Heterogenous Urbanization and Agricultural Green Development Efficiency: Evidence from China

Abstract

Realizing green development in agriculture is fundamental to sustained economic development. As a measure to facilitate the transfer of rural population, urbanization is considered to be strategic in promoting agricultural green development. This paper employs a SBM-DDF–Luenberger method to measure agricultural green total factor productivity (AGTFP) and the agricultural labor surplus in China, and empirically tests the heterogeneous effects of household registration urbanization, permanent residence urbanization, and employment urbanization on the efficiency of agricultural green development. The results reveal that: (1) the average annual growth rate of China’s AGFTP is 4.4374%, which is achieved mainly through improvements in green scale efficiency. (2) The agricultural sector in China is suffering a large surplus of labor force, with an estimation of 20.64 million in 2020. (3) Both household registration urbanization and permanent residence urbanization have a significant promoting effect on agricultural green development efficiency, though the former promotes less. (4) Employment urbanization improves agricultural green development efficiency by providing employment guidance for migrant workers, and employment urbanization of the tertiary industry has a more pronounced improvement effect. The findings suggest that governments remove restrictions on the household registration system and actively guide surplus agricultural laborers to engage in urban service industries to provide an impetus for promoting green agricultural development.

1. Introduction

Along with the constant questioning of the traditional industrialization and urbanization modes, the concept of green development has been proposed for over sixty years. Early in 1962, the American Carson published Silent Spring, a publication on the environmental damage caused by traditional industrial civilization aroused people’s attention to environmental protection [1]. In 1972, the Club of Rome published The Limits to Growth, which questioned the sustainability of western industrialized countries’ high consumption, high pollution growth model [2]. However, the green concept at that time mainly emphasized an end treatment of pollution. Although countries have put forward their own strategic measures around green development, environmental problems brought alongside climate change are challenges encountered by humanity today [3,4].

To build a socialist ecological civilization, it is important for China, the world’s largest emerging economy, to promote green development. In 2002, the United Nations Development Programme released the China National Human Development Report, which suggests that China choose a green development path and achieve harmony between economic development and environmental protection [5]. Agriculture plays a vital role in China’s national economy [6]. It is estimated that with 8% of the world’s arable land and 5% of the world’s fresh water resources, China provides 95% of the food for 18% of the world’s population. Such a great contribution is achieved at a large environmental cost [7]. Traditional agriculture, which relies on heavy inputs in fertilizers and pesticides, has not only seriously affected the ecosystem, also resulting in environmental pollution, soil degradation, a sharp decline in biodiversity, and food safety problems [8,9], but has also increased agricultural carbon emissions, leading to the high-carbon characteristics in agricultural production [10,11]. Promoting agricultural development mode to transmit from the factor-driven extensive growth to the intensive growth that focuses on enhancing green productivity has become a key way for China to tackle climate change and achieve sustainable development [12].

Agriculture green total factor productivity (AGTFP) is an important indicator in the literature to measure the efficiency of green development in agriculture. Its improvement depends greatly on technological progress and resource allocation efficiency under established technological conditions [11,13]. Due to institutional constraints, labor misallocation, a major cause to inefficient agricultural production and agricultural land pollution [14,15], is a prominent bottleneck in China’s agricultural green development. A labor surplus is currently a serious problem faced by China’s agricultural sector [16,17]. The National Bureau of Statistics of China report that China’s employed population in agriculture was 177.15 million in 2020, accounting for 23.6% of the total employed population nationwide, yet the added value in agriculture was RMB 7.8 trillion, accounting for merely 7.76% of the GDP. Then, how to effectively absorb or transfer the redundant labor force in the agricultural sector and reduce the agricultural labor share has become a key step in improving agricultural efficiency and promoting agricultural green development.

Urbanization is an important tool for addressing agricultural labor surplus and promoting agricultural modernization. According to Lewis’ dualistic economic development theory, the wage gap between the modern economic sector and the traditional agricultural sector attracts a direct and steady flow of labor towards cities [18]. This is because urbanization per se reflects the reallocation of population and labor force, and meanwhile, urbanization- and industrialization-induced industrial structure change and technological progress may interact with labor mobility to further advance agricultural modernization. Thus, China has long advocated adjusting and optimizing the employment structure and comprehensively promoting rural revitalization through urbanization and citizenization. We wonder whether urbanization has truly promoted the green development efficiency of China’s agriculture.

It is worth noting that, unlike developed nations where economic development promotes urbanization, China’s urbanization has a strong policy overtone. Due to historical reasons such as the planned economy, urbanization in China can be divided, from the population-type dimension, into two categories, i.e., household registration urbanization and permanent residence urbanization. Household registration urbanization reflects the proportion of the registered urban household population in a region to its total registered residents. In “The Household Registration Ordinance” issued in 1958, agricultural and non-agricultural residents were distinguished explicitly so as to curb population mobility from the rural to urban areas. Since then, China’s household registration system featured a separation in the urban and rural hukou (The household registration system is also well known as the hukou system in China. Thus, our paper uses the term “household registration” and the term “hukou” alternately.) has been formed and gradually strengthened afterwards [19]. Permanent residence urbanization, as opposed to household registration urbanization, categorizes urban and rural population based on residential places, and as a result, reflects the proportion of a region’s urban permanent residents population to its total permanent residents population (The permanent residence population refers to those with one of the following characteristics: (i) those who live in the township street, and have their household registration in the township street or to be determined; (ii) those who live in the township street, and have been away from the township street where their household registration is located for more than half a year; and (iii) those who have their household registration in the township street, and have been away for less than half a year or have been working or studying abroad.). During China’s rapid urbanization, reform of the household registration system has lagged far behind population migration and mobility, resulting in an ever-widening disparity between the urbanization rate of the household registered population and the urbanization rate of the permanent residents population. According to statistics from the seventh census in China, the difference in the two urbanization rates reached 18.49% in 2020, an increase of 2.29 percentage points over 2015. The inconsistency and separation between people and hukou have even become a new normal in the living arrangements of China’s urban population. Then, is there any difference in the effects of household registration urbanization and permanent residence urbanization on agricultural green development efficiency? A timely examination of the actual state of China’s agricultural green development and an evaluation of the development efficacy of urbanization in various dimensions are of great theoretical and practical significance given the strategic context in which green development has evolved into one of the core development goals for modern agriculture. This would offer emerging nations valuable insights on designing new green policies for agriculture and advancing a better urbanization development.

Our paper relates to two strands of the literature. One strand focuses the evaluation of AGTFP with the two common measurement techniques—the stochastic frontier analysis (SFA) method and the data envelopment analysis (DEA) method [20,21]. SFA is applied mostly in the case of a single output without non-desired outputs [22], whereas DEA has advantages in handling data with multiple inputs and outputs. While applying DEA, specific forms of production function are not necessitated, which helps to reduce subjective controversies and makes DEA more commonly used in measuring green TFP [23,24]. The slack-based measure of efficiency in DEA (SBM-DEA) proposed by Tone (2004) takes undesired outputs into account and deals with slacks directly through incorporating them into the objective function [25]. This non-radial feature helps overcome the bias residing in the radially traditional DEA models [26]. As to measurement indicators, studies often select agricultural non-point source pollution such as agricultural water pollution, soil pollution, and carbon emission as the indicators of undesirable output [9,10,27].

The other strand discusses the influencing factors of AGTFP. Some have investigated the effects of precipitation, temperature, and humidity on AGTFP for climate change consideration. It has found that rising temperature and increasing rainfall density lower AGTFP significantly, and that the effects vary in different regions [28,29]. Some have evaluated the technological effects and revealed that the use of digital technology and the popularization of the internet are conducive to the improvement of AGTFP [11,30,31]. Others have explored the impact of rural financial development, and found that the expansion of rural financial scale and the improvement of financial operation efficiency helps boost AGTFP [32]. As a risk protection mechanism, crop insurance not only disperses agricultural business risks, but also improves AGTFP, and this improvement effect is enhanced with the promotion and application of agricultural green technology and the expansion of business scale [33]. Additional studies have examined influences of government environmental regulations and local policies. They point out that environmental regulations have both U-shaped direct effects and spatial spillover effects on AGTFP [34], and that implementing carbon trading pilot policies significantly promote AGTFP [35].

Prior studies have yielded insightful findings on green development and transformation in agriculture, but there are some issues which need further exploration. As can been seen, existing studies have focused either on the measurement of AGTFP [9,10,24,27], or on impacts from climate change, technological progress, capital deepening or institutional elements [28,29,30,31,32,33,34,35]. Rare work has investigated the impact of urbanization on agricultural green development from the standpoints of labor supply and population size. To fill such a gap, we base out study on a Chinese provincial panel data from 2001 to 2020 and a SBM-DEA framework which integrates techniques of directional distance function and Luenberger productivity decomposition to measure AGTFP, as well as the agricultural labor surplus, and then explores the impact of urbanization on agricultural green development efficiency.

Our work contributes to the existing literature in three regards. First, we enrich the research on AGTFP and agricultural labor surplus in China. Due to differences in the selection of measurement methods and indicators, the previous literature fails to achieve consensuses on the measurement results of AGTFP. Ignorance of resource and environmental factors in some studies while measuring agricultural productivity results in an inaccurate assessment on agricultural green development [9,10,24,27]. Meanwhile, few studies in the literature have accounted for agricultural labor surplus out of the estimation of agricultural green TFP. We investigate the temporal-spatial distribution to advance our understandings on the inefficient labor transfer in China’s agriculture.

Second, we complement the existing empirical evidence on influential factors of AGTFP. Although some scholars have observed the directing effects of urbanization on labor migration [36], and the potential impacts of agricultural labor transfer on rural land transfer and agricultural labor productivity [37,38], few, to our knowledge, have directly viewed urbanization as external intervention and evaluated how heterogeneous urbanization, which is explained as urbanization of population from different dimensions, would impact agricultural green development. Unlike prior studies which simply focus on effects of the transfer of agricultural labor force to the non-agricultural sectors, we treat urbanization as an external intervention and examine how household registration urbanization and permanent residence urbanization affect AGTFP. This provides a new perspective to evaluate urbanization effects for developing countries, as well as implementing targeted measures to enhance AGTFP.

Third, we define a term named employment urbanization and examine its effect on AGTFP. Current studies have noticed the labor flows’ impact on urbanization and agricultural development [39,40], but few discuss the employment effects of urbanization for the transferred rural population. We define such effects as employment urbanization from the perspective of the economic foundation and citizenry of agricultural transferred laborers and explore how they impact AGTFP. This provides references for the government and other decision makers on how to promote agricultural green development more efficiently.

The remainder of the article is structured as follows. Section 2 elaborates the theoretical framework and develops hypotheses. Section 3 measures AGTFP in China and articulates stylized facts of agricultural labor surplus. Section 4 investigates the empirical relationship between urbanization and AGTFP. Section 5 conducts further discussions on how to utilize employment measures to transfer agricultural surplus labor force. Section 6 concludes the paper.

2. Theoretical Analysis and Hypotheses Development

2.1. Research on AGTFP: From Technological Progress to Resource Allocation

In recent years, an increasing number of researchers have observed the impact of resource allocation efficiency on TFP. They argue that underdeveloped factor markets and more administrative interventions bring about low resource allocation efficiency for developing nations, and that this in turn hinders the improvement of TFP [17]. To further elaborate this finding, Hsieh and Klenow (2009) conduct a seminal study and demonstrate that if resource allocation efficiency in China and India could reach the level of the United States, TFP in China and India would increase by 30–50% and 40–60%, respectively [41]. For allocation of agricultural factors, Adamopoulos and Restuccia (2020) take the land reform in the Philippines as illustration and explore the potential impact of land market efficiency on agricultural productivity [42]. They find that the Philippines’ restrictions on the cap of land holdings, together with the strict prohibitions on the transfer of land, lead to resource misallocation among farmers, distort farmers’ occupational and technical decisions, and thereby lower farm size and agricultural productivity. A similar conclusion is drawn in Le (2020) by using data from Vietnam and showing that after removing land use restrictions, there would be an increase in agricultural TFP by 37.89% and a decrease in agricultural employment by 5.89% [43]. Likewise, an effective elimination of distortions in the cross-sectoral allocation of land, capital, and labor in agriculture would also improve agricultural efficiency considerably for China. Adamopoulos et al. (2022) find that allocating agricultural inputs effectively would increase China’s TFP by 53.2%, with two-thirds of the gains coming from labor redistribution [17]. Improving factor allocation efficiency and reducing factor misallocation are important ways for developing countries to improve future outputs and productivities [44].

In China, the lag of labor factor market reform and the distortion of labor factor allocation are root causes that impede the transformation of China’s agricultural modernization and green development when compared to production factors such as land, mechanical power, fertilizer, and capital. The numerous institutional constraints in the transformation context have resulted in a large amount of surplus labor force being stranded in rural areas. The high proportion of agricultural employment has complicated land transfer, increased the risk of pesticide and fertilizer abuse, made idled and wasted agricultural land coexist with a finely fragmented operation mode, discouraged the use of capital factors such as machinery into agricultural production, and ultimately, caused difficulties in bringing scale of benefits into play. According to statistics released by the Research Group of the Center for Urban and Small-town Reform and Development in National Development and Reform Commission, the average household operation scale in China is merely 0.6 hectares, which is 40% of that in South Korea (1.5 hectares) and 30% of that in Japan (2 hectares). An even greater gap is observed when compared with that in western countries, among which the United States (169 hectares), France (70 hectares), and Germany (38 hectares) have a respective figure of 282, 117, and 63 times larger than China’s. Agricultural labor productivity in China is also lower than the average level of the middle- and high-income economies, which is only 4.93% of that in the world’s leading agricultural country, the United States, and 5.73% of that in Australia. To sum up, increasing the transfer of surplus agricultural laborers to non-agricultural sectors and breaking down institutional barriers to the free flow of the labor force have become critical prerequisites for achieving economies of scale in China’s per household land, improving agricultural labor productivity, and promoting sustainable agricultural development.

2.2. Urbanization and Improvement in AGTFP: Perspective from Labor Force

Urbanization plays an engine-driven role in the dual structure transformation that regards the non-agriculturalization transfer of rural labor force as the core mechanism. In the first place, urbanization-induced transfer of agricultural surplus laborers might affect AGTFP through the reallocation of factor resources. The factor allocation theory indicates that urbanization proceeds with a direct reduction in agricultural redundant labor force, which helps promote agricultural labor productivity. This is intuitive since the industrial agglomeration and expansion brought by urbanization creates more employment opportunities for the agricultural redundant labor force, thereby accelerating the transfer of agricultural surplus laborers to cities. When outputs are held in an unchanged or increasing status, the reduction of the labor force implies an increase in output per labor in agricultural labor utilization efficiency, and in aggregate TFP.

Moreover, urbanization also raises land intensification and prompts agricultural operation mode to transform towards the moderate-scale operation and green development modes. As is well known, agriculture in China has long faced the rigid resource endowment constraint of a large population with little land. The fine fragmentation of arable land brings about a loss of technical efficiency in agricultural production [42]. Through the reallocation of production factors such as population and land between urban and rural areas, urbanization integrates and revitalizes land resources that were previously scattered and idle, as well as used inefficiently. This makes the moderate scale, industrialized and organized operation mode possible, and is hence conducive to reducing agricultural production costs, enhancing land output rate, guiding peasant households to adjust production structure, and implementing environment-friendly production behaviors. All these gains would have positive effects on the environment, and create conditions for the growth of green TFP in agriculture [45]. Studies have demonstrated that large-scale peasant households are more likely to adopt green farming technologies than small-scale peasant households, and that for every 1% increase in farm production size, fertilizer and pesticide use would decrease by 0.3% and 0.5%, respectively [46,47].

In the second place, labor mobility prompted by urbanization would influence AGTFP through capital deepening and green technological spillovers. Capital accumulation is crucial in the transformation of traditional agriculture and the promotion of green agricultural development efficiency. With the development of urbanization and industrialization, a large number of agricultural surplus laborers moves to non-agricultural industries of high income for employment. This might increase the household income and savings of the migrant agricultural population, and alleviate peasant households’ financial constraints that are caused by the insufficient supply of credit and the lack of formal financing channels in rural China [48]. An acceleration in capital deepening helps peasant households to acquire and increase agricultural capital inputs in a timely manner, thereby improving agricultural production efficiency [49]. In addition, agricultural surplus laborers’ transfer to high-income non-agricultural industries for employment might also accelerate the knowledge and technology spillover, and promote advanced agricultural management concepts and agricultural production technologies to spill over to rural areas. This helps improve farmers’ quality, promote their professional development, provide management, technology, and talent supports for the agricultural sector’s structure upgrading and green development efficiency, and ultimately enhance AGTFP.

In sum, urbanization has the potential to absorb surplus agricultural labor through the direct and indirect roles that production factor reallocation, capital deepening, and technological spillovers play in agricultural green development. Therefore, the first hypothesis is proposed as follows.

H1: 

Urbanization is conducive to the improvement of AGTFP.

2.3. Concept Disparity between Household Registration and Permanent Residence

It is worth noting that the potential urbanization-induced increase in agricultural productivity and green development is deduced under a series of implicit assumptions. For instance, there are complete and perfect factor markets between urban and rural areas, which allow factors to be reallocated under accurate price signals and at lower transaction costs. Urban and rural residents face relatively equal social security resources, and their mobility is mainly driven by the possible obtainable higher economic returns. Labor transfer and population flow between urban and rural areas are synchronized. This implies that the non-agricultural transfer of the rural labor force refers to the migration and integration of rural population into cities [37]. Nevertheless, the applicability of these presumptions and their logical elaboration warrant additional consideration and differentiation for a sizable nation such as China, which is now undergoing institutional transformation.

Due to the household registration system that stems under the planned economy in China, the transfer of agricultural labor force towards non-agricultural sectors was historically characterized by obvious instability and incompleteness. To begin with, some municipalities utilize the household registration threshold and selectively exclude the inflow of labor force with lower literacy and older ages to relieve the burden of local economic development. This practice leads to an insufficient labor mobility [50]. Besides, the welfare distribution and identity recognition issues endogenic in the household registration system also retard labor migration. Without urban hukou, the newly transferred rural population cannot enjoy identical public services and social security as the urban registered population. Their access to occupational chances, wage levels, and social security benefits is also significantly less than that of the local population, resulting in a new dual structure [51]. Studies have shown that the household registration system severely hinders rural labor mobility [52]. Higher hukou barriers ultimately cause low-skilled, rural, and inter-provincial migrants to return home. Under the current institutional system, rural residents would not readily give up rural estates and land that provide a sense of security unless it is confirmed that they could settle down in a desirable host city [53]. This leads to the deficiency in the supply of land transfer. Therefore, unlike what would be predicted in theory, the amount of rural arable land per laborer does not expand significantly. This supports both the improvement in the relative state of agricultural labor productivity and the promotion in the green development of agriculture. These views can also be demonstrated in the statistics that the urbanization rate of the permanent residents population is 63.89%, while that of the household registered population is only 45.4%. Accordingly, the second hypothesis is put forward as follows.

H2: 

Compared to permanent residence urbanization, household registration urbanization has a much lower enhancement effect on AGTFP.

3. Stylized Facts of Agricultural Green Development Efficiency and Labor Surplus in China

In this section, we first present methodology to measure the efficiency of green development in agriculture. To directly control for undesired outputs and slacks and to unravel changes in efficiency components, we set up an SBM-DDF–Luenberger model which combines Tone’s slack-based measure framework with directional distance function and Luenberger’s productivity decomposition techniques. We then examine stylized facts about China’s AGTFP and agricultural labor surplus after obtaining the estimated measurement results.

3.1. Measurement Methodology for Agricultural Green Development Efficiency

3.1.1. SBM-DDF–Luenberger Model

We deal with K decision-making units (DMUs) with the input, desired output, and undesired output matrices $x=(x_{km})\in R_{K\times M}^{+}$, $y=(y_{kn})\in R_{K\times N}^{+}$, and $b=(b_{kl})\in R_{K\times L}^{+}$, respectively. Then, for each objective DMU k’ and period t, we refer to Fukuyama and Weber (2009) who define the slack-based measure framework with directional distance function (SBM-DDF) model as [54]:

$$\begin{array}{l}{\overrightarrow{S}}_V^t(x^{t,k\prime },y^{t,k\prime },b^{t,k\prime };g^x,g^y,g^b)=\underset{s^x,s^y,s^b,\lambda }{Max}\left\{\frac{\frac{1}{M}{\displaystyle \sum _{m=1}^M\frac{s_m^x}{g_m^x}+\frac{1}{N+L}({\displaystyle \sum _{n=1}^N\frac{s_n^y}{g_n^y}+}{\displaystyle \sum _{l=1}^L\frac{s_l^b}{g_l^b}})}}{2}\right.:\\ x_{k\prime m}^t={\displaystyle \sum _{k=1}^K{\lambda }_k^t{x}_{km}^t+s_m^x}, \forall m;y_{k\prime n}^t={\displaystyle \sum _{k=1}^K{\lambda }_k^t{y}_{kn}^t-s_n^y}, \forall n;b_{k\prime l}^t={\displaystyle \sum _{k=1}^K{\lambda }_k^t{b}_{kl}^t+s_l^b}, \forall l;\\ \hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}{\displaystyle \sum _{k=1}^K{\lambda }_k^t=1,{\lambda }_k^t\ge 0}, \forall k;s_n^x\ge 0, \forall n;s_m^y\ge 0, \forall m;s_l^b\ge 0, \forall l;\left.\begin{array}{l}\\ \\ \end{array}\right\}\end{array}$$

(1)

where directional matrices $g^x=(g_m^x)\in R_M^{+}$, $g^y=(g_n^y)\in R_N^{+}$, and $g^b=(g_l^b)\in R_L^{+}$ represent the direction to reduce inputs, increase desired outputs, and lower undesired outputs, respectively; $g=(g^x,g^y,g^b)$; slack matrices $s^x=(s_m^x)\in R_M^{+}$, $s^y=(s_n^y)\in R_N^{+}$, and $s^b=(s_l^b)\in R_L^{+}$ indicate input redundant levels, desired output deficient levels, and undesired output redundant levels, respectively; weight matrix $\lambda =({\lambda }_k^t)\in R_{K\times T}$ reflects the weight of each DMU in constructing the efficient frontier; ${\overrightarrow{S}}_V$ denote the directional distance under varying returns to scale, and if the constraint that the sum of DMU weights equals 1 is removed from model (1), ${\overrightarrow{S}}_C$ is defined to denote the directional distance under constant returns to scale.

Following Grosskopf (2003) [55], the intertemporal Luenberger productivity index TFP_t^t+1 can be decomposed into changes in four elements, i.e., the pure technological efficiency PTE_t^t+1, pure technological progress PTP_t^t+1, scale efficiency SE_t^t+1, and technological scale TS_t^t+1, where TFP_t^t+1 = PTE_t^t+1 + PTP_t^t+1 + SE_t^t+1 + TS_t^t+1.

$$TF{P}_t^{t+1}=\frac{1}{2}\left\{[{\overrightarrow{S}}_C^t\right.(x^{{}_t},y^{{}_t},b^{{}_t};g)-{\overrightarrow{S}}_C^t(x^{{}_{t+1}},y^{{}_{t+1}},b^{{}_{t+1}};g)]+\left. [{\overrightarrow{S}}_C^{t+1}(x^t,y^t,b^t;g)-{\overrightarrow{S}}_C^{t+1}(x^{t+1},y^{t+1},b^{t+1};g)]\right\}$$

(2)

$$PT{E}_t^{t+1}={\overrightarrow{S}}_V^t(x^t,y^t,b^t;g)-{\overrightarrow{S}}_V^{t+1}(x^{t+1},y^{t+1},b^{t+1};g)$$

(3)

$$PT{P}_t^{t+1}=\frac{1}{2}\left\{[{\overrightarrow{S}}_V^{t+1}(x^t,y^t,b^t;g)-{\overrightarrow{S}}_V^t(x^t,y^t,b^t;g)]+[{\overrightarrow{S}}_V^{t+1}(x^{t+1},y^{t+1},b^{t+1};g)-{\overrightarrow{S}}_V^t(x^{t+1},y^{t+1},b^{t+1};g)]\right\}$$

(4)

$$S{E}_t^{t+1}=[{\overrightarrow{S}}_C^t(x^t,y^t,b^t;g)-{\overrightarrow{S}}_V^t(x^t,y^t,b^t;g)]- [{\overrightarrow{S}}_C^{t+1}(x^{t+1},y^{t+1},b^{t+1};g)-{\overrightarrow{S}}_V^{t+1}(x^{t+1},y^{t+1},b^{t+1};g)]$$

(5)

$$\begin{array}{l}T{S}_t^{t+1}=\frac{1}{2}\left\{[({\overrightarrow{S}}_C^{t+1}(x^t,y^t,b^t;g)-{\overrightarrow{S}}_V^{t+1}(x^t,y^t,b^t;g))-\right. ({\overrightarrow{S}}_C^t(x^t,y^t,b^t;g)-{\overrightarrow{S}}_V^t(x^t,y^t,b^t;g))]+\\ \left.[({\overrightarrow{S}}_C^{t+1}(x^{t+1},y^{t+1},b^{t+1};g)-{\overrightarrow{S}}_V^{t+1}(x^{t+1},y^{t+1},b^{t+1};g))-({\overrightarrow{S}}_C^t(x^{t+1},y^{t+1},b^{t+1};g)-{\overrightarrow{S}}_V^t(x^{t+1},y^{t+1},b^{t+1};g))]\right\}\end{array}$$

(6)

3.1.2. Measurement for Agricultural Inputs and Outputs

As stated in the SBM-DDF–Luenberger model, agricultural outputs involve the desired and undesired outputs. Generally, desired output indicators include the gross output and the added value. Since gross output is influenced by intermediate consumption and cannot reflect the real output, we select agricultural added value to measure agricultural desired output. As to the undesired output, we use agricultural carbon emissions as its proxy due to the fact that agricultural undesired output is mainly reflected in agricultural carbon emissions caused by the use of chemical fertilizers, pesticides, agricultural film, diesel, tillage, and irrigation. We follow Chen et al. (2021) to estimate carbon emissions from agricultural production [10].

Concerning agricultural inputs, we select six factors mostly used in agricultural production, i.e., labor, land, mechanical power, fertilizer, agricultural film, pesticides and agricultural water. Among them, labor input is defined as the population engaged in agriculture, which is indirectly obtained by: the population engaged in the primary industry − the population engaged in forestry − (the population engaged in all pastoral areas + the population engaged in semi-pastoral areas/2) − the population engaged in fisheries. Land input is measured by the total sown area. Mechanical power is determined by the total machinery power used in agriculture, which is computed as the gap between the total machinery power used in the primary industry and the sum of the machinery power used in the forestry, livestock, and fishery industries. Other factor inputs such as fertilizers, pesticides, agricultural film, and agricultural water could be directly obtained from the statistical yearbooks.

For data source, all the original data used in our study come from the statistical yearbooks. Specifically, agricultural added value is taken from the China Rural Statistical Yearbook. The population engaged in the primary industry, the total area sown for crops, the total machinery power used in the primary industry, and the total use of agricultural fertilizers are obtained from the China Statistical Yearbook and statistical yearbooks of each province. The population engaged in the forestry, pastoral, and fishery industries are obtained from the China Forestry Statistical Yearbook, China Animal Husbandry Yearbook, and China Fisheries Statistical Yearbook (or Compilation), respectively. In case of interference from price changes, we base the data on the primary industrial output value index, the farming tool price index, and the chemical fertilizer price index to adjust for agricultural added value, mechanical power, and fertilizer use at a 2001 constant price, respectively.

3.2. Stylized Facts of AGTFP in China

Table 1. AGTFP change and its decomposition between 2001 and 2020 (%).

Years	Green TFP	Green PTE	Green PTP	Green SE	Green TS
2001–2002	4.1546	−2.1041	3.4556	7.3638	−3.9556
2002–2003	2.0360	−1.4523	1.7488	4.3758	−2.4148
2003–2004	5.4656	1.0529	2.0695	5.4727	−2.8652
2004–2005	2.2844	−0.5606	1.2907	3.1433	−1.4631
2005–2006	5.7547	−0.7162	2.2084	8.4686	−3.7603
2006–2007	3.7389	−0.0271	1.3515	3.2771	−0.8159
2007–2008	3.1691	0.6962	1.3177	2.3334	−1.1163
2008–2009	1.7451	1.4560	0.2366	0.2040	0.1397
2009–2010	2.2273	−0.6377	1.8638	3.3924	−2.2145
2010–2011	4.3790	−1.2954	2.5562	5.8305	−2.3426
2011–2012	3.2303	0.0637	0.9072	3.6377	−1.2816
2012–2013	3.8030	1.4227	0.8951	2.2820	−0.7190
2013–2014	4.5656	−0.1234	1.9906	3.9429	−1.2218
2014–2015	4.9408	0.2668	1.8362	3.7097	−0.8696
2015–2016	6.3797	0.5436	2.4322	2.6636	0.7149
2016–2017	7.5238	−0.1482	3.1400	6.4305	−1.6402
2017–2018	5.2458	−0.1242	2.3181	3.2316	−0.1590
2018–2019	6.4366	0.5002	2.6483	3.3279	0.0478
2019–2020	7.2301	0.5723	2.7033	4.0051	0.0972
Average	4.4374	−0.0324	1.9458	4.0575	−1.3600

depicts the change and pertinent decomposition of AGTFP in China, where we can observe the following stylized facts. Firstly, the overall AGTFP in China shows an upward trend, with an annually average growth rate of 4.4374%. Secondly, the pure technological progress and scale efficiency in China’s AGTFP grow rapidly at an average rate of 1.9458% and 4.0575% per year, respectively, while the pure technological efficiency and technological scale decline steadily at an average rate of 0.0324% and 1.36% per year, respectively. A further discussion on the change of AGTFP and its four compositions points to the third stylized fact that AGTFP growth in China during the sample period is mainly driven by the scale efficiency increase and a lesser driving force resides in the growth of pure technological progress. This is possibly due to the gradual realization of mechanized production and operation, as well as the promotion and adoption of new production technologies in China’s agriculture. Contrarily, the pure technological efficiency and technological scale inhibit AGTFP growth, which indicates that under current technological conditions, the desired output brought by the same input factors in China’s agriculture is decreasing year by year.

3.3. Stylized Facts of Agricultural Labor Surplus in China

Based on the labor input redundancies estimated in the SBM-DDF–Luenberger model, Figure 1 and

Table 2. Agricultural labor surplus for each province (or district) in China.

Provinces	2001	2003	2005	2007	2009	2011	2013	2015	2017	2020
Beijing	5.75	5.04	5.11	5.47	5.63	5.15	5.07	4.93	5.22	4.70
Tianjin	5.79	5.81	5.95	6.51	6.32	5.62	5.15	4.94	5.22	5.46
Hebei	135.31	130.37	123.94	125.86	130.31	118.50	113.60	112.33	122.05	130.98
Shanxi	54.86	50.39	50.89	55.97	57.54	56.19	56.80	59.51	64.62	66.36
Inner Mongolia	0.00	0.00	0.00	0.00	0.00	0.00	0.00	0.00	0.00	0.84
Liaoning	42.29	41.66	45.18	49.06	47.18	40.27	39.74	41.21	49.87	55.12
Jilin	22.62	18.12	11.10	17.11	26.41	27.42	23.24	18.34	19.06	24.84
Heilongjiang	0.00	0.00	0.00	0.00	0.00	0.00	0.00	0.00	0.00	0.01
Shanghai	6.76	5.24	0.00	0.00	0.00	0.00	0.00	0.00	0.00	0.00
Jiangsu	156.22	122.94	104.29	96.38	81.64	61.48	56.42	47.65	48.32	52.54
Zhejiang	82.76	69.62	64.05	61.44	57.97	40.97	38.76	37.64	36.32	34.23
Anhui	176.49	152.82	146.06	141.86	136.88	132.97	118.52	110.90	118.40	129.41
Fujian	64.57	61.59	57.83	57.34	55.29	54.51	50.30	50.44	55.66	49.99
Jiangxi	68.85	63.68	64.25	68.51	68.19	61.01	43.24	37.77	53.88	56.41
Shandong	260.95	227.97	199.32	202.31	210.36	189.39	172.67	157.26	154.60	144.70
Henan	332.74	299.94	282.05	270.19	258.71	236.79	223.73	228.63	232.20	227.32
Hubei	138.46	137.17	143.38	153.12	153.52	142.47	130.35	110.44	103.86	102.13
Hunan	183.48	167.47	158.72	159.60	154.34	145.34	140.54	136.91	135.84	135.78
Guangdong	143.21	144.80	139.91	152.91	155.48	132.02	129.31	126.07	129.33	131.37
Guangxi	141.06	134.87	132.56	146.23	152.43	146.65	135.44	129.63	135.39	140.39
Hainan	16.44	14.33	18.10	19.76	20.12	19.65	19.34	20.60	23.07	22.89
Chongqing	81.66	59.00	60.29	56.82	54.58	46.65	32.06	17.38	36.84	32.73
Sichuan	219.82	206.90	203.10	196.95	187.53	160.12	146.08	135.09	137.17	141.25
Guizhou	0.00	0.00	0.00	174.22	133.43	114.73	0.00	0.00	0.00	109.40
Yunnan	166.07	162.56	163.61	169.64	168.68	163.56	150.82	147.43	150.34	90.61
Tibet	3.18	2.38	3.55	3.90	7.14	7.39	2.40	3.29	2.27	3.01
Shaanxi	91.20	90.72	87.03	90.13	85.12	75.13	69.48	71.29	77.45	82.98
Gansu	74.57	70.04	70.71	76.50	84.03	76.40	77.23	73.88	78.21	79.75
Qinghai	9.79	8.07	0.00	4.97	0.00	0.00	0.00	0.00	0.00	0.00
Ningxia	7.03	6.44	4.85	5.71	3.42	6.36	6.85	6.50	8.45	9.12
Xinjiang	16.28	12.90	14.98	16.58	16.37	17.59	21.99	19.68	6.71	0.00
Total	2708	2473	2361	2585	2519	2284	2009	1910	1990	2064

illustrate the quo status of agricultural labor surplus in China from the perspective of evolutionary trends and provincial differences, respectively, where the following stylized facts are found. First off, the number of agricultural surplus laborers exhibits a downward trend and a prominent phasal characteristic on the whole. However, a slowly rising tendency has also been seen since 2015. There are 20.64 million agricultural surplus laborers in 2020 countrywide, an increase of nearly 1.55 million since 2015, indicating severe redundancy in agricultural labor input. Second, more than one million redundancies are seen in provinces such as Henan, Shandong, Sichuan, Guangxi, Hunan, Guangdong, Hubei and Anhui, among which Henan has the largest redundancy, reaching an approximately 2.27 million agricultural surplus laborers. Third, in terms of geographical differences, the eastern region has the largest amount of labor surplus, followed in sequence by the central and western regions. As to distinctions of agricultural functional areas, the redundancy in the main grain-producing areas is far greater than that in the main selling areas and the balanced areas. Additionally, significant agricultural labor surplus differences are not only shown among varying geographical regions and agricultural functional areas, but also exist within the same geographical region and agricultural functional area.

4. Empirical Analysis: Relationship between Urbanization and Agricultural Green Development Efficiency

This section is to examine empirically the causality between urbanization and agricultural green development efficiency. We firstly put forward the empirical model, and then elaborate on the pertinent variables and datasets used. After obtaining the estimated baseline relationship, we further conduct robustness and endogeneity tests to confirm its robustness and causality.

4.1. Empirical Model

Our study employs a panel data model with two-way fixed effects, i.e., fixed effects at the provincial and temporal levels, to disclose how urbanization impacts agricultural green development efficiency. The model can be written as

$$lnAGTF{P}_{it}={\beta }_0+{\beta }_{1}URBA{N}_{it}+{\displaystyle \sum Control{s}_{it}}+RE{G}_i+YEA{R}_t+{\epsilon }_{it}$$

(7)

where i and t denote the ith province (or district) and the tth year, respectively; lnAGTFP, URBAN, and Controls represent agricultural green development efficiency, urbanization degree, and the control variable set, respectively; REG and YEAR indicate unobserved fixed effect at the provincial and temporal levels, respectively; and ε denotes the disturbance term.

4.2. Variable Definitions and Data Description

4.2.1. The Dependent Variable

The dependent variable of this paper is agricultural green development efficiency (lnAGTFP), which is defined as AGTFP and estimated by the aforementioned SBM-DDF–Luenberger model. Since estimations from the SBM-DDF–Luenberger model are dynamic indices, we calculate the accumulation levels of these estimations with 2001 as the base year and use their natural logarithm to measure lnAGTFP.

4.2.2. The Independent Variable

This paper regards urbanization as the main independent variable. In accordance with the research aim, we differentiate urbanization into household registration urbanization (UH), which is measured by the share of urban household registered population in the total population, and permanent residence urbanization (UP), which is measured by the proportion of the urban permanent residents population related to the total population.

4.2.3. Control Variables

To control for the additional influential factors of AGTFP, we include a bundle of control variables that reflect the provincial agriculture per se and the provincial macroeconomic status [13,56]. These variables consist of: (i) land quality (LAQA), defined as the ratio of effectively irrigated areas to total sown areas; (ii) agricultural mechanization (lnMECH), measured by the natural logarithm of the number of agricultural mechanized equipment per capita; (iii) rural traffic convenience (lnTRAF), defined by the regional road mileages after excluding expressways, first-class highways, and second-class highways as a percentage of the regional land area; (iv) intra-primary industrial structure (INST), expressed as the output share of a narrowly defined agriculture within the primary sector, and thus regarded as a scaling for the structural change of distinct agricultural types within the broadly defined agriculture; (v) industrialization (IND), measured by the share of the secondary industry in GDP; (vi) economic development (lnECON), defined as the logarithm of provincial GDP per capita; (vii) financial development (FIN), measured by the ratio of provincial year-end loan balance to GDP; (viii) government activities (GOV), quantified by the share of local fiscal expenditure in GDP; (ix) foreign capital utilization (FDI), measured by the ratio of foreign direct investment to GDP; (x) openness (OPEN), measured by the total imports and exports as a share of GDP; (xi) education level (EDU), measured by provincial high school students per 100 persons.

4.2.4. Data Description

In view of data availability, this paper draws on a Chinese provincial panel of data from 2001 to 2020 to assess the impact of urbanization on agricultural green development efficiency. All raw data for the above variables are obtained from the China Population and Employment Statistical Yearbook, China Rural Statistical Yearbook, and China Statistical Yearbook.

Table 3. Variable descriptive statistics.

Variables	N	Mean	S.D.	Median	Min	Max
lnAGTFP	620	0.3290	0.2404	0.2905	−0.1198	1.0507
UH	620	0.3778	0.1694	0.3452	0.1429	0.9666
UP	620	0.5060	0.1541	0.4968	0.2085	0.8960
LAQA	620	0.5693	0.2283	0.6195	0.2034	1.0000
lnMECH	620	1.1265	0.7160	1.1099	−0.9685	3.7147
lnTRAF	620	3.8829	0.7618	3.9352	0.4788	6.0530
INST	620	0.5856	0.1151	0.5759	0.3837	0.9044
IND	620	44.7547	8.5964	46.0000	16.1600	60.1300
lnECON	620	10.1789	0.8199	10.3284	8.0064	12.0090
FIN	620	1.1829	0.4220	1.1018	0.5329	3.0829
GOV	620	0.2357	0.1822	0.1925	0.0767	1.3792
FDI	620	0.0231	0.0203	0.0172	0.0001	0.1465
OPEN	620	0.3007	0.3697	0.1304	0.0127	1.7778
EDU	620	2.9026	0.8495	2.9840	0.8745	4.9308

reports the variable descriptive statistics.

4.3. Empirical Results

4.3.1. Baseline Estimations

Table 4. Baseline estimations for the relationship between urbanization and agricultural green development efficiency.

Variables	(1)	(2)	(3)	(4)	(5)	(6)
UH	0.3176 *** (4.72)	0.2981 *** (4.06)
UP			0.7980 *** (9.63)	0.8823 *** (9.01)
HR					0.0793 *** (2.97)	0.0445 ** (2.34)
Controls	NO	YES	NO	YES	NO	YES
Year FEs	YES	YES	YES	YES	YES	YES
Province FEs	YES	YES	YES	YES	YES	YES
N	620	620	620	620	620	620
R2	0.9182	0.9357	0.9310	0.9426	0.9023	0.9074

Note: Figures in parentheses are robust-adjusted t statistics. ***, ** denote significance at the 1% and 5% confidence levels, respectively.

displays the effects of household registration urbanization (UH) and permanent residence urbanization (UP) on AGTFP. It can be seen that the coefficients for the independent variables UH and UP are significantly positive at the 1% confidence level regardless of whether control variables are included or not, indicating that both household registration urbanization and permanent residence urbanization are favorable to the improvement of AGTFP. This supports the theoretical inference that urbanization would promote agricultural green development efficiency by transferring agricultural surplus laborers with low marginal output outwards [57], reducing ineffective inputs in agricultural production [58], and encouraging large-scale agricultural production and green operation [59]. Therefore, the hypothesis H1 is confirmed.

Moreover, a further comparison between the coefficient for UH in column (2) (or column (1)) and the coefficient for UP in column (4) (or column (3)) shows us that household registration urbanization has a much lower promotion effect on AGTFP than permanent residence urbanization does. This verifies the hypothesis H2.

To unravel the potential logics behind the hypothesis H2, we construct a variable named household registration control (HR) and calculated as the ratio of urban household registered population to permanent residence population to explain the difference between the impact of household registration urbanization on AGTFP and the impact of permanent residence urbanization on AGTFP. Since household registration control reflects the average number of persons holding a local household registration compared to the permanent residents population, it follows that, the smaller the indicator, the fewer permanent residents that hold household registration, and thereby the stricter the household registration control.

The estimated results are reported in columns (5) and (6) of Table 4, where we observe significantly positive coefficients for HR. That is, there is a significantly positive relationship between household registration control and AGTFP. The looser the household registration control, the more AGTFP grows; conversely, the stricter the household registration system, the more detrimental it is to agricultural green development efficiency. This implies that the household registration control is a major cause of the disparity in the impact of household registration urbanization and permanent residence urbanization on agricultural development.

This finding is practically intuitive. Although China has advanced gradual reforms in the household registration system and the role of the household registration system in directly controlling population mobility has also been weakened, it is hard in the short run to eliminate various thresholds and core welfare systems endogenic in the household registration system [60]. A big gap still exists between the rural migrant labor force and local urban residents in job opportunities, wages, and social security benefits [19,61]. The reform lag not only hinders the process of urbanization, but also lowers rural laborers’ long-run expectation on urban-rural transfer, causing distortions between the urban and rural labor markets, as well as unbalanced development among different sectors [62]. This eventually leads to indirect losses in AGTFP.

4.3.2. Robustness Check

Since non-random values and outliers of the urbanization and AGTFP variables may lower reliability of the baseline estimations, we need to conduct robustness tests to examine the robustness of the explanatory power of evaluation methods and indicators. Robustness tests are generally processed by changing settings of a specific parameter and conducting repeated experiments to see whether the empirical results differ from the baseline findings. In this part, we consider two robustness tests which involve changes in the sample periods and in variable values. Detailed information is as follows.

The first robustness test is to remove pre-2005 sample data and reexamine the impact of household registration urbanization and permanent residence urbanization on AGTFP. This is because China’s permanent resident census was firstly initiated in 2005. Although existing studies have deployed various techniques to supplement the missing data and enlarge the sample size, this practice may bias estimations. We present the corresponding estimated results in columns (1) and (2) of

Table 5. Robustness check.

Variables	(1)	(2)	(3)	(4)	(5)	(6)
UH	0.4792 *** (5.73)		0.2977 *** (4.23)
UP		0.6763 *** (3.41)		0.8428 *** (8.91)
L.UH					0.3569 *** (4.31)
L.UP						0.8930 *** (9.62)
Controls	YES	YES	YES	YES	YES	YES
Year FEs	YES	YES	YES	YES	YES	YES
Province FEs	YES	YES	YES	YES	YES	YES
N	465	465	620	620	589	589
R2	0.9029	0.8981	0.9377	0.9464	0.9337	0.9369

Note: Figures in parentheses are robust-adjusted t statistics. *** denote significance at the 1% confidence levels, respectively.

. The second robustness test is to winsorize continuous variables by 1% on both ends to eliminate potential outliers. The estimations are displayed in columns (3) and (4) of Table 5. The third robustness test is to lag independent variables and control variables by one period to adjust for non-synchronization of the baseline relationship. The estimated results are reported in columns (5) and (6) of Table 5.

It is clear that all the coefficients for UH, UP, L.UH, and L.UP remain quantitatively similar to the baseline results. This confirms the robustness of the findings that urbanization positively impacts AGTFP and that household registration urbanization has a much weaker promotion effect.

4.3.3. Endogeneity Analysis

The baseline estimations find that both household registration urbanization and permanent residence urbanization significantly promote AGTFP. One concern is that these findings may suffer from reverse causality. This is because a higher AGTFP often signals a better developed agriculture and a more efficient allocation of agricultural labor resources, both of which encourage surplus laborers to flow to cities and thereby promotes the process of urbanization. To eliminate such a potential reverse causality, we perform two endogeneity tests.

The first endogeneity test involves a two-stage least squared (2SLS) method for instrumental variable estimation. Since the least squared (OLS) estimation holds under a prerequisite assumption that the dependent variables are uncorrelated with the error terms, estimations would be biased once the dependent variables correlate with the error terms. This issue can be addressed with the 2SLS method by performing two regressions using instrumental variables. Specifically, the first stage regression model is to regress the instrumental variables on the endogenous dependent variables, i.e., $IV\stackrel{OLS}{\to }x$, to obtain the fitted values $\widehat{x}$. The second stage regression is to regress the fitted values of the first stage regression on the dependent variables, i.e., $y\stackrel{OLS}{\to }\widehat{x}$.

We lag the variables of household registration urbanization and permanent residence urbanization by two periods to construct instrumental variables. This practice is common and reasonable since urbanization two periods ago is closely related with urbanization today and is unaffected by current AGTFP growth.

Table 6. Endogeneity analysis on reverse causality: 2SLS method.

Variables	(1)	(2)	(3)	(4)
	UH	LNAGTFP	UP	LNAGTFP
UH		0.4458 *** (4.46)
UP				1.5612 *** (8.62)
L2.UH	0.9160 *** (28.78)
L2.UP			0.5597 *** (17.89)
Controls	YES	YES	YES	YES
Year FEs	YES	YES	YES	YES
Province FEs	YES	YES	YES	YES
Under-Identification Test		325.11 [0.0000]		210.34 [0.0000]
Weak-Identification Test		882.15		349.20
N	558	558	558	558
R2	0.9306	0.9162	0.9477	0.9238

Note: Figures in parentheses and brackets are robust-adjusted t statistics and p values, respectively. *** denote significance at the 1% confidence levels, respectively.

reports the results, where columns (1) and (3) show the first-stage estimations, and columns (2) and (4) display the second-stage estimations.

Both coefficients for L2.UH and L2.UP are significantly positive at the 1% confidence level, indicating that each instrumental variable has a considerable explanatory power for the corresponding urbanization variable. This can be further confirmed by the under-identification test, which shows that the Kleibergen-Paap rk LM statistics have p values of 0.0000, and the weak-identification test, which shows that the Cragg-Donald Wald F statistics are larger than the 5% Stock–ogo critical value. In this sense, the significantly positive coefficients for UH and UP, with the former being much smaller than the latter, demonstrate that the documented baseline relationship is not interfered with by reverse causality and is thus credible.

Although we have introduced in the above analysis lagged urbanization as instrumental variables to attenuate the reverse causality problem, the potential omitted variable bias may also cause an endogeneity problem. To overcome this issue, we conduct another endogeneity analysis by utilizing the new urbanization pilot policy implemented in China in 2014 to construct a quasi-natural experiment as a proxy for urbanization. A commonly used policy evaluation method—the difference-in-difference (DID) technique—can be then employed to assess the impact of urbanization on AGTFP. It is implemented by comparing the difference between the control and treatment groups before and after the implementation of the policy [35]. Equation (8) shows a standard DID setting.

$$Y={\beta }_0+{\beta }_{1}du+{\beta }_{2}dt+{\beta }_{3}du\times dt+\epsilon$$

(8)

where du is the grouping dummy variable, which is set to distinguish whether a city has implemented the policy. If the answer is yes, the specified city is classified into the treatment group and du is assigned the value 1; otherwise, it is classified into the control group and du is assigned the value 0. dt is the policy implementation time dummy variable which is set to identify whether a specified time is before or after policy implementation. If the time is before policy implementation, dt takes the value of 0; if the time is after policy implementation, dt takes the value of 1. du × dt is the interaction term of the grouping dummy variable and the policy implementation time dummy, whose coefficient β₃ reflects the policy effect.

The new urbanization pilot policy is a nationwide policy pilot project that aims to explore an effective path for realizing urbanization mode transformation and promoting coordinated economic development between the urban and rural areas. In March 2014, the National Development and Reform Commission of China released the National New Urbanization Plan (2014–2020), in which 62 cities in 25 Chinese provinces were designated as pilot areas [63]. This lays a solid foundation for us to construct a quasi-natural experiment. Firstly, a grouping dummy du is set to distinguish whether a province is a pilot or not. If the answer is yes, the specified province is classified into the treatment group and du is assigned the value 1; otherwise, it is classified into the control group and du is assigned the value 0. It is noteworthy that there are two common ways to determine a pilot province in the literature. One way is that as long as a specified province has one officially designated pilot city, it is recognized as a pilot province. The other way is to compute the ratio of the officially designated pilot cities’ permanent residents population in a given province in the first year after the new urbanization pilot policy was announced to the province’s total population. That is, if the ratio in 2015 is greater than 10%, the specified province is regarded as pilot and thus a member of the treatment group. Secondly, a time dummy dt is set to identify whether a specified year is before or after policy implementation. Since the new urbanization pilot policy and its official implementation plan were announced at the end of 2014, we regard years after 2015 as the policy implementation period and assign the value 1 to dt. For years before 2015, dt is assigned 0. Then, the DID model can be written as

$$LNAGTF{P}_{it}={\beta }_0+{\beta }_{1}d{u}_i\times d{t}_t+{\displaystyle \sum \mathrm{Con}trol{s}_{it}}+RE{G}_i+YEA{R}_t+{\epsilon }_{it}$$

(9)

where du × dt is the new urbanization pilot policy variable and used as an exogeneity proxy variable for urbanization. The coefficient β₁ is our focus.

Table 7. Endogeneity analysis: DID method.

Variables	(1)	(2)
du × dt	0.0694 *** (4.72)	0.0756 *** (6.06)
Controls	YES	YES
Year FEs	YES	YES
Province FEs	YES	YES
N	620	620
R2	0.9476	0.9279

Note: Column (1) displays the result for the first measurement way of du. That is, as long as a specified province has one officially designated pilot city, it is recognized as a pilot province. Jiangsu, Anhui, Zhejiang, Liaoning, Shandong, Hebei, Chongqing, Beijing, Tianjin, Jilin, Heilongjiang, Shanghai, Fujian, Jiangxi, Henan, Hubei, Hunan, Guangdong, Guangxi, Sichuan, Guizhou, Yunnan, Gansu, Qinghai, and Ningxia are among the 25 provinces that are considered pilot. Column (2) displays the result for the second measurement way of du. That is, if the ratio of the officially designated pilot cities’ permanent residents population in a given province in 2015 to the province’s total population is greater than 10%, the specified province is regarded as pilot and thus a member of the treatment group. Figures in parentheses are robust-adjusted t statistics. *** denote significance at the 1% confidence levels, respectively.

reports the estimated results, where columns (1) and (2) display results for the two measurement ways of du, respectively. Obviously, both coefficients for du × dt are significantly positive, implying that urbanization can promote AGTFP. This confirms once again the robustness of the causality between urbanization and agricultural green development efficiency.

5. Further Discussion: Effect of Employment Urbanization on Agricultural Green Development Efficiency

The findings above show us that urbanization is an important means of transferring redundant agricultural labor force and promoting agricultural green development in China. A further dismantled control over household registration and an intensified reform of the household registration system are critical steps toward effectively removing barriers to the flow of redundant agricultural laborers and toward bringing urbanization into full play in promoting green agricultural development efficiency. The household registration system, though, is a fundamental system that affects all facets of society. The reform will be practically difficult, complicated, and prolonged. In this context, this section focuses on how to minimize the deterrence effect of the household registration system on agricultural surplus labor mobility, so that urbanization could play its role in improving green agricultural development efficiency.

Urbanization is a people-centered strategy that seeks to enhance the welfare of the entire society. The key for urbanization to play its role in promoting green agricultural development efficiency lies not just in the conversion of agricultural surplus labor force into urban household registered population, but also in the absorption of this migrant population and in the strengthening of the sustainability of the transfer of surplus agricultural labor. Some studies, however, have shown that the permanent migration of the rural population is not obvious during urbanization [64,65]. In recent years, China’s rural labor backflow phenomenon has become more apparent, and the patterns of the flow are changing [51,66]. In fact, it has been found that only a small portion of migrant workers (“Migrant workers” are a special social group that emerged in China following the reform and opening up and originates in China’s unique household registration system and land system. These workers are generally farmers engaged in non-agricultural work. They maintain rural hukou and properties while working and living primarily in cities, but they are not entitled to the same public services as city dwellers.) from rural areas initially could find suitable jobs in cities [67], and that increasing numbers of migrant rural laborers frequently switch between rural and urban areas due to the unstable employment or their inability to find work [27]. According to a “Survey and Monitoring Report on Migrant Workers” released by the National Bureau of Statistics in 2021, there are 292.51 million migrant workers in China overall, but only 41.5% of those who live in cities believe they are locals (Details on the 2021 “Survey and Monitoring Report on Migrant Workers” can be found at the site of the National Bureau of Statistics: http://www.stats.gov.cn/xxgk/sjfb/zxfb2020/202204/t20220429_1830139.html (accessed on 11 August 2022). Those who have not signed labor contracts or have been hired on a flexible schedule have a substantially lower sense of urban belonging than those who do have steady employment.

To tackle the difficulties in the reform of the household registration system and quicken migrant citizenization, it is necessary during urbanization to create more non-agricultural work opportunities for agricultural surplus laborers, improve the stability of their non-agricultural employment transfer, and satisfy their basic material and emotional needs in urban life. In line of this logic, we define a term called employment urbanization and measure it by the ratio of urban non-agricultural employment to the total employment within a province to explore how urbanization for employment affects agricultural green development efficiency. Since non-agricultural employment includes employment in both the second and tertiary industries, we distinguish employment urbanization (EU) into employment urbanization of the secondary industry (SEU) and employment urbanization of the tertiary industry (TEU) to assess their discrepant effects on agricultural green development efficiency.

Table 8. The effect of employment urbanization on AGTFP.

Variables	(1)	(2)	(3)	(4)
SEU	0.1523 *** (3.82)	0.2520 *** (3.61)
TEU			0.3623 *** (3.94)	0.4188 *** (3.31)
Controls	NO	YES	NO	YES
Year FEs	YES	YES	YES	YES
Province FEs	YES	YES	YES	YES
N	620	620	620	620
R2	0.9174	0.9362	0.9155	0.9338

Note: Figures in parentheses are robust-adjusted t statistics. *** denote significance at the 1% confidence levels, respectively.

displays the estimated effects of employment urbanization on AGTFP. We find significantly positive coefficients for both SEU and TEU, with the latter’s coefficients being much larger than the former. This suggests that employment urbanization in both the secondary industry and the tertiary industry significantly improves agricultural green development efficiency, and that employment urbanization of the tertiary industry has a more pronounced promotion effect. To put it another way, encouraging agricultural surplus laborers to transfer to the tertiary industry and realizing urbanization for employment in the tertiary industry are more advantageous in enhancing the stability and durability of agricultural surplus labor transfer than urbanization for employment in the secondary industry. This makes sense in practice. Migrant workers in China are primarily employed in secondary industries such as manufacturing and construction [68], and in tertiary industries such as wholesale and retail trade, transportation, storage, and postal services, lodging and catering services, and residential services and the repair industry [69]. With the rapid development of the urban economy and the constant upgrading of the industrial structure, the service industry requires an increasing number of workers. New-generation rural immigrants are also more willing to work in tertiary industries such as couriers and housekeepers, which are decent, laid-back, and well-paid, rather than in secondary industries such as processing and construction, where working conditions are harsher and work contents are more monotonous. This eventually leads to the fact that employment urbanization of the tertiary industry promotes agricultural green development efficiency more than the employment urbanization of the secondary industry.

6. Concluding Remarks

Agricultural labor surplus is a major impediment to the green development in China’s agriculture, and urbanization may be a potential way to tackle this problem. However, China’s urbanization has a strong policy overtone, as well as heterogenous features due to the disparity in the household registered population and the permanent residence population. So, what on earth would urbanization in China mean for agricultural green development efficiency? Based on Chinese provincial panel data from 2001 to 2020, this paper sets up a SBM-DDF–Luenberger model to measure AGTFP, and unveils the impact of heterogenous urbanization on agricultural green development efficiency. The following findings are included. Firstly, China’s AGTFP grew at an average annual rate of 4.4374% between 2001 and 2020. This growth was mainly achieved through improved scale efficiency, with increased pure technological progress serving as the secondary driving force. Secondly, China’s agricultural sector had a large surplus of laborers, with an estimated 20.64 million in 2020. Meanwhile, great discrepancies were observed in the labor surplus across regions. The labor surplus in the eastern was greater than in the central and western regions, and the labor surplus in the main grain-producing areas is greater than in the main selling and balanced areas. Thirdly, both household registration urbanization and permanent residence urbanization benefit the promotion of agricultural green development efficiency. Due to the hindrance effect of the household registration system on the migration of agricultural surplus laborers toward cities, household registration urbanization has a reduced promotion effect on AGTFP. Finally, employment urbanization provides an economic foundation for preventing the backflow of agricultural surplus laborers and hastening the process of migrant workers becoming citizens. Employment urbanization in both the secondary and tertiary industries promote agricultural green development by improving the durability and stability of the transfer of agricultural surplus labor force into cities, though the latter has a much more pronounced promoting effect.

Our study has important policy implications for improving the efficacy of agricultural green development and perfecting measures for urbanization development.

To begin with, it is necessary to acknowledge that there are still a large number of surplus laborers in China’s agriculture. Promoting non-agricultural employment for those surplus laborers is key to realize green development in agriculture. Meanwhile, regional heterogeneity should be taken into account when developing strategies to encourage the transfer of agricultural surplus laborers. The eastern regions and the main grain-producing regions should receive special attention in the future when addressing the labor surplus.

Secondly, urbanization has been confirmed in this study to be an effective means to absorb agricultural surplus labor force and foster agricultural green development efficiency. The reform of the household registration system should be continuously promoted in order to accelerate the stripping of the fringe benefits residing in the system and eliminate the hidden shackles of the system on labor mobility between urban and rural areas. In the meantime, governments should actively promote urban high-quality development that places the permanent residence population at the center of social governance and regards the citizenization of migrant agricultural labor force as the core aim. Inclusive social policies concerning welfare distribution, social employment, and humanistic care should be promptly designed and carried out to enhance the absorption, acceptance, and attraction effects of urbanization on the transfer of agricultural workers.

Lastly, another important way to enhance the longevity and stability of the transfer of agricultural surplus laborers involves actively improving the employment service system and encouraging those agricultural workers to work in cities and towns, particularly in the tertiary industry. The government should spare no effort to develop emerging industries and create more job opportunities and employment channels for the transferred agricultural labor force to settle in cities. Investment in rural human capital should also be raised, as should employment guidance, to ensure a rational transfer scale and to encourage innovation and entrepreneurship. All of these practices would help the surplus labor force better adapt to the requirements of economic structural transformation and upgrading, and eventually lead to the realization of agricultural green development.

Although our work advances understandings on the impact of heterogeneous urbanization on agricultural green development efficiency, some limitations should be noted and tackled in future endeavors. First, given that carbon emissions are key contributors to climate warming we mainly choose agricultural carbon emissions as the non-desired output indicator to measure AGTFP. In future studies, we would consider more agricultural pollutants as the non-desired output indicators. Second, this paper uses a Chinese provincial panel dataset due to the data unavailability of major indicators. Future studies would be further polished by collecting city-level samples. Finally, since labor force has a significant mobility characteristic, the spatial effect of employment urbanization on AGTFP deserves further exploration. Despite the limitations, we expect that this paper is enlightening in providing theoretical and practical references for how to employ urbanization to achieve agricultural green development.