ZHU Wei ,QI Li-xia ,WANG Rui-mei

1 College of Economics and Management,China Agricultural University,Beijing 100083,P.R.China

2 School of Public Administration,North China University of Water Resources and Electric Power,Zhengzhou 450045,P.R.China

Abstract Low fertilizer use efficiency is commonly found in China’s agriculture sector. It is critical to understand the drivers of fertilizer use in order to increase fertilizer use efficiency (FUE). The purpose of this paper is to investigate the relationship between farm size and FUE based on farm-level data in China. The results suggest a positive farm size-FUE relationship. The estimated elasticities of farm size for rice,maize,wheat-maize,and rice-maize are similar,and the average elasticity of farm size is around 0.2. Statistically,a 1% increase in farm size is associated with a 0.2%increase in FUE. The positive effect of farm size on FUE is not due to the increase in yield,but the reduction in fertilizer use while keeping yield largely unchanged. The findings suggest that promoting the expansion of farm size may increase FUE,and thus reduce environmental pollution caused by fertilizer overuse.

Keywords:farm size,farm scale,fertilizer use efficiency,positive relationship

1.lntroduction

Fertilizers play an important role in food production worldwide and are essential to feed a growing and increasingly wealthy global population (Li and Zhang 2013;Jiang et al.2018). However,not all of these nutrients applied to crops are used effectively. For example,on average,only 30 to 50% of nitrogen can be assimilated by crop plants (Lehnert et al.2018). These nutrients that crop plants cannot assimilate continuously leach from soil into water or escape into the atmosphere,which can cause various environmental problems (Reay et al.2012). Accordingly,the negative environmental externalities of fertilizer application need to be considered,though the global demand for fertilizers in agriculture has consistently increased to meet the growing demand for food.

Low use efficiency and a high proportion of loss of fertilizers are commonly found in China when farmers attempt to increase yields by fertilizer overuse,leading to serious local,regional,and global pollution (Abler 2015;Zhang et al.2015;Cui et al.2018;Huang and Jiang 2019). China consumes around 28% of global fertilizers on less than 9% of global cropland (FAOSTAT 2017).Additionally,there are more than 200 million smallholder farmers in China,and the average cropping area per labor is around 0.57 ha. Consequently,the agricultural system relies heavily on high-to-excessive inputs (Cui et al.2018). For example,the average rate of fertilizer application is 389.79 kg ha-1in China compared to 97.90 kg ha-1worldwide excluding China (FAO 2017);and the nitrogen use efficiency is only 0.25 in China compared to 0.42 worldwide (Zhang et al.2015). China’s 1st National Census on Pollution Sources showed that the total nitrogen loss from crop cultivation was 1 597 800 tons,and the total phosphorus loss was 108 700 tons,accounting for 33.79 and 25.69% of the total emissions,respectively (MEP et al.2010). In recent years,the environmental pollution and health risk caused by fertilizer overuse have caused serious concerns among the government and the public. To address this problem,the Chinese government has exerted effort to reduce pollution from fertilizer overuse,including the free service of soil testing,engaging smallholder farmers to adopt enhanced management practices,the removal of subsidies to chemical fertilizer manufacturers,and providing subsidies to organic fertilizers (Zhang et al.2016;Ning et al.2017;Cui et al.2018;Jiao et al.2018). To further promote the sustainable development of agriculture in China,it is critical to understand the factors that affect farmers’application of fertilizers.

According to Heisey and Norton (2007),the relative scarcity of agricultural land has been a major underlying cause of the expansion in demand of fertilizers,and more proximate causes include seed varieties and expansion of irrigation or better water control within irrigated systems,declined real fertilizer prices,etc. In addition,soil characteristics,seed varieties,hired labor,applied technology,and education all influence farmers’ use of fertilizers (Qiao et al.2003;Heisey and Norton 2007;Wu 2011;Huang et al.2015;Huang and Ding 2016;Li et al.2017;Mason et al.2017;Morello et al.2018). Another important determinant of farmers’ fertilizer use behavior is farm size,as revealed by a range of studies (Zhang et al.2013;Ju et al.2016;Wu et al.2018;Cheng et al.2019;Hu et al.2019;Ren et al.2019;Zhu and Wang 2020). Most of these studies suggested that large-scale farms on average use less fertilizers than small-scale farms. However,the fertilizer use efficiency (FUE),which is a better indicator of crop assimilation efficiency or nutrients loss rate (Cassman et al.1998),has not been fully discussed in the prior literature. The current study attempts to provide new evidence for this issue. This article investigates the relationship between farm size and FUE among farmers of rice,wheat,and maize by using farm-level agricultural production data of China.

The remainder of the article is structured as follows.The subsequent section presents the data source,variable definition,and estimation methodology. The empirical results are presented followed by a section discussing some factors that limit the expansion of farm size. The last section provides conclusions.

2.Data and methods

2.1.Data source

To investigate the relationship between farm size and fertilizer use efficiency (FUE),this study relied on household survey data from the 2015 China Household Finance Survey (CHFS),conducted by Survey and Research Center for China Household Finance. Although CHFS is a panel dataset,only 2015 CHFS has detailed agricultural production data. That is the reason for which this study only used the data from the 2015 CHFS.

Different crops have different requirements for fertilizers,so the analysis of fertilizer use issues should be discussed for specific crops. A total of 12 035 farm households from 29 provinces in China,who engaged in agricultural production,were interviewed in 2015 CHFS.Of these,6 226 farm households grew grain crops,among which 1 716 farm households also grew cash crops,which means that 4 510 farm households only grew grain crops. Of these,1 130 farm households grew potatoes,sweet potatoes or other grains in addition to rice,wheat,and maize,which may affect the accuracy of fertilizer use analysis. Therefore,3 380 farm households that only grew rice,or maize,or wheat-maize,or rice-maize were selected. Of these,the number of farm households who grew rice,maize,wheat-maize,and rice-maize are 845,1 135,1 057,343,respectively. Wheat-maize and ricemaize represent that farmers rotate two crops on the same farmland in one year.

2.2.Variable definition

According to Cassman et al.(1998),FUE,the dependent variable,is defined as the ratio of grain yield to the quantity of applied fertilizers per unit area,which is the indicator of FUE commonly used in agronomy:

where Y is the grain yield obtained with an applied fertilizer of F;Y0is the grain yield without fertilizer inputs;ΔY is the incremental increase in grain yield that results from fertilizer application;ΔY/F is often called the agronomic efficiency,and it represents the product of uptake efficiency from the applied fertilizers. Therefore,higher FUE means that a higher proportion of fertilizers is converted into grain,and a smaller proportion of fertilizers is lost.

The CHFS investigated farmers’ expenditure on fertilizers rather than the quantity of fertilizers used.Although farmers’ expenditure on fertilizers was used by many studies (Wu et al.2018;Cheng et al.2019)as the proxy variable of the quantity of fertilizers used,considering the possible influence of prices of fertilizers,this paper divided farmers’ expenditure on fertilizers by the average fertilizer price to get the quantity of fertilizers used. PDNDRC (2016) investigated the fertilizer use per unit area and fertilizer expenditure of each crop in each province in 2015,and the average price of fertilizers used by each crop in each province were calculated based on these data.

The major independent variable of interest in this study is farm size,expressed in terms of cropping area of rice,maize,wheat-maize,or rice-maize planted by the farmers. In order to analyze the causality between farm size and FUE,it is necessary to include control variables in the model. Other explanatory variables to be controlled in the model include soil quality,hired labor,farmland rental,irrigation,the number of agricultural labors in the family,farmer’s risk preference,knowledge and skills,operating types,and province where the farm is located.The soil quality was reported by the respondents,and the options are measured from 1 to 5,with 1 representing very good condition and 5 representing very poor condition. The indicators for whether the farmer hired labor,rented farmland,and whether the farm had sufficient irrigation were all set as binary variables,with 1 representing“Yes”,and 0 otherwise. The operating types include agricultural enterprise,agricultural cooperative,family farm,leading farm households,and general farm household. The operating types and provinces were set as dummy variables.

A complex problem is to find the appropriate variables as the proxy variables for farmer’s risk preference,knowledge and skills. There are two questions about the type of financial investment in the questionnaire of CHFS,which can be used to capture farmer’s risk preference.The options for the first question are values of 1 to 5,with 1 representing high risk and 5 representing no risk at all. The options for the second question are values of 0 and 1,with 0 representing high risk and 1 representing low risk. The questionnaire of CHFS contained four more questions that can be used to capture farmer’s knowledge and skills:whether the farmer can correctly calculate the questions about deposit interest and inflation,is engaged in business,and uses online banking. Farmers who are engaged in business have more opportunities to access market,government,and other sources of information,so that they may have more opportunities to access new agricultural technologies (Caffaro et al.2020). Farmers who use online banking may be more interested in new technologies and have enough ability to use new technologies (Li et al.2021). The indicators for whether the farmer can correctly calculate the questions about deposit interest and inflation,is engaged in business,and uses online banking were all set as binary variables,with 1 representing“Yes”,and 0 otherwise. The descriptive statistics for all variables are presented in Table 1.

Table 1 Descriptive statistics for variables used in analysis

2.3.Methods

While little is known about the relationship between farm size and FUE,there are a large body of evidence on the relationship between farm size and productivity,whichthis study can draw on (Wang et al.2015;Muyanga and Jayne 2019;Sheng et al.2019;Yan et al.2019;Yao and Hamori 2019). Following these studies,the model in this paper is as follows:

where subscript i denotes the farm household;FUE,the dependent variable,denotes fertilizer use efficiency;A denotes farm size or cropping area;X are various control variables,including soil quality,irrigation,farmer’s risk preference,etc.;γ and βkare estimated coefficients;and ε is the random error term. Both farm size and FUE are in logarithm in the model,so that the coefficient of farm size is its elasticity. The square-term of farm size has been added into the model to capture the potential non-linear relationship between farm size and FUE.

Eq.(2) can be estimated using the ordinary least squares (OLS) regression,or the weighted least squares to correct for heteroscedasticity. The obtained coefficients of farm size (γ and δ) may be biased from the real relationship between farm size and FUE. This is attributed to the potential endogeneity problem. For example,farmers with more knowledge and skills are more likely to rent more farmland and operate larger farms. Therefore the OLS estimate of the coefficient on farm size reflects both the effect of farm size and the selection effect on the FUE.

To address this potential concern,an instrumental variable regression was adopted to control for unobserved variables. The contractual size of cropland was used as an instrumental variable,which was also used by Wu et al.(2018) and Sheng et al.(2019),to identify farm size. The use rights of farmland were allocated to each farm household according to population characteristics of each farm household and the village endowment under the Household contract responsibility system (Lin 1992;Brandt et al.2004). Therefore,the size of the farmland allocated to each household is unlikely to be correlated with farmer’s knowledge and skills. Meanwhile,because the transfer of farmland use rights is limited by various factors in China (Wu et al.2018),the current farm size and the contractual size are closely related (correlation coefficient is around 0.6). To test the effectiveness of the instrumental variable,this study first conducted the underidentification test and weak identification test;the test results (Table 2) showed that the null hypothesis of underidentification and weak identification both can be rejected.Then,the correlation between the contractual size and the farm size was tested;the Shea’s partial R2and the P value of F-statistic suggested that the correlation between contractual size and farm size was statistically significant. This study then conducted Durbin-Wu-Hausman test on the endogeneity of farm size;the test results showed that for the samples of rice,maize,wheatmaize,the null hypothesis that farm size is the exogenous explanatory variable can be rejected;but for the samples of rice-maize,the null hypothesis cannot be rejected.Therefore,the contractual size could be regarded as a valid instrumental variable for farm size. The instrumental variable regression can capture the effect of farm size on FUE while excluding the selection effect.

Table 2 Test results for instrumental variables

Based on eq.(2),this study first estimated models only containing farm size and square-term of farm size using OLS,then estimated models with control variables to compare these results. Furthermore,a two-stage least squares (2SLS) estimation was implemented by using the contractual size as the instrumental variable,and the results were compared with those from the OLS estimation.

3.Results

The estimation results for the farm size-FUE relationship without control variables are shown in Appendices A and B,and the estimation results with control variables are presented in Table 3 (detailed results shown in

AppendicesC and D). The estimated coefficients of farm size are positive and most of them are statistically significant at the 1% level for rice,maize,wheat-maize,and rice-maize,whether control variables are included or not. Most coefficients of the square-term of farm size are not statistically significant,and thus the square-term was removed from eq.(2) according to Akaike Information Criterion and Bayesian Information Criterion.

The differences between the estimated coefficients of farm size with control variables and those without control variables using OLS are very small,suggesting robust estimated results. The estimated coefficients of farm size,namely elasticities,for rice,maize,wheatmaize,and rice-maize are 0.286,0.234,0.269,0.374,respectively,which indicates that farm size has a positive and statistically significant effect on FUE.

The results of the 2SLS estimation using the contractual size as the instrumental variable to identify farm size are compared with the results of the OLS estimation. While the OLS estimate of the coefficient on farm size reflects both the effect of farm size and the selection effect on the FUE,the 2SLS estimate captures mostly the former. The estimated coefficients of farm size using 2SLS are still significant but smaller in magnitude than those using OLS. This implies that both the farm size and the selection effect play important roles in the positive relationship between farm size and FUE. Overall,the results of the 2SLS estimation still suggest a positive farm size-FUE relationship,though they are not exactly the same as those of the OLS estimation.

Some of the estimated coefficients of control variables are statistically significant. First,for rice growers,the higher the risk preference is,the higher the FUE is.Second,for maize growers,the higher the soil quality is,the higher the FUE is;the number of agricultural labors in the family has a negative impact on FUE;farmers who use online banking have higher FUE than others. Third,for wheat-maize growers,sufficient irrigation has a positive effect on FUE;compared with general farm household,agricultural enterprise’s FUE is higher,but family farm’s FUE is lower. Fourth,for rice-maize growers,those who rent farmland have lower FUE than others;compared with general farm household,agricultural enterprise’s FUE and agricultural cooperative’s FUE are higher.

To examine the mechanism of the effect of farm size on FUE,this study replaced FUE with fertilizer use intensity and grain yield,and re-estimated the models using OLS and 2SLS. The estimation results are presented in Table 4 (detailed results shown in Appendices E and F).The results suggest that farm size has a significant negative effect on the fertilizer use intensity,but has no significant or very small effect on yield. In other words,the positive effect of farm size on FUE is not due to theincrease in yield,but due to the reduction in fertilizer use while keeping the yield basically unchanged.

Table 3 Estimation results for the farm size-FUE relationship

After estimating all the samples of various crops,the average farm size elasticity was obtained. The estimation results are presented in Table 5 (detailed results shown in Appendix G). For all crops,the elasticity of farm size is around 0.2,in other words,holding other factors constant,a 1% increase in farm size is associated with a 0.2%increase in FUE.

In order to investigate the effect of farm size on FUE more comprehensively,this study applied quantile regression to the data. As shown in Table 6 (detailed results shown in Appendices H-L),the estimated coefficients of farm size are positive and statistically significant. For different crops,as the quantile increases,the estimated coefficients of farm size show different trends. The results of quantile regression suggest that the main finding of this study,that there is a positive relationship between farm size and FUE,is robust.

4.Discussion

The main findings of this article are supported by Ju et al.(2016),Wu et al.(2018) and Ren et al.(2019). The expansion of large-scale farming is an important way to modernize agricultural production and ensure sustainable food production from the social perspective. This study analyzed the CHFS data in 2011,2013,2015 (Table 7),and found that while the median of cropping area remained 0.27 hectares,the mean of cropping area rose steadily. Meanwhile,the share of farm households who rented out farmland rose rapidly,but the share of farm households who rented farmland declined. This finding reveals that the farmland is gradually concentrated to large-scale farms,and this will help improve the FUE.

Table 4 Estimation results for fertilizer use intensity and yield

Table 5 Estimation results for all crops

Table 6 Estimation results using quantile regression

Table 7 Survey data of farm households’ farmland

According to 2015 CHFS,the reasons for making decisions on renting out (in) farmland are shown in Appendix M. The main reason for renting out farmland is whether the respondent is engaged in agriculturalproduction for income. When the farmer can get a stable job with higher income in other industries,he/she will cease to engage in agricultural production,and rent out the farmland. The proportion of rural population in China has decreased from 50.44% in 2010 to less than 40% in 2020. The average farm size rises steadily,thereby increasing the FUE and reducing environmental contamination caused by the loss of fertilizers.

Nevertheless,there are also several factors that restrict the free migration of rural population,such as the Household Contract Responsibility System and the Hukou (Household Registration) System (Wu et al.2018). More specifically,rural migrant workers often lack access to urban public services,such as urban public health care,child education,etc.(Cai 2011). This leads to the majority of about 290 million rural workers,who have obtained jobs in urban areas,still keep the ownership of farmland in rural areas as insurance. As shown in Appendix N,58.56% of respondents rented out their farmland to common farmers from local village,and 44.84% of those rented their farmland for free. For all samples,28.26% of respondents rented out their farmland for free. This may be due to the fact that these respondents cannot secure a stable job or life in urban areas,and thus have to keep the ownership of farmland as insurance.

In recent years,the Chinese government has realized the negative impact of farmland fragmentation and small scale on agricultural production,and has actively implemented a series of public policies to promote the development of medium-and large-scale farms. These policies include establishing land transfer service center to activate land rental market,policy supports for land consolidation,and innovated mechanization service for millions of family farms,etc. These driving forces have induced some small-scale farms to scaling up their farms while helping other small-scale farms rent out farmland and moved to off-farm employment,and they also induced the emergence of large-scale farms operated by land cooperatives and companies (Huang and Ding 2016).

5.Conclusion

This study investigates the relationship between farm size and fertilizer use efficiency (FUE) among producers of rice,wheat,and maize by using farm-level agricultural production data across China. The results show that there is a positive relationship between farm size and FUE. The estimated elasticities of farm size for rice,maize,wheatmaize,and rice-maize are similar,and the average elasticity of farm size is around 0.2. Statistically,a 1%increase in farm size is associated with a 0.2% increase in FUE. These estimates are robust to various control variables,to employing the 2SLS estimation by using the contractual size as the instrumental variable,and to quantile regression. The factors that limit the expansion of farm scale have also been discussed. These results suggest that promoting the expansion of farm size can increase FUE,and thus reduce environmental pollution caused by the loss of fertilizers. This finding provides new support for the government to promote the development of medium-and large-scale farms.

It should be noted that this study has several limitations. Due to data limitations,the quality or type of fertilizers was not considered. Some other characteristics of farm households which may affect fertilizer use or yield were not covered in the survey. Future research may use field surveys to obtain more accurate and detailed data to address the above limitations.

Acknowledgements

This work was supported by the National Social Science Fund of China (16AJY013).

Declaration of competing interest

The authors declare that they have no conflict of interest.

Appendicesassociated with this paper are available on

http://www.ChinaAgriSci.com/V2/En/appendix.htm