Heterogeneous Porter Effect or Crowded-Out Effect: Nonlinear Impact of Environmental Regulation on County-Level Green Total Factor Productivity of Pigs in the Yangtze River Basin of China

Abstract

Green total factor productivity (GTFP) is critical to both the economic and ecological objectives of pig breeding. This research utilizes the SBM-ML model to calculate the GTFP of pig breeding in 381 counties within the Yangtze River Basin from 2014 to 2021. Then the GTFP is further decomposed into technical efficiency (MLEC) and technical progress (MLTC) to conduct in-depth exploration. The regression results reveal that: (1) Environmental regulation (ER) has significant double-threshold effects on GTFP, MLEC, and MLTC. (2) MLTC is the main force of GTFP growth, and stronger ER does not always lead to better GTFP growth. (3) GTFP is boosted by mechanization enhancement and industrial agglomeration limitation. (4) Counties in non-provincial capital cities and those closer to the river exhibit greater ER threshold effects. (5) Both pig price and transportation efficiency play a moderating role. (6) Further analysis demonstrates that ER simultaneously reduces pig production capacity and carbon emissions, as well as improves the water quality. And the reduction of ER, although beneficial for capacity, has a significant negative impact on GTFP. Finally, this study concludes with policy recommendations to boost the new quality productivity in the pig industry.

1. Introduction

Green development has become a shared goal worldwide in recent years [1], and its essence is centered on the coordinated development of ecology and economy [2]. Numerous countries worldwide, including the United Kingdom with its Low Carbon Transition Plan, the United States implementing the Clean Energy and Security Act, Japan formulating the 2050 Carbon Neutral Green Growth Strategy, and Germany promoting the Renewable Energy Sources Act, have all actively pursued strategies to advance green development. China is not an outlier in this regard, and ever since the Fifth Plenary Session of the 18th CPC Central Committee introduced the idea of green development, the Chinese government has increasingly prioritized the eco-friendly transformation of the economy.

Pig breeding has long been a fundamental sector of agricultural business in China [3,4,5] and plays a significant role in the rural economy [6]. China holds the title of being the largest consumer and producer of pork globally. As reported by the China National Bureau of Statistics, pork production in China in 2021 reached a staggering 52.96 million tons. This accounts for 59.59% of the total output of livestock and poultry meat in the country and 44.09% of the global proportion, securing the position of China as the top pork producer in the world. Furthermore, China was projected to obtain over 50% of the worldwide pork supply in 2021 [7].

Globally, pollution issues arising from pig breeding have become increasingly severe. Notably, the marine environment of South Korea faces significant threats from inland fecal pollution sources [8]. In Poland, there exists the risk of improper storage and excessive application of pig slurry, which may lead to water, soil, and plant contamination by heavy metals [9]. China, too, confronts environmental pollution challenges stemming from pig farming, which cannot be overlooked. China has experienced a surge in air and water pollution in recent years [10]. As the largest producer of livestock globally [11], China has been recognized for the significant contribution of its livestock industry to air pollution [12] and water pollution [13]. The Second National Pollution Source Census Bulletin reported that in 2017, livestock and poultry farming contributed to water pollution by emitting 10,053,000 tons of chemical oxygen demand, 110,900 tons of ammonia nitrogen, 596,300 tons of total nitrogen, and 119,700 tons of total phosphorus. These emissions accounted for 46.67%, 11.51%, 19.61%, and 37.95% of the national total, respectively. Pig breeding, being a significant component of the livestock sector, has resulted in a range of environmental issues with its rapid growth [4,6,14,15]. For example, the presence of pollutants such as pig manure from pig breeding can lead to a decline in air quality and contamination of water quality [7].

To tackle the pollution issues caused by pig breeding, which cannot be resolved by market forces alone, the Chinese government has implemented a set of environmental regulations (ER) [10]. In 2013, the China State Council issued Regulations on Pollution Prevention and Control of Livestock and Poultry Scale Breeding to tackle the problem of pollution caused by livestock and poultry breeding. These regulations are regarded as the most rigorous water pollution prevention and control measures for pig breeding in China up to now [15]. In 2015 and 2016, the Ministry of Agriculture released two documents titled Guidance Opinions on Promoting the Adjustment and Optimization of Pig Farming Layout in Southern Water Network Areas and Pig Industry Layout Adjustment Program in Southern Water Network Areas. These documents signaled a new phase of ER in the pig industry [3].

The Yangtze River is the primary river that traverses the west–central and east–central regions of China. It serves as the core point for the economic progress of China, contributing to around 50% of the national total economic output [16]. Hence, safeguarding the ecological integrity of the Yangtze River Basin is of immeasurable significance in upholding national environmental security and achieving sustainable development for the Chinese country [17]. Nevertheless, due to the swift economic expansion of China and escalating water contamination, the condition of the Yangtze River Basin ecosystem has persistently worsened, with pig breeding being a notable contributor to water pollution [18]. For this reason, the Chinese government formally passed The Law of the People’s Republic of China on the Protection of the Yangtze River on 26 December 2020, as the first specialized law for the ecological protection of the basin in China, highlighting the commitment of the state to safeguarding the environment of the area.

Encouraging green development of pig breeding has emerged as a significant approach to achieving high-quality growth in the business [19]. Furthermore, there is indisputable evidence about the efficacy of ER in fostering green development [20]. ER is considered to be the non-market mechanism for addressing the issue of pollution externalities [21], and it can successfully manage the pollution issues associated with pig breeding. The drawback lies in the fact that ER impacts the growth of the pig sector to some degree [22], especially in production capacity [15]. In 2019, the China Ministry of Agriculture and Rural Development promulgated the Opinions on Stabilizing Pig Production and Ensuring Market Supply. In 2024, this was followed by the issuance of the Implementation Plan for Pig Production Capacity Control. These documents emphasized the significance of maintaining a stable pig production and ensuring an adequate supply. To incorporate both the objective of green development and the preservation of production capacity in the analysis framework, how can the pig breeding sector effectively embrace this opportunity without causing excessive harm? Which indicators should be utilized to quantify it?

Green total factor productivity (GTFP) offers a potent solution to the aforementioned difficulties. To foster long-term economic growth, a significant reliance is placed on the enhancement of factor productivity [23]. Furthermore, GTFP, a pivotal component of new quality productive forces, embodies innovative allocation of production factors and profound industry transformation, being essential for modern advanced productivity. However, with the implementation of several environmental rules, the traditional total factor productivity is no longer sufficient to estimate the productivity of pig breeding in the Yangtze River Basin. GTFP incorporates resource consumption, pollution emission, and other resource and environmental limitations, aligning with the concept of green development in the modern era [21,24]. It serves as a comprehensive measure of economic and ecological environmental performance [20,25]. A higher level of GTFP means that pig breeding is capable of achieving sustainable development by efficiently utilizing resources and minimizing environmental degradation while also promoting economic growth [19].

Many scholars have conducted extensive research on the impact of ER in different fields, including green total factor productivity [26], green technological innovation [27], outward foreign direct investment [24,25,28], industry size [29], corporate environmental responsibility [30], economic growth and CO₂ emissions [31], corporate ecological and innovation performance [32], and ecological behavior [33]. Meanwhile, GTFP is also increasingly favored by researchers. Its influencing factors include trade openness [21], foreign direct investment [25], green credit [34], clean technology innovation [35], green finance [36], and urban land use [37]. Some scholars use a linear model to study the effect of ER on GTFP, another group believes that the two have a nonlinear relationship and uses the inclusion of a quadratic term to fit it, but most scholars believe that the inclusion of a quadratic term does not reflect the nonlinear relationship well, and so they use a panel threshold model to construct it [1,21,24]. However, there is a gap in the literature on the specific impacts of ER on GTFP in pig breeding. There is a limited amount of research that has focused on investigating ER in the livestock industry [10], and even fewer studies have examined the impact of ER on GTFP in pig breeding. Amid the scarce research available, the GTFP of pig farming was measured in China from 2004 to 2017 and the dual roles of ER were examined including pollution reduction and efficiency enhancement [38], making significant contributions. However, their use of provincial GDP to measure the intensity of ER is less precise than a number of environmental regulatory policies used in this study, and their empirical analysis of ER and GTFP is not thorough, only touching the surface of the impact. By the fixed-window Malmquist–Luenberger (FWML) index, another study measured the GTFP of pig breeding, and, subsequently, the economic benefits and environmental pollution costs under the influence of environmental regulations were analyzed [6]. Compared to [38], they did not delve deeply into the relationship between ER and GTFP in pig breeding or the underlying mechanisms of this impact but predicted pig price using long short-term memory (LSTM) neural networks. Therefore, this paper tries to fill critical gaps in the existing literature by systematically and comprehensively analyzing the effects of ER on GTFP in pig breeding.

Based on panel data from 2014 to 2021, this paper examines the impact of ER on the GTFP of pig breeding across 381 counties in the Yangtze River Basin. This paper devotes a significant portion to the application of the threshold models, with further research also encompassing the double difference model and linear model. When the ER increases, it may prompt pig farming entities to develop new technologies or improve production efficiency, resulting in a significant increase in their green total factor productivity. However, once ER exceeds a certain value, input continues to increase, but output efficiency cannot continue to rise efficiently. Therefore, their green total factor productivity may slightly increase or even decrease. Based on this consideration, this paper studies the nonlinear impact of pig ER on their green total factor productivity by using threshold models. To explore the impact of different degrees of ER on green total factor productivity, mainly through efficiency improvement or technological improvement, this article decomposes green total factor productivity into technical efficiency (MLEC) and technological progress (MLTC). The findings reveal there exist double threshold effects of ER on GTFP, MLEC, and MLTC. On the one hand, as ER increases, it first enhances the promotion of GTFP and MLTC. However, this effect later decreases. On the other hand, the impact on MLEC changes from being insignificant to solid promotion, and it eventually becomes a significant inhibitor. In investigating the mechanism, this paper applies two-panel threshold models to explore the effects of the level of pig mechanization and the level of pig industry agglomeration on ER, respectively, and it is proved that both of them have significant single-threshold effects on ER. So, ER enhances the growth of GTFP through two pathways: by increasing the level of pig mechanization and reducing the level of pig industry agglomeration. Furthermore, the heterogeneity tests and moderating effects in this paper also consistently use panel threshold models. The heterogeneity study reveals that various county characteristics, such as whether it spans the Yangtze River main stem, whether it is located in a provincial capital city, and the distance between the government and the Yangtze River, would result in distinct threshold effects. The investigation of the moderating role reveals that parameters such as pig price and transportation operation efficiency both have moderating effects on the impact of ER on GTFP, MLEC, and MLTC. To address the potential issue of endogeneity between ER and GTFP, this paper employs the environmental regulatory intensity of other counties within the province (ERAOP) as an instrumental variable. Based on the combined consideration of the application of the panel threshold model and the principles of the instrumental variable approach, the endogeneity exploration in this paper is carefully divided into two stages. In the first stage, this paper applies a linear model to deeply analyze the effect of the instrumental variable ERAOP on ER, aiming to accurately estimate and obtain the fitted value of ER after incorporating the environmental regulation factors of other counties in the province. In the second stage, this study turns to the panel threshold model to further explore the specific effects on GTFP centered on the fitted values of ER derived in the previous stage, so as to effectively address and circumvent the potential endogeneity problem of the threshold model.

As can be seen from the above, ER has heterogeneous promoting effects on the green total factor productivity of pigs, but whether it increases pig production capacity while reducing its carbon emissions needs further exploration. Additional evaluations indicate that ER has a suppressive impact on both pig production capacity and carbon emissions. However, the advantages gained from reducing carbon emissions surpass the drawbacks of declining production capacity, hence improving GTFP. Moreover, ER efficiently decreases the concentration of ammonia nitrogen in the Yangtze River, hence improving water quality. In order to further reverse validate the potential conflicts in the impact of ER on GTFP and pig production capacity, this study uses a double difference model to examine the impact of the reduction of livestock prohibition zone policy intensity in certain counties in 2019 on GTFP, MLEC, MLTC, and production capacity. The findings indicate that the enhancement of ER helps to increase GTFP of pig breeding but may have some negative impacts on production capacity, whereas when ER decreases, GTFP decreases despite the significant increase in production capacity.

This research makes significant progress in the following four aspects, enriching the findings of previous studies: (1) County data: This paper uses data from 381 specific counties within the Yangtze River Basin to conduct a more precise and detailed study of how ER in pig breeding affects GTFP. This approach not only overcomes the limitations of using provincial data [19,25,34,39], which makes it challenging to accurately understand the complex relationship between ER and GTFP due to the broad scope of the study but also significantly reduces the statistical error of the data and provides a more detailed and targeted reference for the formulation of environmental regulatory policies. (2) Pig breeding: This paper significantly contributes to the study of how ER affects GTFP in pig breeding. In the existing literature studying the impact of ER and GTFP, most of them focus on the industry [19,34,40,41,42], the manufacturing sector in the industry [20,41], overall agriculture [24,43], and so on. In addition, historical research involving GTFP in pig breeding has only measured and analyzed it [4,5,19,44], whereas there are few studies on the association between ER and GTFP pig breeding. Only one article on dairy farming [1] is more relevant to this paper. However, this article only employed provincial data and the urban–rural income gap as a threshold variable to study the impact without studying its impact mechanism and endogeneity. With the aim of addressing the shortage in pig breeding in the area, this paper uses the SBM-ML model to assess the GTFP of pig breeding in the Yangtze River Basin and further decomposes it into two dimensions: MLEC and MLTC. More importantly, the ER discussed in this paper is not the traditional overall regulation measured by economic indicators, but rather, the ER for pig breeding, which was synthesized using the entropy weight method after searching for policies from the national level to the county level. Through the in-depth analysis of these critical variables, this paper not only reveals the influence mechanism of ER on GTFP of pig breeding but also integrates the dual objectives of economic growth and environmental protection to provide reliable suggestions for the green development of pig breeding. (3) Panel threshold model: This research is the first to integrate the panel threshold model into the analytical framework of ER and GTFP in pig breeding. Previous studies on ER and GTFP in other industries have typically used the threshold model as a supplementary or essential extension of the primary model [39], or very simply to explore whether there is a threshold effect [40], and they seldom explore in depth the nonlinear relationship between ER and GTFP. Different from them, this paper comprehensively and thoroughly applies the threshold model, validating the reliability of the research results from the dimensions of robustness and endogeneity. Furthermore, threshold models are employed in mediation, moderation, and heterogeneity analyses, ensuring consistency and coherence throughout the analytical framework, thereby providing empirical evidence for the application of the threshold model. (4) Policy evaluation: A large number of studies suggest that since 2017, the policy of prohibiting livestock breeding has become increasingly strict, seriously harming the development of the pig industry, especially in 2019. Due to the sharp decline in China’s pig production capacity in 2019, the government decreased the intensity of environmental regulations, but no relevant research has explored the effectiveness of this policy. This paper tries to fill the gap.

The following sections of the paper are structured as follows: Section 2 conducts a theoretical analysis by drawing upon existing literature and formulating the research hypotheses. Section 3 provides an overview of the data sources and research methods employed in this research. Section 4 includes the threshold effect test, robustness examination, impact mechanism, heterogeneity test, moderating effect analysis, endogeneity discussion, and further research. Section 5 summarizes the conclusions of the study and puts forward corresponding policy recommendations.

2. Theoretical Analysis and Research Hypothesis

2.1. Direct Effects of ER on GTFP in Pig Breeding

Economic actions frequently involve externalities, with environmental damage being a particularly notable example. Environmental contamination, a typical example of economic externalities, has detrimental effects on both the environment and society. Market failure occurs when the market system is unable to operate efficiently due to the externalities of environmental damage. Relying solely on the invisible hand of the market to promote spontaneous green production is challenging [20]; governments need to play a regulatory role at this time. In order to achieve this objective, governments typically establish a set of environmental rules with the goal of addressing market inefficiencies and fostering sustainable growth.

Unlike a single indicator, GTFP incorporates multiple significant input and output components into a cohesive analytical framework, providing a more accurate representation of the actual manufacturing process [45]. Compared to traditional total factor productivity, GTFP incorporates environmental considerations, making it a more rational and adaptable measure for the present circumstances [44].

The relationship between ER and GTFP has long been a hot topic of research [20,46]. However, the existing literature lacks an unambiguous conclusion regarding the impact of ER on GTFP [1,26]. There are two or even more distinct conclusions in both theoretical and empirical studies.

Two prominent perspectives in theoretical study are the Compliance Cost Theory and the Porter Hypothesis. First of all, the Compliance Cost Theory is a neoclassical economic perspective that asserts that ER will result in higher expenses for businesses in terms of pollution control. This places an additional burden on enterprises, ultimately having a detrimental effect [39]. Specifically for pig breeding, ER increases the production cost of pig breeders, which leads to the lack of sufficient funds for technological innovation and inhibits the growth of GTFP. However, the Porter Hypothesis suggests that a well-designed ER will generate innovation compensation effects and improve environmental quality without hindering economic performance [26,40]. Applying this theory to pig breeding, it means that ER can incentivize pig breeding subjects to engage in more innovative activities, increase productivity, and make up for the previously mentioned adverse impact of making costs increase, which is manifested in the enhancement of green GTFP. These two theories represent two vastly distinct directions. Still, some scholars have pointed out that both theories have their own rationality, and the reason for the contradiction is that there are both cost increase and innovation compensation caused by ER. Still, the boundary conditions have not been fully explored [20].

Within the realm of empirical studies, there exist three primary perspectives about the influence of ER on GTFP. The first perspective argues that ER positively impacts the improvement of GTFP. Specifically, it was discovered that both command-and-control and market-based ER are correlated with higher GTFP, thus providing support for Porter Hypothesis [38]. Similarly, in a study conducted in [1], it was demonstrated that ER had a substantial positive impact on the growth of GTFP in the dairy farming sector of China. Furthermore, in [47], it was explicitly asserted that ER has a direct positive effect on the increase of GTFP in China, hence providing further evidence in favor of the argument of Porter. Another study showed that the implementation of centralized environmental protection inspections, which is a specialized form of ER, has a substantial positive impact on the growth of GTFP [21]. In contrast, the second perspective argues that ER has a mitigating impact on GTFP. For instance, someone discovered that command-and-control ER has a substantial inhibitory impact on the overall GTFP, which suggests that there is a negative influence on the growth of GTFP to some degree [26]. The last one argues that the positive and negative effects of ER are uncertain or not significant at all. For example, one research contended that the impact of ER on GTFP is contingent upon the overall balance between the positive and negative effects of such regulations [43]. It was also found that ER does not have a significant impact on GTFP in the Yangtze River Economic Belt [47].

As a result, the impact of ER on GTFP lacks an explicit agreement in both theory and empirical studies. It is precisely because of the co-existence of the driving role of ER and the hindering role of ER [20] that the effects of different intensities of ER on GTFP may differ, and there may well be a structural breaking point. Hence, this article proposes the existence of a threshold characteristic in the impact of ER on GTFP in pig breeding. Furthermore, some research has demonstrated the existence of a threshold effect in the relationship between ER and GTFP in other fields [1,20,40]. According to this information, this paper puts forward Hypothesis 1a.

Hypothesis 1a:

The impact of ER on GTFP in pig breeding has a threshold effect.

GTFP can be decomposed into MLEC and MLTC, and there are inconsistent findings on these two decompositions. Some scholars believe that the growth of GTFP comes from MLEC [1,4,5], while other scholars have argued that it is mainly fueled by MLTC [23,24,41]. Another study even suggested that ER has a significant threshold effect on MLTC [18]. However, this research contended that the contrasting results could be attributed to variations in industry diversity. Thus, since MLEC and MLTC are components of GTFP, and since an increase in MLEC and MLTC results in a continuous increase in GTFP [6], it is possible that both factors contribute to the threshold effect of GTFP. Based on the inconsistent findings regarding these factors, this paper proposes Hypothesis 1b and Hypothesis 1c.

Hypothesis 1b:

ER has a threshold effect on MLEC in pig breeding.

Hypothesis 1c:

ER has a threshold effect on MLTC in pig breeding.

2.2. Indirect Mechanisms of ER on Green GTFP in Pig Breeding

Prior research on ER and GTFP has identified several indirect mechanisms, such as promoting technical innovation [48], fostering market concentration, and implementing green market entry obstacles in businesses with high emissions [45].

However, for pig breeding, the mechanistic variables may differ from those of traditional industries. Manure emissions, including CH₄ and N₂O, are significant pollutants during pig breeding [12]. Considering the adverse effects of these pollutants on the environment, the Chinese government has acknowledged the environmental issues associated with animal husbandry and devised a set of policies in response. These policies mandate pig farmers to adopt environmental protection measures, including the implementation of manure runoff control and the separation of liquid and solid components in manure [11]. The introduction of these regulations will unquestionably incentivize farmers to improve the level of mechanization in manure treatment in order to comply with the more rigorous environmental protection criteria. Mechanization has a crucial role in promoting the green transformation of the agricultural sector, particularly in pig breeding [43]. For instance, the greater use of machinery in pig breeding facilitates the adoption of knowledge-intensive approaches to managing manure and wastewater [14]. This, in turn, improves the treatment of pollutants during the pig breeding process and reduces environmental pollution, ultimately achieving environmentally friendly development in the pig industry. Furthermore, enhancing mechanization yields numerous advantages in pig breeding. On the one hand, mechanized farming improves the efficiency of cleaning and transporting sick and dead pigs while enhancing the biosecurity level of farms. On the other hand, it reduces labor costs, improves farming efficiency, and optimally allocates and efficiently uses resources. These extensive advantages collectively enhance the enhancement of MLEC and MLTC, ultimately resulting in an increase in the GTFP in pig breeding. Based on the aforementioned analysis, this paper presents the following hypotheses:

Hypothesis 2a:

ER can enhance the mechanization level of pig breeding, hence promoting the improvement of GTFP.

Industrial agglomeration in pig breeding refers to the phenomenon of a large number of pig breeders being concentrated in a specific geographic area [49]. Due to the rising intensity of ER, several pig farmers may opt to relocate to regions with less harsh regulations to evade the strict environmental standards. This, in turn, may lead to a decline in the level of industrial agglomeration. When the level of industrial agglomeration decreases, the environmental pollution problem caused by the original over-concentration is expected to be alleviated because the area with reduced breeding density will have a larger ecological capacity and less pollution load. This environmental enhancement will incentivize pig farmers to enhance the efficiency of their resource usage and minimize waste, thereby leading to an improvement in GTFP. This is because, in an environment with less environmental pressure, farming entities are more inclined to embrace sophisticated farming technology and management practices in order to achieve environmentally friendly and efficient production. Based on the above analysis, this paper puts forward the following hypotheses:

Hypothesis 2b:

ER can enhance the enhancement of GTFP by decreasing the level of industrial agglomeration in pig breeding.

3. Materials and Methods

3.1. Sample Selection and Data Sources

Considering the reliability and completeness of data acquisition, this paper adopts the balanced panel data of 381 counties within the Yangtze River Basin (from 57 municipalities in Sichuan, Jiangsu, Jiangxi, Yunnan, Hunan, Hubei, Zhejiang, and Anhui Provinces) for the period of 2014 to 2021 as the sample. This paper also incorporates the data from 2013, as the measurement of GTFP necessitates the use of data from the preceding year. In this paper, some indicators are logarithmically processed to improve the effectiveness of data analysis and interpretation of results, and the missing values of relevant indicators in the yearbook are filled in by the moving average method.

The primary source of data for this article is the statistics yearbooks of each prefecture-level city, spanning from 2013 to 2022. The data on ER are sourced from the China Legal Knowledge Resource Database (https://lawnew.cnki.net/kns/brief/result.aspx?dbPrefix=CLKD, accessed on 24 March 2024), while other data are obtained from the National Compendium of Agricultural Cost and Benefit Information, provincial statistical yearbooks, and China Environmental Monitoring Center (https://www.cnemc.cn/jcbg/qgdbsszyb/index_4.shtml, accessed on 8 April 2024).

3.2. Description of Input–Output Indicators

The establishment of appropriate input and output indicators serves as the foundation for the precise and accurate assessment of efficiency [19]. Hence, this paper utilizes pertinent research on GTFP measurement and selects four input indications, one desired output indicator, and one non-desired output indicator.

3.2.1. Input Factors

(1) Labor input. Due to the absence of direct data regarding the number of individuals working in pig breeding, this study uses an indirect approach to estimate the employment figures in pig breeding within counties. So, to quantify the number of workers engaged in pig breeding in each county, this study multiplies the agricultural workforce in that county with the ratio of pork output value to the total agricultural output value. The agricultural workforce in each county is sourced from the municipal statistical yearbook. For those not disclosed, they are calculated using the formula of multiplying the municipal agricultural employment by the ratio of the animal husbandry output value of the districts to the municipal animal husbandry output value.

(2) Capital input. This article uses the perpetual inventory method to calculate the capital stock. Due to the limited availability of statistical data on county-level investment in fixed assets for pigs, the calculation of the annual livestock capital increase is conducted. County livestock capital increase = Municipal investment in agricultural fixed assets × (county livestock output for the year/county agricultural output for the year). The livestock capital stock is then calculated from Equation (1). Then, the capital investment in pig breeding is determined: County Pig Capital Inputs = County livestock capital stock × (County pig output/County livestock output).

$${\mathrm{K}}_{\mathrm{t}}={\mathrm{K}}_{\mathrm{t}-1}\left(1-\delta \right)+{\mathrm{I}}_{\mathrm{t}},$$

(1)

where ${\mathrm{K}}_{\mathrm{t}}$, ${\mathrm{K}}_{\mathrm{t}-1}$ represent the capital stock of the county in the current year and the previous year, respectively, ${\mathrm{I}}_{\mathrm{t}}$ is the amount of new fixed asset investment in the county in the current year, and $\delta$ is the depreciation rate. With reference to [50,51], this paper sets $\delta$ to be 9.6%. And the municipal investment in fixed assets in agriculture, the output value of animal husbandry in the county, and the total output value of agriculture in the county are from county and city level statistical yearbooks.

(3) Energy input. On the one hand, excessive use of energy affects input factors. On the other hand, excessive energy emissions will indirectly affect the undesired output [52]. Compared with traditional pig breeding, energy consumption of modernized pig breeding occupies an essential position in its production process. Therefore, when assessing the GTFP of pig breeding, the important factor of energy inputs must be fully considered. Referring to [50], this paper considers the energy used in the pig farming process as an input indicator of GTFP. However, the statistical yearbook of each prefecture-level city does not provide the specific energy input data for pig breeding, so this paper adopts the following formula to calculate: Energy consumption of pigs = total agricultural energy consumption in the province × (county pig output/province agricultural output). Moreover, the total provincial agricultural energy consumption is taken from the comprehensive energy balance section of the provincial statistical yearbooks and is expressed in tons of standard coal.

(4) Feed input. In this paper, the average feed input per pig in the province where the county is located is multiplied by the total number of pigs stocked and farrowed in the county to measure the level of feed input of pigs in the county. The provincial average feed input level is obtained by summing up the concentrate feed cost, green and rough feed cost, and feed processing cost in the National Compendium of Cost and Benefit of Agricultural Products and averaging the free-range, small-sized, medium-sized, and large-sized feeds.

3.2.2. Output Factors

(1) Desired output. Expected output metrics for pig GTFP measurement in the current literature include live weight per fattening pig [4], the yield of hog production [5], and net hog production [19]. In this paper, the total output value of pork in the county is taken as the desired output, which is explicitly obtained by multiplying the provincial pig price by the pork production in county.

(2) Unexpected output. Carbon dioxide, a key factor in climate change, must be considered in green development [45]. Besides emissions from enteric fermentation and manure management, the indirect emissions and pollution from resource consumption in livestock breeding should also be addressed [11]. Therefore, the CO₂ emissions in this paper include two main parts: CO₂ equivalent emissions from methane and nitrous oxide produced by enteric fermentation and manure management of pigs, and CO₂ emissions from the total energy consumption in pig breeding production and consumption.

Firstly, the carbon emissions generated by the pigs themselves are calculated. In the traditional mode, CH₄ and N₂O are the primary fecal carbon emissions [12]. Therefore, referring to [19], carbon emissions from pigs in this paper can be calculated as Equation (2).

$$\begin{array}{ll}{\mathrm{E}}_{\mathrm{T}\mathrm{o}\mathrm{t}\mathrm{a}\mathrm{l}}& ={\mathrm{E}}_{\mathrm{C}{\mathrm{H}}_4}+{\mathrm{E}}_{{\mathrm{N}}_{2}O}\\ & ={\mathrm{E}}_{\mathrm{C}{\mathrm{H}}_4\left(\mathrm{Abdominal}\mathrm{fermentation}\right)}+{\mathrm{E}}_{\mathrm{C}{\mathrm{H}}_4\left(\mathrm{Excrement}\right)}+{\mathrm{E}}_{{\mathrm{N}}_{2}O}\\ & =25\times \mathrm{A}\mathrm{P}\mathrm{P}\times \alpha +25\times \mathrm{A}\mathrm{P}\mathrm{P}\times \beta +298\times \mathrm{A}\mathrm{P}\mathrm{P}\times \epsilon ,\end{array}$$

(2)

where $\mathrm{A}\mathrm{P}\mathrm{P}$ denotes the annual feeding capacity of pigs, which is calculated by $\mathrm{A}\mathrm{A}\mathrm{P}=\mathrm{D}\mathrm{a}\mathrm{y}\mathrm{s}\_\mathrm{a}\mathrm{l}\mathrm{i}\mathrm{v}\mathrm{e}\times \mathrm{N}\mathrm{A}\mathrm{P}\mathrm{A}/365$. $\mathrm{D}\mathrm{a}\mathrm{y}\mathrm{s}\_\mathrm{a}\mathrm{l}\mathrm{i}\mathrm{v}\mathrm{e}$ denotes the production cycle of pigs, which is taken as 200 days [11,53], and $\mathrm{N}\mathrm{A}\mathrm{P}\mathrm{A}$ is the annual farrowing capacity of pigs. $\alpha$, $\beta$, and $\epsilon$ represent the methane emission coefficients of gastrointestinal fermentation, methane emission coefficients of excreta of pigs, and nitrous oxide emission coefficients of pigs, which are taken as 1, 3.5, and 0.53, respectively, to concerning [11,53].

Next, the following formula is to calculate CO₂ emissions from energy consumption:

$$\mathrm{C}{\mathrm{O}}_2={\displaystyle \sum _{\mathrm{k}=1}^4{\mathrm{E}}_{\mathrm{k}} \times }\mathrm{C}\mathrm{E}{\mathrm{F}}_{\mathrm{k}},$$

(3)

where CO₂ denotes the carbon dioxide emission in tons. E is the four energy sources of coal, coke, gasoline, and diesel, and the data are obtained from the corresponding energy balance sheets of the statistical yearbooks of each province. CEF denotes the carbon dioxide emission factor. The coefficients for coal, coke, gasoline, and diesel are 1.9003, 2.8604, 2.9251, and 3.0959 [54], respectively, in tons C/t.

In this paper, we construct an indicator system as shown in

Table 1. Input–output indicators.

Vector	Indicator	Measure	Unit
Inputs	Labor	Number of people employed in pig breeding.	10,000 persons
	Capital stock	Capital stock for pig breeding.	CNY10,000
	Energy	Pig energy consumption in the county.	10,000 tons of standard coal
	Feed	Average provincial feed input per pig multiplied by the total number of pigs stocked and farrowed in the county.	CNY10,000
Desirable Output	Total pork production	Provincial pig price per head multiplied by county pork production.	CNY10,000
Undesirable Output	carbon dioxide	The total CO2 emissions from the CO2 equivalent of methane and nitrous oxide from pig breeding fermentation and manure management, and CO2 from energy used in county pig breeding production and consumption.	tons

and measure the GTFP of pig breeding in the Yangtze River Basin.

3.3. Description of Variables

3.3.1. Explained Variable: GTFP

GTFP is a key statistic for assessing whether economic development simultaneously achieves growth, energy conservation, and pollution reduction [23]. GTFP is measured using the production frontier approach, which comprises stochastic frontier analysis (SFA) and data envelopment analysis (DEA) [55]. Moreover, the DEA approach is preferred in research because it does not require production function selection [45]. Based on DEA, scholars have proposed a variety of more comprehensive models to measure GTFP through improvement and optimization, such as the EBM model [47,56], the DDF model [5], the SBM model [1,4,21,24,25,34,35,45,53], and so on. The SBM-ML index model addresses the deficiency of the classic total factor productivity measurement method by incorporating environmental restrictions such as resource consumption and pollution emission [21]. Hence, considering the data properties of the samples and the benefits of the SBM-ML model, this study initially employs the SBM model to calculate efficiency values. Subsequently, these efficiency values are utilized to establish the ML index, enabling a comprehensive and precise evaluation of the GTFP.

With reference to [57], the super-efficient SBM model with constant returns to scale with undesired output is constructed in this paper, as shown in Equation (4).

$$\begin{array}{ll}\rho & =\underset{\lambda ,{\mathrm{s}}^{-},{\mathrm{s}}^{+}}{\min }{\displaystyle \frac{1+\frac{1}{\mathrm{m}}{\displaystyle \sum _{\mathrm{i}=1}^{\mathrm{m}}\frac{{\mathrm{s}}_{\mathrm{i}}^{-}}{{\mathrm{x}}_{\mathrm{i}\mathrm{k}}}}}{1-\frac{1}{{\mathrm{s}}_1+{\mathrm{s}}_2}\left({\displaystyle \sum _{\mathrm{r}=1}^{{\mathrm{s}}_1}\frac{{\mathrm{s}}_{\mathrm{r}}^{+}}{{\mathrm{y}}_{\mathrm{r}\mathrm{k}}^{\mathrm{g}}}+}{\displaystyle \sum _{\mathrm{w}=1}^{{\mathrm{s}}_2}\frac{{\mathrm{s}}_{\mathrm{w}}^{-}}{{\mathrm{y}}_{\mathrm{w}\mathrm{k}}^{\mathrm{b}}}}\right)}}\\ \mathrm{s}.\mathrm{t}.& {\mathrm{x}}_{\mathrm{i}\mathrm{k}}\ge {\displaystyle \sum _{\mathrm{j}=1,\mathrm{j}\ne \mathrm{k}}^{\mathrm{n}}{\lambda }_{\mathrm{j}}}{\mathrm{x}}_{\mathrm{i}\mathrm{j}}-{\mathrm{s}}_{\mathrm{i}}^{-};\\ & {\mathrm{y}}_{\mathrm{r}\mathrm{k}}^{\mathrm{g}}\le {\displaystyle \sum _{\mathrm{j}=\mathrm{1},\mathrm{j}\ne \mathrm{k}}^{\mathrm{n}}{\lambda }_{\mathrm{j}}}{\mathrm{y}}_{\mathrm{r}\mathrm{j}}^{\mathrm{g}}+{\mathrm{s}}_{\mathrm{r}}^{+};\\ & {\mathrm{y}}_{\mathrm{w}\mathrm{k}}^{\mathrm{b}}\ge {\displaystyle \sum _{\mathrm{j}=1,\mathrm{j}\ne \mathrm{k}}^{\mathrm{n}}{\lambda }_{\mathrm{j}}}{\mathrm{y}}_{\mathrm{w}\mathrm{j}}^{\mathrm{b}}-{\mathrm{s}}_{\mathrm{w}}^{-};\\ & {\lambda }_{\mathrm{j}}\ge 0(\forall \mathrm{j}),{\mathrm{s}}_{\mathrm{i}}^{-}\ge 0(\forall \mathrm{i}),{\mathrm{s}}_{\mathrm{r}}^{+}\ge 0(\forall \mathrm{r}),{\mathrm{s}}_{\mathrm{w}}^{-}\ge 0(\forall \mathrm{w})\\ & \mathrm{i}=1,2,\cdots ,\mathrm{m}; \mathrm{r}=1,2,\cdots ,{\mathrm{s}}_1; \mathrm{w}=1,2,\cdots ,{\mathrm{s}}_2;\mathrm{j}=1,2,\cdots ,\mathrm{n}(\mathrm{j}\ne \mathrm{k}),\end{array}$$

(4)

where $\lambda$ is a vector of weights, $\mathrm{x}$, ${\mathrm{y}}^{\mathrm{g}}$, and ${\mathrm{y}}^{\mathrm{b}}$ represent inputs, desirable outputs, and undesirable outputs, respectively. ${\mathrm{s}}_{\mathrm{i}}^{-}$, ${\mathrm{s}}_{\mathrm{r}}^{+}$ and ${\mathrm{s}}_{\mathrm{w}}^{-}$ are the slack variables. i, r, and ware numbers ranging from 1 tom, 1 to ${\mathrm{s}}_1$, and 1 to ${\mathrm{s}}_2$, respectively. In this context, ${\mathrm{x}}_{\mathrm{i}\mathrm{k}}$ denotes the $\mathrm{i}\text{-}\mathrm{t}\mathrm{h}$ input of the $\mathrm{k}\text{-}\mathrm{t}\mathrm{h}$ $\mathrm{D}\mathrm{M}\mathrm{U}$; $\rho$ is the efficiency value of the decision-making unit. Condition ${\lambda }_{\mathrm{j}}\ge 0(\forall \mathrm{j})$ denotes the assumption of constant returns to scale.

Nevertheless, the SBM model for assessing efficiency is a static analysis [4], which cannot directly capture the dynamic changes in productivity and can only reflect the relative relationship between counties and production boundaries. Therefore, for panel data spanning multiple time points, the Malmquist Index should be used to analyze the longitudinal changes in productivity [4,25]. Therefore, this paper references the practice of [58] and uses the efficiency value calculated by SBM to construct the Malmquist–Luenberger index, which includes non-expected output as follows.

$$\begin{array}{ll}\mathrm{M}{\mathrm{L}}_{\mathrm{t}}^{\mathrm{t}+1}& ={\left[{\displaystyle \frac{1+{\overrightarrow{\mathrm{D}}}_0^{\mathrm{t}}\left({\mathrm{x}}^{\mathrm{t}},{\mathrm{y}}^{\mathrm{t}},{\mathrm{b}}^{\mathrm{t}};{\mathrm{y}}^{\mathrm{t}},-{\mathrm{b}}^{\mathrm{t}}\right)}{1+{\overrightarrow{\mathrm{D}}}_0^{\mathrm{t}}\left({\mathrm{x}}^{\mathrm{t}+1},{\mathrm{y}}^{\mathrm{t}+1},{\mathrm{b}}^{\mathrm{t}+1};{\mathrm{y}}^{\mathrm{t}+1},-{\mathrm{b}}^{\mathrm{t}+1}\right)}}\times {\displaystyle \frac{1+{\overrightarrow{\mathrm{D}}}_0^{\mathrm{t}+1}\left({\mathrm{x}}^{\mathrm{t}},{\mathrm{y}}^{\mathrm{t}},{\mathrm{b}}^{\mathrm{t}};{\mathrm{y}}^{\mathrm{t}},-{\mathrm{b}}^{\mathrm{t}}\right)}{1+{\overrightarrow{\mathrm{D}}}_0^{\mathrm{t}+1}\left({\mathrm{x}}^{\mathrm{t}+1},{\mathrm{y}}^{\mathrm{t}+1},{\mathrm{b}}^{\mathrm{t}+1};{\mathrm{y}}^{\mathrm{t}+1},-{\mathrm{b}}^{\mathrm{t}+1}\right)}}\right]}^{{\displaystyle \frac{1}{2}}}\\ & ={\left[{\displaystyle \frac{1+{\overrightarrow{\mathrm{D}}}_0^{\mathrm{t}+1}\left({\mathrm{x}}^{\mathrm{t}},{\mathrm{y}}^{\mathrm{t}},{\mathrm{b}}^{\mathrm{t}};{\mathrm{y}}^{\mathrm{t}},-{\mathrm{b}}^{\mathrm{t}}\right)}{1+{\overrightarrow{\mathrm{D}}}_0^{\mathrm{t}}\left({\mathrm{x}}^{\mathrm{t}},{\mathrm{y}}^{\mathrm{t}},{\mathrm{b}}^{\mathrm{t}};{\mathrm{y}}^{\mathrm{t}},-{\mathrm{b}}^{\mathrm{t}}\right)}}\times {\displaystyle \frac{1+{\overrightarrow{\mathrm{D}}}_0^{\mathrm{t}+1}\left({\mathrm{x}}^{\mathrm{t}+1},{\mathrm{y}}^{\mathrm{t}+1},{\mathrm{b}}^{\mathrm{t}+1};{\mathrm{y}}^{\mathrm{t}+1},-{\mathrm{b}}^{\mathrm{t}+1}\right)}{1+{\overrightarrow{\mathrm{D}}}_0^{\mathrm{t}}\left({\mathrm{x}}^{\mathrm{t}+1},{\mathrm{y}}^{\mathrm{t}+1},{\mathrm{b}}^{\mathrm{t}+1};{\mathrm{y}}^{\mathrm{t}+1},-{\mathrm{b}}^{\mathrm{t}+1}\right)}}\right]}^{{\displaystyle \frac{1}{2}}}\times \left[{\displaystyle \frac{1+{\overrightarrow{\mathrm{D}}}_0^{\mathrm{t}}\left({\mathrm{x}}^{\mathrm{t}},{\mathrm{y}}^{\mathrm{t}},{\mathrm{b}}^{\mathrm{t}};{\mathrm{y}}^{\mathrm{t}},-{\mathrm{b}}^{\mathrm{t}}\right)}{1+{\overrightarrow{\mathrm{D}}}_0^{\mathrm{t}+1}\left({\mathrm{x}}^{\mathrm{t}+1},{\mathrm{y}}^{\mathrm{t}+1},{\mathrm{b}}^{\mathrm{t}+1};{\mathrm{y}}^{\mathrm{t}+1},-{\mathrm{b}}^{\mathrm{t}+1}\right)}}\right]\\ & =\mathrm{M}\mathrm{L}\mathrm{T}{\mathrm{C}}_{\mathrm{t}}^{\mathrm{t}+1}\times \mathrm{M}\mathrm{L}\mathrm{E}{\mathrm{C}}_{\mathrm{t}}^{\mathrm{t}+1},\end{array}$$

(5)

where ML is the GTFP index, ${\overrightarrow{\mathrm{D}}}_0^{\mathrm{t}}$ and ${\overrightarrow{\mathrm{D}}}_0^{\mathrm{t}+1}$ denote the distance function with reference to the technology in tandt+1, ${\mathrm{x}}^{\mathrm{t}}$, ${\mathrm{y}}^{\mathrm{t}}$, and ${\mathrm{b}}^{\mathrm{t}}$ represent the inputs, desired outputs, and non-desired outputs in periodt, MLTC is the rate of technical progress, MLEC is the technical efficiency, and $\mathrm{M}{\mathrm{L}}_{\mathrm{t}}^{\mathrm{t}+1}$ can be decomposed into $\mathrm{M}\mathrm{L}\mathrm{T}{\mathrm{C}}_{\mathrm{t}}^{\mathrm{t}+1}$ and $\mathrm{M}\mathrm{L}\mathrm{E}{\mathrm{C}}_{\mathrm{t}}^{\mathrm{t}+1}$. The three values greater than 1 signify an increase in GTFP, MLTC, and improvement in MLEC, respectively. It should be noted that the ML index measures the growth rate of GTFP, reflecting the change in the GTFP of the current year compared with the previous year, which can be transformed into a fixed-base index [35,42], but this paper still adopts the index as the GTFP directly. Moreover, because the index needs to use the data of previous year, in this paper, the ML index of 2014–2021 is obtained by using the data of 2013–2021.

3.3.2. Core Explanatory Variable and Threshold Variable: ER

The implementation of ER is widely recognized as the primary policy measure to achieve sustainable and green development in China [47]. This research establishes ER as both the core explanatory variable and the threshold variable, based on earlier theoretical analysis. Currently, there is a lack of a standardized framework for assessing ER in China [26,39,45], particularly in the domain of animal husbandry [59]. Prior research has assessed ER from three primary viewpoints. Firstly, the size of ER is represented by the behavior of the government and enterprises [27,30,31,33,34,39,45], including, in particular, the number of environmental protection policies [7,60]. Furthermore, the level of ER can be deduced based on the presence of pollutants [43,61]. Finally, the strength of environmental control is determined by the objective qualities of the region, such as the level of economic growth [29]. This research utilizes the number of environmental regulatory policies as a measure of the intensity of ER in pig breeding, considering the level of directness and relevance.

Specifically, this paper draws on the research methodology in [62] and searches for laws and regulations related to pollution control of pig breeding year by year through the General Database of China Legal Knowledge Resource (https://lawnew.cnki.net/kns/brief/result.aspx?dbPrefix=CLKD). The research is conducted using keywords such as farming, pollution, pig, animal husbandry, livestock and poultry, environmental protection, energy consumption, energy saving, green, low carbon, and others. The research results are meticulously categorized based on the organization that issues the policies and regulations. The six levels, namely national, ministry, provincial, departmental, prefecture, and county, represent the diversity of environmental control across different regions.

The entropy weight method (EWM) is currently widely used in several domains and provides a comprehensive evaluation of the pig breeding from multiple perspectives [7]. Therefore, in order to calculate the comprehensive ER intensity, following [63], EWM is applied to calculate the yearly ER intensity of each county. This paper first standardizes the indicators to eliminate scale effects, maximizing positive indicators and minimizing negative ones. Next, these standardized values are used to calculate the weight of each indicator. The entropy value and the coefficient of variation are then determined, from which the final weight of each indicator is derived. Then the overall ER score for each county is obtained by summing these weighted scores.

3.3.3. Control Variables

Aside from ER, there exist other variables that can potentially impact the GTFP of pig breeding. This article analyzes control factors in

Table A1. Specifics of control variables.

Control Variables	Symbol	Metrics
Land carrying capacity	$\mathrm{L}\mathrm{n}\_\mathrm{F}\mathrm{a}\mathrm{r}\mathrm{m}\mathrm{l}\mathrm{a}\mathrm{n}\mathrm{d}$	Sown area in county/total sown area in China; unit: hectares/hectares
Feed production capacity	$\mathrm{F}\mathrm{e}\mathrm{e}\mathrm{d}$	County corn production/total county grain production, unit: ton/ton
Educational attainment	$\mathrm{E}\mathrm{d}\mathrm{u}$	Number of persons with secondary education or higher/total number of persons in the county, the number of persons with secondary education or higher includes secondary schools, high schools and universities; unit: person/person
Industrialization	$\mathrm{L}\mathrm{n}\_\mathrm{I}\mathrm{n}\mathrm{d}\mathrm{u}\mathrm{s}\mathrm{t}\mathrm{r}\mathrm{y}$	Gross industrial output value of counties/China’s GDP; unit: CNY10,000/CNY10,000
Wage income	$\mathrm{W}\mathrm{a}\mathrm{g}\mathrm{e}$	Wage line income of farmers in county/disposable income of farmers in county; unit: CNY1/CNY1
Livestock investment level	$\mathrm{L}\mathrm{n}\_\mathrm{I}\mathrm{n}\mathrm{v}\mathrm{e}\mathrm{s}\mathrm{t}\mathrm{m}\mathrm{e}\mathrm{n}\mathrm{t}$	County livestock fixed asset investment/China’s total livestock fixed asset investment; unit: CNY10,000/CNY10,000
Pig slaughtering capacity	$\mathrm{L}\mathrm{n}\_\mathrm{H}\mathrm{o}\mathrm{g}\mathrm{s}\mathrm{a}$	Number of pigs farrowed in counties/total number of pigs farrowed in China; unit: head/head
Pig slaughtering potential	$\mathrm{L}\mathrm{n}\_\mathrm{H}\mathrm{o}\mathrm{g}\mathrm{s}\mathrm{p}$	County pig inventory/China’s total pig inventory; unit: Head/head
Pig price	$\mathrm{L}\mathrm{n}\_\mathrm{P}\mathrm{r}\mathrm{i}\mathrm{c}\mathrm{e}$	Provincial pig price; unit: CNY/kg
Pork consumption capacity	$\mathrm{L}\mathrm{n}\_\mathrm{P}\mathrm{o}\mathrm{r}\mathrm{k}\mathrm{c}\mathrm{c}$	Provincial per capita rural pork consumption; unit: kg/person
Pork consumption potential	$\mathrm{L}\mathrm{n}\_\mathrm{P}\mathrm{o}\mathrm{r}\mathrm{k}\mathrm{c}\mathrm{p}$	Total population of county at the end of the year/total population of China; unit: person/person
Transportation convenience	$\mathrm{C}\mathrm{o}\mathrm{n}\mathrm{v}\mathrm{e}\mathrm{n}\mathrm{i}\mathrm{e}\mathrm{n}\mathrm{c}\mathrm{e}$	Total county road miles/area of county; unit: kilometers/square kilometers
Transportation efficiency	$\mathrm{L}\mathrm{n}\_\mathrm{T}\mathrm{r}\mathrm{a}\mathrm{f}\mathrm{f}\mathrm{i}\mathrm{c}$	County road freight turnover/area of county; unit: kilometers/square kilometers

Note: $\mathrm{L}\mathrm{n}\_\mathrm{F}\mathrm{a}\mathrm{r}\mathrm{m}\mathrm{l}\mathrm{a}\mathrm{n}\mathrm{d}$, $\mathrm{L}\mathrm{n}\_\mathrm{i}\mathrm{n}\mathrm{d}\mathrm{u}\mathrm{s}\mathrm{t}\mathrm{r}\mathrm{y}$, $\mathrm{L}\mathrm{n}\_\mathrm{i}\mathrm{n}\mathrm{v}\mathrm{e}\mathrm{s}\mathrm{t}\mathrm{m}\mathrm{e}\mathrm{n}\mathrm{t}$, $\mathrm{L}\mathrm{n}\_\mathrm{H}\mathrm{o}\mathrm{g}\mathrm{s}\mathrm{a}$, $\mathrm{L}\mathrm{n}\_\mathrm{H}\mathrm{o}\mathrm{g}\mathrm{s}\mathrm{p}$, and $\mathrm{L}\mathrm{n}\_\mathrm{P}\mathrm{o}\mathrm{r}\mathrm{k}\mathrm{c}\mathrm{p}$ represent the negative logarithms for their original values. $\mathrm{L}\mathrm{n}\_\mathrm{P}\mathrm{r}\mathrm{i}\mathrm{c}\mathrm{e}$, $\mathrm{L}\mathrm{n}\_\mathrm{P}\mathrm{o}\mathrm{r}\mathrm{k}\mathrm{c}\mathrm{c}$, and $\mathrm{L}\mathrm{n}\_\mathrm{t}\mathrm{r}\mathrm{a}\mathrm{f}\mathrm{f}\mathrm{i}\mathrm{c}$ denote their corresponding logarithms.

in Appendix A based on study by [64] and related literature, considering subject characteristics and data availability. This study uses the natural logarithm and adds a negative sign to several indicators since the ratios of some counties to the nation are too small to analyze.

3.4. Empirical Models

3.4.1. Panel Threshold Model

According to the previous theoretical analysis, there is a nonlinear relationship between ER and GTFP. The two main methods for identifying nonlinearities in the current literature are joining the square term and applying a panel threshold model. Adding square terms to identify nonlinearities has the disadvantages of requiring the distribution to be symmetric on both sides of the turning point and failing to identify nonlinear effects that occur in the same direction [45]. The panel threshold model is chosen to study the nonlinear relationship for two reasons: first, it can solve the covariance problem and symmetry requirement in the traditional quadratic term measure and scientifically analyze the relationship between the core variables in different intervals [48]. Second, the theoretical research suggests that the effect of ER on GTFP has both facilitating and inhibiting effects, and the panel threshold model is suitable for fitting this kind of coexistence of positive and negative effects [20].

Therefore, this paper takes ER itself as the threshold variable to explore the threshold effect between ER and GTFP. Based on Hansen’s panel data threshold regression model in 1999 [65], with reference to the improvements made by some scholars [20,21,45,48,52], this paper constructs a single-panel threshold model. Depending on the outcome, double- or triple-panel threshold models can be created similarly.

$${\mathrm{G}\mathrm{T}\mathrm{F}\mathrm{P}}_{\mathrm{i}\mathrm{t}}={\alpha }_0 + {\alpha }_1\mathrm{E}{\mathrm{R}}_{\mathrm{it}}\mathrm{T}\left(\mathrm{E}\mathrm{R} \le {\mathsf{\gamma }}_1\right) + {\alpha }_2{\mathrm{E}\mathrm{R}}_{\mathrm{it}}\mathrm{I}\left(\mathrm{E}\mathrm{R} > {\mathsf{\gamma }}_1\right) + \alpha {\mathrm{X}}_{\mathrm{it}} + {\mathsf{\mu }}_{\mathrm{i}} + {\mathsf{\lambda }}_{\mathrm{t}} + {\epsilon }_{\mathrm{it}},$$

(6)

where $\mathrm{I}\left(·\right)$ is the indicative function, ${\alpha }_1$ and ${\alpha }_2$ are the coefficients of the impact of ER on GTFP in different zones, ${\gamma }_1$ is the threshold; i represents different counties, t denotes different years, $\mathrm{G}\mathrm{T}\mathrm{F}{\mathrm{P}}_{\mathrm{i}\mathrm{t}}$ is the GTFP of pigs in year t of county i, $\mathrm{E}{\mathrm{R}}_{\mathrm{it}}$ is the intensity of ER in year t of county i, X is the set of control variables, such as the carrying capacity of arable land, pig price, etc.; ${\mathsf{\mu }}_{\mathrm{i}}$ represents individual fixed effects, ${\lambda }_{\mathrm{t}}$ represents the time-fixed effects, and ${\epsilon }_{\mathrm{it}}$ is the randomized perturbation term.

Similarly, the threshold effects of the decomposition terms are also explored in this paper, and the panel threshold model is not repeated.

3.4.2. Mechanism Testing Model

Based on the theoretical framework presented in Section 2, the impact of ER on GTFP in pig breeding can be characterized in two ways: (1) ER increases the GTFP of pig breeding by encouraging automation; (2) ER reduces the level of pig industry agglomeration and indirectly increases the GTFP of pig breeding.

This research, with reference to [65], designs two empirical models to investigate the mediating impact of mechanization degree and industrial agglomeration level of pig breeding in the link between ER and GTFP, as well as whether there is a threshold effect:

$$\mathrm{L}\mathrm{n}\_\mathrm{M}\mathrm{a}\mathrm{c}\mathrm{h}\mathrm{i}\mathrm{n}{\mathrm{e}}_{\mathrm{i}\mathrm{t}}={\alpha }_0+{\alpha }_1\mathrm{E}{\mathrm{R}}_{\mathrm{it}}\mathrm{I}\left(\mathrm{E}\mathrm{R}\le {\mathsf{\gamma }}_1\right)+{\alpha }_2\mathrm{E}{\mathrm{R}}_{\mathrm{it}}\mathrm{I}\left(\mathrm{E}\mathrm{R}>{\mathsf{\gamma }}_1\right)+\alpha {\mathrm{X}}_{\mathrm{it}}+{\mathsf{\mu }}_{\mathrm{i}}+{\mathsf{\lambda }}_{\mathrm{t}}+{\epsilon }_{\mathrm{it}}$$

(7)

$$\mathrm{A}\mathrm{g}{\mathrm{g}}_{\mathrm{i}\mathrm{t}}={\alpha }_0+{\alpha }_1\mathrm{E}{\mathrm{R}}_{\mathrm{it}}\mathrm{I}\left(\mathrm{E}\mathrm{R}\le {\mathsf{\gamma }}_1\right)+{\alpha }_2\mathrm{E}{\mathrm{R}}_{\mathrm{it}}\mathrm{I}\left(\mathrm{E}\mathrm{R}>{\mathsf{\gamma }}_1\right)+\alpha {\mathrm{X}}_{\mathrm{it}}+{\mathsf{\mu }}_{\mathrm{i}}+{\mathsf{\lambda }}_{\mathrm{t}}+{\epsilon }_{\mathrm{it}},$$

(8)

where $\mathrm{L}\mathrm{n}\_\mathrm{M}\mathrm{a}\mathrm{c}\mathrm{h}\mathrm{i}\mathrm{n}{\mathrm{e}}_{\mathrm{i}\mathrm{t}}$ and $\mathrm{A}\mathrm{g}{\mathrm{g}}_{\mathrm{i}\mathrm{t}}$ represent the mechanization level of pig breeding, the agglomeration level of pig industry is in the $\mathrm{t}\text{-}\mathrm{t}\mathrm{h}$ county in the $\mathrm{i}\text{-}\mathrm{t}\mathrm{h}$ year, separately. The meanings of the remaining indicators are the same as Equation (6).

When measuring the mechanization level of pig breeding, in order to accurately reflect the degree of its application, this paper not only refers to the data on the total power of agricultural machinery in the municipal statistical yearbook but also customizes the calculation of the total mechanical power of pig breeding in the county through the total power of agricultural machinery in the county and the share of the pig industry in the agricultural industry.

In addition, at present, the location entropy index is widely used to measure the level of industrial agglomeration in the region, but when measuring the level of industrial agglomeration of pig breeding, it is considered that the traditional location entropy index ignores the factor of industrial scale. Therefore, this paper refers to the practice in [49] and improves the traditional location entropy method by using the output value index for calculation. The specific formula is as follows.

$${\mathrm{A}\mathrm{g}\mathrm{g}}_{\mathrm{i}\mathrm{t}}={\displaystyle \frac{{\mathrm{p}}_{\mathrm{i}\mathrm{t}}/{\mathrm{g}}_{\mathrm{i}\mathrm{t}}}{{\mathrm{P}}_{\mathrm{t}}/{\mathrm{G}}_{\mathrm{t}}}}\times {\left[{\log }_2\right(1+{\displaystyle \frac{{\mathrm{p}}_{\mathrm{i}\mathrm{t}}}{{\mathrm{P}}_{\mathrm{t}}}}\left)\right]}^{\delta },$$

(9)

in this equation, $\mathrm{A}\mathrm{g}{\mathrm{g}}_{\mathrm{i}\mathrm{t}}$ represents the industrial agglomeration level of the $\mathrm{i}\text{-}\mathrm{t}\mathrm{h}$ county in yeart. ${\mathrm{p}}_{\mathrm{i}\mathrm{t}}$ and ${\mathrm{g}}_{\mathrm{i}\mathrm{t}}$ represent the pig production value and the gross regional product of the $\mathrm{i}\text{-}\mathrm{t}\mathrm{h}$ county in yeart, separately. ${\mathrm{P}}_{\mathrm{t}}$ denotes the gross pig production value of China as a whole in yeart, and ${\mathrm{G}}_{\mathrm{t}}$ denotes GDP in yeart. The sensitivity $\delta$ is set to 0.3 based on [49].

3.4.3. Instrumental Variable Model

Since the areas with low GTFP of pig breeding in the Yangtze River Basin are relatively more polluted, the government will pay more attention to it and then tend to use higher intensity of ER, so there may be a mutual causality problem between ER and GTFP, and there is also a possibility of omission of variables in the model design. In order to solve the endogeneity problem between ER and GTFP, this paper takes the ER intensity of other counties in the province (ERAOP) as an instrumental variable. When the ER intensity is higher in other counties in the province except one, the county government tends to adopt a stricter ER instrument due to the characteristic of mutual imitation behavior among neighboring county governments, and the ER of other counties in the province has no direct effect on the GTFP of pig breeding in that county, thereby satisfying the exogeneity requirement.

To evaluate the endogeneity of the threshold effect, instead of utilizing the standard instrumental variable technique, the following panel threshold model is constructed.

Stage 1: Linear regression of ER with the instrumental variable ERAOP to determine the fitted value of ER.

$$\mathrm{E}{\mathrm{R}}_{\mathrm{i}\mathrm{t}}={\alpha }_0+\alpha {\mathrm{X}}_{\mathrm{i}\mathrm{t}}+{\beta }_1\mathrm{E}\mathrm{R}\mathrm{A}\mathrm{O}{\mathrm{P}}_{\mathrm{i}\mathrm{t}}+{\mathsf{\mu }}_{\mathrm{i}}+{\mathsf{\lambda }}_{\mathrm{t}}+{\epsilon }_{\mathrm{it}},$$

(10)

where $\mathrm{E}\mathrm{R}\mathrm{A}\mathrm{O}{\mathrm{P}}_{\mathrm{i}\mathrm{t}}$ denotes the ER intensity of other counties in the province where the $\mathrm{i}\text{-}\mathrm{t}\mathrm{h}$ county is located in year t. The statistic is based on the average ER intensities of all counties in the province where a county is located, excluding itself. The remaining indicators have the same significance as the preceding equation.

Stage 2: The obtained estimates of ER are reintroduced into the panel threshold model as follows:

$${\mathrm{G}\mathrm{T}\mathrm{F}\mathrm{P}}_{\mathrm{i}\mathrm{t}}={\alpha }_0+{\alpha }_1\widehat{\mathrm{E}{\mathrm{R}}_{\mathrm{i}\mathrm{t}}}\mathrm{I}\left(\mathrm{E}\mathrm{R}\le {\gamma }_1\right)+{\alpha }_2\widehat{\mathrm{E}{\mathrm{R}}_{\mathrm{i}\mathrm{t}}}\mathrm{I}\left({\gamma }_1\le \mathrm{E}\mathrm{R}\le {\gamma }_2\right)+{\alpha }_3\widehat{\mathrm{E}{\mathrm{R}}_{\mathrm{i}\mathrm{t}}}\mathrm{I}\left(\mathrm{E}\mathrm{R}>{\gamma }_2\right)+\alpha {\mathrm{X}}_{\mathrm{i}\mathrm{t}}+{\mathsf{\mu }}_{\mathrm{i}}+{\mathsf{\lambda }}_{\mathrm{t}}+{\epsilon }_{\mathrm{i}\mathrm{t}}$$

(11)

where $\widehat{E{R}_{\mathrm{i}\mathrm{t}}}$ is the estimated value of ER obtained in Equation (10), and the other letters have the same meanings as in the previous section.

4. Results

4.1. Descriptive Statistics

The descriptive statistics analysis are presented in

Table 2. Descriptive statistics.

Variable	N	Mean	Sd	Min	Max	P50
GTFP	3048	1.109	0.322	0.269	3.963	1.050
MLEC	3048	1.030	0.201	0.247	4.395	1.004
MLTC	3048	1.094	0.316	0.440	2.224	1.021
ER	3048	0.050	0.052	0.007	0.356	0.032
Ln_Machine	3048	0.360	0.314	−5.006	7.486	0.313
Agg	3048	0.181	0.225	0	2.869	0.103
Ln_Farmland	3048	8.142	1.15	0	15.53	7.983
Feed	3048	0.157	0.188	0	1	0.079
Edu	3048	0.047	0.029	0.017	0.177	0.037
Ln_Industry	3048	9.145	1.178	5.521	12.26	9.149
Wage	3048	0.417	0.170	0.045	1.648	0.414
Ln_Investment	3048	8.667	2.128	0	17.10	8.422
Ln_pigsa	3048	7.937	1.409	0	15.94	7.766
Ln_pigsp	3048	7.988	1.451	0	16.04	7.852
Ln_Price	3048	2.913	0.307	2.497	3.586	2.813
Ln_Porkcp	3048	7.835	0.684	5.508	10.29	7.773
Ln_Porkcc	3048	3.175	0.270	2.743	3.718	3.141
Convenience	3048	1.607	1.831	0.335	16.27	1.349
Ln_Traffic	3048	4.804	0.980	1.866	8.035	4.705
ERAOP	3048	0.050	0.050	0.007	0.205	0.033

. There are large differences between the maximum and minimum values of ER, GTFP, MLEC, and MLTC for pig breeding in the Yangtze River Basin, indicating that the differences between the regions of ER and GTFP, as well as MLEC and MLTC in the Yangtze River Basin, are relatively obvious, and that there is an imbalance between the regions. Out of the pig breeding regions, the average value (1.109) of the explanatory variable GTFP is higher than the middle value of 1.050. This suggests that more than half of the pig breeding regions have GTFP indexes that are lower than the average. Furthermore, there is a substantial disparity between the greatest value (3.963) and the mean, indicating that there is much room for improvement in the general level of GTFP in pig breeding regions in the Yangtze River Basin. Nevertheless, the average values of GTFP, MLEC, and MLTC all exceed 1, suggesting that the overall productivity of pig breeding has indeed increased in recent years. This positive trend can be attributed to the significant emphasis on technological innovation in China. In addition to descriptive statistics, this paper conducts multicollinearity tests and correlation analyses to comprehensively examine the potential relationships among variables and their impact on model stability. Specifically, by calculating the variance inflation factor (VIF), this paper systematically assesses the multicollinearity problem. The results show that the VIF values of the vast majority of variables are kept at very low levels, between 1 and 3, and the VIF values of only two variables are slightly higher than 5 but much lower than the commonly regarded multicollinearity warning line of 10, which fully proves that the problem of multicollinearity among the variables in the model has been effectively controlled. In addition, the results of the correlation analysis indicate that the correlation coefficients between all pairs of variables are well below the value of 0.8, a finding that not only indicates that there is no significant multicollinearity problem among the variables but also clearly points out that the variables maintain a high degree of statistical independence. For reasons of space limitation, specific tables are not included.

4.2. Spatio-Temporal Characterization of Green GTFP in Pig Breeding

This paper utilizes the input–output data of 381 county-level units in the Yangtze River Basin in China from 2013 to 2021 to assess the GTFP, MLEC, and MLTC of pig breeding from 2014 to 2021. The SBM super-efficiency model and ML index are employed for this analysis. Meanwhile, this study conducts a comparison of the GTFP of pig breeding in various basins of the Yangtze River in order to examine its spatial distribution characteristics, as shown in

Table 3. GTFP and its decomposition terms in the upper, middle, and lower reaches of the Yangtze River.

Year	The Upper Reaches			The Middle Reaches			The Lower Reaches
	GTFP	MLEC	MLTC	GTFP	MLEC	MLTC	GTFP	MLEC	MLTC
2014	1.3086	1.0398	1.2704	0.9812	0.9850	1.0030	0.9722	0.9351	1.0513
2015	1.3033	1.0124	1.2959	1.1303	1.1315	1.0098	1.1525	1.1514	1.0123
2016	1.1768	1.0108	1.1680	1.1586	1.0545	1.1036	1.1054	0.9958	1.1129
2017	0.8084	1.0499	0.7728	0.9491	1.0975	0.8674	0.8639	1.0494	0.8299
2018	0.9212	1.1169	0.8281	0.9007	1.0151	0.8952	0.9140	1.0234	0.9019
2019	1.5264	0.9407	1.6289	1.4293	0.9198	1.5564	1.5566	0.9444	1.6577
2020	1.5972	1.0432	1.5405	1.3998	1.1199	1.2600	1.3085	0.9745	1.3633
2021	0.6713	1.0883	0.6285	0.8850	1.0927	0.8154	0.7461	1.0192	0.7342
Average	1.1641	1.0377	1.1416	1.1042	1.0520	1.0639	1.0774	1.0116	1.0830

Note: 2014 stands for the index calculated using data from 2013, as well as 2014.

.

Based on the data presented in Table 3, the average values of GTFP and its decomposition term in the upper, middle, and lower reaches of the Yangtze River are all above 1 during the 8 years of the sample period. This suggests that the overall trend of GTFP in pig breeding in these three regions is consistently increasing, which aligns with the findings by [19]. Additional analysis reveals that MLTC is the primary factor driving the trend, which is consistent with the conclusions in [24] regarding GTFP in agriculture overall. The prevalence of MLTC in the pig breeding industry in China can be attributed to the historical predominance of home farming, which relies on outdated technology. Consequently, the implementation of technical development is relatively straightforward, especially with the aid of various legislative initiatives. Conversely, Chinese pig breeding is currently highly efficient, making it challenging to achieve future improvements [38]. Hence, MLTC plays a pivotal role in the development of pig breeding by significantly reducing production costs and enhancing GTFP [41].

Regarding the evolutionary characteristics over time, the GTFP and MLTC in pig breeding exhibit a general M pattern of increasing–decreasing–increasing–decreasing, with growth displaying unstable features of synchronized fluctuations and variations in the pig cycle but less evident patterns in the growth and alterations in MLEC. Spatially, the GTFP growth rate in the upper reaches of the Yangtze River surpasses that in the middle and lower reaches, aligning with the conclusions of [21] about spatial patterns. This could be attributed to the fact that the upstream region is more resource-rich, whereas the middle and lower reaches of the Yangtze River have a greater flow of people and are subject to greater pressure from industrialization and urbanization.

4.3. Threshold Effect Examination

This study uses county-level clustered robust standard errors in all models to ensure regression accuracy. In order to test Hypotheses 1a, 1b, and 1c and verify whether the effect of ER on the GTFP of pig breeding and its decomposition terms (MLEC, MLTC) are asymmetric or not, this paper considers ER itself as the threshold-dependent variable to test the effect of ER on the GTFP and its decomposition terms. Prior to estimating the panel threshold model, it is necessary to test for the presence of threshold effects and determine the number of thresholds. This paper utilizes the triple-panel threshold model to initiate the estimation process, employing bootstrap sampling 500 times to simulate the asymptotic distribution of the F-statistic and the critical value. Finally, this study constructs the corresponding p-value based on these simulations. The test results in

Table 4. Results of the triple-threshold effect test.

Explained Variable	Threshold Type	F-Statistic Value	p-Value	Critical Value
				10%	5%	1%
GTFP	Single	58.90	0.0000	10.150	12.166	15.213
	Double	62.14	0.0000	9.393	12.019	16.323
	Triple	32.25	0.4480	61.797	71.503	96.060
MLEC	Single	19.99	0.0000	8.820	10.426	14.811
	Double	33.91	0.0020	11.329	13.159	19.316
	Triple	26.77	0.3560	57.703	68.720	112.363
MLTC	Single	124.30	0.0000	10.124	12.276	16.488
	Double	72.73	0.0000	10.589	12.353	18.875
	Triple	45.59	0.3460	62.767	74.226	83.335

Note: All panel threshold model bootstrap counts are 500, and threshold-dependent variable is ER.

indicate that the F-statistics for both single- and double-threshold effects of GTFP and its decompositions have passed the significance test at the 1% level. However, all the triple-threshold effects fail the significance test. This suggests that there is a double-threshold effect of ER on GTFP, MLEC, and MLTC. In other words, the direction and extent of the impact of ER may vary depending on the level of ER intensity.

In order to test the double threshold more precisely, this paper conducts the test of the double-threshold effect, and

Table 5. Results of the double-threshold effect test.

Explained Variable	Threshold Type	F-statistic Value	p-Value	Critical Value
				10%	5%	1%
GTFP	Single	58.90	0.0000	10.150	12.166	15.213
	Double	62.14	0.0000	9.393	12.019	16.323
MLEC	Single	19.99	0.0000	9.473	11.021	14.543
	Double	33.91	0.0000	10.402	13.598	19.699
MLTC	Single	124.30	0.0000	10.434	12.472	16.413
	Double	72.73	0.0000	10.277	12.861	17.198

Note: All panel threshold model bootstrap counts are 500, and threshold-dependent variable is ER.

still reports the relevant results and also calculates the corresponding thresholds of ER. Additionally,

Table 6. Threshold and confidence interval results for the double-threshold effect.

Explained Variable	Threshold Type	Threshold Value	95% Confidence Intervals
GTFP	First threshold	0.0234	[0.0226, 0.0235]
	Second threshold	0.0238	[0.0218, 0.0240]
MLEC	First threshold	0.0258	[0.0257, 0.0271]
	Second threshold	0.0261	[0.0260, 0.0269]
MLTC	First threshold	0.0235	[0.0122, 0.0355]
	Second threshold	0.0243	[0.0234, 0.0247]

displays the thresholds and 95% confidence intervals of the three explanatory variables.

Table 7. Double-threshold effect regression results.

	(4)	(5)	(6)
Variable	GTFP	MLEC	MLTC
$\mathrm{E}\mathrm{R}\times \mathrm{I}\left(\mathrm{E}\mathrm{R}<{\gamma }_1\right)$	3.012 ***	0.125	4.167 ***
	(0.871)	(0.795)	(0.493)
$\mathrm{E}\mathrm{R}\times \mathrm{I}\left({\gamma }_1\le \mathrm{E}\mathrm{R}\le {\gamma }_2\right)$	8.898 ***	12.153 ***	10.112 ***
	(0.895)	(3.363)	(0.707)
$\mathrm{E}\mathrm{R}\times \mathrm{I}\left(\mathrm{E}\mathrm{R}>{\gamma }_2\right)$	0.838 ***	−0.229 **	1.174 ***
	(0.110)	(0.105)	(0.088)
Ln_Farmland	0.039 ***	0.018 **	0.017 ***
	(0.008)	(0.008)	(0.003)
Feed	−0.283 ***	−0.024	−0.270 ***
	(0.094)	(0.057)	(0.076)
Edu	2.350 *	1.683	2.968 ***
	(1.210)	(1.064)	(0.476)
Ln_Industry	−0.046 **	−0.013	−0.030 ***
	(0.018)	(0.015)	(0.010)
Wage	0.092	0.093	−0.015
	(0.090)	(0.092)	(0.055)
Ln_Investment	0.004	0.007	0.004
	(0.009)	(0.008)	(0.005)
Ln_hogsa	−0.007	−0.014	0.015 ***
	(0.013)	(0.010)	(0.005)
Ln_hogsp	0.013	0.013	−0.003
	(0.013)	(0.009)	(0.006)
Ln_Price	0.732 ***	0.753 ***	0.231 *
	(0.187)	(0.157)	(0.136)
Ln_Porkcp	−0.051	−0.103 **	0.034
	(0.056)	(0.045)	(0.027)
Ln_Porkcc	0.380 ***	0.087 **	0.233 ***
	(0.047)	(0.044)	(0.045)
Convenience	−0.010	−0.013	0.009
	(0.010)	(0.009)	(0.007)
Ln_Traffic	−0.006	0.040 ***	−0.042 ***
	(0.012)	(0.011)	(0.011)
Year FE	YES	YES	YES
County FE	YES	YES	YES
Constant	−1.741 **	−0.821	−0.503
	(0.790)	(0.583)	(0.410)
N	3048	3048	3048
R2	0.640	0.100	0.828

Note: Values in parentheses indicate robust standard errors for clustering at the county level, and *, **, and *** indicate significance at the 10%, 5%, and 1% levels, respectively.

presents the comprehensive estimation findings of the threshold effects. To explore the threshold effect of ER itself on GTFP, Columns (1)–(3) show the results with control variables. Specifically, Column (1) underscores the double-threshold effect of ER on GTFP, with all three ER interval coefficients passing a 1% significance level. When ER is less than 0.0234, the estimated coefficient of GTFP is 3.012, which indicates that ER significantly contributes to the improvement of GTFP. This may be related to the lower sensitivity of the initial pig farming entities in the Yangtze River Basin to environmental protection requirements; they still have room for other economic activities under lower environmental pressure and lack sufficient incentives to prioritize environmental protection. When ER falls within the range of 0.0234 to 0.0238, the estimated coefficient of GTFP increases to 8.898. This indicates that the impact of ER on GTFP is enhanced during the second threshold interval, which reflects that with the gradual tightening of environmental policies in the Yangtze River Basin, pig farmers have emphasized more on the investment in environmental protection technology and efficiency improvement to effectively promote GTFP. Nevertheless, when the level of ER surpasses 0.0238, although the coefficient of GTFP remains positive at this threshold, it decreases to 0.838. This suggests that the positive impact of ER on GTFP is significantly weakened within this range. This may be due to the fact that within the Yangtze River Basin, farming entities, when faced with extremely stringent ER, need to invest a large number of resources in order to meet compliance requirements, which squeezes funds for other innovative and productive activities, leading to less efficient use of resources. This discovery contradicts the widely held belief that the higher the ER intensity, the larger the GTFP-enhancing impact. In contrast, an excessively high ER intensity can restrict its ability to enhance GTFP, as observed in the studies conducted by [1,20].

As seen in Column (2), the initial MLEC threshold interval fails the significance test with a coefficient of 0.125. This suggests that lower ER does not directly improve MLEC of pig breeding. At the 1% level, ER is favorably significant with an estimated coefficient of 12.153 when above the initial threshold (0.0258), demonstrating a significant positive contribution to MLEC under moderate ER intensity. This transition highlights a pivotal economic mechanism: Under moderate ER intensity, pig breeders are compelled to innovate and enhance resource utilization efficiency, thereby positively contributing to MLEC. This phenomenon reflects the economic incentive structure where stricter ERs drive breeders towards more efficient practices. When ER surpasses the second threshold (0.261), the effect of ER on MLEC changes to a negative effect, with a coefficient of −0.229, which is significant at the 5% level. This structural shift from a facilitator to an inhibitor of MLEC improvement underscores the economic costs associated with heightened ER compliance. This is likely because, with higher ER circumstances, pig breeding enterprises must devote more resources to meet strict ERs, which may reduce R&D and production expenditure and hinder MLEC.

Finally, as shown in Column (3), the double-threshold impact of MLTC passes the 1% significance test in all three intervals with estimated coefficients of 4.167, 10.112, and 1.174. The promoting impact of ER on MLTC is usually beneficial, but its strength rises and falls. MLTC not only enables pig farmers to meet the requirements of ER but also helps them to improve their core competitiveness and form the scale development of pig breeding. Therefore, this paper argues that when the main body of pig farming is faced with ER, it will prefer to conduct green technological innovation to promote the growth of MLTC, which explains why when the intensity of ER is low, ER is significant for MLTC and not significant for MLEC. As ER increases, farming businesses confront more environmental strain, forcing them to invest more in technical innovation to discover more ecologically friendly and efficient production methods. However, when ER reaches a certain level, the space for MLTC may be limited to a certain extent, and the investment funds for technical innovation are squeezed out by the huge operating costs, which leads to a slowdown in the rate of MLTC. The trends of MLTC and GTFP coefficients are the same, which proves once again that MLTC is the main driving force for the improvement of GTFP.

Therefore, both statistically and economically, these results corroborate Hypotheses 1a, 1b, and 1c.

Figure 1 and Figure 2 illustrate plots of the double-threshold effect of ER on GTFP and its decomposition terms.

4.4. Robustness Examination

In order to test the robustness of the threshold effect, the following four robustness examinations are conducted.

4.4.1. Add Control Variables

Considering the influence of omitted variables on the model estimation results, this paper incorporates other factors that may affect GTFP, such as the urbanization level and pig industry structure into the panel threshold model and re-estimates the model. The urbanization level (urban) is measured by the ratio of the year-end urban resident population to the resident population of the county, and the pig industry structure is measured by the ratio of the total pig output value of the county to its total agriculture output value, with the results shown in Columns (1)–(3) of

Table 8. Robustness examination with added control variables and winsorization.

	Adding Control Variables			Winsorization
	(1)	(2)	(3)	(4)	(5)	(6)
Variable	GTFP	MLEC	MLTC	GTFP	MLEC	MLTC
Urban	−0.031	−0.043	−0.095
	(0.075)	(0.076)	(0.065)
Structure	0.456	0.439	−0.023
	(0.299)	(0.279)	(0.033)
$\mathrm{E}\mathrm{R}\times \mathrm{I}\left(\mathrm{E}\mathrm{R}<{\gamma }_1\right)$	2.852 ***	−0.089	4.167 ***	3.442 ***	0.080	10.931 ***
	(0.859)	(0.792)	(0.493)	(0.868)	(0.817)	(0.805)
$\mathrm{E}\mathrm{R}\times \mathrm{I}\left({\gamma }_1\le \mathrm{E}\mathrm{R}\le {\gamma }_2\right)$	8.537 ***	12.355 ***	10.175 ***	9.260 ***	12.115 ***	4.395 ***
	(0.904)	(3.333)	(0.706)	(0.907)	(3.364)	(0.411)
$\mathrm{E}\mathrm{R}\times \mathrm{I}\left(\mathrm{E}\mathrm{R}>{\gamma }_2\right)$	0.804 ***	−0.264 **	1.178 ***	0.936 ***	−0.247 **	1.785 ***
	(0.111)	(0.108)	(0.089)	(0.112)	(0.113)	(0.116)
Other Control	YES	YES	YES	YES	YES	YES
Year FE	YES	YES	YES	YES	YES	YES
County FE	YES	YES	YES	YES	YES	YES
Constant	−1.506 *	−0.606	−0.546	−1.750 **	−0.813	0.231
	(0.799)	(0.591)	(0.416)	(0.789)	(0.583)	(0.396)
N	3048	3048	3048	3048	3048	3048
R2	0.646	0.114	0.828	0.641	0.010	0.830
First Threshold	0.0234	0.0258	0.0235	0.0234	0.0258	0.0238
Second Threshold	0.0240	0.0261	0.0243	0.0238	0.0261	0.0472
p-value (Double)	0.0000	0.0000	0.0000	0.0000	0.0020	0.0000

Note: Values in parentheses indicate robust standard errors for clustering at the county level, and *, **, and *** indicate significance at the 10%, 5%, and 1% levels, respectively.

. The result in Column (1) indicates that even after accounting for other control variables, ER continues to have a significant positive impact on the GTFP of pig breeding, and this effect is still influenced by the double-threshold effect of ER. More precisely, the impact of ER on GTFP is enhanced once it surpasses the initial threshold but diminishes once it surpasses the second threshold. This pattern aligns with the original panel threshold model. Upon further examination of the results in Columns (2) and (3), it is evident that, with the exception of one insignificant coefficient, the impact of ER on MLEC and MLTC remains consistent with the original model in terms of both coefficient size and significance. This further confirms the robustness and reliability of the estimation results obtained from the original panel threshold model.

4.4.2. One Hundredth of a Percent Up-Down Winnowing

Due to the significant differences in the level of ER among Yangtze River Basin counties, this study performed winsorization of the ER variables at the 1% and 99% levels to prevent outliers and extreme values from negatively affecting the empirical analyses. The results of the threshold effects after winsorization are shown in Table 8 and Columns (4)–(6). GTFP, MLEC, and MLTC continue to show double-threshold effects, with the two thresholds of the three variables passing the 1% significance test. The regression model after winsorization is compatible with the original regression in terms of coefficient significance, demonstrating the robustness of the empirical investigation of this paper.

4.4.3. Delete Counties with Extremely Small Sum of Pig Slaughter and Inventory

During data collection, it is discovered that the number of pigs slaughtered and stocked in some counties decreases to zero year after year, which can be attributed to a variety of factors, including industrial restructuring, the implementation of environmental policies, and changes in market supply and demand. At the same time, some counties practice small-scale pig farming, which may inject bias into the GTFP calculation. To ensure the reliability of the research results in this paper, the pig stock and farrowing in 2014 and 2021 are ranked together, the data from all years in the bottom 1% of counties are deleted, and the threshold effect test is repeated, with the results shown in Columns (1)–(3) in

Table 9. Robustness examination of removing counties with low pig inventory and in cities that are smart city pilots.

	Removing Counties with Low Pig Slaughter and Inventory			Removing Smart City Pilots
	(1)	(2)	(3)	(4)	(5)	(6)
Variable	GTFP	MLEC	MLTC	GTFP	MLEC	MLTC
$\mathrm{E}\mathrm{R}\times \mathrm{I}\left(\mathrm{E}\mathrm{R}<{\gamma }_1\right)$	2.971 ***	0.038	9.783 ***	2.098 **	−2.051 ***	0.456
	(0.861)	(0.777)	(0.858)	(0.984)	(0.490)	(0.649)
$\mathrm{E}\mathrm{R}\times \mathrm{I}\left({\gamma }_1\le \mathrm{E}\mathrm{R}\le {\gamma }_2\right)$	8.759 ***	12.004 ***	3.915 ***	9.217 ***	2.183 *	7.687 ***
	(0.866)	(3.365)	(0.442)	(0.928)	(1.205)	(0.713)
$\mathrm{E}\mathrm{R}\times \mathrm{I}\left(\mathrm{E}\mathrm{R}>{\gamma }_2\right)$	0.868 ***	−0.221 **	1.569 ***	0.914 ***	−0.491 ***	1.046 ***
	(0.109)	(0.105)	(0.133)	(0.124)	(0.112)	(0.088)
Control	YES	YES	YES	YES	YES	YES
Year FE	YES	YES	YES	YES	YES	YES
County FE	YES	YES	YES	YES	YES	YES
Constant	−2.272 ***	−1.075 **	−0.009	−1.410	−0.773	−0.278
	(0.556)	(0.442)	(0.388)	(0.987)	(0.701)	(0.507)
N	3016	3016	3016	2592	2592	2592
R2	0.657	0.105	0.833	0.647	0.097	0.843
First Threshold	0.0234	0.0258	0.0238	0.0234	0.0414	0.0197
Second Threshold	0.0238	0.0261	0.0472	0.0240	0.0432	0.0238
p-value (Double)	0.0000	0.0020	0.0000	0.0000	0.0000	0.0000

Note: Values in parentheses indicate robust standard errors for clustering at the county level, and *, **, and *** indicate significance at the 10%, 5%, and 1% levels, respectively.

. It is clear that ER continues to have a significant double-threshold effect on GTFP, MLEC, and MLTC, and the results are mostly consistent with the initial regression, demonstrating the resilience of the original regression.

4.4.4. Delete Counties in the Smart City Pilots

With the acceleration of urbanization, China has long been devoted to easing and fundamentally solving urbanization-related difficulties such as resource scarcity, environmental pollution, and traffic congestion. In December 2012, the China Ministry of Housing and Urban–Rural Development announced the official launch of the national smart city pilot program [66], with the goal of achieving intelligent and refined urban management through the use of advanced information technology. A list of three batches of smart city pilots was released. Due to the potential positive impact of smart city pilots on GTFP in pig breeding, we especially investigate this component in our study. As a result, this paper cuts off data from counties participating in the smart city pilot and re-examines the threshold effect test. Columns (4)–(6) in Table 9 reveal that the double-threshold effect of pig GTFP and its decomposition term remains significant. It can be seen that the overall trend in Columns (4)–(6) remains consistent with the original regression results. This conclusion demonstrates that, while the smart city pilot policy has some impact on pig GTFP, the main pattern of the threshold effect does not change dramatically when this policy element is removed, demonstrating the robustness of the original threshold regression results.

Overall, the four robustness tests described above effectively validate the resilience and dependability of the original double-threshold model results.

4.5. Mechanism Examination

In order to examine the mechanism, this study starts with the three-threshold test, and the results demonstrate that both have only one threshold. As a result, this research runs a single-threshold model test on the two mediating variables, and the results are reported in

Table 10. Results of the single-threshold effect of the mechanism.

	(1)	(2)
Variable	Ln_Machine	Agg
$\mathrm{E}\mathrm{R}\times \mathrm{I}\left(\mathrm{E}\mathrm{R}\le {\gamma }_1\right)$	2.359 ***	−0.616 ***
	(0.632)	(0.181)
$\mathrm{E}\mathrm{R}\times \mathrm{I}\left(\mathrm{E}\mathrm{R}>{\gamma }_1\right)$	0.275 ***	−0.077 *
	(0.067)	(0.043)
Control	YES	YES
Year FE	YES	YES
County FE	YES	YES
Constant	0.135	−0.4752
	(0.949)	(0.298)
N	3048	3048
R2	0.141	0.164
First threshold	0.0257	0.0397
p-value (Single)	0.0060	0.0020

Note: Values in parentheses indicate robust standard errors for clustering at the county level, and * and *** indicate significance at the 10% and 1% levels, respectively.

. Column (1) shows the single-threshold effect of ER on the level of mechanization in pig breeding. When ER is smaller than 0.0257, it has a considerable positive effect on $\mathrm{L}\mathrm{n}\_\mathrm{M}\mathrm{a}\mathrm{c}\mathrm{h}\mathrm{i}\mathrm{n}\mathrm{e}$ at the 1% level, with a coefficient of 2.359. This suggests that ER can effectively enhance mechanization in pig breeding when the intensity of ER is modest, hence promoting GTFP growth. The behavior of pig farmers in the Yangtze River Basin shows that moderate ER drives mechanization investments. This not only aids compliance but also boosts productivity and efficiency, enhancing the growth of GTFP and illustrating the economic benefits of balancing regulation and technological adaptation. When ER exceeds the threshold of 0.0257, the coefficient decreases from 2.359 to 0.275, which is still significant at the 1% level. This indicates that the boosting effect of ER on pig mechanization sharply weakens, thereby also weakening its promotion of GTFP. This trend is consistent with previous studies of this paper, further confirming the reasonableness of the regression results. The explanation for this trend could be that as ER increases, farmers face increased cost pressures, making them more cautious about mechanized investment and then delaying the increase in mechanization level. On the contrary, Column (2) shows the results of the single-threshold effect of ER on the level of agglomeration in pig breeding. The first stage (ER ≤ 0.0397) and second stage (ER > 0.0397) are adversely significant at the 1% and 10% levels, with values of −0.616 and −0.077, respectively, indicating that ER inhibits the increase of the level of agglomeration in the pig industry as a whole. And this inhibition was gradually alleviated as ER increased. The reason for this is that when ER strengthens, some pig farmers may relocate to less controlled areas in order to evade rigorous regulation, resulting in a decrease in industrial agglomeration levels. When the amount of industrial agglomeration reaches a particular threshold, the less agglomerated the industry and the more conducive it is to GTFP [49]. In essence, when ER is initially weak, it effectively spurs GTFP growth by fostering industrial agglomeration. However, as ER intensifies, its effectiveness diminishes, aligning with the paper’s core finding.

To summarize, ER can improve GTFP by increasing pig breeding mechanization and preventing industrial agglomeration. These two mechanisms represent distinct elements of MLEC and MLTC, respectively. Among them, increased mechanization can directly lead to innovation in production technology and efficiency, whereas reduced industrial agglomeration can help to reduce resource waste and environmental pollution, ultimately improving the green development level of the overall industry. These regression results support Hypothesis 2a and 2b.

4.6. Regional Heterogeneity Test

Due to the large differences in ER intensity across counties, this paper examines regional heterogeneity in the sample counties using two classifications to further investigate the threshold effects of ER on pig breeding GTFP, as well as the decomposition terms MLEC and MLTC in different regions.

4.6.1. Impact of Whether the County Is Located in a Provincial Capital City

Due to the significant difference between the economic development levels of provincial capital cities and non-provincial capital cities, the sample is categorized by whether the location of the county is a provincial capital city or not, and this paper, respectively, conducts the threshold effect test. The specific regression results are shown in

Table 11. Heterogeneity test results categorized by whether or not counties are located in a provincial capital city.

	(1)	(2)	(3)	(4)	(5)	(6)
Variable	GTFP	GTFP	MLEC	MLEC	MLTC	MLTC
$\mathrm{E}\mathrm{R}\times \mathrm{I}\left(\mathrm{E}\mathrm{R}<{\gamma }_1\right)$	−27.801 ***	10.115 ***	−13.294	−1.873 ***	−7.095 ***	5.403 ***
	(7.538)	(1.137)	(11.381)	(0.507)	(1.906)	(0.531)
$\mathrm{E}\mathrm{R}\times \mathrm{I}\left({\gamma }_1\le \mathrm{E}\mathrm{R}\le {\gamma }_2\right)$	−9.040 ***	3.577 ***	−6.552 *	2.144 *	−8.254 ***	0.903 ***
	(2.221)	(0.404)	(3.789)	(1.249)	(1.805)	(0.141)
$\mathrm{E}\mathrm{R}\times \mathrm{I}\left(\mathrm{E}\mathrm{R}>{\gamma }_2\right)$	−1.659	1.218 ***	0.079	−0.460 ***	−2.324 ***	1.243 ***
	(1.046)	(0.121)	(0.872)	(0.111)	(0.559)	(0.098)
Control	YES	YES	YES	YES	YES	YES
Year FE	YES	YES	YES	YES	YES	YES
County FE	YES	YES	YES	YES	YES	YES
Constant	−0.673	−4.575 ***	−4.113	−1.360 **	1.363	−0.642
	(4.487)	(0.873)	(3.633)	(0.659)	(1.528)	(0.565)
N	304	2744	304	2744	304	2744
R2	0.611	0.674	0.222	0.106	0.854	0.830
First threshold	-	0.0249	-	0.0417	0.0354	0.0243
Second threshold	-	0.0899	-	0.0432	0.0363	0.1464
p-value (Double)	-	0.0160	-	0.0000	0.0000	0.0080

Note: Values in parentheses indicate robust standard errors for clustering at the county level, and *, **, and *** indicate significance at the 10%, 5%, and 1% levels, respectively.

. It should be noted that Columns (1) and (3) do not pass the threshold effect test, and the regressions in Columns (2)–(6) all pass the double-threshold effect test. Columns (1), (3), and (5) describe the regression results for counties in provincial capital cities, whereas Columns (2), (4), and (6) describe the regression results for counties in non-provincial capital cities. Specifically, the effect of ER on GTFP and its decomposition terms (including facilitating and inhibiting effects) is less favorable in counties located in provincial capitals, most likely because provincial capitals are typically regional economic centers, and their economic structures may be more diverse, with service and high-end manufacturing industries accounting for a larger share and traditional pig breeding accounting for a smaller share. As a result, the influence of ER on pig breeding GTFP and its decomposition terms may be less pronounced, causing the threshold test to fail in provincial capitals with a service and high-end industrial economy. The suppression of ER on MLTC in provincial capital cities in Column (5) could be attributed to the fact that the environmental capacity of provincial capital cities is relatively small due to their dense population and frequent economic activities, which necessitate stricter control of pollutant emissions. This may result in harsher environmental limits and greater treatment costs for pig rearing in the counties around provincial capital cities, impeding MLTC.

4.6.2. Impact of the Distance of County Governments from the Yangtze River

The distance between the county and the Yangtze River will be a significant concern for the government when developing environmental policies. Due to the availability and accuracy of the data, the distance of county governments from the Yangtze River can be used to approximate the distance of county centers from the Yangtze River. As a result, this paper divides the distance between county governments and the Yangtze River (including the main stem and tributaries) into three intervals of straight-line distance less than 300 m, 300–1000 m, and more than 1000 m, and it tests the double-threshold effect in each interval. The specific results are shown in

Table 12. Heterogeneity test results categorized by the distance of county governments from the Yangtze River.

	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)
Variable	GTFP	GTFP	GTFP	MLEC	MLEC	MLEC	MLTC	MLTC	MLTC
$\mathrm{E}\mathrm{R}\times \mathrm{I}\left(\mathrm{E}\mathrm{R}<{\gamma }_1\right)$	1.028	8.210 ***	−11.748 ***	−0.515	−0.371	−7.266 **	5.890 ***	10.893 ***	−5.491 ***
	(1.490)	(1.536)	(3.564)	(1.207)	(0.861)	(3.118)	(0.861)	(1.263)	(1.832)
$\mathrm{E}\mathrm{R}\times \mathrm{I}\left({\gamma }_1\le \mathrm{E}\mathrm{R}\le {\gamma }_2\right)$	9.442 ***	2.449 ***	3.206	−4.615 ***	16.315 ***	2.172	0.925 ***	4.232 ***	12.474 ***
	(1.783)	(0.735)	(2.356)	(1.203)	(4.038)	(1.575)	(0.217)	(0.641)	(2.135)
$\mathrm{E}\mathrm{R}\times \mathrm{I}\left(\mathrm{E}\mathrm{R}>{\gamma }_2\right)$	0.950 ***	1.090 ***	0.418	−0.127	−0.345 ***	−0.275	1.325 ***	1.645 ***	0.557 ***
	(0.211)	(0.227)	(0.289)	(0.258)	(0.122)	(0.246)	(0.186)	(0.194)	(0.142)
Control	YES	YES	YES	YES	YES	YES	YES	YES	YES
Year FE	YES	YES	YES	YES	YES	YES	YES	YES	YES
County FE	YES	YES	YES	YES	YES	YES	YES	YES	YES
Constant	0.612	−4.005 ***	−2.754 *	1.737	−1.895	−2.592 *	−0.058	−0.697	1.229
	(1.555)	(1.341)	(1.530)	(1.251)	(1.215)	(1.319)	(0.772)	(0.545)	(0.937)
N	848	1392	808	848	1392	808	848	1392	808
R2	0.682	0.705	0.574	0.166	0.158	0.101	0.831	0.838	0.830
First Threshold	0.0197	0.0238	-	0.0363	0.0257	-	0.0252	0.0238	0.0122
Second Threshold	0.0252	0.0432	-	0.0412	0.0261	-	0.1464	0.0472	0.0143
p-value (Double)	0.0060	0.0000	-	0.0240	0.0000	-	0.0160	0.0000	0.0000

Note: Values in parentheses indicate robust standard errors for clustering at the county level, and *, **, and *** indicate significance at the 10%, 5%, and 1% levels, respectively.

. Columns (3) and (6) pass only the single-threshold test, whereas the rest pass the double-threshold test. Columns (1), (4), and (7) depict the regression results for county governments less than 300 m from the Yangtze River; Columns (2), (5), (8), and (3), (6), (9) show the regression results for distances between 300 and 1000 m and more than 1000 m, respectively. Columns (1) and (2) suggest that low-intensity ER is more effective in remote areas, whereas high-intensity ER has a greater impact in closer counties. It is possible that pig breeding restrictions are stronger in counties closer to the Yangtze River, so when the ER strength is weaker, it has no positive effect on the GTFP of pig breeding. Columns (4) and (5) demonstrate that, between the first and second ER thresholds, ER significantly reduces MLEC in counties near the Yangtze River while increasing it in distant counties. This variation could be attributed to differences in ER enforcement and regulatory stringency by geographic location. Finally, Columns (7)–(9) demonstrate that at low ER intensity, ER most strongly promotes MLTC in mid-range districts but suppresses it in farther counties. When ER reaches the first threshold, all three coefficients are positive at the 1% level, and the boosting effect grows with distance.

In summary, counties farthest from the Yangtze River struggle to manage additional environmental demands due to initially low technological levels, hindering GTFP growth. However, once ER intensity exceeds the first threshold, the effect becomes positive. Counties near the Yangtze River face higher environmental challenges and resource expenditures, requiring more intensive ER for significant outcomes. Moderately distant counties benefit most from ER, balancing resource access, technological innovation, and environmental pressure.

4.7. Moderating Effect

In this paper, we investigate the moderating effects of the control variables of pig price and transportation efficiency in the ER on pig GTFP, MLEC, and MLTC, respectively. The specific results are as follows.

4.7.1. Pig Price-Moderating Effect

The fluctuation of pig prices has an impact on all aspects of pig production and consumption [6]. In order to investigate the regulatory role of pig prices, after centering the ER and pig price (Ln_Price), this paper generates the interaction term Interact1 ($\mathrm{I}\mathrm{n}\mathrm{t}\mathrm{e}\mathrm{r}\mathrm{a}\mathrm{c}\mathrm{t}1=\mathrm{c}\_\mathrm{E}\mathrm{R}\times \mathrm{c}\_\mathrm{L}\mathrm{n}\_\mathrm{P}\mathrm{r}\mathrm{i}\mathrm{c}\mathrm{e}$), and then conducts the threshold effect test. The three regressions all pass the double-threshold effect at the 1% level, and specific results are presented in Columns (1)–(3) of

Table 13. The moderating role of pig price and operational efficiency of transportation in the effects of ER and GTFP, MLEC, and MLTC.

	Pig Price			Transportation
	(1)	(2)	(3)	(4)	(5)	(6)
Variable	GTFP	MLEC	MLTC	GTFP	MLEC	MLTC
$\mathrm{E}\mathrm{R}\times \mathrm{I}\left(\mathrm{E}\mathrm{R}<{\gamma }_1\right)$	2.675 ***	−8.103 ***	5.182 ***	2.861 ***	0.122	3.554 ***
	(0.845)	(1.404)	(0.577)	(0.829)	(0.808)	(0.475)
$\mathrm{E}\mathrm{R}\times \mathrm{I}\left({\gamma }_1\le \mathrm{E}\mathrm{R}\le {\gamma }_2\right)$	7.182 ***	−2.527 ***	5.332 ***	8.770 ***	10.800 ***	7.216 ***
	(1.026)	(0.528)	(0.483)	(0.852)	(2.199)	(0.605)
$\mathrm{E}\mathrm{R}\times \mathrm{I}\left(\mathrm{E}\mathrm{R}>{\gamma }_2\right)$	0.558 ***	−0.781 ***	0.850 ***	0.544 ***	−0.149	0.799 ***
	(0.098)	(0.147)	(0.073)	(0.118)	(0.130)	(0.083)
$\mathrm{I}\mathrm{n}\mathrm{t}\mathrm{e}\mathrm{r}\mathrm{a}\mathrm{c}\mathrm{t}\times \mathrm{I}\left(\mathrm{E}\mathrm{R}\le {\gamma }_1\right)$	−5.179 ***	0.135	−4.392 ***	−0.527 ***	0.397	−1.379 ***
	(1.142)	(1.444)	(0.693)	(0.201)	(0.242)	(0.165)
$\mathrm{I}\mathrm{n}\mathrm{t}\mathrm{e}\mathrm{r}\mathrm{a}\mathrm{c}\mathrm{t}\times \mathrm{I}\left({\gamma }_1\le \mathrm{E}\mathrm{R}\le {\gamma }_2\right)$	−11.691 ***	4.614 **	−14.259 ***	2.114 ***	9.802 ***	3.113 ***
	(1.927)	(1.793)	(1.004)	(0.426)	(2.597)	(0.409)
$\mathrm{I}\mathrm{n}\mathrm{t}\mathrm{e}\mathrm{r}\mathrm{a}\mathrm{c}\mathrm{t}\times \mathrm{I}\left(\mathrm{E}\mathrm{R}\ge {\gamma }_2\right)$	−2.204 ***	−0.459	−2.688 ***	−0.637 ***	0.142	−0.725 ***
	(0.412)	(0.641)	(0.261)	(0.135)	(0.143)	(0.084)
Control	YES	YES	YES	YES	YES	YES
Year FE	YES	YES	YES	YES	YES	YES
County FE	YES	YES	YES	YES	YES	YES
Constant	−1.942 **	−1.542 **	−0.421	−1.308	−1.070 *	−0.143
	(0.780)	(0.598)	(0.395)	(0.797)	(0.603)	(0.391)
N	3048	3048	3048	3048	3048	3048
R2	0.653	0.104	0.838	0.644	0.105	0.833
First Threshold	0.0217	0.0158	0.0218	0.0234	0.0253	0.0235
Second Threshold	0.0249	0.0405	0.0252	0.0238	0.0261	0.0249
p-value (Double)	0.0000	0.0020	0.0000	0.0000	0.0040	0.0000

Note: Values in parentheses indicate robust standard errors for clustering at the county level, and *, **, and *** indicate significance at the 10%, 5%, and 1% levels, respectively.

. In Column (1), the coefficients of Interact1 reveal that pig prices moderate the relationship between ER and GTFP negative in all three threshold ranges and pass the 1% significance test. Due to the positive main impact coefficient, pig prices significantly reduce the promotion effect of ER on GTFP, moderating it negatively. The results in Column (3) indicate that pig prices significantly, negatively moderate the relationship between ER and MLTC. However, in Column (2), only the coefficient between the first and second thresholds is significantly positive at the 5% level, suggesting a significant positive moderating effect of pig prices on MLEC. On the one hand, higher pig prices may encourage farmers to increase the number of sows in their herd [6] and adopt more advanced technology and management practices to improve production efficiency and reduce costs, thereby offsetting some of the potential production pressure from ER. On the other hand, higher pig prices may attract new resources, increasing efficiency further. However, exploring and applying new technologies takes time and money, so during high pig prices, pig breeding entities may prefer to expand their production scale or improve process utilization rather than invest in technological innovations and R&D. Furthermore, pig prices fluctuate due to oversupply and undersupply in the pork market [6], and improving efficiency does not always lead to farmer profitability, so GTFP and MLTC are moderately negatively affected by pig price.

4.7.2. Operational Efficiency of Transportation Moderating Effect

Also, the paper investigates the moderating role of transportation operational efficiency. Following centering, the interaction term $Interact2$ ($Interact2=\mathrm{c}\_\mathrm{ER}\times \mathrm{c}\_\mathrm{Ln}\_\mathrm{Traffic}$) is constructed, and the double-threshold model test is performed, with the specific regression findings shown in Table 13. Except for MLEC, which passes at 5%, all pass the double-threshold effect at 1%. In Column (4), unlike the first moderating variable, traffic operation efficiency can significantly contribute to the improvement of ER on GTFP between the first and second thresholds. However, it has a negative moderating effect in the other two segments. It indicates that when ER intensity is moderate, improving traffic operation efficiency might significantly contribute to the favorable influence of ER on GTFP. The likely reason is that improving transportation operation efficiency reduces logistics costs and environmental pollution, hence encouraging the promotion and implementation of green production methods in pig breeding. A similar conclusion can be drawn for MLTC in Column (6), demonstrating the positive influence of transportation efficiency in supporting MLTC. In Column (5), for MLEC, the moderating effect of traffic operation efficiency is only significant in the second interval, indicating a considerable positive moderating effect. The explanation for this is that, as ER intensity increases, pig breeding entities are under more pressure to reduce emissions, and improving traffic operation efficiency can help enterprises optimize resource allocation and production efficiency. Improvements in transportation efficiency can create more options for businesses to optimize resource allocation and production efficiency, hence helping to improve MLEC.

4.8. Endogeneity Test

Table 14. Threshold regression results with the addition of instrumental variables.

	(1)	(2)
Variable	ER	GTFP
ERAOP	1.004 ***
	(0.011)
$\mathrm{E}\mathrm{R}\times \mathrm{I}\left(\mathrm{E}\mathrm{R}\le {\gamma }_1\right)$		5.569 ***
		(0.830)
$\mathrm{E}\mathrm{R}\times \mathrm{R}\left({\gamma }_1\le \mathrm{E}\mathrm{R}\le {\gamma }_2\right)$		3.602 ***
		(0.508)
$\mathrm{E}\mathrm{R}\times \mathrm{I}\left(\mathrm{E}\mathrm{R}\ge {\gamma }_2\right)$		1.216 ***
		(0.142)
Control	YES	YES
Year FE	YES	YES
County FE	YES	YES
Constant	−0.029	−1.601 **
	(0.038)	(0.787)
N	3048	3048
R2	0.934	0.637
First Threshold		0.0311
Second Threshold		0.0464
p-value (Double)		0.0100

Note: Values in parentheses indicate robust standard errors for clustering at the county level, and ** and *** indicate significance at the 5% and 1% levels, respectively.

reports the threshold effects after the addition of ERAOP. Column (1) is the first stage, and Columns (2)–(4) are the second one. First, by the regression in the first stage, it is found that ERAOP has a positive effect on ER at the 1% significance level. Furthermore, ERAOP is not a weak instrumental variable since its F-value is 2374.995, which is substantially higher than 10. The competition effect and imitation behavior among surrounding governments may explain why ERAOP can make a considerable contribution to the ER. When adjacent counties in the province have stricter ER criteria, a county may boost its own ER to stay competitive. Secondly, from the regression in the second stage, this paper discovers that the threshold effects of ER on GTFP in pig breeding remain significant. Although the threshold coefficients in each segment differ significantly, all segments pass the test at the 1% significance level and have more significant effects than those without instrumental factors. By introducing instrumental variables, this article mitigates the endogeneity problem between ER and GTFP to some extent, thus improving the reliability and accuracy of the regression results.

4.9. Further Analysis

4.9.1. Impact of ER on Pig Production Capacity and Carbon Emissions

Carbon emissions from livestock account for a significant component of global carbon emissions [11]. Based on the findings of this article, it is obvious that under ER, pigs’ GTFP increases. Under ER, carbon emissions as an undesired output are relatively reduced. However, at the same time, ER tends to limit pig production [10,15,22]. This paper explores the impact of ER regarding capacity and carbon emissions in the sample counties.

Columns (1)–(4) of

Table 15. Results of linear regression of ER on capacity, carbon emissions, and water quality.

	(1)	(2)	(3)	(4)	(5)	(6)
Variable	Ln_Production	Ln_Production	Ln_Emission	Ln_Emission	NH3-N	NH3-N
ER	−0.964 ***	−0.896 ***	−1.245 ***	−1.139 ***	−0.714 *	−0.726 ***
	(0.205)	(0.260)	(0.180)	(0.218)	(0.330)	(0.229)
Control	No	YES	No	YES	No	YES
Year FE	YES	YES	YES	YES	YES	YES
County FE	YES	YES	YES	YES	YES	YES
Constant	2.631 ***	5.802 ***	1.372 ***	5.826 ***	0.314 ***	1.581 **
	(0.010)	(1.298)	(0.009)	(1.410)	(0.019)	(0.515)
N	3048	3048	3048	3048	96.	96
R2	0.912	0.919	0.910	0.917	0.779	0.807

Note: Due to the unavailability of data, only part of the sample was explored in the fifth and sixth columns. Values in parentheses indicate robust standard errors for clustering at the county level, and *, **, and *** indicate significance at the 10%, 5%, and 1% levels, respectively.

show the results of the regression of ER on capacity and carbon emissions. Columns (1) and (3) indicate baseline regression without control variables, while Columns (2) and (4) include the effects of the corresponding control variables. Column (2) includes control variables such as arable land carrying capacity, urbanization level, wage income, education level, feed production capacity, transportation operation efficiency, pig industry structure, pig mechanization level, pork consumption capacity, county GDP level, and pig price. Control variables introduced in Column 4 include arable land carrying capacity, urbanization level, wage income, education level, feed production capacity, pig industry structure, pig price, and transportation accessibility. Observing Columns (1) and (2), it is discovered that regardless of whether control variables are added or not, all of them are significantly negative at the 1% level of significance, indicating that ER will limit the reduction in pig production capacity, which is consistent with the findings in [10,15,22]. Similarly, Columns (3) and (4) indicate that ER has a suppressive effect on carbon emissions at the 1% significance level. Furthermore, the absolute value of the ER coefficient on production capacity is lower than that of ER on carbon emissions, implying that ER suppresses pig production capacity less than it decreases pig carbon emissions. This article concludes that implementing ER reduces pig production capacity while also reducing CO₂ emissions during the pig breeding process. However, because ER has a larger suppression effect on carbon emissions, this policy has enhanced the GTFP of pig breeding in general. In other words, ER reduces pig production capacity, but this loss is less advantageous than the reduction in carbon emissions generated by ER.

4.9.2. Impact of ER on Water Quality

In recent years, the Chinese government has established environmental laws and regulations to minimize water pollution, especially in the livestock sector [13,67]. The sample area studied in this paper is located in the Yangtze River Basin, so it is necessary to explore whether the ER of pig breeding has improved the water quality. Water contamination is measured by ammonia nitrogen (NH₃-N), and high levels indicate nitrogen pollution [13]. The higher the NH₃-N content, the worse the water contamination and quality [67]. Based on this, this article examines how environmental rules affect water quality in sample counties.

The results of the regression of ER on NH₃-N are shown in Columns (5) and (6) of Table 15. Column (5) shows the data without control variables, and the coefficient of ER on NH₃-N is considerably negative at the 10% level, at −0.714, indicating that ER inhibits NH₃-N rise irrespective of other factors. Based on previous research in [13,68], the variables of urbanization, population density, crop density, industrialization, pig breeding density, and annual precipitation are controlled, and the regression results are considered, as shown in Column (6). After controlling for covariates, ER remains negatively significant on NH₃-N and passes the 1% significance test. Similar to [10,67], this finding strongly suggests that ER can diminish NH₃-N in water even after correcting for other parameters. The reason for this may be that ER compels pig farms to develop wastewater treatment facilities, use contemporary treatment technology and techniques, and ensure that wastewater discharges satisfy criteria. The deployment of these procedures has reduced the discharge of pollutants such as NH₃-N, thus improving the water quality of the Yangtze River.

4.9.3. Impact of Restricted Zone Policy Adjustment in 2019 on GTFP, MLEC, MLTC, and Capacity

All counties in this study have successively adopted the restricted zone policy, which is a command-and-control ER to promote green economic growth in pig breeding. In 2019, due to African swine fever and the abuse of the restricted zone policy, pig production capacity in China declined sharply. In order to increase production capacity, the government relaxed the restricted zone policies in some counties. The Ministry of Ecology and Environment opposed exceeding legal and regulatory restrictions on pig farming in the name of environmental protection, which marked a decrease in ER intensity. This paper employs double-difference models to examine how the restricted zone alteration affects GTFP, MLEC, MLTC, and production capacity.

Table 16. Results of linear regression of ER on GTFP, MLEC, MLTC and production.

	(1)	(2)	(3)	(4)
Variable	GTFP	MLEC	MLTC	Ln_Production
DID	−0.042 ***	−0.045 ***	−0.004	0.096 **
	(0.015)	(0.012)	(0.008)	(0.048)
Control	YES	YES	YSE	YES
Year FE	YES	YES	YES	YES
County FE	YES	YES	YES	YES
Constant	−1.500 *	−1.457 **	0.217	6.156 ***
	(0.826)	(0.625)	(0.484)	(1.277)
N	3048	3048	3048	3048
R2	0.641	0.140	0.813	0.918

Note: $DI{D}_{\mathrm{it}}=Trea{t}_{\mathrm{it}}\times Tim{e}_{\mathrm{it}}$ represents the restricted zone tool for the $i\text{-}th$ county in $t\text{-}th$ year, and $Treat=1$ if a county has adjusted its restricted zone policy in 2019, and 0 otherwise; if the time is in 2019 and later, $Time=1$, and the rest of the indicators have the same meaning as in the previous section. Values in parentheses indicate robust standard errors for clustering at the county level, and *, **, and *** indicate significance at the 10%, 5%, and 1% levels, respectively.

shows double-difference modelling results with control variables from prior models. Columns (1) to (3) illustrate DID regression on GTFP, MLEC, and MLTC after the restricted zone regulation change in 2019. The regression coefficients of DID on all three are negative, but only GTFP and MLEC pass the 1% significance level test, while MLTC does not. It demonstrates that the restricted zone policy adjustment has a strong inhibitory effect on the increase of GTFP and MLEC in pig breeding. After lowering ER intensity, DID increases the production capacity of the pig industry by 0.096 at the 5% level, according to Column (4). Combined with prior findings, this article concludes that increasing ER will improve the GTFP of pig breeding while potentially reducing output capacity. However, lowering ER, although increasing manufacturing capacity, also diminishes GTFP. This discovery is a significant reference point for policymakers, who must strike a balance between fostering economic growth and environmental protection in order to achieve the long-term development of pig breeding.

5. Conclusions and Policy Implications

This research conducts an in-depth investigation of the impact of ER on GTFP using panel data from 381 counties in the Yangtze River Basin between 2014 and 2021. During the research process, this paper applies the panel threshold model and conducts a robustness examination, as well as the threshold effect test, including mechanism analysis, exploration of regional heterogeneity, examination of the moderating role, and endogeneity treatment to ensure the comprehensiveness and accuracy of the research results. This research makes the following main conclusions based on our rigorous empirical analysis:

(1) First, this study conducts a nonlinear investigation using the panel threshold model, with ER serving as the threshold variable. The results reveal that as ER intensity increases, the promotion effect on GTFP and MLTC of pig breeding initially rises and then falls, whereas the influence on MLEC varies from negligible to significant positive effect, then to significant negative effect. Based on this, this paper strongly believes that ER has a positive impact on GTFP of pig breeding in the Yangtze River Basin, with the core of this impact being to promote technological innovation among pig breeders to improve MLTC, rather than to improve MLEC by optimizing management level and production realization ability. Furthermore, this article indicates that the intensity of ER should not be maximized but should be controlled within an acceptable range in order to contribute to the green and long-term development of pig breeding.

(2) Secondly, from the perspective of mechanism analysis, this paper argues that the ways for ER to enhance GTFP may include two aspects: First, by increasing the mechanization level of pig breeding in the Yangtze River Basin, we can achieve large-scale development, optimize resource allocation, and reduce carbon emissions, thus enhancing GTFP and achieving the goal of ecological environmental protection. Second, by reducing the level of industrial agglomeration in pig breeding, the concentration of environmental pollution will be reduced, improving the utilization rate of regional resources and creating favorable conditions for breeding entities that can seize green development opportunities, thus increasing their GTFP.

(3) After considering heterogeneity, moderating effects, and further analysis, this paper finds that the threshold effect of ER on GTFP exhibits different characteristics due to regional differences. Specifically, under low-strength ER, compared with the counties that did not cross the Yangtze River main stem, the counties that crossed the Yangtze River main stem showed more significant enhancement in MLTC but also stronger suppression of MLEC, resulting in a limited enhancement effect on the overall GTFP. In addition, the effect of ER on GTFP is more pronounced in non-capital city counties where pig breeding is more developed. At the same time, when the ER intensity is low, the promotion of GTFP is stronger in areas where the county government is farther away from the Yangtze River, but the opposite is true beyond a certain ER intensity. Moreover, the moderating factors such as pig price and transportation operation efficiency also show significant differences in the moderating effects of GTFP, MLEC, and MLTC in different ER intervals. Finally, increasing ER reduces production capacity but significantly reduces carbon emissions, and the environmental benefit of reducing carbon emissions outweighs the economic benefit of losing production capacity, resulting in the final increase of GTFP. However, decreasing ER will substantially boost production capacity while decreasing GTFP. Especially noteworthy is that the ER effectively improves water quality in the Yangtze River Basin and implements the ER of the Yangtze River Basin. This means that the core objectives of ER in the Yangtze River Basin have been achieved.

On the basis of the above conclusions, the paper makes the following three policy recommendations: (1) Optimizing ER policy design. The Chinese government should keep ER within a reasonable range to allow for scientific formulation and timely adjustment of policy intensity. The empirical results show that ER intensity does not improve the higher it is, but there exists an intermediate range, which has the most significant promotion effect on GTFP of pig breeding. In addition, due to regional heterogeneity, the threshold effect of ER on GTFP varies significantly among regions with different characteristics, so the policy formulation needs fully consider the actual situation of each region and formulate and implement the ER policy according to the local conditions. (2) Facilitating the improvement of MLEC in pig breeding. From the results of SBM-ML, it can be seen that the main factor driving the improvement of GTFP in pig breeding is MLTC, so the improvement of MLEC needs to be strengthened. For instance, the Chinese government can encourage the main body of breeding to adopt advanced breeding technology and management mode, optimize the allocation of resources, and improve production efficiency by providing technical support, financial support, and talent training. In addition, the government can also establish a technical exchange and cooperation platform to promote the sharing of experience and technological innovation among pig breeders so as to jointly promote the green transformation of the pig breeding industry as well as the development of new quality productive forces in response to the advocacy of China. (3) Provide suitable environmental subsidies to pig farmers. Further analysis reveals that ER reduces pig production capacity, resulting in lower farmer income. To balance the relationship between environmental conservation and economic development, the Chinese government should provide suitable environmental subsidies to pig farmers. These subsidies can not only compensate farmers for economic losses caused by ER but also serve as an incentive mechanism to encourage farmers to pay more attention to environmental protection within their own means, actively implement environmental protection measures, reduce pollution emissions, and practice green farming. At the same time, the government should tighten oversight and review of subsidized funding to ensure that funds are used transparently and effectively.