Does Farmland Rental Contribute to Reduction of Agrochemical Use? A Case of Grain Production in Gansu Province, China

Abstract

As a consequence of the new strategy to boost productivity capacity and ensure food security, China’s farmland rental market is developing rapidly, and its impacts on agricultural productivity have been extensively studied. However, the impacts of farmland rental on food safety have not been considered. The aim of this study was to determine the causal effects of farmland rental on fertilizer and pesticide use in wheat and maize production and evaluate the potential effects of this activity on food safety. Survey data obtained from 900 households in eight counties in Gansu province were used in this study, and the propensity score matching (PSM) method was employed to solve selection bias problems with the data. The results showed that farmland rental significantly reduced fertilizer and pesticide use in wheat and maize production, implying a potential reduction in heavy metal contamination of food and drinking water as well as less pesticide residues-remaining in food and contamination of environment. Also, households renting land were more likely to adopt new agricultural technologies and management methods and to acquire more agricultural acknowledges and information than those not renting land and renting out land. Thus, farmland rental is a benefit to the application of new agricultural technologies and management methods, to rational use of agrochemicals, and finally to food safety and environmental conservation. Policies such as encouraging farmland rental, enhancing education of farmers, improving technological innovation, and providing better information transfer should help ensure not only “enough food” but also “safe food”.

1. Introduction

For a long time, decreasing farmland and rapidly growing populations have made food security a growing concern worldwide [1,2]. As the largest developing country with the largest population and grain consumption in the world, China has successfully created the “Miracle in China” by feeding 22% of the global human population with less than 9% of the world’s arable land over the past several decades, indicating a substantial contribution to global food security [3,4]. However, with the reduction of arable land and the increase of population and grain consumption, China’s food security is increasingly acute [5].

To ensure food security, the central government of China is promoting the formation of large-scale and mechanized farms through farmland rental [5,6]. With substantial support from the government and promotion from the booming off-farm labor market, China’s farmland rental market is growing rapidly [7,8]. Meanwhile, issues related to the impacts of farmland rental on agricultural productivity and food security have received special attention by economists and policy makers. Various studies have shown the positive effects of farmland rental and consolidation on food security due to its improving allocative and technical efficiency and reducing land fragmentation [9,10,11,12].

Food security requires an access to “sufficient, safe, and nutritious food” [2], i.e., food security not only means “sufficient food” but also “safe food”. In particular, the quality of food products, especially their safety, is an increasing focus for governments, companies, and international trade bodies after the 2008 melamine problem in China. Food safety was ranked first in the top five safety issues that were of concern to the Chinese population in 2011 [13]. Thus, understanding the role of farmland rental in food safety is of fundamental importance not only for agriculture policymakers but also for society stability.

China’s agriculture development strongly depends on the use of many kinds of agrochemicals, especially fertilizers and pesticides. China is now the largest consumer of fertilizers and pesticides in the world, using over one-third of the world’s fertilizers and pesticides [14,15]. Overuse of fertilizers and pesticides is widespread in China. The massive use of fertilizers and pesticides not only causes serious contamination of soil and water, but it can be harmful to human health with heavy metals and pesticide residues in food [1,16,17]. Thus, China’s intensive use of fertilizers and pesticides is threating food safety and the environment [14,18].

Theoretically, farmland rental will increase farm size and promote the application of new technologies and management based on scientific knowledge, which may improve use efficiency of fertilizer and pesticides, thereby reducing the risks of agriculture production on food safety. On one hand, the enlargement of farm size and reduction of farmland fragmentation caused by farmland rental will promote the application of new agricultural machines and technologies [19]. On the other hand, the increase of agricultural production cost and risk caused by farmland rent will urge households renting land to acquire more information and acknowledges than those without renting land to reduce this risk. This will finally cause a more rational agrochemical use of households renting land than those without renting land. In addition, the insecure land right of rented farmland will also have effects on agrochemical use. Gao et al. [20] found households had different behaviors on manure and chemical fertilizer use between contracted farmland and rented farmland. Therefore, those changes along with farmland rental will cause fundamental impacts on agrochemical use in agricultural production.

Unfortunately, previous studies have mainly focused on the impacts of farmland rental on “sufficient food” with the impact on “safety food” receiving little attention [6,21]. In this study, we investigated the effects of farmland rental on fertilizer and pesticide use in grain production. Results of this study should provide insights on the effects of farmland rental on food safety that can be helpful for policymakers.

2. Background

In the past several decades, the household contract responsibility system (HCRS) of farmland starting in 1978 has greatly promoted the economic growth and the transition of rural China [22]. This land system was compatible to the backward agricultural productivity and simple economic structure of China at that time [23]. For a long time, it not only aroused rural households’ enthusiasm for agricultural production and improved agricultural productivity, but also optimized the agricultural production structure and food consumption structure of residents in China [24]. However, with the development of agricultural productivity, the shortcoming of HCRS is increasingly outstanding. The agricultural development of China is facing many new challenges, such as declining farmland area, aging farmers, and deteriorating ecological environment [5,21].

Developing large-scale and mechanized farms is considered a promising approach to address the challenges facing China’s agriculture [6,25]. For this purpose, Chinese government has openly encouraged farmland rental and large-scale farming by a series of agricultural policies, including subsidies on leased land, allowance for purchasing agricultural machinery, farmland certification, agricultural insurance, and farmland collateral [26]. Furthermore, farmland rental and large-scale farming were also encouraged and emphasized in the Chinese Number 1 Document of 2012 to 2019. In addition, the release of surplus labor from rural areas to urban industrial sectors also contributed to the booming of farmland rental market. China’s farmland rental market grows rapidly, while only 5.2% of contracted farmland was leased in 2007, but this ratio rose to 35.1% in 2016 [27].

The rapid growth of farmland rental market in China attracted considerable attention of academia. Some studies focused on the understanding of the determinants of farmland rental. Plenty of studies on the determinants of farmland rental found that off-farm employment, migration, tenure security, and agricultural ability had significantly positive impacts on the development of the farmland rental market [7,28,29,30,31]. Additionally, the productive heterogeneity among farmers, household attributes, contract condition, credit market development, institutional buildings, and property rights also played critical roles on farmland rental [8,32]. Some other studies explored the effects of farmland rental on agricultural production and farmers’ income. The impacts of farmland rental and large-scale farming on agricultural investments to land of farmers, agricultural productivity and efficiency, occupational diversification, farmers’ income and income inequality, as well as farmers’ welfare, were extensively studied [6,20,33,34,35,36].

Although the impacts of farmland rental and large-scale farming on agricultural production have been extensively studied, the casual relationship between farmland rental and agrochemical use in agricultural production is still unclear. Thus, it is important to clarify this relationship.

3. Econometric Approach

To estimate the effects of a treatment, the difference in an individual’s outcome with and without treatment should be estimated. However, it is not possible to observe both outcomes for the same individual at the same time. Ideally, experimental data could provide information on the counterfactual situation that would solve the problem of causal inference [37]. As this is not the case, there is a problem of “missing data” [38]. Thus, the problem of selection bias arises when taking the mean outcome of nonparticipants as an approximation [39]. In our study, households renting land and renting out land were usually different from those not renting land, even in the absence of farmland rental. For instance, households active on farmland rental market might be more economically oriented or economically rational than those not renting land. The propensity score matching (PSM) approach is one possible solution to selection bias, and it has become a popular approach to estimate causal treatment effects, since it shows a close link to the experimental context [39] and isolates the treatment effects from other socio-economic determinants [37].

The basic idea of PSM is to find in a large group of nonparticipants that are similar to the participants in all relevant pretreatment characteristics. Then, the differences in outcomes of this are well selected and thus act as an adequate control group, and those of the participants can be attributed to the treatment.

Firstly, PSM reduces the high level of dimensionality of households by comparing households with probability of farmland rental based on relevant characteristics [40,41]. A definition of the conditional probability of involvement of rural households i in farmland rental can be expressed as a logit form:

$$\mathrm{P}\left(X_i\right)=Pr\left(T_i=1|X_i\right)=\frac{\exp \left(X_i^{\prime }{\beta }_i\right)}{1+\exp \left(X_i^{\prime }{\beta }_i\right)}$$

(1)

where $\mathrm{P}\left(X_i\right)$ is the conditional probability of household i renting land or renting out land and is defined as the propensity score (p-score); $T_i$ is a binary variable for renting land or renting out land for rural household i, and it represents the treatment here; $X_i$ is the vector of exogenous variables; ${\beta }_i$ is a vector of parameters to be estimated. All variables are defined in the next section. The p-score estimation ranks households according to their own behavior toward renting land or renting out land. Thus, this study estimated the effects of farmland rental among groups of farmers with similar behaviors.

In our study, households renting land (estimating the effects of renting land) and households renting out land (estimating the effects of renting out land) were considered as treatment groups, and those not involved in farmland rental were considered as control groups. Then, each household in the treatment group was matched with a household in the control group according to their p-scores. Three of the most widely-used matching methods, nearest neighbor matching (NNM), radius-based matching (RBM), and kernel-based matching (KBM) [42], were used in this study. To make two samples with similar p-scores from different matched groups, NNM entails backward and forward searching to identify one or more samples in the control group with the closet p-scores for the samples in the treatment group; RBM entails searching all samples of the control group occurring within a certain radius according to their p-scores for the samples in the treatment group; KBM entails matching each sample in the treatment group with a weighted average of all weighted controls that are inversely proportional to the distance between p-scores of the samples in the treatment and the control groups [42].

After matching, the balancing hypothesis should be tested to check whether the two households from different groups are, in fact, similar. Households with the same or similar p-score are assumed to have the same distribution of $X_i$, irrespective of their farmland rental status in the balancing hypothesis. Moreover, the common support condition that bounds the p-score away from zero and one and excludes the tails of the distribution of p-score is imposed to improve the quality of matches [37]. The p-score density functions of two groups should be proximate, thus indicating similar characteristics of $X$ in the two groups after matching according to the common support assumption. Two household groups after matching are considered to behave similarly when the balancing hypothesis and the common support assumption are simultaneously satisfied [43].

The main feature of PSM is creating the conditions of a randomized experiment to evaluate a causal effect as in a controlled experiment. Thus, the conditional independence assumption is made that farmland rental is random and unrelated to fertilizer (chemical fertilizer here) and pesticide cost with controlled $X$. The conditional independence assumption is more plausible than in the case of ordinary least square (OLS), since farmland rental is assumed to be random within households having the same behavior towards farmland rental [37]. Then, the impact of renting land or renting out land on fertilizer and pesticide cost of wheat and maize production is:

$$f\left(X_i\right)=E\left(Y_i^1-Y_i^0|X_i\right)=E\left(Y_i^1|T=1, X_i\right)-E\left(Y_i^0|T=0, X_i\right)$$

(2)

where $f\left(X_i\right)$ is the difference of fertilizer or pesticide cost of wheat and maize production between a household i from treatment group and its matched household from control group; $Y_i^1$ and $Y_i^0$ represent the fertilizer or pesticide cost of wheat and maize production of a household i from treatment group and those of its matched household from control group, respectively; $T=1$ and $T=0$ indicate a household from the treatment group and from the control group, respectively.

To facilitate comparison, the cost of fertilizer and pesticide were used to represent the use of fertilizer and pesticide in this study. There are several reasons for using agrochemical cost to represent agrochemical use. Firstly, it is improper to use the real agrochemical use amount, since different households used different agrochemical types with different ingredients and concentrations. The amounts of agrochemicals with different ingredients and concentrations cannot be directly compared. Secondly, in the study area, almost all of households brought their agrochemical from local stores of the nearest towns, and the prices of agrochemicals were stable and transparent. In addition, only several households had a huge farm size after renting land, thus households in our study did not get the point of reducing the input price by large volume purchase. For example, only three households in our study had a farm size over one hundred mu (the mu unit is widely used as the unit of farmland in China, and 1 mu is 1/15 hectare). Hence, agrochemical cost should be a good index to represent agrochemical use in our study.

The average effect of farmland rental is:

$$\mathrm{ATT}=E\left[f\left(X_i\right)\right]$$

(3)

where ATT is the average effect of fertilizer or pesticide cost of wheat and maize production resulting from farmland rental. It is worth noting that small sample bias remained in the statistical analysis of this study. To reduce this bias, the bootstrap method was employed to estimate standard errors of ATTs.

4. Survey and Data

The data for this study were derived from a survey of rural households conducted in eight selected counties of Gansu province, China (Figure 1). Gansu province is a typical grain production area in a less-well developed region of China with relatively poor agricultural production conditions and low level of urbanization. It has a land area of about 4.7% of the total area of China and is one of the major grain producing provinces in Northwestern China. The planted grain crops area of Gansu is about 2.5% of the national level during 2015, which is larger than that of most other provinces in Northwestern China [44]. In addition, the farmland rental market in Gansu province has developed rapidly. For example, only 4.1% of contracted farmland had been leased in September 2010, while 24.6% of contracted farmland had been leased in September 2016 [45]. Thus, the study of Gansu province may provide some hints for not only Northwestern China but also for some less developed areas similar to Gansu in other countries.

Eight counties with different economic development levels located in different parts of Gansu province were selected as sample regions (Figure 1). These counties contain the majority of landform types and planting models in Gansu province, such as Minle county in the Hexi Corridor and Huining county on the Loess Plateau. The face-to-face questionnaire survey was conducted with the household heads during April to September 2015 by students of Lanzhou University after proper training, and sample households were randomly selected. The questionnaire contained demographic information, planting and breeding situation, inputs and outputs of crops, farmland types and area, farmland rental situation in 2014, and willingness to take part in farmland rental. A total of 900 useable questionnaires was collected (144 from Xifeng county, 157 from Yuzhong county, 123 from Minle county, 28 from Gulang county, 30 from Lingtai county, 137 from Jingning county, 132 from Huining county, and 149 from Min county).

Wheat and maize were selected to be included in this study, since they were two main grain crops taking up 56.54% of the total planted area in the eight counties. The farmland rental and the planting situations of wheat and maize in the study area are reported in

Table 1. Farmland renting and crop planting situations.

Planting Situations	Renting Land	Renting Out Land	Not Renting Land	Total
Households planting wheat	84	71	324	479
Households planting maize	78	65	309	452
Households planting both wheat and maize	62	48	226	336
Households not planting either wheat or maize	14	121	170	305
Total	114	209	577	900

Note: “Total” in the row is the number of households with different farmland rental status without regard to the crop planting situation; “Total” in the column is the number of households with different crop planting situations without regard to their farmland rental status.

. There were 479 households that planted wheat, and 84 of them rented and 71 of them rented out some land. The number of households that planted maize was 452, and 78 of them rented and 65 of them rented out some land. Of the 900 surveyed households, 114 were renting land and 209 renting out land, i.e., 12.67% and 23.22%, respectively. The farmland rental market in Gansu province was still at its infancy in 2014.

Some descriptive information of the study area and the definition of variables is reported in

Table 2. Descriptive information of the study area.

Variable	Mean	SD	Min.	Max.	Observations
Renting land	0.127	0.333	0	1	900
Renting out land	0.232	0.422	0	1	900
Head age	50.761	11.141	21	81	900
Age square	2700.681	1154.397	441	6561	900
Occupation	0.811	0.392	0	1	900
Education	6.143	3.809	0	16	900
Health condition	0.757	0.429	0	1	900
Labor ratio	0.660	0.231	0	1	900
Income	11.648	9.879	0.180	101.155	900
Off-farm income	0.689	0.323	0	1	900
Land area	2.585	2.136	0.057	16.500	900
Irrigation	0.360	0.431	0	1	900
Landscape	0.718	0.293	0	1	900
Block size	1.746	0.882	0.584	3.946	900
Farmland reallocation	1.113	0.952	0	3	900
Farmland rental policy	0.490	0.500	0	1	900
Xifeng	0.160	0.367	0	1	900
Yuzhong	0.174	0.380	0	1	900
Minle	0.137	0.344	0	1	900
Gulang	0.031	0.174	0	1	900
Lingtai	0.033	0.180	0	1	900
Jingning	0.152	0.359	0	1	900
Huining	0.147	0.354	0	1	900
Min	0.166	0.372	0	1	900

Note: The mu unit is widely used as the unit of farmland in China, and 1 mu is 1/15 hectare; renting land = 1 if a household rented in some land, otherwise = 0; renting out land =1 if a household rented out some land, otherwise = 0; head age is the age of household head; head age square is the square of head age; occupation = 1 if the household head was farming, otherwise = 0; education is the education year of household head; health condition =1 if the health condition of household head was good, otherwise = 0; labor ratio is the quotient of labor numbers to family size and family labors, including all family members who were able-bodied, not in school, and between 16 and 65 years old [7]; income is the annual income per capita, measured in thousand yuan; off-farm income is the quotient of off-farm income to total income of a household; land area is the original owned farmland area per capita (the contracted farmland), measured in mu; irrigation is the ratio of irrigable land at village levels; landscape is the ratio of farmland in level areas at village levels; block size is the average block size at village levels, measured in mu; farmland reallocation is the frequency of which the farmland has been reallocated by any kind of reallocation since the start of Household Contracted Responsibility System (HCRS); farmland rental policy is the answer of households to the question, “Do you know that the Chinese Central Government is encouraging farmland rental?”.

. According to previous studies, exogenous variables $X$ including variables of farmer characteristics (head age, age square, occupation, education, health condition, labor ratio, income, and off-farm income), variables of farmland characteristics (land area, irrigation, landscape, and block size), some other factors (farmland reallocation and farmland rental policy) and county dummies (Xifeng, Yuzhong, Minle, Gulang, Lingtai, Jingning, and Huining) were expected to influence farmland rental behaviors of households [7,29,46].

The descriptive information of wheat and maize production is presented in

Table 3. Fertilizer and pesticide use and output of wheat and maize production.

Variable	Mean	SD	Min.	Max.	Observations
Wheat analysis
Planted area	4.580	13.920	0.300	300.000	479
Output per mu	296.718	98.933	45.000	570.000	479
Fertilizer cost per mu	100.754	47.786	0.000	260.000	479
Pesticide cost per mu	11.745	13.886	0.000	73.000	479
Fertilizer cost per kg wheat	0.360	0.224	0.000	1.895	479
Pesticide cost per kg wheat	0.041	0.050	0.000	0.322	479
Maize analysis
Planted area	4.213	4.968	0.100	80.000	452
Output per mu	579.153	159.784	220.000	1050.000	452
Fertilizer cost per mu	126.914	61.756	0.000	330.000	452
Pesticide cost per mu	16.566	16.873	0.000	95.000	452
Fertilizer cost per kg maize	0.230	0.137	0.000	1.222	452
Pesticide cost per kg maize	0.030	0.033	0.000	0.259	452

Note: 1 mu is 1/15 hectare; planted area is measured in mu; output per mu is measured in kg; all the input costs are measured in yuan.

. The average planted wheat and maize area was 4.580 mu and 4.213 mu, respectively. The average output of wheat was 296.718 kg∙mu⁻¹, which is approximately in line with the data of Rada et al. [6], and the average output of maize was 579.153 kg∙mu⁻¹. The data suggest a significant variability in fertilizer and pesticide use as well as crop outputs in the study area.

5. Empirical Results

5.1. Estimation Results of the Propensity Score

The Logit model was used to estimate propensity scores of rural households for renting land and renting out land. A condition number test was made to test the multicollinearity of explanatory variables before estimating models. The condition index of explanatory variables in the Logic models of renting land (149.85) and renting out land (140.80) were both lower than 30 [47], indicating there was no multicollinearity in the Logic regressions (

Table 4. Estimation of the propensity score for renting land and renting out land.

Variable	Renting Land		Renting Out Land
	Coefficient	Robust Standard Error	Coefficient	Robust Standard Error
Intercept	−11.882 ***	3.770	−0.046	2.062
Head age	0.403 ***	0.147	0.112	0.077
Age square	−0.004 ***	0.002	−0.001	0.001
Occupation	0.579	0.460	−1.749 ***	0.261
Education	−0.050	0.350	−0.047	0.033
Health condition	0.004	0.322	−0.502 *	0.279
Labor ratio	0.166	0.546	−0.486	0.502
Income	0.014 *	0.010	0.019 *	0.010
Off-farm income	−1.022 ***	0.374	0.815 ***	0.146
Land area	−0.026	0.072	0.136 **	0.053
Irrigation	1.172 *	0.627	0.720	0.64
Landscape	2.121 **	1.050	−1.151	0.981
Block size	0.310	0.503	−1.110 **	0.442
Farmland reallocation	−0.556 ***	0.179	−1.121 ***	0.172
Farmland rental policy	1.170 ***	0.304	1.395 ***	0.254
Xifeng	−2.960 *	1.661	2.863 **	1.419
Yuzhong	−0.241	0.588	0.084	0.592
Minle	−0.163	0.833	−0.628	0.869
Gulang	0.063	0.993	−0.145	1.147
Lingtai	0.340	0.724	−0.790 *	1.054
Jingning	0.277	0.678	1.155 **	0.662
Huining	0.386	0.660	−1.602	0.738
Observations	691		786
Pseudo R2	0.171		0.337
Log pseudolikelihood	−256.598		−301.945
Condition index	149.85		140.80
AUC	0.785		0.866

Notes: *, ** and *** indicate statistical significance at 10%, 5% and 1% levels, respectively. AUC= area under the receiver operating characteristic curve.

). Robust standard errors were applied to reduce the heteroscedasticity in the regressions. The pseudo R² value, which was widely used to measure the good fitness of a model, was 0.171 (renting land) and 0.337 (renting out land) in this study, showing the models had good fit with micro data [42]. Additionally, the AUC indicator (area under the receiver operating characteristic curve) greater than 0.7 (0.785 for renting land and 0.866 for renting out land) could be considered a good indicator that the model’s specifications were appropriate in this study [48].

Results of Logit formulations for the propensity scores of renting land and renting out land are reported in Table 4. Some farmer and farmland characteristics significantly influenced farmland rental. For farmer characteristics, an inverted U-shaped relationship was found between renting land behavior and household head age with a peak point of 50.430 years old. Households whose heads were not involved in farming were less likely to rent out their land (−1.749). Off-farm income ratio was negatively (−1.022) correlated with renting land behavior but positively (0.815) correlated with renting out land behavior. These results are in line with the study of Kung [28] and Huang et al. [7]. For farmland characteristics, contracted land area per capita was positively (0.136) correlated with renting out land behavior of the household, indicating households lacking labors were more likely to rent out their land than those with sufficient labors. The significant positive coefficients of irrigation (1.172) and landscape (2.121) indicated a high demand for high quality farmland. The significant negative coefficient (−1.110) of block size indicated households tended to rent out their land with small area. The significant negative effect of farmland reallocation (−0.556 for renting land and −1.121 for renting out land) and significant positive effect of rural households’ perceptions of farmland rental policy (1.170 for renting land and 1.395 for renting out land) on households’ renting land and renting out land behaviors indicate the importance of farmland right security and government support on the development of farmland rental market in China. The location of rural households also had some effects on households’ farmland rental behaviors.

5.2. Tests of Sample Matching

Estimating p-scores enabled similar samples from the treatment and the control groups to be obtained after sample matching. Three matching methods (NNM, RBM, and KBM) were used to estimate the effects of farmland rental on fertilizer and pesticide costs of wheat and maize production in this study, and results from different methods were the same (Table 7). Thus, only the test results for the matching quality with NNM are reported here.

Table 5. Balancing hypothesis test showing characteristics of variables before and after matching in wheat analysis.

Variable	U M	Renting Land			Renting Out Land
		Mean of Treatment Group	Mean of Control Group	Bias (%)	Mean of Control Group	Mean of Treatment Group	Bias (%)
Head age	U	49.440	50.985	−15.6	50.000	50.985	−8.1
	M	49.617	49.512	1.1	49.941	50.457	−4.7
Age square	U	2508.000	2730.900	−21.9	2610.800	2730.900	−10.0
	M	2526.100	2525.500	0.1	2609.400	2660.700	−4.6
Occupation	U	0.964	0.883	31.0 **	0.845	0.883	−10.9
	M	0.963	0.954	3.5	0.868	0.834	9.7
Education	U	6.595	6.556	1.1	6.592	6.556	1.0
	M	6.728	6.963	−6.4	6.574	6.863	−7.6
Health condition	U	0.833	0.716	28.3 **	0.718	0.716	0.5
	M	0.852	0.824	6.7	0.735	0.717	4.1
Labor ratio	U	0.674	0.648	12.1	0.686	0.648	16.1
	M	0.673	0.656	7.9	0.683	0.684	−0.5
Income	U	12.147	11.781	2.7	11.351	11.781	−3.8
	M	10.907	12.108	−8.9	11.035	11.328	−2.6
Off-farm income	U	0.571	0.707	−43.1 ***	0.736	0.707	11.5
	M	0.592	0.576	5.0	0.732	0.736	−1.3
Land area	U	2.819	2.697	6.4	2.803	2.697	5.5
	M	2.861	2.709	7.9	2.765	2.789	−1.3
Irrigation	U	0.452	0.342	24.5 **	0.209	0.342	−32.9
	M	0.456	0.485	−6.4	0.218	0.218	0.1
Landscape	U	0.749	0.769	−7.3	0.793	0.769	9.7
	M	0.748	0.756	−2.9	0.802	0.797	1.6
Farmland reallocation	U	1.655	2.042	−50.6 ***	2.155	2.042	12.8
	M	1.662	1.632	4.0	2.177	2.169	0.9
Farmland rental policy	U	0.679	1.083	−50.4 ***	0.479	1.083	−79.3 ***
	M	0.679	0.757	−9.7	0.657	0.683	-6.8
Block size	U	0.690	0.404	59.8 ***	0.394	0.404	−2.0
	M	0.674	0.662	5.7	0.338	0.339	−0.2
Xifeng	U	0.060	0.250	−54.5 ***	0.324	0.250	16.3
	M	0.062	0.056	1.8	0.338	0.338	−0.1
Yuzhong	U	0.071	0.130	−19.4	0.056	0.130	−25.4 *
	M	0.074	0.108	−9.	0.059	0.063	−1.5
Minle	U	0.214	0.105	30.1 ***	0.127	0.105	6.8
	M	0.610	0.207	0.8	0.132	0.135	−0.8
Gulang	U	0.036	0.068	−14.5	0.000	0.068	−38.1 **
	M	0.037	0.034	1.4	0.000	0.000	0.0
Lingtai	U	0.071	0.062	3.9	0.028	0.062	−16.2
	M	0.049	0.040	3.7	0.029	0.032	−1.2
Jingning	U	0.286	0.154	32.0 ***	0.423	0.154	61.7 ***
	M	0.296	0.291	1.5	0.397	0.378	4.5
Huining	U	0.202	0.173	7.5	0.028	0.173	−49.5 ***
	M	0.210	0.210	0.0	0.029	0.035	−2.0
Min	U	0.060	0.059	0.4	0.014	0.059	−23.9
	M	0.062	0.056	2.6	0.015	0.019	−2.4
Treated observations	U	84			71
	M	74			62
Control observations	U	324			324
	M	302			264
Total observations	U	408			395
	M	376			326

Note: U and M indicate before and after matching, respectively; *, ** and *** indicate that the difference between the treated and control groups is significant at 10%, 5%, and 1% levels, respectively; bias is the standardized difference [50]. The same in Table 6.

and

Table 6. Balancing hypothesis test showing characteristics of variables before and after matching in maize analysis.

Variable	U M	Renting Land			Renting Out Land
		Mean of Treatment Group	Mean of Control Group	Bias (%)	Mean of Control Group	Mean of Treatment Group	Bias (%)
Head age	U	48.821	51.638	−30.2 **	51.108	51.638	−4.7
	M	49.014	48.840	1.8	51.083	51.872	−7.1
Age square	U	2445.900	2789.400	−35.1 **	2736.700	2789.400	-4.5-
	M	2467.300	2460.800	0.7	2742.800	2820.000	6.7
Occupation	U	0.949	0.880	24.6 *	0.815	0.880	−18.0
	M	0.945	0.946	−0.4	0.850	0.830	−5.5
Education	U	7.423	6.641	22.6 *	6.369	6.641	−7.3
	M	7.370	7.537	−4.8	6.500	6.378	3.3
Health condition	U	0.876	0.741	26.1 *	0.692	0.741	−10.8
	M	0.849	0.821	6.9	0.733	0.706	6.0
Labor ratio	U	0.649	0.655	−2.7	0.664	0.655	3.8
	M	0.650	0.639	5.1	0.671	0.659	5.5
Income	U	11.459	11.805	−2.9	12.288	11.805	3.9
	M	11.508	11.951	−3.8	11.989	11.801	1.5
Off-farm income	U	0.597	0.668	−21.7 *	0.717	0.668	17.0
	M	0.590	0.597	−2.2	0.713	0.716	-1.1
Land area	U	2.944	2.825	6.3	2.923	2.825	4.9
	M	2.926	2.893	−1.7	2.571	2.754	−9.2
Irrigation	U	0.446	0.344	23.7 *	0.222	0.344	−31.1 **
	M	0.463	0.463	−0.1	0.222	0.255	−8.5
Landscape	U	0.744	0.693	16.8	0.794	0.693	36.2 **
	M	0.746	0.746	−0.2	0.805	0.806	−0.5
Block size	U	1.731	2.085	−47.7 ***	2.195	2.085	12.4
	M	1.751	1.742	1.2	2.235	2.197	4.3
Farmland reallocation	U	0.731	1.256	−58.0 ***	0.462	1.256	−94.4 ***
	M	0.731	0.718	1.4	0.500	0.694	−8.3
Farmland rental policy	U	0.756	0.453	65.0 ***	0.308	0.453	−30.2 **
	M	0.577	0.513	2.7	0.349	0.367	−4.0
Xifeng	U	0.077	0.217	−40.2 ***	0.338	0.217	27.3 **
	M	0.082	0.083	−0.4	0.367	0.350	3.7
Yuzhong	U	0.103	0.188	−24.3 *	0.108	0.188	−22.6
	M	0.110	0.109	0.2	0.117	0.139	−6.4
Minle	U	0.141	0.052	30.4 ***	0.062	0.052	4.2
	M	0.137	0.153	−5.5	0.050	0.069	−8.2
Gulang	U	0.051	0.068	−7.0	0.000	0.068	−38.1 **
	M	0.055	0.055	−0.3	0.000	0.000	0.0
Lingtai	U	0.051	0.006	26.8 ***	0.000	0.006	−11.4
	M	0.000	0.006	−3.3	0.000	0.000	0.0
Jingning	U	0.269	0.168	24.5 ***	0.415	0.168	56.2 ***
	M	0.288	0.274	3.3	0.400	0.370	6.7
Huining	U	0.308	0.301	1.5	0.077	0.301	−59.5 ***
	M	0.329	0.319	2.0	0.067	0.071	−1.2
Treated observations	U	78			65
	M	76			62
Control observations	U	309			309
	M	282			226
Total observations	U	387			374
	M	358			288

report the results of balancing hypothesis tests to the sample matching with NNM. The results indicated that significant differences based on characteristics of $X$ between the treatment groups and the corresponding control groups did not exist after matching [49], and all the biases after matching were below 10% [50,51]. Thus, the balancing hypothesis was satisfied in this study.

The kernel density functions before and after matching for the treatment and the control groups are shown in Figure 2. It was apparent from the figures that differences in the density of functions between treatment groups and their corresponding control groups were highly significant prior to conducting the matching procedure. After matching, the distribution density functions of treatment groups and their corresponding control groups were very similar, and there was an evident decrease in their deviations. This evident contrast indicated that the common support assumption was satisfied in this study [37,42].

5.3. Effects of Farmland Rental on Fertilizer and Pesticide Cost

The estimated results of the effects of farmland rental on fertilizer and pesticide costs of wheat and maize production are presented in

Table 7. Matching estimates for effects of renting land and renting out land on fertilizer and pesticide uses.

Variable	Renting Land			Renting Out Land
	NNM	RBM	KBM	NNM	RBM	KBM
Output per mu	−14.599	−15.210	−14.077	−37.904 ***	−37.848 ***	−38.706 ***
	(13.785)	(11.961)	(11.454)	(13.622)	(12.716)	(12.495)
Fertilizer cost per mu	−16.281 **	−14.738 **	−14.789 **	−6.765	−8.290	−8.960
	(7.495)	(6.653)	(6.551)	(6.971)	(6.169)	(6.177)
Pesticide cost per mu	−4.239 *	−3.240 *	−3.236 *	−7.325 ***	−6.723 ***	−6.623 ***
	(2.174)	(1.957)	(1.840)	(1.949)	(1.342)	(1.411)
Fertilizer cost per kg wheat	−0.046	−0.056	−0.067	0.052	0.031	0.020
	(0.085)	(0.063)	(0.062)	(0.040)	(0.038)	(0.046)
Pesticide cost per kg wheat	−0.008	−0.056	−0.006	−0.017 ***	−0.017 ***	−0.017 ***
	(0.009)	(0.008)	(0.007)	(0.006)	(0.005)	(0.006)
Output per mu	−84.631	−91.262	−87.233	−93.306 *	−75.290 *	−74.900 *
	(59.621)	(59.331)	(59.094)	(47.373)	(43.170)	(42.652)
Fertilizer cost per mu	−32.981 ***	−36.982 ***	−36.985 ***	−8.992	−9.114	−9.666
	(11.201)	(10.361)	(9.984)	(9.894)	(8.570)	(8.385)
Pesticide cost per mu	−5.502 **	−5.916 **	−5.775 **	−7.708 ***	−7.139 ***	−7.218 ***
	(2.785)	(2.608)	(2.338)	(2.913)	(2.355)	(2.300)
Fertilizer cost per kg maize	−0.060 ***	−0.060 ***	−0.061 ***	0.017	0.009	0.008
	(0.021)	(0.020)	(0.021)	(0.023)	(0.022)	(0.023)
Pesticide cost per kg maize	−0.010 **	−0.011 ***	−0.010 ***	−0.010 **	−0.010 **	−0.012 **
	(0.005)	(0.003)	(0.004)	(0.005)	(0.005)	(0.004)

Note: 1 mu is 1/15 hectare; standard errors were derived from bootstrap method with 10,000 replications; *, ** and *** indicate significance at 10%, 5% and 1% levels, respectively; outputs are measured in kg and all costs are measured in yuan. NNM= nearest neighbor matching, RBM= radius-based matching, KBM= kernel-based matching.

. Overall, renting land significantly reduced fertilizer and pesticide costs per mu for both wheat and maize production, meanwhile significantly reducing the fertilizer and pesticide costs per kg output of maize production. Renting out land significantly reduced the output per mu, pesticide cost per mu, and pesticide cost per kg output of both wheat and maize production.

Households renting land used significantly less fertilizer and pesticides per unit of land area for both wheat and maize production than those not renting land (Table 7). The “nearest-neighbor” casual effects of renting land on fertilizer and pesticide costs per mu of wheat production were significant and equal to −16.281 yuan and −4.239 yuan, respectively, which were the average differences between fertilizer and pesticide cost per mu for wheat production for similar pairs of households belonging to different groups (renting-land group and not-renting-land group). In other words, the average fertilizer and pesticide costs per mu of wheat production of households renting land were 16.281 yuan and 4.239 yuan less than those of households not renting land, respectively. Similarly, the average fertilizer and pesticide costs per mu of maize production of households renting land were 32.981 yuan and 5.502 yuan less than those of households not renting land, respectively. This resulted in a reduction of 0.060 yuan and 0.010 yuan in fertilizer and pesticide cost per kg maize, respectively, for households renting land compared with those not renting land.

Surprisingly, households renting out land also significantly reduced their pesticide cost per mu and output per mu of wheat and maize production, but not fertilizer cost, compared with those not renting land (Table 7). Meanwhile, reduction in outputs per mu was also observed in both wheat and maize production of households renting out land. The average pesticide costs per mu of wheat and maize production of households renting out land were 7.325 yuan and 7.708 yuan less than those of households not renting land with NNM, respectively. The significantly lower average outputs per mu of wheat and maize production of households renting out land, which were 37.904 kg and 93.306 kg lower than those of households not renting land, respectively, might partially be attributed to the reduction of pesticide use in wheat and maize production. The reduction on pesticide use also caused a reduction of 0.017 yuan and 0.010 yuan in pesticide cost per kg outputs of wheat and maize, respectively, for households renting land compared with those not renting land.

5.4. Situation of Adoption of New Agricultural Technologies and Management Methods

Figure 3 shows the situation of adoption of new agricultural technologies and management methods in the study area. The results indicated that households adopting new agricultural technologies or management methods took up 29.79% of the renting-land group, 5.08% of the renting-out-land group, and 10.14% of the not-renting-land group. The percentage of households in the renting-land group adopting new technologies and management methods was almost three times that of households in the not-renting-land group, while the percentage of households adopting new technologies and management methods of renting-out-land group was only around half of that of the not-renting-land group.

6. Discussion

This study shed new light on food safety by determining the effects of farmland rental on the use of fertilizer and pesticides in grain production. To overcome the selection bias problem of the survey data, propensity score matching (PSM) was employed in this study, which perfectly removed selection bias through creating a context close to experimental conditions [39].

Our study clearly showed that households renting land significantly reduced their fertilizer and pesticide use in wheat and maize production without decreasing crop outputs by much. This is partly explained by the findings of Ju et al. [52] and Pan et al. [14] that increasing farm size sharply reduces the use of fertilizer for agricultural production. One possible explanation is that the increase in farm size enables households to apply some new farming technologies and management methods based on scientific knowledge such as the 4R fertilization technology, which involves applying the right kind and amount of fertilizer at the right time by using the right application methods [53]. Moreover, households renting land often had higher expectations of profit from farming than those not renting land, and this encouraged them to be more specialized. Thus, a more rational amount of fertilizer and pesticide was used by households renting land. Indeed, this is confirmed by the results that households renting land had adopted a significantly higher percentage of new agricultural technologies and management methods than those not renting land. However, it is still worth noting that only 29.79% of the renting-land group had adopted new agricultural technologies and management, suggesting the adoption of new agricultural technologies and management methods adoption is limited in the study area. This may be attributed to the low level of economic development and poor education in the study area, hence farmers’ education about technology should be further improved.

On the other hand, households renting out land also significantly reduced their pesticide use of wheat and maize production but not fertilizer use. The numerous incidents of China’s grain food safety, such as the cadmium rice event in 2013 and the rice adulteration problems in 2015 [54], have weakened the public’s confidence in the safety of grain foods sold at markets. Thus, some households renting out land still have small patches of land for growing subsistence crops for their own use [55], which receive reduced amounts of agrochemicals or even no agrochemicals. Therefore, less pesticide use in wheat and maize production of households renting out land was observed. However, subsistence crop growth was not sensitive to fertilizer use, since the harmful effects of chemical fertilizer to human health are indirect. This caused the non-responsiveness of fertilizer use to renting out land. In addition, households renting out land often devoted more time and effort to off-farm working than to farming [9], which was also supported by the result that the percentage of households adopting new technologies and management methods of the renting-out-land group was much lower than that of other groups. Together with the reduction of pesticide use, there was a dramatic reduction in outputs for wheat and maize production among households renting out land.

In general, farmland rental significantly reduced fertilizer and pesticide use for production of both wheat and maize, but renting land and renting out land go different ways. Households renting land reduced their fertilizer and pesticide use by rational use of agrochemicals and by applying new agricultural technologies and management methods, while those renting out land reduced their pesticide use due to concerns about grain food safety and planted food for subsistence.

Our results are based on data from a single year, and this is a limitation to understanding the effects of changes in farmland rental on agrochemical use over time. However, our results still provide important references for farmland rental and food safety for policymakers. Besides, this study only determined the impacts of farmland rental on agrochemical use, which is an indirect indicator of food safety. Thus, the impacts of farmland rental to direct indicators of food safety such as pesticide residues and heavy metal content of the food should be further studied.

7. Conclusions and Implication

The causal treatment effect of farmland rental on agrochemical use of grain production was estimated in this study. The results indicate that farmland rental can greatly reduce the agrochemical use of grain production. Additionally, the mechanism of households’ renting and renting out land behaviors influencing agrochemical use was also analyzed. The study found that households renting land mainly reduced agrochemical use through rational agrochemicals usage and new agricultural technology and management method applications, while those renting out land mainly reduced agrochemical use with concerns of subsistence grain safety.

The results of this study have important implications for farmland rental, food security, and environmental conservation. As a conclusion, our findings suggest that farmland rental contributes to increased food safety and environmental conservation by reducing fertilizer and pesticide use. The development of China’s farmland rental markets will continually benefit not only agricultural productivity but also food safety. With the further development of farmland rental markets in China, farmers’ knowledge on agricultural technologies and management will become increasingly important to the improvement of productivity and food safety in China. To ensure agricultural productivity and food safety, support for farmer education and training should be increased further.