Pork Consumption Patterns among Rural Residents in China: A Regional and Cultural Perspective (2000–2020)

Abstract

Pork is a principal component of the food supply for residents in China, acting as a primary source of animal protein. Analyzing the factors affecting pork consumption among rural Chinese residents is critical for understanding trends in the pig market and the direction of price control. China encompasses an immense geographical area (ranked third in the world) and is home to a globally acknowledged cultural diversity. This results in China having a diverse range of regional and cultural characteristics. These factors may profoundly influence the meat consumption habits of rural residents in China in a real-world context. In econometric decision making, traditional fixed effects models are incapable of estimating regional and cultural factors as these come under individual effects. In this study, leveraging a distinct fixed-effect filter model recently developed, we estimate these regional and cultural variables to verify whether they are statistically significant. This approach adds a fresh perspective to traditional consumption theory. This study uses panel data from 31 provinces, municipalities, and autonomous regions registered in national data as the research object. The detailed findings include: (1) Rural residents in coastal provinces show a stronger inclination towards pork consumption compared to those in inland areas. (2) The ethnic feature of provinces also presents a significant impact, with provinces reflecting Mongolian, Zhuang, Tujia, and Yi ethnic characteristics pointing to higher pork consumption. This contrasts with provinces characterized by Yao, Li, and Muslim ethnicities, which tend towards the opposite. Based on these empirical findings, this paper provides policy suggestions for optimizing the layout of the pig industry, which will offer multi-dimensional regulatory directions for optimizing the pig industry layout in China.

1. Introduction

There is a proverb in China: “Food is the first necessity of the people, and stability in swine horticulture ensures peace for the world”. This proverb underscores, to a certain extent, the significance of pork products for Chinese consumers. China has the highest total consumption of pork worldwide (Source: Organization for Economic Co-operation and Development, OECD, 2021). As one of the most critical commodities in the “food basket” of the populace, the apparent consumption of pork in China reached 56.59 million tons in 2021 (Source: U.S. Department of Agriculture, USDA, 2022). As president Xi Jinping pointed out, it is crucial to develop a broad perspective of food and understand the trends in the people’s food consumption patterns to better meet their needs for a better life [1].

From a historical viewpoint, China’s per capita grain consumption continues to decrease, while the consumption of various livestock products is rapidly increasing (Source: National Bureau of Statistics of China). The public’s demand for a nutrient-rich, diverse, and healthy diet is not only an objective development trend but also a crucial task in response to people’s ardent expectations. Still, numerous challenges exist. Regular challenges like the “Pig Cycle”, due to the high degree of marketization of livestock products, are prone to price fluctuations with the impact often deeper than that for grain. Irregular challenges include exogenous input and endogenous spread of animal diseases. Severe disease outbreaks often lead to livestock deaths and large-scale culling, hurting market confidence and affecting the pork supply. In particular, after the nationwide outbreak of African swine fever in 2018, the Ministry of Agriculture and Rural Affairs, along with other departments, issued a series of targeted guidelines on severe issues affecting people’s livelihood such as a decline in pig production capacity and a significant rise in pork prices. The Central Document No. 1 of 2019–2023 focuses on the stability of supply in the pig sector and disease prevention [2,3,4,5,6].

With the continuous development of society and the economy, noticeable changes have been seen in the dietary structure of the Chinese people. In particular, demand for livestock products among urban and rural residents continually grows [7], with a noticeable surge in pork consumption in rural regions [8]. The per capita consumption of livestock products among Chinese residents is rapidly increasing. As for the rural consumption dimension, according to data from the “China Statistical Yearbook”, the per capita consumption of pork, beef, mutton, poultry, eggs, and dairy products by rural residents in 2000 was 13.28 kg, 0.52 kg, 0.61 kg, 2.81 kg, 4.77 kg, and 1.06 kg, respectively. These figures in 2021 were 25.4 kg, 1.5 kg, 1.2 kg, 12.4 kg, 13.0 kg, and 9.3 kg, respectively. The per capita consumption of grain by rural residents decreased from 250.23 kg to 170.8 kg over the same period [9]. Chinese residents’ consumption structure has undergone significant changes. Specifically, the per capita pork consumption among rural residents nearly doubled within this time span (25.4/13.28 ≈ 1.913). Therefore, the research topic of this paper is meaningful. The proportion of pork in nationwide meat consumption consistently remains over 50%. In the foreseeable future, significant structural variations in meat consumption are unlikely in the context of steady consumer preferences and the standardization of the production sector. Therefore, pork will continue to have a dominant presence in meat consumption. Moreover, given China’s vast territory, large population, and diverse culture, overall judgment and calculations for the pig sector are necessary. It is imperative to consider regional heterogeneity resulting from differing regions and settlements to avoid misjudgment caused by various discrepancies on the macro-consumption situation and trend of pork in China. Current econometric studies primarily focus on time-variant variables such as income, price, and output.

There are studies suggesting that income is an essential factor influencing the per capita meat consumption level and structure [10]. The results from LA/AIDS model calculations also indicate that increasing urban and rural residents’ income can significantly increase the consumption of poultry and pork [11]. On the price front, research shows that a distinct agricultural price magnification effect exists in China’s pig sector from production procurement to retail [12]. This price magnification effect ultimately impacts end consumers and their decisions. Moreover, as lifestyle concepts change, residents’ preference for other meats gradually emerges. This could be attributed to the fact that, in China, people widely accept a variety of media messages claiming that “pork, rich in saturated fatty acids, is not beneficial to human health”. Concurrently, the “Dietary Guidelines for Chinese Residents”—revised and compiled by the Chinese Nutrition Society (a foundational document for health education and public policy in China)—advocates for “diversity in food and balanced diet”, and recommends “a moderate intake of fish, poultry, eggs, and lean meat” [13]. The price level of other substitutes will also affect pork consumption and its market price changes [14].

In terms of production, the outbreak of African Swine Fever in 2018 resulted in serious impairment to China’s pork industry chain, characterized by a significant reduction in pork production capacity, which led to severe market fluctuations [15].

Even though these discrepancies in rural residents’ pork consumption were identified through data collection and field investigations, the root causes have not been investigated further. This leaves a gap in supporting macro-control by the country and the pinpointing of accurate policy implementation, especially evidence from mathematical aspects, which needs heightening. Alongside the consideration of time-variant influences, other factors could be tied to regional features and individual characteristics that do not change over time. A core focus of our research is to unearth these time-invariant individual characteristics (akin to “traits”) that impact consumers’ pork consumption. Some studies have also directed their inquiries towards this aspect.

For instance, Adalja et al. employed a combined hypothetical and non-hypothetical analysis of data to study consumer behaviors in Maryland, exhibiting how local residents displayed a predilection for local products based on factors like trust [16]. A study based on a survey conducted in Shandong Province, China, used the Logit-ISM model to confirm that residents’ willingness to consume antibiotic-free pork (a breeding method without the use of antibiotic drugs, considered a green breeding behavior) was significantly influenced by variables such as place of residence and gender [17]. Lu et al., focusing on traceability information of pork hindquarters, categorized it into four types: “breeding traceability information”, “slaughtering and processing traceability information”, “transportation and sales traceability information”, and “government-approved traceability information”. Using the k-modes clustering method to explore the consumption preferences of residents in Jiangsu Province, they found distinct levels and differences in consumers’ preferences for traceable pork attributes [18]. Chen et al.’s study on the preference for traceable pork attributes among 328 consumers in Wuxi, Jiangsu Province, China, indicated that consumers were willing to accept a higher price if the product included animal welfare information [19]. Consumers’ purchasing behavior is intensely influenced by cultural, societal, personal, and psychological characteristics; each group and society possesses a distinct culture, and this culture significantly influences purchasing behavior [20]. A research conducted in Germany, Sweden, and Denmark involving approximately 1600 consumers identified animal welfare as an ethical orientation affecting consumers’ purchases of meat products, which may be related to the difference of cultural factors [21]. In a study examining the “de-pork” movement among young people in Denmark and Sweden, Mathilde and Klaus used structural equations to verify that young people’s concern for animals and the environment had a negative impact on this demographic group’s pork consumption. Their willingness to buy and eat pork was primarily influenced by attitude and habit intensity [22]. In Hungary, the majority of consumers (71% determined by latent class, 65% determined by random parameter latent class) attached more importance to the Geographical Indications (GIs) on sausages [23]. Denver et al.’s research found that the willingness of consumers to pay for pork that has either not been treated with antibiotics or has undergone minimal antibiotic treatment is influenced by individual factors such as geography and gender [24]. A research limitation pertains to the source of these individual factors—whether they come from consumers or products. These investigations validate, to a certain extent, that individual factors do influence pork consumption among consumers. However, the current academic literature seldom features research incorporating regional and cultural factors as primary explanatory variables in econometric models.

In our research, we discovered a trend demonstrating stark differences in pork consumption habits amongst rural residents across various regions of China. Coincidentally, the diversity of geographical features and cultural attributes across these regions captured our attention. Upon recognizing that these geographical and cultural characteristics diverged at the provincial level, we ventured a hypothesis. Could these variances relate to the diverse dietary habits seen across the different provinces of China?

In panel data, variables that do not change with time cannot be effectively measured using log models and fixed effect models due to the multicollinearity arising from the time-invariant variables in the model calculation [25]. If we can estimate the various individual fixed effects that conventional econometric methods cannot determine, this would enrich the understanding of the pork consumption patterns among rural residents. This paper takes advantage of the fixed-effect filter model [26] and focuses on estimating the time-invariant variables affecting pork (with the time-variant variables as control variables).

This will augment the explicative potential of factors influencing pork consumption in rural regions, which are grounded on original research outcomes. Simultaneously, it provides a quantified corroborative assessment of time-invariant elements affecting the consumptive structure. Specifically, this study aims to statistically ascertain whether religious culture and regional factors have a significant impact on pork consumption among rural residents in China, thereby filling the gap in the existing research regarding econometric measurements. This will provide new insights for research in consumption patterns.

2. Theoretical Framework and Hypothesis

In response to the lack of individual effect analyses in empirical research on consumption in academia, we aim to enrich the explanatory dimensions of rural residents’ pork consumption. We constructed the following theoretical framework and research hypotheses.

2.1. Urban Influence on Rural Consumption Values

China’s cities have undergone significant transformations in rural areas. Since the beginning of the new century, the strong influence of modern values from urban areas has also impacted relatively stable rural regions, leading to a reshaping of the consumption structure of rural residents [27]. Long-standing habits of rural consumption have been changed due to increased personnel mobility and widespread information dissemination, reducing information asymmetry. The binary urban–rural structure is shifting towards urban–rural integration and convergence. However, urban development in China exhibits regional differences, showing a trend from west to east where urbanization rates gradually increase, the proportion of the tertiary industry expands, and infrastructure networks gradually improve. This implies geographical heterogeneity in the diffusion of urban influences centered on industrialization and marketization.

2.2. Resource Endowment’s Constraints on Consumption Structure

Resource endowment is an essential factor to consider when engaging in production and living in a region. From the perspective of studying economic cycles based on industrial structures, economists generally believe that the quantity and efficiency of resource endowment are key influencing factors in shaping and organizing industrial structures [28]. Based on their local resource endowments, regions establish comparative advantages in different production stages. Consequently, local industries and their layout and changes also affect local production and consumption activities. For example, the total national output of aquatic products was approximately 66.9029 million tons in 2021, with the consumption of these aquatic products concentrated in major fishing areas and economically developed regions (these areas’ natures can intersect). In places like Guangxi, the per capita annual consumption of aquatic products exceeds 30 kg, while in areas like Zhejiang and Shanghai, consumption can exceed 26 kg. In contrast, in inland regions such as Tibet and Qinghai, the per capita annual consumption is less than 2 kg (National Bureau of Statistics of China, 2022). Although aquatic products can be distributed through cold chains, the industrial layout and regional differences to some extent determine the local residents’ consumption structure. Under the premise that per capita intake has a maximum bearing capacity, the consumption of aquatic products will inevitably encroach on the consumption of other food items. Therefore, resource endowment affects the local consumption structure through its influence on the industrial layout in different regions.

2.3. Influence of Ethnic Culture on Consumption Preferences

Consumption behavior reflects culture, and culture also shapes different consumption preferences [29]. In China, this is particularly evident in the heterogeneity of consumption behavior among ethnic minority groups [30]. They often exhibit a strong preference for brands and products that carry symbols of their own ethnicity. The influence of their own ethnic culture enables them to retain individuality in a market economy, and this individuality becomes fixed and does not easily change with migration [31]. At the same time, this heterogeneity solidifies ethnic sentiments, providing a sense of belonging in social cognition. Therefore, ethnic culture to some extent shapes consumers’ preferences, and for ethnic minorities, these cultural differences may lead to differences in consumption structure.

For this study, a theoretical framework based on the above-mentioned three points has been developed, as illustrated in Figure 1.

Based on the theoretical framework above, the research hypotheses are proposed as follows:

H1: 

Regional factors have differential effects on pork consumption among rural residents.

H2: 

Ethnic cultural attributes have differential effects on pork consumption among rural residents.

The article will next introduce the theoretical background and application scenarios of the fixed-effect filter model, systematically explaining the advantages of using this method. The study will refer to the approach used by Jennifer (2023) [32] and others, incorporating a Pooled Ordinary Least Squares (OLS) model (combined with the Least Squares Dummy Variables (LSDV) method) and a Hausman–Taylor IV model for comparative analysis and discussion of empirical results. Finally, the article will reflect on the methods used and provide policy recommendations, further discussing unresolved issues, optimizing model methods, and proposing targeted policy suggestions based on the current empirical results.

3. Variable Selection and Data Description

3.1. Variable Selection

The selection of independent variables in this study includes variables from existing research in the field, certain statistical indicators listed in yearbooks, and exploratory variables based on research from other domains. All data can be classified into two types: time-variant and time-invariant variables. Time-variant variables refer to those that exhibit obvious changes over time, such as per capita disposable income, prices of meat products, etc. These variables are treated as control variables in the analysis. On the other hand, time-invariant variables are those derived from individual effects, such as coastal dummy variables and ethnic dummy variables. This study focuses on estimating the coefficients of these explanatory variables. Additionally, this study notes that some ethnic minorities may show a trend of migration and dispersion. However, within the timeframe of this research, this phenomenon has not resulted in significant population sizes in non-native provinces. Therefore, for the purpose of the statistical analyses, these minor changes are also treated as time-invariant variables.

Table 1. Variable selection and descriptive statistics.

Category	Variable	Description	Unit	Mean	St. dev.	Min	Max
Response variable	cons_pork	Per capita pork consumption	kg/person	13.54	6.813	0.590	31.03
Explanatory variable—geographical factor	sea	Whether coastal or not	dummy variable	0.355	0.479	0	1
Explanatory variable—ethnic factors	menggu	Mongolian or not	DV	0.0320	0.177	0	1
	chaoxian	Korea or not	DV	0.0970	0.296	0	1
	zang	Tibetan or not	DV	0.127	0.334	0	1
	miao	Hmong or not	DV	0.161	0.368	0	1
	li	Li or not	DV	0.0320	0.177	0	1
	man	Manchu or not	DV	0.161	0.368	0	1
	zhuang	Zhuang or not	DV	0.0970	0.296	0	1
	yi	Yi or not	DV	0.0970	0.296	0	1
	tujia	Tujia or not	DV	0.129	0.335	0	1
	buyi	Buyi or not	DV	0.0320	0.177	0	1
	dong	Dong or not	DV	0.0650	0.246	0	1
	yao	Yao or not	DV	0.0970	0.296	0	1
	bai	Bai or not	DV	0.0320	0.177	0	1
	hani	Hani or not	DV	0.0320	0.177	0	1
	dai	Dai or not	DV	0.0320	0.177	0	1
	muslim	Muslim or not	DV	0.129	0.335	0	1
Control variable—income	newdpi	Per capita disposable income	CNY (thousand)	7.964	5.956	1.331	34.91
Control variable—price	price_pig	Pork price	CNY/kg	21.35	9.951	7.570	57.66
	price_b	Beef price	CNY/kg	40.58	23.82	10.58	118.7
	price_m	Mutton price	CNY/kg	40.73	22.55	10.51	109.7
	price_poul~y	Poultry price	CNY/kg	15.27	5.556	6.490	36.18
Control variable—consumption habits	lagcons_g	Grain consumption (lag.1)	kg/person	185.6	44.62	81.06	303.6
	lagcons_v	Vegetable consumption (lag.1)	kg/person	90.70	33.41	13.90	199.4
	lagcons_bm	Beef and mutton consumption (lag.1)	kg/person	2.357	3.928	0.0800	18.70
	lagcons_po~y	Poultry consumption (lag.1)	kg/person	4.048	3.358	0	19.42
	lagcons_e	Egg consumption (lag.1)	kg/person	5.294	3.160	0.410	14.38
	lagcons_s	Seafood consumption (lag.1)	kg/person	5.301	5.363	0	21.69
Control variable—output	output_pig	Output of pig	10,000	2051	1764	12.60	7471
	output_cat~e	Output of cattle	10,000	151.1	140.8	0	755.1
	output_mut~n	Output of lamb	10,000	898.5	1167	13.10	6674
	output_pou~y	Output of poultry	10,000	35,000	39,000	0	250,000
Control variable—yield	yield_pork	Yield of pork	10,000 tons	157.3	132.1	0.800	541.3
	yield_b	Yield of beef	10,000 tons	21.01	21.66	0	109.3
	yield_m	Yield of mutton	10,000 tons	13.14	17.66	0.200	113
	yield_poul~y	Yield of poultry	10,000 tons	59.24	61.58	0.100	357.1
	yield_d	Yield of dairy	10,000 tons	98.76	159.9	0	945.7
Control variable— disease	disease	Epidemic year or not	DV	0.167	0.373	0	1

Note: 1. To prevent misunderstanding, the following specifications are made for ethnic dummy variables: nouns mentioned in the descriptions refer to ethnic minorities within the territory of China and do not represent those in other countries. 2. DV stands for dummy variable. 3. In official Chinese consumption statistics, beef and mutton are combined and counted together.

presents specific variable description information.

Rural residents’ per capita pork consumption is the dependent variable. As for the control variables, income level is a crucial factor influencing consumption behavior. Income not only increases the overall meat consumption but also leads to changes in the meat consumption structure [33,34]. In this study, per capita disposable income was chosen to represent rural residents’ income level. Additionally, some studies have considered the previous period’s consumption as an important explanatory variable for the current consumption [35,36,37]. To account for rural residents’ tendency to purchase food items, the study extends the one-period lagged consumption combination to represent their consumption habits.

The epidemic dummy variable is based on the Bric Agricultural Database’s standards, with years of severe outbreaks coded as 1 and the rest as 0. The coastal dummy variable represents whether a province is located near the coast or not (1 for coastal provinces and 0 otherwise), providing insights into geographical differences among provinces. The ethnic dummy variable includes the fifteen ethnic minorities with the highest population rankings in China (some, like the Hui and Uighurs, are grouped into the Muslim ethnic dummy variable). Based on the National Census Yearbook’s population data and relevant government information, this study set the ethnic dummy variable to 1 for provinces with more than three hundred thousand people belonging to that ethnic group, implying that the ethnic population has reached a significant scale in that province, which may influence its consumption structure [38]. Thus, the variable’s value does not represent the ethnic group’s national proportion but rather the proportion of provinces with that ethnic group. The inclusion of numerous ethnic dummy variables aims to depict the impact of ethnic characteristics on residents’ consumption from various perspectives, while minimizing potential misinterpretations due to chance. The Muslim ethnic dummy variable considers provinces with a significant population of Muslims. This study paid special attention to the religious beliefs of the Muslim ethnic group, believing that it significantly influences the meat consumption structure in those provinces. The interpretation of the mean value is the same as for the ethnic dummy variable.

3.2. Data Source and Description

The data for this article mainly come from official statistics such as the “China Rural Statistical Yearbook”, “China Population Statistical Yearbook”, “China Livestock and Veterinary Yearbook”, as well as data from the National Development and Reform Commission and the National Bureau of Statistics. The time span covers the years from 2000 to 2020. The data were cross-referencing with official endorsed data websites such as the General Administration of Customs website and China Business Network to ensure data reliability.

In empirical research, missing observations may affect the results of the model, leading to debates about whether the results are biased. Due to various reasons, such as time constraints and statistical subject issues, some data may have missing observations. The following explains the situation of missing observations in the dataset before and after processing:

(1) The epidemic depth index, established by the Brueckner Agricultural Database to measure the scope, severity, and spread rate of animal epidemics, effectively reflects the animal epidemic situation. Some studies also employ this variable in empirical research [39,40,41,42], confirming its reliability. However, since the earliest statistics for this variable only date back to 2009, this study did not have observations from 2000 to 2008 when incorporating the national epidemic depth dummy variable into the model. Although this may affect some models’ degrees of freedom and fail to encompass significant losses from diseases like classical swine fever and foot-and-mouth disease, we believe that it will not have a significant impact on the research strategy. This is because the variable coverage years include African swine fever (ASF), allowing for the consideration of the impact of epidemic years on pork consumption, rather than denying the objective existence of epidemic situations.

(2) This study treated the previous period’s consumption of various food items as an explanatory variable set, resulting in the loss of 31 observations for each province over one year. However, this treatment, which sacrifices some degrees of freedom, conforms to consumers’ psychological regularities. The food consumption combination in the previous period can represent, to some extent, the consumption habits of the region’s residents. While individual consumption combinations may undergo sudden changes, expanding the analysis to the entire region renders any potential errors caused by significant fluctuations negligible.

(3) Due to data collection reasons, all meat prices were missing and the Tibet Autonomous Region’s observations could not influence the overall estimation results in models involving meat price variables. Nevertheless, considering that Tibet’s rural population accounts for only 4‰ of the national total and that the region’s high-altitude terrain restricts market and personnel mobility, we deemed the impact on the overall empirical results to be extremely limited.

4. Research Methods and Estimation Strategies

4.1. Pooled OLS Model

4.1.1. Theoretical Analysis of the Pooled OLS Model

Without considering individual effects, both time-variant and time-invariant variables can be treated equally as factors influencing the dependent variable. In this case, a Pooled OLS (Pooled Ordinary Least Squares) model can be used. However, in this study, there is no reason to assume that the same intercept and slope can capture all provinces and other groups classified by time-invariant variables (e.g., coastal and non-coastal provinces). This approach often overlooks the nature of panel data, leading to biased estimation results.

The model is set as follows:

$$Y_{it}=\alpha +X_{it}^{\prime }\beta +Z_i^{\prime }\gamma +{\epsilon }_{it}$$

where $\alpha$ represents the intercept, which is the same due to the absence of considering individual effects. $X_{it}^{\prime }$ and $Z_i^{\prime }$ are matrices of time-variant and time-invariant variables, respectively, with subscripts distinguishing between the two types of variables. ${\epsilon }_{it}$ is the random disturbance term that is not correlated with any of the considered variables, assuming no endogeneity issues.

4.1.2. Estimation Strategy for the Pooled OLS Model

This study will report the results of the Pooled OLS model along with robust standard errors for comparison and supplementation to the Fixed Effects Filter (FEF) model. In the basic panel regression, the explanatory variables include per capita disposable income, a set of meat price variables, a set of consumption habit variables, coastal dummy variable, and epidemic dummy variable for empirical analysis. Model 1 employs robust standard errors in the regression and corresponds to Pooled OLS_1. Model 2 clusters provinces and uses cluster-robust standard errors, corresponding to Pooled OLS_2. Model 3 uses the Least Squares Dummy Variable (LSDV) method with dummy variables for provinces and years. Model 3 excludes the epidemic and coastal dummy variables because the former can be captured by year dummy variables, and the latter can be captured by province dummy variables. Moreover, excluding the epidemic dummy variable using the LSDV method can effectively expand the sample size, considering significant diseases like classical swine fever and foot-and-mouth disease that severely affected China’s pig industry in the early 21st century.

4.2. Hausman–Taylor IV Model

4.2.1. Theoretical Analysis of the Hausman–Taylor IV Model

In the presence of correlation between some explanatory variables and the random disturbance term, the Hausman–Taylor IV model can simultaneously estimate time-variant and time-invariant coefficients. Based on the binary classification of whether variables are time-variant, the model introduces the exogeneity assumption as a criterion for classifying explanatory variables [43].

The model is set as follows:

$$Y_{it}=X_{1,it}^{\prime }{\beta }_1+X_{2,it}^{\prime }{\beta }_2+Z_{1,i}^{\prime }{\gamma }_1+Z_{2,i}^{\prime }{\gamma }_2+u_i+{\epsilon }_{it}$$

In this model, $X_{1,it}^{\prime }$ and $Z_{1,i}^{\prime }$ are matrices of time-variant and time-invariant exogenous variables, respectively, while $X_{2,it}^{\prime }$ and $Z_{2,i}^{\prime }$ are matrices of time-variant and time-invariant endogenous variables.

Assuming no correlation between any explanatory variables and ${\epsilon }_{it}$:

$$E\left(X_{1,it}^{\prime }\ast {\epsilon }_{it}\right)=0; E\left(X_{2,it}^{\prime }\ast {\epsilon }_{it}\right)=0; E\left(Z_{1,i}^{\prime }\ast {\epsilon }_{it}\right)=0; E\left(Z_{2,i}^{\prime }\ast {\epsilon }_{it}\right)=0$$

For exogenous variables:

$$cov\left(u_i,X_{1,it}^{\prime }\right)=0; cov\left(u_i,Z_{1,i}^{\prime }\right)=0$$

For endogenous variables:

$$cov\left(u_i,X_{2,it}^{\prime }\right)\ne 0; cov\left(u_i,Z_{2,i}^{\prime }\right)\ne 0$$

According to the definition of instrumental variables and the law of expected iteration, exogenous variables can be their own instrumental variables. For time-variant endogenous variables $\left(X_{2,it}^{\prime }\right)$, $(X_{2,it}-{\overline{X}}_{2,it})$ can be used as an instrumental variable, and the cube of dispersion is used as an instrumental variable in the study of Lewbel (1997) [44]. For time-invariant endogenous variables ($Z_{2,i}^{\prime })$, the mean of the exogenous variable ${(\overline{X}}_{1,it})$ can be used as the instrumental variable, but the conditions that need to be met are:

$$\mathit{dim}\left(X_{1,it}\right)\ge \mathit{dim}\left(Z_{2,i}\right)$$

The model uses instruments for endogenous variables and relies on instrumental variables to obtain consistent estimates for both time-variant and time-invariant coefficients. However, the choice and identification of instrumental variables, particularly for time-invariant endogenous variables, can be challenging in practice. The model’s effectiveness depends on the availability of appropriate instrumental variables and the exogeneity assumption’s validity.

4.2.2. Estimation Strategy for the Hausman–Taylor IV Model

The analysis based on the Hausman–Taylor model incorporated three sets of control variables, among which, the common elements were per capita disposable income, a pandemic dummy variable, and a set of meat price variables. For Model 1, following the approach of Lu Yanping et al. (2020) [35], a set of consumption habit variables was introduced, resulting in HT-IV1. Models 2 and 3, respectively, incorporated the variables of slaughter volume and production volume, correlating to HT-IV2 and HT-IV3. For meat products, the slaughter volume generally has a high correlation with production volume. However, the food production variable set in Model 3 encompasses dairy products. Given the availability of data, it cannot be ruled out that rural residents might obtain animal protein from dairy products. Hence, it is essential to differentiate them into two distinct equations.

4.3. Fixed-Effect Filter Model

4.3.1. Theoretical Analysis of the FEF Model

The fixed-effect filter model discards the assumption of categorization of endogenous and exogenous variables, ensuring superior small sample properties and avoiding the issue of low estimation efficiency, which could arise when using instrumental variables obtained by transforming the variables themselves. The traditional fixed-effect model can estimate unbiased statistics of time-variant variables, but it cannot estimate the effects of time-invariant variables. Individual fixed effects can be decomposed into several time-invariant variables, and fixed effects are controlled in conventional models. The FEF model allows coefficients to be decomposed and estimated from the set of fixed effects, effectively revealing characteristic information, and intuitively displaying their degree of influence on the dependent variable.

The first step is to decompose and separate the individual fixed effects through the fixed effect model. Firstly, a standard fixed effect model is established as follows:

$$Y_{it}=\alpha +\beta X_{it}^{\prime }+{\epsilon }_{it}$$

Time-variant variables are eliminated within the model, yielding the estimated coefficient $\widehat{\beta }$ of the time-variant variable matrix and residuals ${\widehat{u}}_{it}$.

The second step involves dividing into time-invariant variable dimensions. In this case, the division is based on provinces (Provin). A time average for ${\widehat{u}}_{it}$ is calculated, resulting in the residual time mean ${\overline{\widehat{u}}}_{it}$. By regressing the time-invariant variables onto ${\overline{\widehat{u}}}_{it}$, we obtain the estimated coefficient $\widehat{\gamma }$ and intercept $\widehat{\alpha }$ for the time-invariant variables.

$$\widehat{\gamma }={\left[\sum _{i=1}^N\left(Z_i-\overline{Z}\right){\left(Z_i-\overline{Z}\right)}^{\prime }\right]}^{-1}\cdot \sum _{i=1}^N\left(Z_i-\overline{Z}\right)\left(\overline{{\widehat{u}}_i}-\overline{\widehat{u}}\right)$$

It should be noted that:

$$\overline{{\widehat{u}}_i}=T^{-1}\sum _{i=1}^T{\widehat{u}}_{it}; \overline{\widehat{u}}=N^{-1}\sum _{i=1}^N\overline{{\widehat{u}}_i}$$

The FEF model discards the assumption of variable exogeneity, permitting both time-variant and time-invariant variables to be effectively included and estimated in the model. This will aid in progressing the mathematical verification of the existence of time-invariant factors in pork consumption in our country, enriching the consumer-based explanation mechanisms. In the existing literature, the research is based on fixed regions, thereby eliminating the need to consider the implications of fixed effects, which constitutes an objective reality to circumvent this issue [33,36,45]. The FEF model enables the simultaneous accurate estimation of time-variant and time-invariant variables, capturing market variation and consumer preferences, thereby expanding the number of variables in empirical consumption research, while considering both operability and data obtainability. Although similar results can be provided in the HT-IV model, due to the issues with instrument variable selection and estimation efficiency, this model lacks universality and cannot provide unbiased information for decision makers in various specific problems.

4.3.2. Estimation Strategy the Fixed-Effect Filter (FEF) Model

In the FEF model, two equations are set up with a difference in the choice of control variables in step one. Model 1 includes a set of meat price variables and a set of consumption habit variables as control variables in constructing the standard fixed-effect model. Model 2 replaces the consumption habit variables with a set of food production variables. The reason for this change is that consumption habits and production represent two aspects that are closest to end consumers and directly influence rural residents’ pork consumption decisions. While the output quantity is highly related to the quantity of pork produced, it overlooks the positive impact of advancements in farming techniques on the average weight of carcasses. Moreover, the food production data scientifically expands the simple intake of meat to the intake of nutrients (such as protein), fully considering the existence of an upper limit for the periodic intake of nutrients by humans. Therefore, it is necessary to consider the impact of multiple foods on single-category meat consumption. Both equations share the same time-invariant variables as step two, which includes coastal dummy variables and dummy variables for various ethnic groups. As the analysis controls for years, epidemic dummy variables are not included in the model. However, it is essential to note that the fixed effects for each province represent a combination, and the decomposition work requires ensuring orthogonality among these variables; otherwise, it could lead to endogeneity issues, affecting the estimation of time-invariant variables. Nevertheless, this does not affect the estimation of time-variant variables in step one. The subsequent text will further discuss this endogeneity issue. Each model reports the results of both steps separately, denoted as _step1 and _step2. The Model 1 estimation results correspond to FEF1, and Model 2 corresponds to FEF2.

5. Empirical Results and Discussion

5.1. Model Estimation Results

The analysis of the model results used the Mixed Effect Model, the Hausman–Taylor IV Model, and the fixed-effect filter model. The first two models will demonstrate, from certain perspectives, the estimation advantages of the FEF model and the necessity of incorporating individual effects into rural residents’ pork consumption behavior explanation.

5.1.1. Pooled OLS Model Estimation Results

The estimation of the Mixed Effect Model, using robust standard errors and clustering provinces, shows that most of the control variables remain significant. This suggests that the control variables used in the study can adequately capture rural residents’ meat consumption behavior in China. Furthermore, after using the LSDV method to capture individual and time effects, the model’s R² significantly improves. This, to some extent, confirms that when studying consumption behavior, focusing solely on the impact of time-variant variables on consumption may overlook other dimensions of consumers’ behavior.

Table 2. Regression results of pooled OLS.

Variable	(1)	(2)	(3)
	Pooled OLS_1	Pooled OLS_2	LSDV
newdpi	0.458 ***	0.458 **	0.242 ***
	(0.075)	(0.178)	(0.038)
price_pig	−0.116 ***	−0.116 ***	−0.243 ***
	(0.034)	(0.037)	(0.046)
price_b	−0.241 ***	−0.241 **	−0.069 ***
	(0.040)	(0.091)	(0.019)
price_m	0.115 ***	0.115	0.054 ***
	(0.042)	(0.105)	(0.015)
price_poultry	0.832 ***	0.832 ***	−0.025
	(0.070)	(0.163)	(0.035)
lagcons_g	−0.005	−0.005	−0.010 ***
	(0.009)	(0.022)	(0.003)
lagcons_v	0.109 ***	0.109 ***	0.018 ***
	(0.008)	(0.022)	(0.006)
lagcons_bm	−0.566 ***	−0.566 **	0.112
	(0.077)	(0.233)	(0.071)
lagcons_poultry	−0.409 ***	−0.409	0.237 ***
	(0.093)	(0.251)	(0.087)
lagcons_e	−0.585 ***	−0.585 *	0.221 ***
	(0.119)	(0.326)	(0.070)
lagcons_s	0.289 ***	0.289	0.133 **
	(0.064)	(0.171)	(0.055)
sea	−1.872 ***	−1.872	-
	(0.583)	(1.599)
disease	−2.564 ***	−2.564 ***	-
	(0.640)	(0.570)
Provinces FE	-	-	Yes
Years FE	-	-	Yes
Constant	1.120	1.120	12.718 ***
	(2.092)	(4.493)	(1.160)
Observations	360	360	596
R2	0.575	0.575	0.975

Note: *** p < 0.01, ** p < 0.05, * p < 0.1, Applied robust standard errors.

shows the estimates for the pooled OLS model.

5.1.2. Hausman–Taylor Model Estimation Results

The Hausman–Taylor model demonstrates the possibility of estimating fixed effects, and the equation that includes the consumption habit variables shows excellent estimation efficiency and effects. This is because among the time-variant variables, per capita disposable income, meat prices, and consumption habits are the most relevant factors for rural residents’ consumption decisions. Consumers in the market cannot obtain all the production-side information for a product, such as output quantity and slaughter quantity. This aligns with the theoretical logic and empirical evidence presented in related literature. However, due to the instrument matrix being derived from the variables themselves, the weak instrument problem severely impacts the model’s estimation efficiency. This is evident in the non-significant estimates for the output quantity and slaughter quantity, which contradicts existing empirical results. This discrepancy might stem from the model’s strict assumptions and the inability to effectively identify exogenous variables. The HT-IV model is not a universally effective method for estimating individual effects.

Table 3. Regression results of Hausman–Taylor model.

Variable	(1)	(2)	(3)
	HT-IV1	HT-IV2	HT-IV3
newdpi	0.076 **	0.084 ***	0.093 ***
	(0.031)	(0.031)	(0.031)
disease	−0.268 *	−0.291	−0.277
	(0.160)	(0.180)	(0.179)
price_b	−0.077 ***	−0.052 ***	−0.053 ***
	(0.013)	(0.013)	(0.012)
price_m	0.045 ***	0.036 ***	0.036 ***
	(0.010)	(0.011)	(0.010)
price_pig	0.006	−0.004	−0.004
	(0.008)	(0.010)	(0.010)
price_poultry	−0.042	−0.015	−0.025
	(0.031)	(0.032)	(0.031)
lagcons_g	−0.013 ***
	(0.003)
lagcons_v	0.014 ***
	(0.005)
lagcons_bm	0.095
	(0.092)
lagcons_poultry	0.047
	(0.089)
lagcons_e	0.171 **
	(0.067)
lagcons_s	0.255 ***
	(0.066)
output_pig		0.000
		(0.000)
output_cattle		−0.001
		(0.001)
output_mutton		−0.001
		(0.000)
output_poultry		0.000
		(0.000)
yield_pork			0.000
			(0.002)
yield_b			−0.009
			(0.010)
yield_m			−0.026
			(0.024)
yield_poultry			0.006 *
			(0.004)
yield_d			0.003 **
			(0.001)
provinces FE	Yes	Yes	Yes
Constant	14.078 ***	22.874 ***	18.782 ***
	(0.940)	(1.953)	(1.090)
Observations	360	360	360

Note: *** p < 0.01, ** p < 0.05, * p < 0.1; Applied robust standard errors.

shows the estimates for the Hausman–Taylor model.

5.1.3. FEF Model Estimation Results

The FEF model provides unbiased estimates for geographical and ethnic characteristics using coastal and ethnic dummy variables. Additionally, in the two equations of the FEF model, there were many missing values for poultry meat production in the output information. However, since poultry meat serves as a substitute for pork, removing this variable would significantly impact the estimation results. Therefore, the final choice was to use the results from FEF1 as the core source of explanation.

After decomposing the fixed effects for each province, the empirical results show that most ethnic dummy variables (including the Muslim ethnic dummy variable) and the coastal dummy variable have a highly significant impact on pork consumption. This provides important information for macro-regional planning in the pork industry and the prioritization of investments in rural farms from the consumer side. The estimation results for the coastal dummy variable in the FEF model are almost significantly positive. If a province has coastal characteristics, it is reasonable to assume that the coastal ports in the eastern region of China create a distinct trade advantage, leading to vibrant markets and higher per capita income levels. Consequently, under otherwise identical conditions, rural residents in these coastal regions tend to have higher pork consumption levels. It is also worth noting that coastal residents may substitute seafood consumption for pork. By examining basic data information, it is evident that coastal provinces generally have higher seafood consumption compared to inland provinces. However, such negative effects on pork consumption are fully offset by the positive effects, as hypothesis H1 is supported, indicating that regional factors influence pork consumption among rural residents, with coastal regions having higher consumption levels.

Similarly, if a province has any of the ethnic characteristics (this study selected the fifteen largest ethnic groups in terms of population), it will also significantly influence local consumption patterns. In the FEF model estimation, the two equations yield slightly different estimation results in terms of numerical values and significance levels, but these differences do not affect the signs of the estimated coefficients. The impact of ethnic characteristics from ethnic groups such as Mongolian, Tibetan, Li, Zhuang, Yi, Tujia, Yao, and Muslim ethnicities is significant at the 1% level, and the direction of impact aligns with ethnic customs and religious habits. In particular, in the results of FEF1 and FEF2, it is observed that the Bai, Hani, and Dai ethnic dummy variables do not provide estimated coefficients. This is due to the fact that, according to the division criteria of ethnic dummy variables, the populations of these three ethnic groups are only substantial in Yunnan Province, which is essentially equivalent to setting a dummy variable for Yunnan Province, causing multicollinearity, and leading to the removal of these variables by the software. However, this does not affect the basic conclusion of the correlation between ethnic characteristics and local dietary differences. H2 is supported, indicating that the impact of ethnic cultural attributes on pork consumption among rural residents varies.

As for the control variables, variables such as per capita disposable income, prices, consumption habits, and output quantities significantly affect pork consumption in different models. Based on the results estimated by FEF1_step1, per capita disposable income is significant at the 1% level. An increase of one thousand yuan in per capita disposable income leads to a rise of 0.504 kg in pork consumption among rural residents. This is consistent with logical expectations: the increase in per capita disposable income raises the purchasing willingness of rural residents. Before reaching the level of food satiation caused by physiological inhibition [46], the overall pork consumption among rural residents will continue to increase with the rise in per capita disposable income. In terms of nutrient acquisition, other meat sources such as poultry, beef, and mutton, which also serve as sources of protein and fat, show a significant negative correlation with pork consumption. Additionally, the impact of the pandemic on consumption is also significant, as the inclusion of pandemic dummy variables in the model results in a reduction in pork consumption among rural residents. To eliminate the potential bias from overlooking certain variables, an Oster test (2019) [47] was conducted after estimating the FEF1 model to test the potential omitted variables and their impact on regression results.

Table 4. Regression results of fixed-effect filter model.

Variable	(1)	(2)	(3)	(4)
	FEF1_Step 1	FEF1_Step 2	FEF2_Step 1	FEF2_Step 2
sea		2.680 ***		3.547 ***
		(0.201)		(0.223)
menggu		12.114 ***		16.351 ***
		(0.201)		(0.223)
chaoxian		0.157		8.595 ***
		(0.226)		(0.284)
zang		4.664 ***		4.816 ***
		(0.196)		(0.416)
miao		1.766 **		1.165
		(0.756)		(0.911)
li		−1.910 ***		−0.937 ***
		(0.129)		(0.130)
man		−0.064		−3.309 ***
		(0.129)		(0.130)
zhuang		8.133 ***		10.472 ***
		(0.781)		(1.002)
yi		2.455 ***		1.264 ***
		(0.261)		(0.460)
tujia		4.200 ***		10.763 ***
		(0.639)		(0.977)
buyi		−9.638 ***		0.307
		(0.647)		(0.686)
dong		6.918 ***		−1.371
		(0.998)		(1.060)
yao		−11.595 ***		−12.135 ***
		(0.658)		(0.781)
o.bai		-		-
o.hani		-		-
o.dai		-		-
muslim		−0.982 ***		−1.211 ***
		(0.261)		(0.418)
newdpi	0.504 ***		0.384 **
	(0.163)		(0.159)
price_pig	−0.110		−0.375
	(0.229)		(0.366)
price_b	−0.325 ***		−0.207
	(0.115)		(0.145)
price_m	0.030		0.172
	(0.132)		(0.146)
price_poultry	0.995 ***		0.698 ***
	(0.171)		(0.197)
lagcons_g	−0.024 *
	(0.014)
lagcons_v	0.109 ***
	(0.020)
lagcons_bm	−0.561 **
	(0.254)
lagcons_poultry	−0.489
	(0.309)
lagcons_e	−0.836 ***
	(0.245)
lagcons_s	0.274
	(0.204)
yield_pork			0.037 ***
			(0.009)
yield_b			−0.132 **
			(0.056)
yield_m			0.016
			(0.097)
yield_poultry			−0.030 *
			(0.016)
yield_d			0.003
			(0.007)
Constant	9.522	−2.572 ***	7.355	−3.994 ***
	(7.640)	(0.172)	(10.393)	(0.196)
Observations	596	596	521	521
R2	0.579	0.797	0.441	0.819

Note: *** p < 0.01, ** p < 0.05, * p < 0.1 for applied clustered standard errors.

shows the estimates for the fixed-effect filter model.

To exclude the potential bias caused by overlooking certain variables, an Oster test was conducted after estimating the FEF1 model to test potential omitted variables and their impact on regression results. For partially randomly selected variables, the delta values for different explanatory variables were greater than 1, indicating that the model is robust.

Table 5. Oster bound test results.

Variables	(1)	(2)	(3)	(4)
	Newdpi	Price_Poultry	Lagcons_Bm	Lagcons_e
$\widehat{\beta }$	0.418 **	0.892 ***	−0.561 **	−0.836 ***
	(0.173)	(0.188)	(0.254)	(0.245)
$\mathrm{Oster} \mathrm{bound} (\widehat{\beta },{\beta }^{ *})$	(0.418, 2.292)	(0.892, 2.470)	(−0.575, −0.455)	(−1.395, −0.778)
Rmax value	0.764	0.764	0.764	0.764
$\mathrm{Delta} \mathrm{value} (\delta$)	10.302	1.110	2.072	2.427
Not included 0	Yes	Yes	Yes	Yes

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

shows the results of the Oster boundary test.

5.2. Discussion of Empirical Results

Several studies have identified time-invariant factors that could engender disparities in dietary tendencies across diverse regions through comparative data analysis and empirical inquiries. For instance, it is posited that aspects of folklore and culture play a determinative role, to an extent, in shaping dietary diversity [48]. This premise is accentuated by data obtained from the National Bureau of Statistics, which illuminate regional variance in pork consumption among rural dwellers. For illustration, between the years of 2017 and 2020, pork consumption averaged 34.86 kg amongst rural inhabitants in Sichuan Province, contrasting starkly with the 1.39 kg in the Xinjiang Uighur Autonomous Region. Relevant investigations carried out by researchers affirm that a predilection for pork consumption is predominant in provinces like Guangxi, Guizhou, and Sichuan [49]. Our rigorous econometric findings offer compelling empirical support for the statistically significant phenomena identified by these studies through field research. As such, this research constitutes a pivotal augmentation to the corpus of scholarship on pork consumption patterns among rural residents.

The estimations may prompt some debates stemming from the inherent flaws in the theoretical approach and model setup. The following discussion addresses these issues and attempts to provide explanations: Firstly, regarding the decomposition of the fixed effects, this study only performed decomposition based on coastal and ethnic characteristics. We acknowledge that these do not represent the entirety of fixed-effect combinations but were selected based on data availability and their relevance to the research question, as evident in the theoretical framework diagram. Fixed effects that were not decomposed were captured by the intercept term, and the estimation results show that the intercept term is significant at the 1% level, suggesting the possible existence of time-invariant variables that influence pork consumption among rural residents. The existing literature and experience indicate that regional and ethnic characteristics can influence consumption through different resource endowments and preference impacts. However, it is essential to note that the decomposition of fixed effects should not be based on a more-is-better mindset, as too many time-invariant variables could lead to severe endogeneity problems, which is one of the important conditions mentioned in step two of the FEF model to ensure orthogonality among time-invariant variables.

Secondly, there is an issue with the conceptual definition of time-variant and time-invariant variable sets (as reflected in control and explanatory variables). According to data from the “National Population Census Statistical Yearbook”, minority groups may undergo population mobility through non-agricultural employment transfers [50]. The sixth national population census data show that 16.5386 million minority individuals migrated to cities. Additionally, the population growth rate of minority groups in China is higher than that of the Han population. According to the results of the seventh population census announcement, the population of various minority groups increased by 10.26% compared to the sixth census in 2010, while the Han population increased by 4.93%. This suggests that ethnic characteristics might undergo changes over the long-term historical dimension. This could lead some scholars to argue that the total number or proportion of the minority population could also be chosen as variables, transforming them into time-variant variables. However, this change would not fundamentally overturn the basic conclusion that ethnic characteristics influence consumption preferences.

Thirdly, there is a possibility of ambiguous representation with dummy variables. Since dummy variables represent various fixed effects and these fixed effects matrices are linear combinations of provincial dummy variables, a single ethnic dummy variable might simultaneously represent other fixed effects. For example, the Korean ethnic dummy variable (although significant to different extents in different equations) was set to 1 for Heilongjiang, Jilin, and Liaoning provinces and 0 for other provinces. This could lead to the interpretation that it only represents the concept of the three northeastern provinces, as historically and experientially, the three provinces might be considered as a single entity due to similar geographical and cultural conditions. In this case, the Korean ethnic characteristic may not be the main contributor. Two solutions are proposed: first, summarizing and quantifying fixed features that are subject to debate (such as geographical and cultural conditions) on a nationwide scale. However, this approach might lead to endogeneity problems arising from correlations between time-invariant variables, as discussed earlier. The second solution is to exclude such debated variables from the equations, as the FEF model is designed to provide unbiased estimates for time-invariant variables under the assumption conditions. The purpose of the article is to provide insights on the consumer side for industry planning, and the model results can offer indications of whether certain provinces require increased investments. If variable names and meanings are set scientifically, they can represent all possible fixed features to derive estimation results.

Lastly, an integral limitation to recognize in our study lies in its specific demographic focus which significantly concentrates on the rural population in China. Given the dynamic socio-cultural milieu of China’s rural regions, these areas offer a compelling ground for investigating dietary patterns, particularly the consumption of pork, which is our study’s primary focus. However, the principle inherent in this single demographic focus is that our findings may have less immediate relevance for populations outside this rural context. Primarily, the urban populace exhibits substantially different socio-economic characteristics, dietary habits, and lifestyle customs. The urban–rural distinction inflicts considerable disparity in income levels, access to commodities, dietary preferences, and many health indicators significantly influenced by diet. Given these disparities, the consumption patterns and the impact of the examined cultural and geographical variables on urban residents may differ from our study findings. By extension, countries other than China with varying socio-economic profiles, culinary cultures, and structures of agricultural produce might observe different consumption behaviors. The caution to be taken while extrapolating our findings to foreign contexts stems from the uniqueness of China’s rural context and the essential socio-cultural nuances informing the diet patterns therein. While our study provides valuable insights into the dietary practices within China’s rural regions, we suggest conducting similar studies in urban habitats or other nations and comparing these findings with our own. This extended approach will not only validate our research across different contexts but also advance a more globally representative understanding of dietary behaviors and their influencers.

6. Conclusions and Recommendations

To explain the pork consumption patterns among rural residents in China, it is essential to focus on interpreting various characteristics and clarify the impact or correlation of fixed effects on consumption, based on understanding the role of common time-variant variables in meat consumption. In clear terms, the econometric results support both Hypothesis 1 and Hypothesis 2. That is, regional and cultural factors exert a heterogeneous impact on the consumption of pork among rural residents. The sustainable development of the swine sector relies on a rational and scientific industrial layout. Building on the empirical research from the consumer side, the following policy recommendations and future research directions are proposed.

6.1. Policy Recommendations

We should continue to reinforce the strategic position of pork in the overall meat consumption in China, while also understanding the dynamic evolution of pork in the dietary structure of residents. This requires targeted industrial development and adopting a holistic approach to food production by optimizing agricultural facilities. The government should actively coordinate and rationally plan the national land resources to ensure the dynamic stability in the quantity and structure of land used for swine farming. This will integrate the swine sector into a food production system that harmonizes crop cultivation, animal husbandry, and fishery development.

We should emphasize the impact of regional characteristics on consumption and develop regions with comprehensive comparative advantages in consumption. This entails appropriately increasing investment in the swine sector in northern, eastern coastal, and southwestern provinces, which are the main markets for pork. Unlike weather-dependent grain crops, the swine production sector can cater to market demands and rationally consider the relationship between market demand and risk. Therefore, in regions with consumption potential, there should be active expansion of the scale of farming to reduce cost losses in the circulation process. This will promote rapid development of swine farming in northern, eastern, and southwestern regions.

We should establish a robust mechanism for industry collaboration between the swine sector and other production sectors and expedite the geographical clustering of the swine industry chain. Funding should be prioritized for optimizing the links in the industry chain and supportive policies should be implemented for smooth and effective market integration of swine farming, slaughtering, and processing sectors. These efforts will boost farmers’ income through swine and related industries, ensure market supply, stabilize market prices, and increase investments in animal disease prevention and control. This will prevent severe market disruptions caused by unexpected “black swan” events and thereby enhance the confidence of both breeders and consumers.

6.2. Future Research Directions

We successfully validated the significant influence of regional and cultural (religious) characteristics on pork consumption among rural residents in China, showing heterogeneity at the provincial level. Simultaneously, this study illustrates the clear advantages of the fixed-effect filter model over the traditional fixed-effect model which cannot estimate individual effects. Our future work primarily focuses on several aspects.

Firstly, we aim to explore and continuously validate the standards for dummy variable processing concerning individual effects.

Secondly, we plan to extend the scope of individual effects in the model because, for different provinces, individual effects may include not only regional and cultural characteristics but possibly also features such as weather and long-term policy impacts. More specifically, regional characteristics may not only include coastal traits but may also involve average altitude, mountain proportions, and so forth. Cultural characteristics may not be constrained to religious features but may include folklore, lifestyle, and more. While the features mentioned above are not necessarily factors impacting rural pork consumption, they could provide future research directions, offering a fresh perspective for consumption theory. We will emphasize the need to incorporate a broader array of supply-side elements into our research agenda.

It is worth noting that we relied on sizeable datasets to generate and substantiate our results, employing a wide variety of sources to gather a comprehensive collection of data. However, this vast data collection task is not without inherent problems. In particular, potential inaccuracies might stem from inconsistent data collection methodologies across different sources, inherent biases in the way data are reported, potential missing values, and the difficulty of ensuring uniformity in data reporting standards.

Further, given our focus on the rural populace of China, data collection could be additionally challenging due to less uniform record keeping, potentially impacting the accuracy of our data. Additionally, while we have been as thorough as possible in collecting data pertinent to our study, there remains a possibility of unwittingly missing potential variables of relevance. The FEF model is not exempt from its inherent weakness. The issues pertaining to endogeneity persist as a notable concern. Considering these limitations, future researchers examining these interests are encouraged to further perfect data collection methodologies, strive for larger and more inclusive databases, and extend the time period of analysis for a more nuanced perspective.