable. This paper constructed two VAR models: negative emotion and information search behavior (Model 1), negative emotion and violent criminal behavior (Model 2). Figures 5 and 6 display the Inverse roots results of the two VAR models. As shown in the Figures, no root is outside the unit circle, confirming that the estimated VAR models satisfy the stability condition. Inverse roots of AR Characteristic Polynomial in model 1(negative emotion and information search behavior). Inverse roots of AR Characteristic Polynomial in model 2 (negative emotion and violent criminal behavior). Granger causality test and optimal lag date establishment As depicted in Figure 3, all test results, with the exception of the Final Prediction Error (FPE), converge upon a conclusion that the optimal lag order for both Model 1 and Model 2 is 3. First of all, AIC (Akaike Information Criterion) and SC (Schwarz Criterion) are usually used as the core observation indicators of the optimal lag order in the VAR model70,71, which should be considered mainly. Secondly, if the lag order above order 3 is selected for model 1 according to the selection method of HPE, the model will need to estimate more parameters, and then lose more sample size. In conclusion, according to the principle of the most concise lag order, the third-order lag was selected to be included in the construction of VAR model. Table 3Result of optimal lag order test.LagLogLLRFPEAICSCHQModel 10–12905.96NA327279.818.3743218.3818018.377121–12058.901690.51268565.6917.2969217.3187217.182612–12041.0735.5391496643.8117.2820317.3183617.168503–11994.3193.0431690937.4317.23450*17.28536*17.11323*Model 20–5229.338NA5.5175107.3836817.3911007.3864531–4909.904637.51603.5349976.9384676.9607236.9467822–4820.604177.96893.1340256.8180726.8551676.8319303–4786.06268.74312*3.001793*6.774964*6.826896*6.794366**Indicates lag order selected by criterion. Result of optimal lag order test. *Indicates lag order selected by criterion. The optimal lag time, determined by both the Akaike Information Criterion (AIC) and Schwarz Information Criterion (SC) in models 1 and 2, was found to be 3. Therefore, VAR (Lag = 3) was selected for the test, and the results are presented in Table 4. Table 4Granger Causality Test.Null HypothesisF-StatisticProb.LagModel1: Negative emotions does not Granger Cause Information Searching Behaviors33.6063***0.00003Model 2: Negative emotions does not Granger Cause Violent Crime Behaviors3.33127*0.01893Model1: Max lag(= 3) selected according to Akaike information criterion (AIC) by building VAR model. AIC(Lag = 1), 17.29692; AIC(Lag = 2), 17.28203 AIC(Lag = 3),17.23450; SC(Lag = 1), 17.31872; SC(Lag = 2), 17.31836; SC(Lag = 3),17.28536.Model 2: Max lag(= 3) selected according to Akaike information criterion (AIC) by building VAR model. AIC(Lag = 1), 6.938467; AIC(Lag = 2), 6.818072; AIC(Lag = 3), 6.774964; SC(Lag = 1), 6.960723; SC(Lag = 2), 6.855167; SC(Lag = 3), 6.826896.*significant at 0.05; **significant at 0.01; ***significant at 0.001. Granger Causality Test. Model1: Max lag(= 3) selected according to Akaike information criterion (AIC) by building VAR model. AIC(Lag = 1), 17.29692; AIC(Lag = 2), 17.28203 AIC(Lag = 3),17.23450; SC(Lag = 1), 17.31872; SC(Lag = 2), 17.31836; SC(Lag = 3),17.28536. Model 2: Max lag(= 3) selected according to Akaike information criterion (AIC) by building VAR model. AIC(Lag = 1), 6.938467; AIC(Lag = 2), 6.818072; AIC(Lag = 3), 6.774964; SC(Lag = 1), 6.960723; SC(Lag = 2), 6.855167; SC(Lag = 3), 6.826896. *significant at 0.05; **significant at 0.01; ***significant at 0.001. The probability that negative emotion does not have Granger Causality of information search behavior is only 0.0000. Hence, negative emotions can predict information search behaviors in advance. Additionally, the probability that negative emotion does not have Granger Causality of violent crime behavior is only 0.0189 (< 0.05). Therefore, negative emotions can predict violent crime behaviors in advance. The research hypothesis 1 was confirmed in both models. Verification of the factors influencing the prediction effect Table 5 Indicates that at the 1% significance level, all variables related to types, issues, participants, and stakeholders are stationary and meet the conditions for modeling. Unconstrained VAR models were constructed for each variable and risk perception, and the AR root estimation method was employed to test the stationarity of the estimated results of the VAR model. After the test, no root was found outside the unit circle, confirming that the estimated VAR model satisfies the stability condition. Table 5Augmented Dickey-Fuller test of the factors.FactorsVariablest-StatisticProb.*Test critical values1% level5% level10% levelTypesNegative comments–17.249820.0000–3.434624–2.863315–2.567763Negative feelings–36.970400.0000–3.434618–2.863312–2.567762Issuespolicy formulation–17.249820.0000–3.434624–2.863315–2.567763policy implementation–36.744610.0000–3.434618–2.863312–2.567762policy effectiveness–18.887330.0000–3.434624–2.863315–2.567763ParticipantsInstitution emotions–17.610060.0000–3.434624–2.863315–2.567763Individual emotions–26.637450.0000–3.434618–2.863312–2.567762StakeholdersKey stakeholders–25.652440.0000–3.434618–2.863312–2.567762Non-key stakeholders–6.0164460.0000–3.434664–2.863333–2.567773Marginal stakeholders–18.718950.0000–3.434624–2.863315–2.567763 Augmented Dickey-Fuller test of the factors. To assess Hypotheses 2 to 5, Granger causality analysis was conducted to evaluate the overall predictive impact of the variables representing ten influencing factors on public risk-coping behavior. Table 6 presents the analysis results for public risk-coping behavior, as indicated by information search behavior. The findings revealed that among the two variables representing negative emotion types, negative comments (A1, 8.30***) and negative feelings (A2, 4.11**), the probability that negative comments were not the Granger cause of risk-coping behavior was less than 0.001. In contrast, the probability that negative feelings were not the Granger cause of risk-coping behavior was less than 0.01. Thus, negative comments can better predict public risk-coping behavior, confirming Hypothesis 2. Table 6Granger Causality Tests of different emotion variables and information search behaviors.A1A2B1B2B3C1C2D1D2D3Lag3333333333SC17.1913.6416.6114.0315.3816.0216.4116.6313.8015.75F8.30***4.11**10.36***0.890.780.3316.38***14.01***0.0480.17P0.0000.0070.0000.4460.5030.8030.0000.0000.9860.914*significant at 0.05; **significant at 0.01; ***significant at 0.001. Granger Causality Tests of different emotion variables and information search behaviors. *significant at 0.05; **significant at 0.01; ***significant at 0.001. Regarding the analysis of issues, policy formulation (B1, 10.36***) was found to predict risk-coping behavior, while policy implementation and policy effectiveness were not identified as Granger causes of risk-coping behavior. The probabilities were 0.446 (B1) and 0.503 (B2). Therefore, public risk-coping behavior cannot be predicted by negative emotions related to these two issues, supporting Hypothesis 3. Analysis of participants indicated that negative emotions from institutions (C1, 0.33) could not predict public risk-coping behavior, while negative emotions from individuals (C2, 16.38***) might be the cause of public risk-coping behavior. Thus, Hypothesis 4 was validated. In the variables related to stakeholders, the probability that negative emotions from key stakeholders (D1, 14.01 ***) were not the Granger cause of public risk-coping behavior was less than 0.001. This suggests that negative emotions from key stakeholders can predict risk-coping behavior, while negative emotions from non-key stakeholders (D2, 0.048) and marginal stakeholders (D3, 0.17) cannot predict risk-coping behavior. Therefore, Hypothesis 5 was confirmed. Table 7 presents the analysis results for public risk-coping behavior represented by violent crime behavior. The outcomes indicated that among the two variables representing negative emotion types, negative comments (A1, 9.09***) and negative feelings (A2, 1.90), the probability that negative comments were not the Granger cause of risk-coping behaviors was less than 0.001. In contrast, the probability that negative feelings were not the Granger cause of risk-coping behavior was greater than 0.05. Therefore, negative comments can better predict public risk perception, thus confirming Hypothesis 2. Table 7Granger Causality Tests of different emotion variables and violent crime behaviors.A1A2B1B2B3C1C2D1D2D3Lag55555556165SC6.673.0976.043.474.835.465.926.163.085.15F9.09***1.908.54***3.28**2.85*3.66**10.36***7.54***1.92*5.33***P2.E-080.09096.E-080.00600.01440.00279.E-106.E-080.01567.E-05*significant at 0.05; **significant at 0.01; ***significant at 0.001. Granger Causality Tests of different emotion variables and violent crime behaviors. *significant at 0.05; **significant at 0.01; ***significant at 0.001. Concerning the three variables related to relocation issues, the predictive effect of policy formulation (B1, 8.54***, p < 0.001) was superior to policy implementation (B2, 3.28**, p < 0.01) and policy effectiveness (B3, 2.85*, p < 0.05). The probability that negative emotions on the policy formulation issue were not the Granger cause of risk-coping behavior was lower, validating Hypothesis 3. Among the two variables representing negative emotions from different participants, emotions from institutions (C1, 3.66**) and emotions from individuals (C2, 10.36***), the probability that negative emotions from individuals were not the Granger cause of risk-coping behavior was less than 0.001. In contrast, the probability that negative emotions from institutions were not the Granger cause of risk-coping behavior was less than 0.01. Thus, negative emotions from individuals can better predict public risk-coping behavior, confirming Hypothesis 4. In the variables related to stakeholders, the probability that negative emotions from key stakeholders (D1, 7.54***, P = 6.E-08) were not the Granger cause of public risk perception was the least. This suggests that negative emotions from key stakeholders can better predict risk-coping behavior than negative emotions from non-key stakeholders (D2, 1.92*, P < 0.05) and marginal stakeholders (D3, 5.33***, P = 7.E-05). Therefore, Hypothesis 5 was verified. Discussions and conclusions Summary of Major findings and contributions The present research verified the impact of negative emotions on public risk-coping behaviors1–4, but fewer studies focused on the prediction of negative emotions on public risk-coping behaviors. This study verified that negative emotions can effectively predict public information searching behaviors and violent crime behaviors using the case of relocation policy of Beijing Municipal Government. Our findings further explored the factors influencing the prediction effect. The results indicated that: (1) The publics’ negative emotions expressed online towards the relocation policy can predict the public risk-coping behaviors in advance; (2) The negative comments can better predict the public risk-coping behaviors than the negative feelings; (3) The negative emotions on issue of relocation policy formulation can better predict the public risk-coping behaviors than on issues of relocation policy effectiveness and relocation policy implementation. (4) The negative emotions from individuals can better predict public risk-coping behaviors than from institutions. (5) The negative emotions from key stakeholder can better predict the public risk-coping behaviors than from non-key stakeholder or marginal stakeholder. The theoretical contributions of this study encompass the following points. 1)Validation of the predictive relationship between negative emotions and public risk-coping behaviors. This finding is consistent with existing literature on the impact of emotions on decision-making and behavior, as exemplified by Lerner & Keltner, who argued that emotional states can significantly influence an individual’s information processing and decision-making processes72. 2) Revealing differences in predictive effectiveness among different types of emotions and emotional issues. This aligns with Schwarz & Clore’s “Emotion as Information” hypothesis, which posits that different types of emotions carry distinct information and subsequently affect an individual’s cognition and behavior73. 3) Emphasizing the role of information sources (individuals vs. institutions) in prediction. This echoes Granovetter’s theory of the strength of weak ties, suggesting that weak connections may have greater influence in the dissemination of information, albeit with a focus on the emotional influence aspect74. 4) Highlighting the importance of key stakeholders in predicting risk-coping behaviors. This discovery offers a novel perspective on stakeholder theory, underscoring the significance of identifying and prioritizing key stakeholders in public policy and risk management, which is consistent with Mitchell, Agle & Wood’s discussions on stakeholder classification and influence75. These findings have two practical contributions: (1) This study promotes the prediction of public risk-coping behaviors in the relocation policy. In the process of the relocation policy, in order to effectively reduce the public risk perception, which determines the successful implementation of the policy, the relocation policy related departments should pay attention to the negative public opinion on the network in a timely manner, because the negative emotions can effectively predict the pubic risk-coping behaviors. If the content of the policy is adjusted according to the negative emotions, the policy risk can be reduced. In the practice of relocation policy, we should pay attention on the negative emotions from the key stakeholders and individuals, the negative emotions on issue of policy formulation and the negative comments. (2) This study adds a case of the relocation policy of Beijing Municipal Government to the research fields of urban relocation policies in China. In the process of rapid economic development, China has to face the problems of urban relocation and it is of great significance that the relocation policies are successfully implemented. This study also further improves on our understanding of relocation policy from the perspective of risk-coping behaviors. Simultaneously, it provides a case reference for comparisons with urban relocations in other cities both in China and globally. Based on the above findings and contributions, three policy recommendations can be formulated as follows: Firstly, to enhance the prediction and management of public risk-coping behaviors during relocation policies, relevant departments should establish a real-time monitoring system for negative public opinions and emotions expressed online. By promptly addressing these negative sentiments, especially those from key stakeholders and individuals, as well as those related to policy formulation and negative comments, policy risks can be mitigated. Secondly, to improve the effectiveness of relocation policies, policy-makers should incorporate a stakeholder-centric approach, prioritizing engagement and communication with key stakeholders to gather insights and tailor policies that better align with their concerns. Finally, policymakers should strive for transparency and education campaigns to build trust and reduce negative emotions, thereby fostering a more positive environment for successful relocation policies. By integrating these recommendations, China can better navigate the challenges of urban relocation amidst rapid economic development and ensure the successful implementation of relocation policies. Limitations and suggestions for Future Research The main limitations of this study include the following aspects. On the one hand, the primary limitation of this study lies in the inherent constraints associated with manual coding. The identification of negative emotion types and issues related to negative emotions relies on human judgment, introducing the possibility of deviations. To address this concern, the study implemented a rigorous approach by training two independent coders. However, the potential for subjectivity remains. Future research endeavors could explore the application of big data analysis methods to achieve more objective identifications, mitigating the limitations associated with manual coding. On the other hand, although this paper verified the predictive relationship between online negative emotions and risk coping behaviors through a panel Granger causality test, which concluded that the causality between the two in a statistical sense is not equal to the actual causality, the perspective of this study is relatively novel and still has some reference value. However, there is still an endogeneity problem to be solved in this study, and further validation by using instrumental variables and other methods can be attempted in future studies. In addition, whether there exists some potential commonality and interaction effect among different indicators is also a topic worthy of in-depth study, and further useful additions can be made regarding indicators related to negative public sentiment, such as economic indicators, information dissemination speed, media attention, and public opinion bias. In future research, we will expand the sample size of social media data, increase comparisons with other national or regional studies, and adopt both qualitative and quantitative research methods to reduce data bias. Author contributions Conceptualization, Q.X.; methodology, Q.X. and H.W.; software, Q.X. and R.Z.; validation, Q.X., H.W. and R.Z.; formal analysis, Q.X.; investigation, Q.X. and H.W.; resources, Q.X.; writing—original draft preparation, Q.X.; writing—review and editing, Q.X. and R.Z.; visualization, Q.X.; supervision, Q.X.; All authors have read and agreed to the published version of the manuscript. Funding This research was funded by the National Social Science Fund of China, grant number 21CGL045. Data availability Online negative emotions data are publicly available from the Boryou Technology (https://www.boryou.com/). Public information searching behaviors data are publicly available from the Baidu search index (http://index.baidu.com). Violent crime behaviors data obtained from Weibo accounts such as “@ Ping An Beijing,” and the Hebei Provincial Public Security Department, “@ Hebei Public Security Network Spokesman”. The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request. Declarations Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. References Associated Data Data Availability Statement Online negative emotions data are publicly available from the Boryou Technology (https://www.boryou.com/). Public information searching behaviors data are publicly available from the Baidu search index (http://index.baidu.com). Violent crime behaviors data obtained from Weibo accounts such as “@ Ping An Beijing,” and the Hebei Provincial Public Security Department, “@ Hebei Public Security Network Spokesman”. The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request.