Negative Media Sentiment about the Pig Epidemic and Pork Price Fluctuations: A Study on Spatial Spillover Effect and Mechanism

Abstract

As the media have continued to pay increasing attention to pig epidemic events, some local pig epidemic events may have a large degree of negative impact on the pork market and the whole pig industry chain, leading to pork price fluctuations. Strengthening pig epidemic control, monitoring media reporting sentiment, and stabilizing pork price fluctuations are important measures to improve the economy and people’s livelihood. This paper sets out to identify the relationship between the negative media sentiment about the pig epidemic and the market risk of pork prices within a setting with pig epidemic risk. Based on the provincial panel data of China from January 2011 to December 2022, this paper uses the spatial panel Durbin model to investigate the impact of negative media sentiment about the pig epidemic on pork price fluctuations from the perspective of local and spillover effects, and further discusses the mechanism of consumer sentiment. The empirical results show that: (1) The negative media sentiment about the pig epidemic significantly exacerbates pork price fluctuations, and there is a single threshold effect, which is weakened after crossing the threshold value. (2) The negative media sentiment about the pig epidemic has a significant positive spillover effect on pork price fluctuations, showing the characteristics of “being a neighbor”. The spatial spillover effect shows a significant spatial attenuation feature and an inverted U-shaped change with the inflection point at 1400 km. (3) The effect is related to the heterogeneity of media reputation. The local aggravation effect of local media’s negative sentiment on pork price fluctuations is greater than that of central media and information network platforms. In terms of the spatial spillover effect, the negative sentiment of the information network platforms has the strongest effect on the aggravation of pork price fluctuations in neighboring regions. (4) The mechanism study finds that the negative media sentiment about the pig epidemic positively affects pork price fluctuations through the path of “consumer sentiment”. Therefore, this research recommends that the government department should strengthen the supervision of media sentiment about the pig epidemic and reasonably guide consumer sentiment to stabilize the pork market.

1. Background and Introduction

In recent years, there have been frequent outbreaks of sudden pig epidemics in China, from swine pasteurellosis in 2009 and foot-and-mouth disease in 2013 to African swine fever in 2018, which severely shook pork prices and severely affected the pig industry. Due to the African swine fever, the stock of sows and hogs has continued to decrease. China’s Ministry of Agriculture and Rural Affairs reports that the stock of sows decreased from 31.3 million to 19.0 million by August 2019, while the stock of hogs fell from 320.8 million to 190.9 million. In 2019, the total output of pork in China was 42.55 million tons, down 21.30% from the previous month, and the total consumption of pork hit the lowest level in nearly 15 years. Pork prices continued to rise in 2019 and remained high, reaching a record high of 8.49 USD per kilogram in February 2020. The pig epidemic is defined as a specific unforeseeable risk event. Since pork dominates the Chinese diet, online media, which serve as an information distribution center and public opinion platform, also extensively report the news of the pig epidemic and fluctuations in pork prices [1]. The media coverage of the pig epidemic plays two functions “information supply” and “emotional contagion” in the disclosure process of the event. On the one hand, providing information about the pig epidemic can increase the amount of information required by consumers and reduce the information risk in consumption decision making. On the other hand, to capture the public’s attention, the media tend to spread emotional information and opinions in shocking language. Many studies question the objectivity of media coverage and believe that media may form reporting bias through media sentiment based on the purpose of attracting attention [2]. Media reporting sentiment may be positive, negative, or neutral, and has two characteristics. First, media sentiment has cognitive, communication functions, and behavior guidance. Secondly, compared with rational information, emotional information from media is more likely to trigger cognitive resonance in economic people. The authority of media information sources and consumers’ perception of pig epidemic risk are the main factors that affect consumers’ behavioral decisions [3]. Changes in purchasing behavior decisions of consumer groups are often considered to cause further significant economic shocks [4], especially directly or indirectly leading to price fluctuations in the pork market [5].

The external impact of the sudden pig epidemic exacerbates the contradiction between supply and demand in the pork market and has a significant impact on pork price fluctuations [6]. In the process of the information of pig epidemic risk events from the information source through the intermediary disseminator to the final receiver, the risk signal is constantly strengthened, leading to the amplification of the risk of pig epidemic [7]. Media sentiment, as an important part of market sentiment, is a key formation mechanism for the amplification of the risk of the pig epidemic [8]. The media collect risk information and spread it to the public through media coverage [9]. The negative media sentiment causes a social response, and the social response generates new information through the perception and behavior of consumers and spreads it [10]. The intensity of media sentiment can promote the dissemination of information, especially the emotional “arousal” that stimulates public information-sharing behavior [11].

As an important source of information on animal epidemic risk and food health [12], media coverage has multifaceted and multilayered influences on the agricultural market, especially on the demand and prices of agricultural commodity, which are also topics of interest to many scholars [13,14]. Some studies have included media coverage in the demand equation and examined the impact of media coverage on meat demand after animal outbreaks and food safety incidents [15]. Yadavalli et al. [3] found that media coverage of lean finely textured beef (LFTB) had an impact on consumers’ demand for pork and beef for two weeks or more, and the impact gradually decreased. Browning et al. [16] examined the heterogeneity of consumer responses to health-related media coverage and found that 16% of consumers would respond to high-fat fish media coverage. The difference in the influence of media coverage on consumers’ decision-making behavior is potentially related to consumers’ risk perception level, media usage pattern, and acceptance of media information [5].

Scholars have discussed the role of media coverage of emergencies on price fluctuations in agricultural commodity, but the research only focused on the broiler and beef markets [17,18]. Hassouneh et al. [19] pointed out that when the avian flu crisis broke out, the increase in media coverage of avian flu would lead to a strong rise in the price level and fluctuations of Egypt’s poultry industry chain. Unfortunately, few existing studies have further discussed the impact of media reporting sentiment on the price fluctuations in agricultural commodity. Only some scholars have taken food safety and quality events as research objects [20] and built the media coverage index of agricultural commodity quality and safety based on web crawler technology, and found that the impact of negative media coverage on meat and poultry prices was greater than that on grain prices. Moreover, the impact is greatest in the outbreak period of agricultural commodity quality safety incidents and least in the recovery period [21]. Wen et al. [22] explored the impact of food safety scandals on the price transmission of production and sales in China’s pork supply chain from the perspective of media coverage and pointed out that the impact effect had asymmetric threshold characteristics.

As a specific relationship within a certain geographical unit, media sentiment about the pig epidemic and pork price fluctuations have a certain spatial effect. Scholars investigated the dynamic evolution process of the spatial distribution of the risk of the pig epidemic and found that the impact of the pig epidemic events could easily spread to the whole region [23]. The risk diffusion of inter-regional transport of live animals made the risk of the pig epidemic spread in spatial dimensions, and the spatial distribution pattern of the overall incidence of African swine fever showed aggregation [24]. The fluctuations in pork prices show a positive spatial spillover effect, with significant local spatial agglomeration characteristics [25]. With the deepening of the research, scholars found that the media coverage of the outbreak of the pig epidemic had the characteristics of “cross-domain” distribution, “dynamic” evolution, and “social networking” transmission [26]. In some regions, the pig epidemic events may cause the public to doubt the pork product market through the emotional coverage of the media and conclude that “the world is as black as a crow”, resulting in the risk crisis of pork food in neighboring regions and damaged demand of the pork market, forming a contagious effect. Both the risk of pig epidemic events and the sentiment conveyed by the media coverage show a spatial aggregation effect [27]. The sentiment conveyed by the media triggers the reputation effect and regional spillover effect of animal epidemic emergencies [28], which may cause a spatial spillover impact on the trans-regional pork market.

The existing studies generally ignore the amplification effect of media coverage as an information transmission mechanism on pork price fluctuations and the differences in the text content and the emotional tendency of media coverage of the pig epidemic. Secondly, there is a lack of research on the spatial spillover effect of negative media sentiment about the pig epidemic on pork price fluctuations. Ignoring the amplifying effect of negative media sentiment in time and space may lead to bias in model estimation.

So naturally, there are such questions: in the pork market, will the negative media sentiment about the pig epidemic affect pork price fluctuations? If so, does this impact have a spatial spillover effect? Further, what are the influencing mechanisms? Therefore, based on signal transmission theory and behavioral economics theory, this paper explores the impact and spatial spillover effect of negative media sentiment about the pig epidemic on pork price fluctuations. The main marginal contributions of this paper are: First, we try to add the information transmission path into the traditional analysis framework of fluctuations in agricultural product prices and introduce the media sentiment effect into the field of pig epidemic, to explore the impact of negative media sentiment about the pig epidemic on pork price fluctuations, and provide a new perspective for the study of pork price fluctuations. Apply text mining and machine learning text analysis technology to effectively identify the media sentiment about the pig epidemic, to expand valuable data sources for the research of behavioral economy. Second, we scientifically and reasonably calculate the spatial spillover effect of negative media sentiment about the pig epidemic on pork price fluctuations and accurately locate the spatial attenuation boundary to provide general conclusions and empirical evidence for the spatial spillover effect of negative media sentiment about animal epidemics on fluctuations in agricultural product prices. Finally, based on the sentiment analysis of social media, this paper investigates the consumer sentiment about the pig epidemic and dug into the intermediary transmission mechanism of consumer sentiment in negative media sentiment about the pig epidemic and pork price fluctuations. Exploring these problems is of great significance for implementing strict prevention and control of pork market risk under the pig epidemic situation, ensuring the elastic matching between the pig production side and the consumer demand side, realizing the dynamic, timely, accurate, and efficient supply and demand connection of pork products, and calming pork price fluctuations.

2. Theoretical Analysis and Methods

2.1. Theoretical Hypotheses

When pig epidemic events occur, the media conveys the risk signal of the pig epidemic to consumers and influences the pork market by changing consumer cognition and behavior. On the one hand, network media, as an intermediary channel for information release and dissemination, can reduce the cost of consumers’ access to epidemic information and alleviate the disadvantage of consumers in pig epidemic information [29]. On the other hand, after the outbreak of the pig epidemic, the official media as the source of the signal first expose information, then all kinds of media forwarded reports. However, media coverage can be biased in two ways. One is supply-side factors, mainly from ownership or government pressure. The second is the demand-driven media bias. Due to competition, media compete for attention to increase advertising revenues [30]. Some media coverage is highly subjective and mixed with polarizing descriptions of the pig epidemic. Information polarization caused by content or attitude polarization can form public opinion encirclement in a short time. As opinion leaders, some media, with their authority and influence, can directly enhance consumers’ uncertainty and negative sentiment about the pig epidemic to varying degrees and strengthen their perception of the risk of the pig epidemic [31].

The pig epidemic event goes through multiple rounds of feedback and circulation in the signal source, information mechanism, and psychological mechanism to form its risk amplification mechanism. First of all, the information source of the pig epidemic event enhances consumers’ perception of the risk of the pig epidemic through emotional media coverage. At the same time, the emotional information of the pig epidemic becomes a new signal source through the feedback loop of the Internet, which is propagated and amplified again [32]. Secondly, after receiving the negative sentiment of the pig epidemic information source, consumers’ psychology generates negative sentiments such as panic and anger about the pig epidemic and pork food through emotional inspiration due to the unknown risk to their health [33], and make the cognitive assessment, forming emotional reactions such as rejection, forwarding and comment [34]. The social contagion effect of emotional response may accumulate to trigger the negative sentiment of consumer groups and spread the information emotionally. The information rift diffusion feedback becomes a new signal source. So many cycles lead to the polarization reaction of consumer groups [35]. Finally, based on the risk amplification theory, the pig epidemic event is transmitted through negative media sentiment, resulting in the “amplification” of the epidemic risk and ripple effect over time [36]. Consumer groups amplify the risk perception of the pig epidemic, change the total demand of the pork market, and the imbalance between supply and demand leads to pork price fluctuations. Therefore, we can propose the first hypothesis:

H1.

Negative media sentiment about the pig epidemic amplifies the risk of pig epidemic events and exacerbates pork price fluctuations.

The imbalance of the regional distribution of pork consumption and pig breeding promotes horizontal price transmission between regions, and pork prices show spatial convergence and spillover effects [25]. Media reporting sentiment has the function of signaling. The public opinion environment of the pig epidemic formed by media coverage makes it impossible to escape the influence of media sentiment on consumer behavior decisions within a certain spatial range. Risk transmission is manifested as cross-regional risk diffusion of “multi-place joint transmission”, so, logically, pork price fluctuations have a spatial spillover effect. According to the theory of psychological safety zone, as a major participant in the pork market, consumers have their own familiar space, namely the psychological safety zone [37]. Consumers will screen, sort, and further process the information in the market according to the degree of psychological safety [38]. When consumers find that the information about an emergency pig epidemic event is in their psychological safety space and related to it, they will choose to actively participate in the information dissemination; otherwise, they tend to take an indifferent attitude [39]. The media sentiment carried in the pig epidemic information is constantly transmitted and accumulated in different levels of safe space. While affecting consumers’ awareness of the epidemic risk, the transmitted consumer sentiment will also have different degrees of the resonance effect, thus forming an “opinion environment” of different intensities [40], which will trigger consumers’ convergence behavior and ultimately affect pork price fluctuations. In other words, driven by media sentiment contagion, consumers in different safe spaces will dance together, and the sentiment linkage effect will appear. Consumers make decision behaviors based on a strong opinion environment, and consumers in a certain region tend to make the same decision on pork purchase behavior, resulting in a cross-space pork price fluctuations linkage phenomenon.

As the space safety range radiates around, the speed and intensity of media sentiment contagion gradually decrease, the sentiment linkage between the target space region and the information source region becomes weaker, the degree of psychological security of consumers is weakened, and the pig epidemic information in the remote space region is increasingly ignored, thus forming a ring ladder of the information transmission mechanism. As a result, the spillover effect of pork price fluctuations in the two regions gradually decreases(i.e., spatial attenuation). From this, the second hypothesis can be put forward:

H2.

The negative media sentiment about the pig epidemic has obvious spatial spillover to the fluctuations in pork prices, and this spatial spillover effect has the geographical boundary of spatial attenuation.

If the above-mentioned price fluctuation effect of media sentiment contagion exists, an important question is, how does negative media sentiment spread to pork price fluctuations? Based on the existing research of behavioral economics, this paper argues that there may be an influence path of “consumer sentiment”. The public opinion of the pig epidemic event is sudden and volatile, often affecting the whole body and accompanied by the butterfly effect. Once the pig epidemic event breaks out, the release, comment, and forwarding of online media such as Weibo and WeChat will naturally arouse the attention of consumers and form the gathering of consumer sentiment, resulting in the phenomenon of “tornado” [41].

From the perspective of “consumer sentiment”, behavioral economics research shows that media sentiment can induce consumers’ emotional attitudes and further affect their consumption decision-making behaviors by interfering with their expectations and sentiments [42]. When externally stimulated by the negative media sentiment about the pig epidemic, consumers tend to release their sentiment through online comments after arousing their psychological reaction, indicating their positive, negative, or neutral attitude toward the epidemic event, resulting in more emotional information transmission [43]. Positive media coverage of the pig epidemic is more likely to infect and stimulate the positive sentiment of consumers, while negative media coverage induces consumers to have pessimistic expectations of the pork market and generates negative sentiment of consumers [44]. Meanwhile, the positive feedback effect enables the positive and negative sentiment to trigger group emotional resonance through the sentiment infection mechanism in the opinion environment. As a result, online public opinions rise and spread [45], generating strong consumer sentiment. Based on the stimulus-sentiment-behavior conduction path of cognitive emotion theory [42], consumers first evaluate pig epidemic events through risk perception, including perceived severity and perceived susceptibility, two aspects [46], analysis, and other ways. Secondly, different evaluation results produce different valence sentiments and then affect consumer behavior patterns. On the one hand, the negative sentiment generated by the negative evaluation produces strong emotional responses, and the limited rational consumption behavior based on cognitive psychological bias expands the negative impact of the epidemic event and forms the group polarization behavior [47]. Liu et al. [48] found that the impact of public online negative sentiment on agricultural product prices had time-varying, lag, and life cycle characteristics. On the other hand, the positive sentiment generated by the positive evaluation is further transmitted to the consumer group, resulting in group reactions and behavioral decisions. The demand behavior of consumer groups affects the balance of supply and demand in the pork market and finally results in pork price fluctuations. From this, we can propose the third hypothesis:

H3.

Consumer sentiment plays an intermediary effect in the process of negative media sentiment about the pig epidemic exacerbating pork price fluctuations, that is, negative media sentiment stimulates consumer sentiment, thus changing the total demand of the pork market and aggravating pork price fluctuations.

2.2. Data Sources

Based on the availability of data, this paper sets the research object as the interprovincial panel data of 30 provinces, autonomous regions, and municipalities of China from January 2011 to December 2022, excluding Tibet. The price data came from Chinese agricultural big data (http://www.agdata.cn/, accessed on 3 January 2023). The negative media sentiment about the pig epidemic and consumer sentiment were collected by compiling a crawler program. A natural language processing algorithm (SnowNLP) was used to analyze the text contents and calculate the emotional tendency score. Data on pig epidemics were obtained from Official Veterinary Bulletin (http://www.moa.gov.cn/gk/sygb/, accessed on 3 January 2023). Urban disposable income data were collected from the National Bureau of Statistics (https://data.stats.gov.cn/, accessed on 3 January 2023).

2.3. Variable Selection

2.3.1. Pork Price Fluctuation

Pork price fluctuation (pork) selects the market price of bone-free pork. To eliminate the influence of price and seasonal factors, it first takes the CPI in January 2011 as the base period to deflate all price variables and makes an X-12 seasonal adjustment [49]. Then it takes the natural logarithm of the adjusted data for the first-order difference, and the absolute value is taken after multiplying by 100. Following the previous study [50], the calculation method of pork price fluctuation is shown in Equation (1), where p_t is the t-period pork price.

$$por{k}_t=\left|100\times \ln (p_t/p_{t-1})\right|$$

(1)

2.3.2. Negative Media Sentiment about the Pig Epidemic

The steps to measure the negative media sentiment about the pig epidemic (media) are as follows: First, the Python crawler program is written to capture the microblog transcripts of media coverage of the pig epidemic on Sina Weibo from January 2011 to December 2022 (with seven types of pig epidemics such as swine fever and African swine fever as keywords). Secondly, the emotional analysis of the microblog data is carried out one by one. Emotional dictionaries are generally used in the financial field for emotional analysis, but there may be a big deviation in the construction of emotional dictionaries of network media texts. The sentiment analysis method of deep learning technology has higher accuracy when dealing with different language situations. Therefore, this paper uses the naive Bayes algorithm program of natural language processing (SnowNLP) to analyze the sentiment of microblog posts reported by the media about the pig epidemic one by one and classifies microblog posts into positive, neutral and negative posts according to the emotional tendency of the text. Finally, based on the study of Lu et al. [51], this paper calculates the monthly relative negative media sentiment index according to the following methods:

$$Medi{a}_{it}=\ln (\frac{1+N_{it}^{NEG}}{1+N_{it}^{POS}})$$

(2)

where $N_{it}^{NEG}$ is the number of negative media coverage of the pig epidemic in province i in month t. $N_{it}^{POS}$ is the number of positive media coverage of the pig epidemic in province i in month t. According to the study of Wen et al. [52], the new information in the emotion variable value can more accurately reflect the impact of media reporting sentiment on the pork market. Therefore, in this paper, the first-order difference of monthly media sentiment level value ($Medi{a}_{it}$) is adopted to measure the monthly fluctuations in negative media sentiment(Δ$Medi{a}_{it}$), indicating the negative media sentiment about the pig epidemic (media).

2.3.3. Consumer Sentiment

In this paper, a Python crawler program is written to capture the comment text of microblog posts of media coverage of the pig epidemic from January 2011 to December 2022, and the natural language processing algorithm is used to emotionally divide each microblog comment one by one, and the comments are divided into positive, neutral and negative categories according to the emotional tendency of text. According to the practice of Wang et al. [53], the negative sentiment of consumers(consumer) is measured as follows:

$$consume{r}_t=\frac{\mathrm{Number} \mathrm{of} \mathrm{negative} \mathrm{reviews}-\mathrm{Number} \mathrm{of} \mathrm{positive} \mathrm{reviews}}{\mathrm{Number} \mathrm{of} \mathrm{reviews}}\times 100\%$$

(3)

where $consume{r}_t$ is the negative sentiment of consumers and represents the relative proportion of negative reviews of media coverage in the total number of reviews during the t period.

2.3.4. Control Variable

(1) Piglet price fluctuation (Piglet), the upstream product of the pig industry chain, is measured by piglet market price fluctuations. (2) Pig price fluctuation (Pig), the middle product of the pig industry chain, is measured by the price fluctuations in the live pig market to be slaughtered. (3) Corn price fluctuation (Corn), corn accounts for more than 50% of the pig feed ratio, which is reflected by fluctuations in corn market prices. (4) Soybean meal price fluctuation(SBM), as reflected by soybean meal market price fluctuations. (5) Beef price fluctuation(Beef), the main alternative product price, is measured by the price fluctuations in the boneless beef market. (6) Mutton price fluctuation(Mutton) is measured by fluctuations in bone-in Mutton market prices. (7) Chicken price fluctuation(Broiler), measured by fluctuations in white strip chicken market prices. (8) Fluctuation in disposable income of urban residents(Income), as reflected by the fluctuations in disposable income of urban residents. (9) Pig epidemic(EPI). The external shock of the pig epidemic may have a significant impact on pork prices. We use the sum of the number of pig epidemic deaths and the number of forced culls to measure the severity of the pig epidemic. Descriptive statistics of variables are shown in

Table 1. Statistical characteristics.

Variable Name	Code	Mean	SD	Min	Max
Pork price fluctuation	Pork	4.879	4.822	0.0028	49.86
Negative media sentiment about the pig epidemic	Media	0.364	0.687	0	5.649
Consumer sentiment	Consumer	0.108	0.148	−0.066	0.454
Pig price fluctuation	Pig	6.325	5.924	0.0050	46.75
Piglet price fluctuation	Piglet	7.558	8.880	0.0008	134.9
Corn price fluctuation	Corn	1.938	1.952	0	35.14
Soybean meal price fluctuation	Sbm	2.561	2.979	0	57.19
Beef price fluctuation	Beef	1.217	1.326	0.0007	13.90
Mutton price fluctuation	Mutton	1.556	1.468	0.0004	14.61
Chicken price fluctuation	Chicken	2.193	2.685	0	49.23
Fluctuation in disposable income of urban residents	Income	0.654	0.460	0	5.448
Pig epidemic(Unit)	Epidemic	99.54	163.9	0	1236

.

2.4. Methods

2.4.1. Benchmark Model

To address the impact of negative media sentiment about the pig epidemic on pork price fluctuations, the benchmark regression model is constructed as follows:

$$por{k}_{it}^{}={\mathsf{\varphi }}_0+{\mathsf{\varphi }}_{1}medi{a}_{it}+{\mathsf{\varphi }}_{2}control{s}_{it}+{\mathsf{\epsilon }}_{it}$$

(4)

where i represents the region. t is time. $por{k}_{it}$ represents pork price fluctuation. $medi{a}_{it}$ expresses negative media sentiment about the pig epidemic. $control{s}_{it}$ denotes the control variable that affects pork price fluctuations.

2.4.2. The Threshold Model

To explore the nonlinear relationship between negative media sentiment about the pig epidemic and pork price fluctuations in different segments, this paper takes negative media sentiment as the threshold variable to conduct a threshold test. Following Yu et al. [54], the single threshold model is set as follows:

$$\begin{array}{l}por{k}_{it}^{}={\mathsf{\tau }}_0+{\mathsf{\tau }}_{1}medi{a}_{it}\cdot I(medi{a}_{it}\le {\gamma }_1)+{\mathsf{\tau }}_{2}medi{a}_{it}\cdot I(medi{a}_{it}>{\gamma }_1)\\ +{\mathsf{\tau }}_{3}control{s}_{it}+{\mathsf{\epsilon }}_{it}\end{array}$$

(5)

where ${\gamma }_1$ represents the threshold value for negative media sentiment about the pig epidemic.

2.4.3. Spatial Panel Durbin Model

Each spatial economic unit does not exist in isolation but interacts with the neighboring economic units in space through various connections, which is manifested as spatial dependence and overflow characteristics in geography. The negative media sentiment about the pig epidemic discussed in this paper has a strong spatial externality, and there may be mutual influence between the pork price fluctuations in different provinces. Therefore, it is necessary to increase the testing of spatial effects on traditional econometric models when exploring the relationship between negative media sentiment and pork price fluctuations. The spatial panel Durbin model (SPDM) includes spatial lag terms of both explanatory and dependent variables, taking into account the spatial effects of explanatory and dependent variables. Therefore, following Pan et al. [55], an SPDM model is constructed in this paper, which considers the impact of spatially lagged negative media sentiment about the pig epidemic and spatially lagged term of pork price fluctuation on pork price fluctuation, as shown in Equation (6) below:

$$\begin{array}{l}por{k}_{it}=\alpha +\rho {\displaystyle \sum _{j=1}^N{w}_{ij}}por{k}_{it}+{\beta }_{1}medi{a}_{it}+{\beta }_2{X}_{it}\\ +{\mathsf{\theta }}_1{\displaystyle \sum _{j=1}^N{w}_{ij}medi{a}_{it}+{\mathsf{\theta }}_2}{\displaystyle \sum _{j=1}^N{w}_{ij}{X}_{it}}+{\mu }_i+{\gamma }_t+{\mathsf{\epsilon }}_{it}\end{array}$$

(6)

where ρ represents the impact of local pork price fluctuation on pork price fluctuation in neighboring regions; media is the matrix of negative media sentiment about the pig epidemic; X is the control variable matrix; θ is the regression coefficient of the spatial lag term; w is the spatial geographical weight matrix; α is the model intercept term; ${\mathsf{\mu }}_i$, ${\mathsf{\gamma }}_t$ and ${\mathsf{\epsilon }}_{it}$ are individual effect, time effect, and random error term, respectively.

2.4.4. Spatial Weight Matrix

The correlation of geographical distance is the most direct manifestation of spatial dependence and spillover. In this paper, the inverse Euclidean distance between the centroids of two provinces is calculated by the latitude and longitude to describe the spatial dependence characteristics between the provinces, as shown in Equation (7) below:

$$W_{ij}(1)=\left\{\begin{array}{l}1/d_{ij} d_{ij}<d\\ 0 d_{ij}\ge d\end{array}\right.$$

(7)

Considering that the economic development of each province determines the consumption level and directly affects the total demand for pork, this paper uses GDP and distance to build a gravity model, as shown in Equation (8) below:

$$W_{ij}(2)=\left\{\begin{array}{l}(\overline{E_i}\times \overline{E_j})/d_{ij}^2 i\ne j\\ 0 i=j\end{array}\right.$$

(8)

where $\overline{E_i}$ and $\overline{E_j}$ respectively represent the per capita GDP of the two provinces. According to the gravity matrix, the relationship between different spatial units is not only directly affected by geographical distance but also related to regional economic development.

2.4.5. Mediator Effect Model

To further clarify whether consumer sentiment plays a mediating effect between negative media sentiment about the pig epidemic and pork price fluctuations, this paper draws on the test method of Wen et al. [56] and sets up the mediating effect model as follows based on the benchmark model:

$$consume{r}_{it}^{}={\lambda }_0+{\lambda }_{1}medi{a}_{it}+{\lambda }_{2}control{s}_{it}+{\mathsf{\epsilon }}_{it}$$

(9)

$$por{k}_{it}^{}={\mathsf{\phi }}_0+{\mathsf{\phi }}_{1}medi{a}_{it}+{\mathsf{\phi }}_{2}consume{r}_{it}+{\mathsf{\phi }}_{3}control{s}_{it}+{\mathsf{\epsilon }}_{it}$$

(10)

where $consume{r}_{it}$ is the intermediary variable, representing consumer sentiment; other variables are the same as in Model (4). Model (9) tests the impact of negative media sentiment about the pig epidemic on consumer sentiment. The sign expectation ${\lambda }_1$ is positive, indicating that negative media sentiment about the pig epidemic triggers negative consumer sentiment. Model (10) tests the impact of negative media sentiment about the pig epidemic on pork price fluctuations when controlling negative consumer sentiment variables.${\mathsf{\phi }}_1$ is the direct effect of negative media sentiment about the pig epidemic on pork price fluctuations; the product of ${\lambda }_1$ and ${\mathsf{\phi }}_2$ represents the mediating effect of negative consumer sentiment.

3. Results

3.1. Negative Media Sentiment about the Pig Epidemic and Pork Price Fluctuations

The results of the multicollinearity test and the correlation test(

Table A1. Estimates of multicollinearity test.

Variables	VIF	1/VIF
Media	1.11	0.900
Pig	1.33	0.754
Piglet	1.30	0.771
Corn	1.15	0.866
Sbm	1.16	0.862
Beef	1.33	0.753
Mutton	1.30	0.767
Chicken	1.05	0.951
Income	1.02	0.983
Epidemic	1.04	0.966
Mean VIF	1.18

and

Table A2. Correlation coefficient matrix.

	Pork	Media	Pig	Piglet	Corn	Sbm	Beef	Mutton	Chicken	Income	Epidemic
Pork	1	0.2941 ***	0.7499 ***	0.3866 ***	0.0645 ***	0.0924 ***	0.0619 ***	0.0035	0.1691 ***	0.0333 **	−0.0273 *
Media	0.2756 ***	1	0.2552 ***	0.1967 ***	−0.0140	0.0928 ***	0.0183	−0.0237	0.0829 ***	−0.1110 ***	0.0875 ***
Pig	0.8347 ***	0.2468 ***	1	0.4026 ***	0.0681 ***	0.1348 ***	0.0556 ***	0.0059	0.1383 ***	0.0202	0.0370 **
Piglet	0.4056 ***	0.1882 ***	0.4532 ***	1	0.1822 ***	0.1659 ***	0.0786 ***	0.0752 ***	0.0590 ***	0.1094 ***	0.0290 *
Corn	0.0541 ***	−0.0083	0.0452 ***	0.1210 ***	1	0.1956 ***	0.0015	−0.0127	0.0153	0.1074 ***	−0.0375 **
Sbm	0.1157 ***	0.0877 ***	0.1109 ***	0.1256 ***	0.3431 ***	1	0.0144	0.0075	−0.0328 **	0.0652 ***	0.0432 ***
Beef	0.1342 ***	0.0324 **	0.0850 ***	0.0700 ***	0.0296 *	0.0643 ***	1	0.4221 ***	0.1483 ***	0.0877 ***	0.0139
Mutton	0.0347 **	−0.0095	0.0153	0.0686 ***	−0.0085	0.0154	0.4736 ***	1	0.0911 ***	0.0538 ***	0.0604 ***
Chicken	0.1760 ***	0.0723 ***	0.1248 ***	0.0601 ***	0.0204	0.0128	0.1699 ***	0.0952 ***	1	0.0500 ***	0.0137
Income	−0.0002	−0.0580 ***	0.0019	0.0432 ***	0.0785 ***	0.0291 *	0.0377 **	0.0390 **	0.0651 ***	1	0.0297 *
Epidemic	0.0371 **	0.1578 ***	0.0676 ***	0.0324 **	−0.0306 **	0.0500 ***	0.0157	0.0561 ***	−0.0161	0.0062	1

Note: *** p < 0.01, ** p < 0.05, * p < 0.1.

in Appendix A) indicate that there is no serious multicollinearity in the empirical model. The results of the reliability test (

Table A3. Estimates of variable reliability analysis.

Variable	Cronbach’s α Coefficient
Negative media sentiment about the pig epidemic	0.8886

in Appendix A) suggest that the internal consistency of the indicators is good and the reliability is high. The results of the stationarity test (

Table A4. Estimates of unit root test.

Variables	LLC	IPS	Fisher-ADF	Breitung
Pork	−41.7441	−42.5952	155.6715	−8.4690
	(0.0000)	(0.0000)	(0.0000)	(0.0000)
Media	−19.3095	−27.8018	141.9882	−16.1777
	(0.0000)	(0.0000)	(0.0000)	(0.0000)
Pig	−54.7928	−54.2729	175.3367	−8.6936
	(0.0000)	(0.0000)	(0.0000)	(0.0000)
Piglet	−35.8054	−36.1127	135.5601	−9.0907
	(0.0000)	(0.0000)	(0.0000)	(0.0000)
Corn	−45.7350	−41.9542	139.5824	−12.3776
	(0.0000)	(0.0000)	(0.0000)	(0.0000)
Sbm	−53.0135	−50.6868	150.3343	−10.2994
	(0.0000)	(0.0000)	(0.0000)	(0.0000)
Beef	−40.3161	−41.6626	134.3540	−16.6654
	(0.0000)	(0.0000)	(0.0000)	(0.0000)
Mutton	−40.1693	−43.3153	134.8924	−16.7603
	(0.0000)	(0.0000)	(0.0000)	(0.0000)
Chicken	−33.4112	−34.6728	120.4541	−16.4672
	(0.0000)	(0.0000)	(0.0000)	(0.0000)
Income	−27.6815	−35.7543	150.9095	−8.7811
	(0.0000)	(0.0000)	(0.0000)	(0.0000)
Epidemic	−18.3515	−25.3143	131.0701	−14.5611
	(0.0000)	(0.0000)	(0.0000)	(0.0000)

Note: Data in brackets are p-values of statistics.

in Appendix A) suggest that all variables in this paper are original order stationary variables without unit roots, which can be directly added to the panel regression model for analysis.

In this paper, the traditional panel model is first used to investigate the relationship between negative media sentiment about the pig epidemic and pork price fluctuations. The fixed-effect OLS model is used to estimate Equation (4), and the results are shown in column (1) of

Table 2. Estimates of the benchmark regression and the threshold effect test.

Variables	(1) FE	(2) FGLS	(3) The Threshold Effect
Media	0.4899 ***	0.4919 ***
	(0.0795)	(0.0562)
Media × d1			4.9502 ***
			(0.5792)
Media × d2			0.5280 ***
			(0.0606)
Pig	0.6455 ***	0.6616 ***	0.6419 ***
	(0.0183)	(0.0075)	(0.0076)
Piglet	0.0137	0.0115 **	0.0097 *
	(0.0082)	(0.0051)	(0.0051)
Corn	0.0033	0.0232	0.0068
	(0.0222)	(0.0209)	(0.0219)
Sbm	0.0427 ***	0.0296 **	0.0372 ***
	(0.0151)	(0.0137)	(0.0143)
Beef	0.2283 ***	0.2396 ***	0.2342 ***
	(0.0496)	(0.0336)	(0.0344)
Mutton	−0.0329	−0.0610 **	−0.0385
	(0.0370)	(0.0293)	(0.0310)
Chicken	0.1328 ***	0.1246 ***	0.1248 ***
	(0.0301)	(0.0152)	(0.0152)
Income	−0.0721	0.0061	−0.0492
	(0.0768)	(0.0820)	(0.0861)
Epidemic	−0.0005	−0.0008 ***	−0.0007 ***
	(0.0003)	(0.0002)	(0.0003)
cons	−0.0180	−0.0250	−0.0184
	(0.1453)	(0.0958)	(0.1024)
N	4320	4320	4320
R2	0.7206		0.7246
F	370.1028		987.3873

Note: Standard errors are shown in parentheses. ***, **, and * indicate statistically significant differences at the 1%, 5%, and 10% levels, respectively.

. The influence coefficient of negative media sentiment about the pig epidemic on pork price fluctuations is 0.4899, which is statistically significant. To solve the problem of heteroskedasticity, intra-group autocorrelation, and cross-sectional correlation (

Table A5. Estimates of heteroscedasticity, intra-group autocorrelation, and cross-sectional correlation tests.

Heteroscedasticity Test		Intra-Group Autocorrelation Test		Cross-Sectional Correlation Test
Statistic	p-value	Statistic	p-value	Statistic	p-value
215.66	0.0000	F (1,29) = 19.754	0.0001	13822.505	0.0000

in Appendix A), the feasible generalized least squares (FGLS) is used to estimate the benchmark model, and the results are shown in column (2) of Table 2. The influence coefficient of negative media sentiment about the pig epidemic is 0.4919, which is statistically significant. The findings suggest that the negative sentiment conveyed by media coverage of the external impact of the pig epidemic would interfere with the pork market and has a significant positive impact on pork price fluctuations. Specifically, the polarized information transmitted by the negative media sentiment about the pig epidemic strengthens the feedback and circulation of the epidemic information through the Internet, constantly amplifying the risk of the pig epidemic. Consumers’ fear and risk cognition of pork food form emotional contagion among groups, which easily leads to the polarization of the group’s pork consumption behavior. The downturn in pork market demand, combined with the shortage of pork supply caused by the pig epidemic, has exacerbated the pork price fluctuations. This result verifies hypothesis H1.

The negative media sentiment is a relative ratio. Considering the nonlinear influence of negative media sentiment about the pig epidemic on pork price fluctuations in different intervals, this paper takes negative media sentiment about the pig epidemic as the threshold variable to carry out threshold effect analysis. Firstly, we examine the form of the panel threshold model, simulate the likelihood ratio statistics 2000 times using the Bootstrap sampling method, and estimate the threshold value and related statistics. The results in

Table 3. Estimates of threshold effect test.

Test	Threshold Value	F-Value	Prob	The 10% Threshold	The 5% Threshold	The 1% Threshold
Single threshold test	0.4055 ***	95.78	0.0000	15.4674	17.9012	23.5449
Double threshold test	0.4520	9.31	0.1110	9.6583	12.3066	18.2152
Three threshold test	1.6094	3.99	0.8990	18.1383	21.6141	27.2817

Note: *** indicates statistically significant differences at the 1% level.

show that the F statistic of a single threshold is significant at the level of 1%, while the F statistic of double and triple thresholds is not significant. Therefore, it is believed that there is a single threshold effect on negative media sentiment about the pig epidemic, and the total sample is divided into two sections according to the negative media sentiment about the pig epidemic. Interval 1: media ≤ 0.4055; interval 2: media > 0.4055. The single threshold model is set as Equation (5).

Column (2) of Table 2 presents the influence of negative media sentiment about the pig epidemic on pork price fluctuations in different threshold ranges. When media ≤ 0.4055, the negative media sentiment about the pig epidemic has a significantly elevated effect on pork price fluctuations. When media > 0.4055, the exacerbation effect of negative media sentiment decreases to 0.5280. The finding indicates that when the negative media sentiment about the pig epidemic is greater than the threshold value, the frequency of recurrence and circulation of the epidemic information source decreases, the risk amplification effect of negative media sentiment about the pig epidemic attenuates, consumers from the established perception of the safety of the pig epidemic, and the marginal change of group polarization behavior decreases. The total demand of the pork market is relatively flat, which reduces the aggravating effect on pork price fluctuations. Overall, the negative media sentiment about the pig epidemic has a threshold effect on pork price fluctuations, and with the negative media sentiment about the pig epidemic as the threshold variable, the aggravating effect of the negative media sentiment about the pig epidemic on pork price fluctuations changes from strong to weak after crossing the threshold value.

3.2. Spatial Spillover Effect of Negative Media Sentiment about the Pig Epidemic on Pork Price Fluctuations

This paper further uses the spatial econometric model to estimate Equation (6). The Wald test and LR test both significantly reject the null hypothesis, indicating that only using the spatial lag model and the spatial error model to investigate the spatial spillover effect may be biased, and the results of the Husman test significantly reject the null hypothesis at the 1% confidence level. Therefore, this paper selects the spatial panel Durbin model (SPDM) under fixed effect to analyze the spillover effect of negative media sentiment about the pig epidemic on pork price fluctuations. Column (1) in

Table 4. Estimation results of spatial spillover effect.

	Variables	(1)	(2)	(3)	(4)
		Inverse Distance Matrix	Matrix of Gravity	Inverse Distance Matrix	Matrix of Gravity
	Media	0.1612 ***	0.1802 ***	0.1352 **	0.1505 ***
		(0.0530)	(0.0531)	(0.0530)	(0.0531)
	Pig	0.5673 ***	0.5641 ***	0.5711 ***	0.5680 ***
		(0.0103)	(0.0103)	(0.0103)	(0.0103)
	Piglet	0.0192 ***	0.0176 ***	0.0169 ***	0.0146 ***
		(0.0053)	(0.0052)	(0.0052)	(0.0052)
	Corn	0.0209	0.0247	0.0255	0.0300
		(0.0229)	(0.0228)	(0.0229)	(0.0228)
	Sbm	0.0459 ***	0.0487 ***	0.0455 ***	0.0480 ***
		(0.0163)	(0.0163)	(0.0163)	(0.0163)
	Beef	0.1451 ***	0.1537 ***	0.1436 ***	0.1491 ***
		(0.0317)	(0.0315)	(0.0317)	(0.0316)
	Mutton	0.0427	0.0345	0.0470*	0.0407
		(0.0261)	(0.0261)	(0.0262)	(0.0262)
	Chicken	0.0556 ***	0.0613 ***	0.0521 ***	0.0570 ***
		(0.0135)	(0.0135)	(0.0135)	(0.0134)
	Income	0.0547	0.0742	0.0446	0.0615
		(0.1020)	(0.1004)	(0.1019)	(0.1003)
	Epidemic	−0.0000	−0.0000	0.0000	0.0000
		(0.0002)	(0.0002)	(0.0002)	(0.0002)
	W×Media	0.3393 ***	0.3085 ***	0.2439 ***	0.2092 ***
		(0.0742)	(0.0725)	(0.0750)	(0.0732)
	W×Pig	−0.2460 ***	−0.2255 ***	−0.2356 ***	−0.2150 ***
		(0.0139)	(0.0134)	(0.0140)	(0.0135)
	W×Piglet	−0.0139 **	−0.0110 *	−0.0270 ***	−0.0239 ***
		(0.0068)	(0.0066)	(0.0070)	(0.0067)
	W×Corn	−0.0212	−0.0352	−0.0114	−0.0244
		(0.0279)	(0.0270)	(0.0279)	(0.0269)
	W×Sbm	−0.0287	−0.0314 *	−0.0258	−0.0284
		(0.0190)	(0.0186)	(0.0190)	(0.0186)
	W×Beef	0.0204	0.0132	0.0003	−0.0013
		(0.0411)	(0.0385)	(0.0413)	(0.0385)
	W×Mutton	−0.1246 ***	−0.1138 ***	−0.0961 **	−0.0870 **
		(0.0378)	(0.0356)	(0.0381)	(0.0358)
	W×Chicken	0.0478 **	0.0428 **	0.0362 *	0.0335 *
		(0.0188)	(0.0179)	(0.0189)	(0.0179)
	W×Income	−0.1252	−0.1651	−0.1089	−0.1479
		(0.1207)	(0.1186)	(0.1207)	(0.1185)
	W×Epidemic	−0.0013 ***	−0.0012 ***	−0.0013 ***	−0.0011 ***
		(0.0003)	(0.0003)	(0.0003)	(0.0003)
	ρ	0.5089 ***	0.4860 ***	0.4878 ***	0.4636 ***
		(0.0120)	(0.0117)	(0.0125)	(0.0122)
Direct Effect	Media	0.2555 ***	0.2690 ***	0.2157 ***	0.2246 ***
		(0.0571)	(0.0573)	(0.0544)	(0.0546)
Indirect Effect	Media	0.6460 ***	0.5761 ***	0.5482 ***	0.4743 ***
		(0.1033)	(0.0962)	(0.1252)	(0.1168)
Total Effect	Media	0.9015 ***	0.8451 ***	0.7639 ***	0.6989 ***
		(0.1329)	(0.1264)	(0.1527)	(0.1451)
	L.WPork			0.0686 ***	0.0740 ***
				(0.0087)	(0.0087)
	N	4320	4320	4290	4290
	R2	0.7240	0.7212	0.7369	0.7354

Note: Standard errors are shown in parentheses. ***, **, and * indicate statistically significant differences at the 1%, 5%, and 10% levels, respectively.

presents the result of the inverse distance matrix. Under the direct effect, the coefficient of negative media sentiment about the pig epidemic is 0.2555, which is significant at the 1% level. The finding indicates that the negative media sentiment about the pig epidemic in the local region will significantly exacerbate the pork price fluctuations in the local region. Meanwhile, compared with the fixed-effect panel model, the coefficient is reduced by 0.2344, indicating that if the spatial spillover effect is ignored, the positive influence of negative media sentiment on pork price fluctuations would be overestimated. Under the indirect effect, the coefficient of negative media sentiment about the pig epidemic is 0.646, which is significant at the 1% level. The finding suggests that the increase in negative media sentiment about the pig epidemic in the local region will increase pork price fluctuations in the neighboring region. On the one hand, there is a risk of spatial spread of the pig epidemic. On the other hand, in the Internet era, information spreads rapidly and widely. As public opinion leaders, some media release authoritative information, and the negative reporting sentiment is easily transmitted to consumers through the Internet. Consumers in this region will spread their sentiment to consumers in neighboring regions and strengthen the “overreaction” of consumers in neighboring regions. Risk aversion and herding effect lead consumers in neighboring regions to observe and learn the behavior of local consumers, reduce pork purchases, and resume pork purchases during the decline stage of the pig epidemic. The fluctuations of pork market demand in the neighboring areas eventually lead to the spatial spillover effect of pork price fluctuations.

As regional economic development has a direct impact on the total demand for pork, column (2) presents the spatial spillover result of the gravity model matrix. Both direct and indirect effects of negative media sentiment about the pig epidemic are significantly positive, and the intensity of the indirect effect decreases. Compared with the traditional static spatial panel model, the dynamic spatial panel model (SDPD) has advantages in solving the endogenous problems caused by the complexity of spatial metrology models and spatial weights. Column (3) presents the result of the dynamic spatial model using the inverse distance spatial matrix. The spatial lag term of pork price fluctuation is significantly positive at the 1% level, and the spatial spillover coefficient is further reduced to 0.5482, indicating that the static spatial panel model is prone to overestimate the spatial spillover effect of negative media sentiment about the pig epidemic on pork price fluctuations. Column (4) reports the result of the dynamic spatial model using the gravity model matrix. Both direct and indirect effects are significantly positive. These results verify hypothesis H2.

Table A7. Estimates of robustness test.

Variables	Index Reconstruction			Replace the Sentiment Analysis Algorithm
	(1) FE	(2) SPDM	(3) SDPD	(4) FE	(5) SPDM	(6) SDPD
Media	0.9602 ***	0.4317 ***	0.3484 ***	0.7327 ***	0.2773 ***	0.2215 ***
	(0.1846)	(0.1024)	(0.1025)	(0.1127)	(0.0766)	(0.0767)
L.Pork
L.WPork			0.0743 ***			0.0735 ***
			(0.0087)			(0.0087)
W×Media		0.5096 ***	0.2980 **		0.4470 ***	0.2873 ***
		(0.1374)	(0.1391)		(0.1037)	(0.1051)
ρ		0.4867 ***	0.4645 ***		0.4853 ***	0.4636 ***
		(0.0117)	(0.0122)		(0.0118)	(0.0122)
Direct effect		0.5936 ***	0.4722 ***		0.4074 ***	0.3249 ***
		(0.1144)	(0.1105)		(0.0843)	(0.0811)
Indirect effect		1.0483 ***	0.7964 ***		0.8451 ***	0.6635 ***
		(0.1925)	(0.2331)		(0.1417)	(0.1726)
Total effect		1.6420 ***	1.2686 ***		1.2525 ***	0.9884 ***
		(0.2634)	(0.3003)		(0.1908)	(0.2191)
N	4320	4320	4290	4320	4320	4290
R2	0.7203	0.7206	0.7345	0.7206	0.7217	0.7351

Note: Standard errors are shown in parentheses. *** and ** indicate statistically significant differences at the 1% and 5% levels, respectively.

and

Table A8. Estimates of instrumental variable.

Variables	Model 1	Model 2
Media	4.3532 ***	1.1960 ***
	(0.3682)	(0.2576)
Control variable		YES
Fixed Effect	YES	YES
N	4320	4320
R2	0.7131	0.7157
The first stage regression results	0.2828 ***	0.2271 ***
	(0.0131)	(0.0137)
KP-F test	467.584	274.991

Note: Standard errors are shown in parentheses. *** indicates statistically significant differences at the 1% level.

in Appendix A present the results of the robustness test and the endogeneity test of the spatial Durbin model respectively. The findings suggest that the spatial Durbin model constructed in this paper and its conclusions are robust.

3.3. Attenuation Boundary of the Spatial Spillover Effect of Negative Media Sentiment about the Pig Epidemic on Pork Price Fluctuations

According to the first Law of Geography hypothesis, the spatial dependence decrease with the increase in the geographical distance between provinces. Due to the attenuation of information spatial transmission and local protectionism, the spatial impact of negative media sentiment about the pig epidemic may have a certain regional boundary. To reveal the change of spillover effect of negative media sentiment about the pig epidemic on pork price fluctuations with distance attenuation, the space matrix of the distance threshold value is set as follows:

$$W_{ij}(3)=\left\{\begin{array}{l}1/d_{ij} d_{ij}<d\\ 0 d_{ij}\ge d\end{array}\right.$$

(11)

In this study, the minimum distance between provinces is 75.01 km, and the longest is 3597 km. Therefore, the initial value of the threshold distance spatial weight matrix is set to 100 km, and the progressive distance is set to 100 km. Considering that when the distance threshold is greater than 3000 km, only a few areas remain, and there is relatively large noise in the result of the spatial spillover effect, so only the result within 3000 km is adopted. Based on the threshold distance spatial weight matrix, the SPDM model has been used to re-estimate the spatial spillover effect of negative media sentiment about the pig epidemic on pork price fluctuations in different spatial distances.

Figure 1 presents the variation characteristics of the visual-spatial spillover effect under the distance constraint. With the increase in distance, the spatial spillover effect of negative media sentiment about the pig epidemic shows significant changes. The spatial spillover effect of negative media sentiment about the pig epidemic on pork price fluctuations depends on the geographical distance, and the spatial spillover effect and geographical distance show an inverted U-shaped relationship. The inflection point of the inverted U-shaped curve is 1400 km. When the spatial distance between two regions is less than 1400 km, the spillover effect coefficient fluctuates from 0.036 to 1.575, and the positive spatial spillover effect increases with the increase in geographical distance. The spillover effect on pork price fluctuations in surrounding areas is greater than that in adjacent areas, indicating that negative media sentiment significantly amplifies the risk of the pork market within the regional radiation range. When the spatial distance between two regions is greater than 1400 km, the positive spatial spillover effect decreases with the increase in geographical distance. The attenuation boundary of the spatial spillover effect of negative media sentiment about the pig epidemic is about 2300 km, and the spatial spillover effect coefficient rapidly attenuates close to 0 near 2300 km, while the significance level drops sharply. It indicates that the region beyond the attenuation boundary is hardly affected by the spillover effect of negative media sentiment.

3.4. Heterogeneity Analysis of Media Reputation Level

This paper examines whether the spillover effect of negative media sentiment about the pig epidemic on pork price fluctuations depends on the level of media reputation signal. On the one hand, due to professionalism, the media with a high reputation can release more objective and fair information on the pig epidemic and correctly guide consumers’ judgment on the pork market. On the other hand, when consumers have an accurate judgment and a high degree of recognition of the reputation level of the media, they rely more on the pig epidemic coverage released by the media and then adjust their consumption behavior decisions. This study uses the coverage and authority of media to measure the level of media reputation. According to the level of media reputation, media can be divided into three categories: central media, local media, and information network platforms. The central media have the highest authority and are well known by the vast majority of consumers, with better editing resources and a better team of journalists. Local media also have the right to edit and strong editing ability. Information network platforms do not have the right to collect and edit but can only reprint the coverage of regular news media. To attract traffic and audiences, they have a relatively large motive to publish subjective coverage, such as Tencent news, Sina news, etc. Therefore, compared with information network platforms, the reputation level of central media and local media is higher. In this paper, the spatial spillover effect is analyzed by subsample regression according to three types of media.

Columns (1), (3), and (5) in

Table 5. Heterogeneity analysis of media reputation level.

Variables	Central Media		Local Media		Information Network Platforms
	(1) SPDM	(2) SDPD	(3) SPDM	(4) SDPD	(5) SPDM	(6) SDPD
Media	0.3484 ***	0.3234 ***	0.3711 ***	0.3382 ***	0.1666 ***	0.1356 **
	(0.1050)	(0.1046)	(0.0783)	(0.0781)	(0.0544)	(0.0543)
L.WPork		0.0805 ***		0.0778 ***		0.0744 ***
		(0.0085)		(0.0086)		(0.0086)
W×Media	0.1974	0.0680	0.1879 *	0.0904	0.3203 ***	0.2271 ***
	(0.1378)	(0.1379)	(0.1017)	(0.1018)	(0.0735)	(0.0740)
ρ	0.4938 ***	0.4676 ***	0.4900 ***	0.4653 ***	0.4864 ***	0.4636 ***
	(0.0116)	(0.0121)	(0.0117)	(0.0122)	(0.0117)	(0.0122)
Direct Effect	0.4347 ***	0.3868 ***	0.4558 ***	0.4086 ***	0.2567 ***	0.2122 ***
	(0.1187)	(0.1133)	(0.0850)	(0.0806)	(0.0586)	(0.0558)
Indirect Effect	0.5413 ***	0.4170 *	0.5398 ***	0.4552 ***	0.5841 ***	0.4909 ***
	(0.1993)	(0.2381)	(0.1371)	(0.1641)	(0.0975)	(0.1181)
Total Effect	0.9760 ***	0.8038 ***	0.9956 ***	0.8637 ***	0.8408 ***	0.7032 ***
	(0.2758)	(0.3083)	(0.1841)	(0.2070)	(0.1284)	(0.1470)
N	4320	4290	4320	4290	4320	4290
R2	0.7172	0.7342	0.7197	0.7359	0.7206	0.7352

Note: Standard errors are shown in parentheses. ***, **, and * indicate statistically significant differences at the 1%, 5%, and 10% levels, respectively.

, respectively, show the estimation results of the negative sentiment about the pig epidemic of central media, local media, and information network platforms on pork price fluctuations using the static spatial model of gravity model matrix. Columns (2), (4), and (6) in Table 5, respectively, report the estimation results of central media, local media, and information network platforms using the dynamic spatial model of gravity model matrix. In terms of direct effect and total effect, the influence coefficient of local media is the highest. The finding suggests that local media has greater influence and trust on local consumers than central media. Local media’s negative coverage of the pig epidemic in the region has a more obvious impact on the risk perception and purchasing decisions of local consumers and a greater impact on fluctuations in local pork prices. In terms of indirect effects, negative coverage of information network platforms has the largest spillover effect on pork price fluctuations. The finding suggests that information network platforms republish a large number of coverage from official media, combined with the wide coverage of mobile phone clients, which can quickly transmit negative media sentiment to more consumers in neighboring areas and have a more obvious positive spillover effect on pork price fluctuations in neighboring areas. To sum up, the negative sentiment of the media with a high reputation has a more obvious effect on the aggravation of pork price fluctuations within the region and the whole. When formulating policies to stabilize pork prices, it is necessary to strengthen the objective and fair supervision of epidemic information released by local media and control negative coverage, to restrain the aggravating effect of pork price fluctuations.

4. Mechanism Research: The Mediating Effect of “Consumer Sentiment”

The above empirical analysis proves that negative media sentiment has a significant aggravating effect on pork price fluctuations. Therefore, in the next step, this paper needs to explore through which influence path negative media sentiment about the pig epidemic is transmitted to pork price fluctuations. According to the above theoretical analysis, negative media sentiment may exacerbate pork price fluctuations through the mechanism of consumer sentiment. The mediation effect model is tested according to Equations (9)–(10).

Table 6. Estimates of intermediate effect test.

	(1)	(2)	(3)
Variables	Pork	Consumer	Pork
Media	1.5814 ***	0.0927 **	1.5764 ***
	(0.4842)	(0.0358)	(0.4414)
Consumer			0.3244 ***
			(0.0721)
Pig	0.7672 ***	0.0102 ***	0.6489 ***
	(0.0629)	(0.0022)	(0.0389)
Piglet	−0.0448	−0.0015	−0.0400
	(0.0306)	(0.0014)	(0.0242)
Corn	0.3306	−0.0005	0.2503
	(0.2104)	(0.0085)	(0.1693)
Sbm	−0.0285	0.0012	−0.0022
	(0.0822)	(0.0024)	(0.0787)
Beef	−0.1581	−0.0239 **	−0.4054 **
	(0.2603)	(0.0089)	(0.1823)
Mutton	−0.0902	0.0175 *	0.1584
	(0.2496)	(0.0104)	(0.1683)
Chicken	0.7184	−0.0267 *	0.4284
	(0.6281)	(0.0142)	(0.3630)
Income	−0.0288	−0.0007	−0.0518 *
	(0.0377)	(0.0016)	(0.0292)
Epidemic	0.0019	0.0001	0.0060
	(0.0051)	(0.0001)	(0.0038)
cons	−0.3311	0.2387 ***	0.2423
	(1.4533)	(0.0358)	(1.0842)
R2	0.9008	0.6944	0.9413
F	47.3029	36.0593	92.8722

Note: Standard errors are shown in parentheses. ***, **, and * indicate statistically significant differences at the 1%, 5%, and 10% levels, respectively.

presents the results of mediating effects. Column (1) examines the overall effect of negative media sentiment about the pig epidemic on pork price fluctuations. The coefficient of media is 1.5814, which is significantly positive at the 1% level. In column (2), the coefficient of media is 0.0927. The finding suggests that negative media sentiment about the pig epidemic has a significant positive impact on consumer sentiment, and the more negative media sentiment about the pig epidemic is, the more negative consumer sentiment is. The results of column (3) show that negative media sentiment and negative consumer sentiment can positively influence pork price fluctuations at the same time, which is significant at the 1% level, and the coefficient of media is significantly smaller than the coefficient in column (1). The findings suggest that negative consumer sentiment plays a partial mediating effect in the process of negative media sentiment about the pig epidemic affecting pork price fluctuations. That is to say, negative media sentiment about the pig epidemic exacerbates pork price fluctuations by enhancing the transmission mechanism of negative consumer sentiment. It has also been found in other agricultural price studies that the uncertainty of sudden major events significantly affects the spillover effect between public sentiment and agricultural product prices, such as Akyildirim et al. [57]. Negative media sentiment builds a powerful opinion environment for consumers and can significantly affect consumers’ decision-making environment and cognitive response. When the media sentiment about the pig epidemic is relatively negative, consumers may arouse negative fear through the mechanism of emotional contagion, reduce pork purchasing decisions, herd effect forms group polarization behavior, then changes in the total demand of the pork market aggravate pork price fluctuations. These results verify hypothesis H3, that there is a “consumer sentiment” influence path. The results of sensitivity analysis(Figure A1 and

Table A9. Sensitivity results.

Rho at which ACME = 0	0.6319
R2_M*R2_Y* at which ACME = 0:	0.3993
R2_M~R2_Y~ at which ACME = 0:	0.0197

Note: ACME is the average causal mediation effect. * indicates statistically significant differences at the 10% level.

in Appendix A) indicate that under the assumption of sequence ignorability, the conclusions of this paper are reasonable to a certain extent, and the results of the study of mediation effect, direct effect, and total effect are robust.

5. Discussion

This study provides a new research perspective on pork price fluctuations and provides general conclusions and empirical evidence for the spatial spillover effect of pork price fluctuations in the animal epidemic. The pork prices are determined by both the supply and demand in the pork market. As for the external impact of the pork price fluctuations, the negative media sentiment about the pig epidemic amplifies the risk of the epidemic and affects pork price fluctuations. To cater to the psychological expectation of the public’s preference for negative information, the media often give up the neutral stance in reporting the pig epidemic and use emotionally biased reporting strategies to convey negative media sentiment and opinions to the public. Previous studies have shown that the public tends to magnify the effect of negative information in the face of epidemic information, which can easily determine the influence trend and effect of comprehensive information [58]. The feedback and circulation of negative information as well as emotional contagion, make it easier for consumers to have collective resonance with negative sentiments and views of media. Box-Steffensmeier et al. [59] claimed that there is a differential impact of media sentiment on message engagement, information spread, and public reaction. Based on the stimulus-cognitive-behavior model of signal transmission theory, the negative sentiment stimulation of media may induce the risk cognition, emotion, and instinctive self-protection response of individual consumers to the pig epidemic, and cross contagion through the Internet, causing excessive behavior of the converging consumer group [60], which may lead to changes in the market demand for pork, the imbalance between supply and demand eventually exacerbates pork price fluctuations.

6. Conclusions

Based on the perspective of behavioral economics, this paper uses the pork market data from January 2011 to December 2022 through the crawler program to obtain media coverage of the pig epidemic, using the natural language processing technology based on the machine learning method for the emotion analysis of media coverage, to investigate the impact and spillover effect of negative media sentiment about the pig epidemic on pork price fluctuations, and further explore whether consumer sentiment has the mediating effect between negative media sentiment about the pig epidemic and pork price fluctuations. The following conclusions are obtained:

The negative media sentiment about the pig epidemic has a significant positive impact on pork price fluctuations, and the impact has a single threshold effect. After the negative media sentiment crosses the threshold, the aggravating effect on pork price fluctuations changes from strong to weak.

The negative media sentiment about the pig epidemic has a positive spatial spillover effect on pork price fluctuations in neighboring areas, showing the characteristics of “mutual loss”. However, due to the constraints of spatial distance and administrative boundary, with the increase in geographical distance between the two regions, the spatial spillover effect of negative media sentiment shows an inverted U-shaped change. When the spatial distance between two regions is greater than 1400 km, which is the inflection point of the inverted U curve, the spatial spillover effect of negative media sentiment about the pig epidemic on pork price fluctuations shows obvious spatial attenuation characteristics. Moreover, the spatial geographic attenuation boundary is about 2300 km. When the spatial distance between the two regions is greater than 2300 km, the spillover effect of pork price gradually declines to zero.

The effect of negative media sentiment on pork price fluctuations is heterogeneous in media reputation. The negative sentiment of the central media, local media, and information network platforms about the pig epidemic not only increases pork price fluctuations in the local region but also aggravates pork price fluctuations in neighboring regions. Among them, the negative sentiment of local media has the most significant local aggravating effect on pork price fluctuations, and the negative sentiment of information network platforms has the strongest spillover aggravating effect on pork price fluctuations in neighboring areas.

Mechanism research finds that consumer sentiment plays a part in the mediating effect in the process of negative media sentiment about the pig epidemic exacerbating pork price fluctuations, forming the transmission path and resonance mechanism. To be specific, negative media sentiment about the pig epidemic trigger negative consumer sentiment, which leads to changes in the total demand in the pork market, thus exacerbating pork price fluctuations.

Based on the above conclusions, the following suggestions are put forward: First, public opinion management departments of the government should establish a regulation system for emotions conveyed by multiple media, prudently supervise the emotional texts of media coverage and verify their objectivity, avoid excessive and untrue negative coverage of the pig epidemic, then control pork price fluctuations from the perspective of public opinion. Secondly, we should pay attention to the spatial spillover effect of media sentiment. Local governments should establish a joint prevention and control mechanism for inter-regional pork price fluctuations warning, public opinion information sharing, and media sentiment control. Third, public opinion management departments of the government should reasonably guide consumer sentiment by releasing accurate information on the pig epidemic and emergency plans, and establish a mechanism for monitoring and channeling consumer sentiment, to avoid negative consumer sentiment leading to consumer group polarization behavior. Finally, pig farmers can take the media sentiment about the pig epidemic as a signal of demand in the pork market, then adjust their production and management decisions and breeding scale, to achieve an elastic matching with pork market demand and minimize production and operation losses. Animal husbandry management departments should make use of pork price fluctuations prediction information to correctly guide the decision-making behavior of market participants, guide the production activities of pig farmers and the consumption behavior of consumers, and reduce the negative impact of the pig epidemic.

Combined with the possible shortcomings of this paper, future studies can be improved from the following aspects: First, this study used macro data to explore the key nodes of the transmission path of negative media sentiment about the pig epidemic and pork price fluctuations. Future studies can use survey data to explore the impact of negative media sentiment about the pig epidemic on micro-individual consumers’ risk perception, emotion, and decision-making behavior regarding pork consumption and more carefully depict the mechanism of media sentiment on pork price fluctuations. Second, the monthly provincial panel data in this paper may not be precise enough. Future studies can construct daily municipal high-frequency data to more accurately depict the spatial spillover effect of negative media sentiment on fluctuations in agricultural product prices. Third, due to the limited sample size, this study only discussed the influence mechanism of media sentiment on pork price fluctuations after 2011. Future studies can extend the sample observation time as much as possible to explore the differences in the influence effect of negative media sentiment about the pig epidemic on pork price fluctuations under different epidemic nodes in the long term.