Can Livestock Raising Alleviate Farmland Abandonment?—Evidence from China

Abstract

Farmland abandonment is a global phenomenon. Changes in socioeconomic factors in China impact the traditional crop–livestock system; however, studies on the relationship between livestock raising and farmland abandonment are insufficient. This study used the farmer behavior decision-making model to analyze the impact of livestock raising on farmland abandonment and its mechanism. Based on 6707 samples from the 2016 database of the China Labor-force Dynamic Survey, the Logit and Tobit models were used to empirically analyze the relationship between livestock raising and farmland abandonment at the national level and different terrains (plain, hill and mountain) in China. The results showed farmland abandonment in 15.63% of rural households, and a farmland abandonment ratio of 6.24%. The spatial distribution of farmland abandonment was high in the south and low in the north. Livestock raising households accounted for 9.45%, and the influence coefficient of livestock raising on farmland abandonment was negative but not significant. Livestock raising would significantly increase the ratio of rural households with farmland abandonment by 3.9% and 10% in plain areas and hilly areas, respectively, and decrease the ratio in mountain areas by 11.4%. The abandonment ratio due to livestock raising increased by 21.46% in hilly areas and decrease by 41% in mountain areas. For every 1% increase in livestock scale, the ratio of households with farmland abandonment in plain and hilly areas increased by 0.05% and 0.07%, respectively, and in mountain areas decreased by 0.09%. The abandonment ratio in hilly areas increased by 0.02% and in mountain areas decreased by 0.05%. The effects of raising livestock on farmland abandonment differed across terrains and thus require different measures for alleviating. Plain areas and hilly areas could combine livestock raising and crop planting between different households to improve farmland production capacity, and mountain areas could moderately develop livestock raising to alleviate farmland abandonment.

1. Introduction

Farmland abandonment can be explained as the reduction or cessation of agricultural management activities on special land [1,2,3]. This phenomenon has become a problem in many developed and developing countries worldwide [1,4,5,6]. Farmland was abandoned during the period of economic growth after the 1950s in developed countries, mostly Europe, the United States, Australia, and Japan [4,6,7,8]. Farmland abandonment was first noted in China from the late1980s to the early 1990s [9], and it gradually became widespread after the mid-1990s [9,10]. Abandonment has received substantial attention from researchers and policy makers due to its impact on the environment, food security, and socioeconomic factors, particularly in marginal areas such as hilly and mountainous areas [11,12,13,14,15,16]. Li and Li [17] found that the ratio of farmland abandonment was 14.32% in Chinese mountainous regions through a questionnaire survey; Zhang et al. [10] found that the spatial distribution of farmland abandonment is resembles a “T” rotated 90°. The Chinese law for land management and rural land contract prohibit anyone from abandoning farmland. According to the “Guiding Opinions on the Overall Planning and Utilization of Abandoned Land to Promote the Development of Agricultural Production” from 2019, the administrative department should take active measures to alleviate farmland abandonment. Furthermore, local governments, such as the provincial governments of Sichuan and Chongqing and county government of Fushun in Sichuan, Hualong in Qinghai, and Liulin in Shanxi, are concerned about the issue and have formulated relevant policies.

Farmland abandonment is an extreme outcome of marginalization, the essence of which is the continuous decline in farmland economic importance [18,19,20,21,22]. Rural labor migration induced by urbanization and industrialization leads to farmland marginalization and abandonment [9,23,24,25]. The value of farmland assets keeps falling during marginalization at the household level [26,27,28], which leads to family livelihood strategies being adjusted and agricultural management activities being reduced or even abandoned [13,28,29,30]. Smallholder farming systems are an important part of Chinese agriculture [31,32]. According to the “Third National Agricultural Census” in China, smallholders accounted for more than 98% of agricultural operators, operating and managing 70% of the farmland in China, and smallholder practitioners accounted for 90% of agricultural practitioners. Furthermore, smallholder farming systems are the main form of livestock farms [30,31,33,34], and small and medium-scale livestock raising (LR) currently play a key role in the supply of animal products in China [33,35]. The rural household crop–livestock mixed agricultural system is an important agricultural practice in rural China [33,36,37]. Traditional small and medium-scale LR typically needs to match the corresponding farmland scale at the household level [31,33,36,37,38]. Crop planting (CP) can provide feed for LR, and livestock can provide animal power and manure for CP, thereby raising land production capacity of land and income without increasing additional input [31,33,34,36]. The demand and economic benefits from livestock products are important driving factors affecting the development of the livestock industry [33,39]. With economic development and an increase in income, people eat more animal products to obtain protein and improve their diet [34,40,41], and traditional crop–livestock systems are being transformed into specialized and industrial crop and livestock farming systems [33,36,39,40]. Therefore, restructuring rural production factors caused by China’s socioeconomic transition leads to system decoupling to some extent at the household level [33,36,37,39,42] and rural households transforming into different types [13,33], particularly in flat and high-quality agricultural areas. The relationship between CP and LR needs to be examined, particularly, in the vast territory of China, as well as in different regions. In the context of the changing structure and function of the traditional crop–livestock systems, it is necessary to examine whether LR can continue to improve the land production capacity. Studies should also examine whether raising livestock affects households’ land resource allocation and the related spatial differences.

The crop–livestock system is transitioning from smallholders to modern agriculture [33,36], and the impact of LR on land use is changing. Farmland abandonment occurs owing to land use change [4,15], and there are different opinions about the impact of LR on farmland abandonment. On one hand, some studies argue that LR aggravates farmland abandonment due to the time squeeze of the CP labor force [33,39]. On the other hand, some studies argue that LR can reduce the cost of corn planting and increase the additional value of farmland, thus alleviating farmland abandonment [4,13]. However, the current information about the influence of LR on abandonment is extremely brief, and is largely qualitative with a lack of quantitative analysis. Moreover, promoting the use of farmland through industrial integration driven by LR is an important part of China’s rural revitalization strategy. Theoretical analysis, experience summary, and empirical study can reveal whether continuous promotion of LR effectively can alleviate farmland abandonment and the relevant mechanisms. Therefore, this study used large sample data from the 2016 China Labor-force Dynamic Survey (CLDS) as a study sample to analyze the impact of LR on farmland abandonment and its mechanism. We focused on the impact of LR on farmer abandonment behavior and abandonment extent to provide policy suggestions for farmland protection and rural development.

2. Materials and Methods

2.1. Data Sourcing

As the main participant in agricultural activities, rural households and their production decisions have a strong impact on land use [13,19]. Rural households adjusting their land use behavior are the most important direct inducers of farmland abandonment [19]. Therefore, this study focused on farmland abandonment and decision-making behavior at the household level. The data for this study were obtained from the 2016 CLDS [43], which was first collected and sorted by the Center for Social Science Survey of Sun Yat-sen University in 2012, and every two years subsequently. The database is directed toward the labor force population, covering aspects such as household production, agricultural labor migration, and land use. To ensure data representativeness, the 2016 CLDS database includes 27 provinces across the country (except Hong Kong, Macao, Taiwan, Shanghai, Tibet, Qinghai, and Hainan), and the farmland in these 27 provinces accounts for more than 98% of the total farmland land in the country (Figure 1). Therefore, CLDS samples can be considered representative national rural land surveys. The data used in this study were surveyed in 2016 and represent 401 villages, 14,226 households, and 21,086 individuals. This study focused on farmland abandonment and LR in rural areas. To reduce the statistical error from missing indicates, a series of operations was applied to clean the data. First, 5952 urban household samples were excluded. Second, 1511 rural households that lacked farmland were eliminated. Third, 56 rural households that lacked other key indicators, such as head age, were also eliminated. Finally, there were 6707 valid sample households (Figure 1).

2.2. Theoretical Hypothesis

Neoclassical economics postulate that rural households are rational, and their production decisions regarding labor and agricultural arrangements stem from the pursuit of maximizing household benefits [20,44]. Labor force arrangement is at the core of family production decision-making. Farmers compare the differences between various industries or departments, allocate labor force to maximize their possible benefits, and arrange family agricultural production and land use. These factors are at the core of the household’s decisions [19,29,42,45]. Moreover, rural household resource allocation can also be influenced by multiple factors such as farmers’ emotional dependence and socioeconomic status and government policy support.

Theoretically, well-functioning crop–livestock systems contribute to cost savings and increased revenue because the cost of input factors in the system is lower than in the market. Plant grain and straw can provide feed for LR and reduce the use of industrial feed and LR costs. LR can provide animal power in the cultivation process and reduce labor input, and livestock manure provides necessary nutrition for crop growth, rather than chemical fertilizer. LR improves farmland comprehensive production capacity, increases household income and the opportunity cost of off-farm employment, and alleviates labor migration and farmland abandonment. However, with the development of the market economy, rural households in different regions have gradually evolved into different types [13,16,33,46]. If the market purchase cost of production input factors is lower than the internal cost of the crop–livestock system, the traditional LR of smallholder system transforms into specialized and large-scale farming, CP time is occupied, and some people are engaged full-time in raising, completely abandoning CP. In this case, the crop–livestock system will be decoupled, and abandonment will occur if the farmland transfer market is defective. According to existing empirical studies, the decoupling of the crop–livestock system is occurring in China; however, it is not comprehensive. Therefore, scholars have not formed a unified opinion on the relationship between LR and farmland abandonment [31,33,35,36,39,41,42], and the existing studies ignore the possible heterogeneity of LR impact on farmland abandonment.

Based on theoretical analysis and relevant empirical studies, this study proposes the following hypotheses regarding the relationship between LR and farmland abandonment.

Hypothesis 1 (H1).

LR alleviates rural household farmland abandonment and has a significant negative impact on household farmland abandonment.

Farmland abandonment relies on the relationship between labor input and minimum labor demand for a normal outcome. If the land transfer market is not complete, farmland will be abandoned when labor input is less than the minimum demand for farmland [4,15,47,48,49]. Rural households are rational and motivated to pursue maximum profits. They allocate labor by comparing the potential benefits and opportunity costs of labor input in different departments. The agricultural production function is characterized by diminishing marginal returns according to the postulated law of diminishing marginal returns [20,44]. Rural households would pursue the maximization of their own profit under the constraints of agricultural production, family labor, and labor markets, and farmers fall within the wage ratio of non-agricultural labor. The family income function is as follows.

$$I=f\left(L_{farm}\right)+w{L}_{non-farm}$$

(1)

where $f\left(L_{farm}\right)$ is the household agricultural production function, which is determined by production price and labor input. Non-agricultural income is a function of market wage 𝑤 and labor input $L_{off-farm}$.$L_{farm}$ and $L_{off-farm}$ are constrained by the total amount of labor available to households,$L_{farm}+L_{off-farm}=L_t$. Bringing constraints into the household income function, we can obtain $I=f\left(L_{farm}\right)+w\left(L_{off-farm}=L_t-L_{farm}\right)$. The derivative of the family income function is used to obtain labor allocation when the income is maximized, and the following expression is obtained:

$$\frac{\partial \left[f\left(L_{farm}\right)+w\left(L_t-L_{farm}\right)\right]}{\partial L_{farm}}=0$$

(2)

$$\frac{\partial f\left(L_{farm}\right)}{\partial L_{farm}}-w=0$$

(3)

The results show that when the marginal income of a household agricultural labor input is equal to the market wage rate, the household labor allocation is in an equilibrium state of income maximization. Therefore, when the household labor force and labor market are stable, allocation of the household labor force primarily depends on the characteristics of the agricultural production function.

Based on the above theoretical analysis, if the crop–livestock system is running well, crop planting and livestock raising (CPLR) households have higher crop profits than CP-alone households, and the former $f\left(L_{farm}\right)$ has higher elasticity than the latter. In Figure 2, the top curve represents the agricultural labor production function of CPLR households, the middle curve represents the CP labor production function of CPLR households, and the bottom curve represents the labor production function of CP households. For CP households, when the marginal income from farm production is equal to the wage in the labor market, agricultural labor input is $L_3$, non-agricultural labor input is $L_t-L_3$, and the total household income is ${\int }_0^{L_3}{f}_{CP}\left(L_{farm}\right)+w\left(L_t-L_3\right)$. With non-agricultural wage rate change from 𝑤 to 𝑤’, the agricultural labor input $L_1$ is in equilibrium. If $L_1<L_m$, $L_m$ is the minimum labor demand for normal land use, and rural households would abandon their farmland. For CPLR households, when the non-agricultural wage rate changes, the agricultural labor input changes from $L_6$ to $L_4$, and farmland CP’ changes from $L_5$ to $L_2$. If $L_2>L_m$, rural households would not abandon their farmland. If $L_2<L_m$, rural households would abandon their farmland. However, if $L_2>L_1$, CPLR households show less extensive farmland abandonment than CP-alone households, and LR can alleviate farmland abandonment.

Hypothesis 2 (H2).

When the crop–livestock system is transitioning, LR aggravates rural households’ farmland abandonment and exerts a significant positive impact on households’ farmland abandonment.

The crop–livestock system is changing at the household level in China, which will result in LR transforming the industry and occupying more labor input in some households. As seen in the H1 analysis, when the equilibrium state of the CPLR household maximizes income, the CP marginal income is equal to the LR marginal income. As shown in Figure 3a, the tangent point A of the equal-income curve 𝑰 and the production possibility frontier (PPF1) of CPLR households is the equilibrium state when income is maximized. At this time, labor configuration in the case of income level Y1 is represented by point C in Figure 3b, the labor input for CP is $L_{CP}x$, and LR is $L_{LR}$. With the continuous improvement in socioeconomic development level, the demand for animal protein gradually increases. If the cost of LR in the market is lower than the household’s own production cost, then the production function of LR changes and the possible frontier of the CPLR household changes from PPF1 to PPF2. At this time, the equal-income curve is tangent to the possible production boundary PPF2 at point B, the agricultural income curve is Y2, the equilibrium point of household labor allocation is D, and the labor input of CP is $L_{CP}^{\prime }$, LR is $L_{LR}^{\prime }$. $L_{CP}^{\prime }<L_{CP}$ and $L_{LR}^{\prime }>L_{LR}$. Therefore, LR occupies more household labor and crowds out the CP household labor input. If $L_{CP}^{\prime }<L_m$, farmland abandonment is occurring or increasing, and LR can aggravate this process.

2.3. Variable Selection and Definition

2.3.1. Dependent Variable

This study focused on whether rural households abandoned farmland and the extent of abandonment. Therefore, there were two dependent variables: whether rural households abandoned farmland (1 = Yes, 0 = No) and the ratio of abandoned farmland (%).

2.3.2. Core Independent Variable

The core variable of this study is households’ LR (primarily cattle, mules, donkeys, and horses), specifically considering the raising scale as the core variable. However, the types of LR by different rural households differ and are difficult to compare quantitatively, and there may be endogeneity between CP and LR. Therefore, this study used the total price at the time of livestock purchase to represent the LR scale.

2.3.3. Control Variable

To ensure the reliability of the fitting results, referring to previous empirical research, this study started from the characteristics of the head of the household (age, education, and health status), family management characteristics (agricultural labor input, per capita household income, farmland area, total price of agricultural fixed assets, and whether rural households obtain land ownership certificates), and village characteristics (land titling, irrigation measures, land quality, village location, schools, and terrain), introducing a series of control variables, as listed in

Table 1. Definition and summarized statistics for variables. Note: USD ^a: US Dollar.

Variables	Definitions	Total		Plain		Hill		Mountain
		Mean	S.D	Mean	S.D	Mean	S.D	Mean	S.D
Abandonment/not	If rural household has abandoned farmland 1 = Yes, 0 = No	0.156	0.363	0.107	0.31	0.201	0.401	0.191	0.393
Abandonment ratio	Ratio of abandoned farmland (%)	10.104	27.245	7.251	23.798	12.906	30.22	11.951	28.911
Livestock raising	Ln (Total price of livestock, USD a)	0.68	2.122	0.465	1.8	0.511	1.861	1.157	2.652
Head age	Age of household head	55.26	12.321	55.662	12.097	56.209	12.192	53.807	12.646
Head age2	(Age of household head)2	3205.47	1386.198	3244.511	1361.413	3308.047	1403.845	3055.026	1396.056
Head education	Education year	6.917	3.372	7.22	3.297	6.969	3.272	6.409	3.511
Head health	1 = Very healthy; 2 = Relatively healthy; 3 = Generally healthy; 4 = Relatively unhealthy; 5 = Very unhealthy	2.583	1.077	2.442	1.082	2.635	1.066	2.752	1.05
Ln (per capita income)	Ln (Total revenue divided by total number of people)	6.404	1.295	6.58	1.246	6.368	1.289	6.169	1.334
Per capita farmland scale	Per capita farmland area (ha)	0.116	0.578	0.14	0.474	0.083	0.457	0.111	0.781
Ln (Agricultural assets)	Ln (Total value of agricultural fixed assets, CNY)	0.861	2.11	0.998	2.241	0.824	2.072	0.686	1.914
Average labor input of farmland	Quantity of labor input per unit of farmland (number/ha)	7.05	13.021	6.423	12.516	7.274	12.896	7.804	13.82
Land titling	Whether rural household obtained land ownership certificate (1 = Yes; 0 = No)	0.508	0.5	0.512	0.5	0.436	0.496	0.564	0.496
Grain for Green	Whether Grain for Green in village (1 = Yes; 0 = No)	0.435	0.496	0.223	0.416	0.513	0.5	0.69	0.463
Irrigation	Are there irrigation facilities in the village (1 = Yes; 0 = No)	0.673	0.469	0.706	0.455	0.701	0.458	0.597	0.491
Land quality	1~5 (1 is the worst and 5 is the best)	4.675	.808	4.661	.787	4.598	0.904	4.764	0.739
Distance	Distance from village to county town (km)	28.141	26.839	20.452	18.444	26.431	21.889	41.346	35.301
School	Is there a school (1 = Yes; 0 = No)	0.701	0.458	0.662	0.473	0.696	0.46	0.765	0.424
Terrain	1 = Plain; 2 = Hill; 3 = Mountain	1.847	0.846	-	-	-	-	-	-

:

2.3.4. Econometric Model

To analyze the actual impact of LR on farmland abandonment in depth, an econometric model was needed for estimation and testing. According to the above theoretical analysis, the farmland abandonment behavior of rural households is affected not only by LR, but also by household head characteristic variables, family variables, and village environmental variables. The specific Logit and Tobit models were used to estimate the impact of LR on farmland abandonment and its extent. The model formulae are as follows:

$$FL{A}_l={\alpha }_0+{\alpha }_{1}LR+{\alpha }_2{X}_i+{\epsilon }_1$$

(4)

$$FL{A}_t={\beta }_0+{\beta }_{1}LR+{\beta }_2{X}_i+{\epsilon }_2$$

(5)

where $FL{A}_l$ represents rural household farmland abandonment/no abandonment; $FL{A}_t$ is the ratio of abandoned farmland; $LR$ is the scale of livestock; $X_i$ is the control variable; ${\epsilon }_1$ and ${\epsilon }_2$ are error term; ${\alpha }_1$, ${\alpha }_2$, ${\beta }_1$, ${\beta }_2$ are the parameters of the models that need to be estimated.

2.4. Uncertainties and Shortcomings

The random sample did not survey farmers who had relocated and abandoned their farmland, and the absence of these samples would lead to sampling error. The survey results are based on farmers’ subjective perceptions and expressions, and misreporting and underreporting will affect the accuracy of the results. Moreover, China is a vast country and 6707 samples might still be insufficient, especially with the lack of samples in some provinces, such as Xinjiang and Qinghai, which may also cause errors.

3. Results

3.1. Descriptive Statistical Analysis Results

Table 1 shows the descriptive statistical analysis results for the dependent, independent, and control variables. As shown in Table 1, farmland abandonment was seen in 15.63% of rural households, the total abandoned farmland share was 6.24%, and the average economic scale of LR was 2451.80 USD per household. The average age of the household head was 55.25 years old, and the average length of education was 6.92 years. Household, on average, held 7.05 labor force per ha of farmland with a per capita income of 1259.50 USD and per capita farmland area of 0.14 ha. The current market value of all agricultural assets was 597.92 USD, and 50.75% of households had obtained a land ownership certificate. The overall share of CPLR households was 7.91%, the share of only LR households was 1.54%, and the share of only CP households was 90.55%. Households in plain areas accounted for 45.60%, households in hilly areas accounted for 26.06%, and households in mountain areas accounted for 29.34%.

3.2. Spatial Distribution of Farmland Abandonment and LR

China has a vast territory, and therefore, farmland abandonment in different provinces may have different characteristics. The spatial distribution of farmland abandonment and LR in different regions was obtained by comparing the ratio of households with farmland abandonment, ratio of abandoned farmland, and the ratio of LR households in different provinces.

As shown in Figure 4a, farmland abandonment behavior in rural households was primarily distributed in the southeast coastal provinces and southwestern mountainous provinces. The highest ratio of rural households with farmland abandonment was in Guangdong (34.49%), followed by Chongqing (31.18%) and Gansu (25.62%); the provinces with a relatively lower ratio were primarily distributed in North China and Northeast China. Among them, Heilongjiang had the lowest ratio (0.73%), followed by Jilin (0.78%), Hebei (1.22%), and Liaoning (1.6%).

The farmland abandonment ratio obtained from the weighted average farmland area (Figure 4b) showed similar spatial distribution characteristics to the ones seen in Figure 4a. The highest farmland abandonment ratio was in Guangdong (30.99%), followed by Chongqing (20.14%) and Gansu (16.33%); the lowest ratio of farmland abandonment was in Heilongjiang (0.18%), followed by Liaoning (0.31%), Jilin (0.39%), Inner Mongolia (0.69%), and Hebei (0.72%). Spatial distribution of the ratio of rural households with farmland abandonment and the ratio of farmland abandonment showed that farmland abandonment presents a spatial pattern, with high ratio in the south and low in the north. Provinces with more serious abandonment have obvious topographical fluctuations and farmland fragmentation. The farmland abandonment ratio in rural households is low where the area is flat and has good conditions for large-scale production.

As shown in Figure 4c, the provinces with a high ratio of LR households were primarily distributed in the western region, with the highest ratio being in Xinjiang (52%), followed by Guangxi (35.39%), Inner Mongolia (25.88%), Guizhou (15.48%), Gansu (14.74%), and Hubei (13.69%). The ratio of LR households was higher in the west and lower in the east, showing different characteristics from the spatial distribution of farmland abandonment. From the comparison of different provinces, as shown in Figure 4d, it can be found that the relationship between LR and farmland abandonment is not a simple linear relationship. This because abandonment is affected by multiple factors. Provinces with a high ratio of abandoned farmland households also displayed a high ratio of farmland abandonment. However, there were provinces where the ratio of LR households was high and abandonment was not serious, as well as provinces where the ratio of LR households was high and abandonment was serious. Therefore, the relationship between LR and farmland abandonment must be tested further using an econometric model.

3.3. Econometric Model Analysis Results

3.3.1. National Level Analysis Results

The regression fitting results for the impact of LR on farmland abandonment are presented in

Table 2. Estimation results of the Logit and Tobit models.

Variables	Logit Model	Tobit Model
	Model 1	Model 2
Livestock raising	−0.002	−0.784
	(0.013)	(0.756)
Head age	−0.052 ***	−3.394 ***
	(0.018)	(1.064)
Head age2	0 ***	0.032 ***
	(0)	(0.009)
Head education	0.014	0.541
	(0.011)	(0.623)
Head health	0.222 ***	12.613 ***
	(0.034)	(1.982)
Ln (Per capita income)	−0.015	−0.353
	(0.028)	(1.563)
Per capita farmland scale	−0.107 ***	−4.845 ***
	(0.024)	(1.108)
Ln (Agricultural assets)	−0.044 ***	−2.925 ***
	(0.016)	(0.856)
Average labor input of farmland	−0.390 ***	−17.619 ***
	(0.071)	(3.344)
Land titling	−0.243 ***	−15.429 ***
	(0.07)	(4.022)
Grain for Green	0.151 **	7.537 *
	(0.076)	(4.364)
Irrigation	0.056	4.385
	(0.075)	(4.300)
Land quality	−0.080 *	−5.382 **
	(0.042)	(2.426)
Distance	−0.002 *	−0.171 **
	(0.001)	(0.081)
School	0.058	5.365
	(0.078)	(4.415)
Terrain	0.302 ***	17.468 ***
	(0.047)	(2.682)
Constant	−0.944	−43.806
	(0.595)	(35.296)
LR	252.78 ***	256.95 ***
Log likelihood	−2780.467	−8111.837
Pseudo R2	0.044	0.016
Numbers of observations	6707	6707

Note: () is the robust standard error; *, **, and *** denote coefficients significant at 10%, 5%, and 1% levels, respectively.

. The coefficient of LR influence on whether rural households abandoned farmland and the extent of abandonment was negative, which was consistent with the expectation that LR could alleviate the farmland abandonment, but no results were significant. Therefore, theoretical hypothesis H1 has not been confirmed, and rural households raising livestock cannot significantly alleviate the occurrence of farmland abandonment and scale expansion, which is inconsistent with the conclusions of previous studies [35,42]. The impact of LR on farmland abandonment may have internal heterogeneity in China, and all samples need to be further estimated and analyzed using subsample regression.

3.3.2. Analysis Results in Different Terrains

From the above regression results at the national level, it can be seen that the impact of LR on farmland abandonment may be heterogeneous. According to the hypothesis of farmland marginalization [17], farmland abandonment varies significantly among different terrains. Therefore, it is necessary to perform subsample regression analysis on different terrains. As shown in Figure 5, in plain, hilly, and mountain areas, the ratios of households with farmland abandonment were 10.73%, 20.14%, and 19.06% in mountain areas, the ratios of farmland abandonment were 4.89%, 10.16%, and 6.33%, and the ratios of LR households were 6.39%, 7.09%, and 16.21%.

Table 3. Estimation results of the Logit and Tobit models.

Variables	Logit Models			Tobit Models
	Plain	Hill	Mountain	Plain	Hill	Mountain
	Model 3	Model 4	Model 5	Model 6	Model 7	Model 8
Livestock raising	0.049 **	0.072 ***	−0.089 ***	2.083	2.305 *	−4.631 ***
	(0.024)	(0.025)	(0.022)	(1.544)	(1.375)	(1.091)
Head age	−0.096 ***	0.007	−0.072 **	−6.533 ***	0.426	−4.155 **
	(0.031)	(0.037)	(0.031)	(1.975)	(1.981)	(1.63)
Head age2	0.001 ***	0	0.001 ***	0.056 ***	−0.009	0.045 ***
	(0)	(0)	(0)	(0.017)	(0.017)	(0.015)
Head education	−0.034 *	−0.011	0.061 ***	−2.234 *	−0.744	3.097 ***
	(0.019)	(0.021)	(0.018)	(1.186)	(1.099)	(0.971)
Head health	0.318 ***	0.304 ***	0.074	19.687 ***	15.534 ***	3.966
	(0.061)	(0.065)	(0.06)	(3.738)	(3.506)	(3.131)
Ln (Per capita income)	0.077	−0.02	−0.08 *	5.677 *	−0.408	−4.429 *
	(0.052)	(0.051)	(0.045)	(3.141)	(2.728)	(2.297)
Per capita farmland scale	−0.011	−0.357 ***	−0.134 ***	−0.537	−12.467 ***	−7.021 ***
	(0.024)	(0.09)	(0.041)	(1.276)	(3.24)	(1.926)
Ln (Agricultural assets)	−0.063 **	−0.087 ***	−0.003	−4.259 ***	−5.35 ***	−0.422
	(0.027)	(0.03)	(0.027)	(1.594)	(1.522)	(1.389)
Average labor input of farmland	−0.166	−0.394 ***	−0.756 ***	−7.796	−16.851 ***	−34.74 ***
	(0.107)	(0.119)	(0.145)	(5.318)	(5.632)	(6.368)
Land titling	−0.296 **	−0.238 *	−0.01	−15.09 *	−14.527 **	−4.759
	(0.128)	(0.13)	(0.121)	(7.766)	(6.947)	(6.342)
Grain for Green	0.183	0.139	−0.232 *	9.273	2.473	−11.737 *
	(0.153)	(0.141)	(0.132)	(9.275)	(7.506)	(6.912)
Irrigation	0.414 ***	0.107	−0.081	23.526 ***	6.761	−4.516
	(0.148)	(0.139)	(0.132)	(8.738)	(7.482)	(6.904)
Land quality	−0.209 ***	0.02	0.199 *	−16.366 ***	1.154	11.919 **
	(0.067)	(0.077)	(0.105)	(4.268)	(4.003)	(5.313)
Distance	−0.011 ***	0.005 *	0.001	−0.689 ***	0.192	0.034
	(0.004)	(0.003)	(0.002)	(0.243)	(0.149)	(0.109)
School	0.909 ***	−0.665 ***	−0.004	52.775 ***	−30.876 ***	−2.311
	(0.161)	(0.132)	(0.144)	(9.411)	(7.278)	(7.385)
Constant	−0.206	−1.145	−0.094	−0.932	−76.421	−4.411
	(1.068)	(1.198)	(1.033)	(65.727)	(64.527)	(54.758)
LR	150.97 ***	120.40 ***	111.64 ***	156.96 ***	100.13 ***	121.33 ***
Log likelihood	−944.085	−817.822	−902.817	−2590.891	−2613.861	−2793.067
Pseudo R2	0.074	0.069	0.058	0.029	0.019	0.021
Numbers of observations	2991	1748	1968	2991	1748	1968

Note: () is robust standard error; *, **, and *** denote coefficients significant at 10%, 5%, and 1% levels, respectively.

shows the regression results in different terrains. It can be seen that the impact of LR on farmland abandonment was different in different terrains and heterogeneous at the national level. According to the tested statistics models, Model 4, 5, and 8 were significant at the 0.01 level, Model 3 and 7 were significant at the 0.05 level and 0.1 level, and Model 6 failed the significance test. The regression results of models in plain areas and hilly areas were consistent with hypothesis H2 and the regression results of models in mountainous areas were consistent with hypothesis H1. For the plain areas, keeping all other variables constant, the core variable coefficient of Model 3 was significantly positive. The probability farmland abandonment in rural households increased by 0.05% for every 1% increase in the LR scale, while the coefficient of Model 7 was positive and failed to pass the significance test, indicating that LR in plain areas will promote rural households to abandon farmland, but it will not continue to aggravate abandonment. For the hilly areas, the core variable coefficients of Models 4 and 7 were significantly positive at the 0.01 and 0.1 levels, and with every 1% increase in the LR scale, the probability of farmland abandonment in rural households increased by 0.07% and the farmland abandonment ratio increased by 0.02%. Regression results in mountain areas were completely different from those in plains areas and hills. The core variable coefficients of Models 5 and 8 were significantly negative at the 0.01 levels and for every 1% increase in the LR scale, the probability of farmland abandonment in rural households decreased by 0.09% and farmland abandonment ratio decreased by 0.05%.

LR in different terrains evidently had different effects on farmland abandonment. Meanwhile, the absolute value of the marginal effect of core variable was mountain areas > hilly areas > plain areas. The mountain areas were more strongly affected than plain areas and hilly areas. The plain areas have flat terrain, convenient transportation, and market-oriented division of work to promote different types of rural households. Farmland scale management can use land transfer and socialized services to alleviate the impact of LR squeezing labor input. The hilly areas have obvious topography, and the substitution effect of mechanical input is relatively limited when rural households change into different types. Therefore, under decoupling of the crop–livestock system, LR plays a significant role in promoting farmland abandonment. The mountain areas have more obvious topography, poor traffic, lower mechanization use, and lower use of chemical fertilizers. The purchase price of animal protein and feed is affected by traffic location, which is relatively higher than in other regions. Therefore, rural households are more inclined to maintain the crop–livestock mixed agricultural system for nutrition, manure, and animal power, and the extent of rural household differentiation is lower than in other areas. LR could alleviate farmland abandonment.

3.3.3. Robustness Analysis

To test the robustness of the regression results, the discrete data of the core variables were replaced with binary data (whether rural household raised livestock, 1 = Yes, 0 = No), and the Logit and Tobit model regression estimations were performed at the national and sub-terrain region levels (

Table 4. Robustness test of Logit models and Tobit models.

Variables	Logit Models				Tobit Models
	Total	Plain	Hill	Mountain	Total	Plain	Hill	Mountain
	Model 9	Model 10	Model 11	Model 12	Model 13	Model 14	Model 15	Model 16
Livestock raising	−0.012	0.428 *	0.669 ***	−0.786 ***	−6.888	18.149	21.462 *	−40.996 ***
	[0.002]	[0.039]	[0.1]	[0.115]	(6.815)	(14.056)	(12.364)	(9.776)
Control variables	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
LR	252.77 ***	150.73 ***	120.86 ***	110.75 ***	256.90 ***	156.81 ***	100.33 ***	120.10 ***
Log likelihood	−2781.175	−944.466	−817.59	−903.078	−8111.864	−2590.965	−2613.758	−2793.344
Pseudo R2	0.043	0.074	0.069	0.058	0.016	0.029	0.019	0.021
Numbers of observations	6707	2991	1748	1968	6707	2991	1748	1968

Note: [] is the marginal effect of LR (livestock raising); () is robust standard error; * and ***denote coefficients significant at 10% and 1% levels, respectively.

). There was no significant relationship between LR and farmland abandonment at the national level, and the significance results of core variables in the different terrains were consistent with Table 3. The core variable coefficient was negative in mountainous areas and positive in other areas. Specifically, keeping all other variables constant, in the plain areas, the marginal effect of Model 10 was 0.04 at the 0.05 significance level and LR could increase the ratio of households with farmland abandonment by 3.9%. In hilly areas, the respective marginal effects of Model 11 and 15 were 0.1 and 0.215 at the 0.01 and 0.05 significance levels; LR could increase the ratio of households with farmland abandonment by 10% and the ratio of farmland abandonment by 21.46%. In mountain areas, the respective marginal effect of Model 12 and 16 were 0.115 and 0.41 at the 0.01 significance level; LR could decrease the ratio of households with farmland abandonment by 11.5% and the ratio of farmland abandonment by 41%. Therefore, the positive and negative impact coefficient of LR on farmland abandonment and their significance are consistent with previous results. Overall, the regression results can be considered robust.

4. Discussion

At present, there are two ways to obtain information on farmland abandonment: remote sensing extraction and questionnaire surveys. Remote sensing was used to analyze the spatial distribution and temporal evolution of farmland abandonment [50,51,52], and a questionnaire survey was used to analyze the formation mechanism and impact of farmland abandonment [2,3,13,17,53]. Based on theoretical analysis, this study used 6707 household samples from 27 provinces of China to reveal the impact of LR on farmland abandonment by comparing the spatial distribution characteristics of farmland abandonment and LR with the econometric model estimation. Short qualitative descriptions and small-scale studies have not reached a single conclusion on whether LR promotes or alleviates farmland abandonment [33,39,42]. The difference in this study was that analysis of the national level and different terrains revealed the impact of LR on farmland abandonment, and that the mechanism of action was more varied and area-specific. The results at the national level did not confirm hypothesis H1, and the effect of LR on farmland abandonment was not significant; the results in plain areas and hilly areas effectively confirmed hypothesis H2, where LR could promote farmland abandonment. In mountainous areas, the results were consistent with hypothesis H1, where LR can effectively alleviate farmland abandonment. The different relationship between LR and farmland abandonment in different terrains are a key reason for the insignificant results at the national level. This difference is induced by the difference in farmland resource endowment, rural household livelihood, and transportation location in different terrains, leading to different behavioral choices when rural households pursue maximum benefits.

Farmland abandonment, as an extreme manifestation of marginalization, is the result of changes in household land use behaviors under declining farmland economic returns. For example, when the income from three staple foods (rice, wheat, and maize) in China began to decline since the mid-1990s [9,28], farmland abandonment increased [9,10]. Therefore, increasing farmland economic income is an important way to alleviate farmland abandonment. At present, the government and academia have two approaches to explore the solution for farmland abandonment: First, policies and institutions are adjusted to improve the efficiency of resource allocation, such as abolishing agricultural taxes, encouraging land transfer, separation of rights, and industrial integration [13,54,55,56]. However, the implementation of these measures depends on the extent of marketization. The economic marketization of China’s rural areas has not yet been fully developed [32,53,55,57], and the existing strategic adjustments may not be able to achieve the expected results. Second, the government improves land capital investment, such as terracing, soil improvement, improving irrigation facilities and other land consolidation projects, and agricultural biotechnology innovation [43,57,58,59]. Land consolidation projects can effectively improve farmland quality and increase its scale. Technological innovation can improve product quality and increase the added value of the farmland. Both can effectively increase rural household income, but the implementation effects of both are affected by the response of rural household behavior. If rural households do not reasonably use farmland after land consolidation and adopt new technologies, farmland marginalization would not be effectively alleviated. Currently, the comprehensive production capacity of LR is continuously increasing, as it transforms into specialized and industrial types [33,36,39,40], and it plays an important role in increasing household income. However, to protect farmland, LR is subjected to different regional conditions and has different impacts on land use. Therefore, the authority should adopt different LR development strategies in different terrain areas, and make use of the market mechanism to reallocate land, particularly in plain areas and hilly areas. In addition, land abandonment may bring about the improvement of the environment and alleviate poverty. Therefore, which places and which lands should be given timely treatment when the abandonment occurs should be the focus of future research and government attention.

In this study, spatial distribution of farmland abandonment in China based on household survey data was similar to previous results [10,17], indicating that household data can effectively reflect farmland abandonment in China. With the advancement of economic marketization, rural households have gradually differentiated into different types, and the relationship between LR and farmland abandonment is also dynamic. This relationship must be explained and analyzed using long-time panel data. However, this study only used the cross-sectional data of 27 provinces in 2016 to analyze the impact of LR on farmland abandonment due to data constraints. In China, there are various types of farmland and this study only used different topographic villages to indirectly characterize the impact of livestock raising on abandonment under different topographic features. Hilly and mountainous villages containing sloping land, terraces, and flat land lack more detailed analysis. Moreover, the rural household model in this study was based on traditional economic assumptions and did not incorporate external factors such as environment and institutions into the analysis. Although rural household behavior is affected by multiple factors, these factors will change their behavior by influencing the expectation of income, and the limitations of model based on economic assumptions do not affect the interpretation of rural household behavior. Therefore, in future studies, remote sensing and questionnaire surveys can be combined to analyze the dynamic characteristics of farmland abandonment and influencing factors at multiple scales, and build integrated analytical models under natural, economic, and institutional complex systems.

5. Conclusions

According to the above theoretical and empirical analyses of the farmland abandonment in China, the following conclusions can be drawn:

(1) At the national level in China, abandoned farmland households accounted for 15.63%, and the farmland abandonment area was 6.24%. The spatial distribution pattern was high in the south and low in the north. LR households accounted for 9.45% of all samples, with high spatial distribution of high in the west and low in the east. At the national level, the impact coefficient of LR on farmland abandonment was negative, but not significant, and the alleviation effect was limited.

(2) In the different terrains, the ratio of households with farmland abandonment was hilly areas > mountain areas > plain areas (20.14% > 19.06% > 10.73%), the ratio of farmland abandonment was hilly areas > mountain areas > plain areas (10.16% > 6.33% > 4.89%), and the ratio of LR households was mountain areas > hilly areas > plain areas (16.21% > 7.09% > 6.39%). For every 1% increase in LR, the ratio of households with farmland abandonment increased by 0.05% and 0.07% in the plain areas and hilly areas, respectively, and decreased by 0.09% in mountain areas. The ratio of farmland abandonment increased by 0.02% in hilly areas and decreased by 0.05% in mountain areas. LR increased the ratio of households with farmland abandonment by 3.9% and 10% in plain and hilly areas, respectively, and decreased by 11.4%, in mountain areas, and the ratio of farmland abandonment increased by 21.46% in hilly areas and decreased by 41% in mountain areas.

(3) Different terrains were affected by topography and location, the impacts of LR were different, and the absolute value of the marginal effect showed a trend of mountain areas > hilly areas > plain areas, requiring different alleviation measures.