2.1 Emerging large-scale production units in China

With rapid industrialization and urbanization, livelihoods and socioeconomic structures in rural China have undergone tremendous changes. Many young rural workers have left traditional farming to join non-farming industries, resulting in a weaker agricultural sector, run by part-time farmers (Yang et al., Reference Yang, Wang, Jin, Chen, Riedinger and Peng2016). In response, the Chinese government has introduced policies encouraging migrant farmers to transfer their farmland operation rights to LSPUs.

LSPUs are ‘large-scale and market-oriented agricultural operation organizations or individuals, engaging in specialized production, intensive management and social services’ (Chen, Reference Chen2014) and mainly include professional farmers, family farms, agricultural cooperatives and agricultural enterprises (Song and Zhao, Reference Song and Zhao2014). Professional farmers, family farms and cooperatives originate and develop out of small farms rather spontaneously, while agricultural enterprises emerge through the investment of commercial capital in the agriculture industry. Due to different structural and organizational characteristics, investment decision-making processes of agricultural cooperatives and agricultural enterprises are more complex and specific. In this paper, the focus is on professional farmers and family farms, which we pool as LSPUs.

LSPUs acquire farmland by either accessing it directly from small-scale farmers or renting it via intermediary organizations (Yu et al., Reference Yu, Wu, Verburg, van Vliet, Yang, Zhou and Tang2013; Wang and Zhang, Reference Wang and Zhang2017). On the one hand, LSPUs rent plots from their relatives and acquaintances based on kinship or geopolitical networks, especially in areas where both land and labor markets are underdeveloped. Plots acquired through this type of transfer, however, are often small-scale and scattered; moreover, the contract conditions in this case are generally agreed upon by both sides verbally and the lessors can reclaim their land at any time (Meng, Reference Meng2012). In a more formalized land market, on the other hand, LSPUs may choose to access farmland from intermediary organizations which not only integrate scattered plots and save LSPUs transaction costs by reducing their number of transactions with individual small farmers, but also guide the signing of written contracts and rent out land through more formal contractual arrangements.

2.2 Contract type and soil-improving investments

The soil-improving investments of particular interest in this paper are organic fertilization, green manure planting and straw application, all of which improve soil quality and enhance productivity over more than one agricultural season. These investments are relationship-specific in that they are incorporated into the soil and thus are not transferable once applied; in addition, unlike fixed investments such as the construction of irrigation infrastructures which can be distinguished and verified to some extent, these three investments are non-contractible (Jacoby and Mansuri, Reference Jacoby and Mansuri2008; Abdulai and Goetz, Reference Abdulai and Goetz2014).

The content and type of contracts through which farmland is transferred to LSPUs reflect the levels of land tenure security (Place et al., Reference Place, Roth, Hazell, Bruce and Migot-Adholla1994; Guriev and Kvasov, Reference Guriev and Kvasov2005). Our theoretical analysis on the relationship between contract type and soil-improving investments is embedded in the broader framework of tenure security effects. Specifically, we distinguish two effects through which contract type in farmland transfer may contribute to long-term farm investments (Beekman and Bulte, Reference Beekman and Bulte2012). First, through an assurance effect, LSPUs are more willing to invest with a written contract since their land operational rights are legally endowed and protected in this situation. In case of insecure tenure where the tenancy arrangement for a particular plot is oral, the rights and obligations of the contracting parties are less clear; hold-up problems may arise due to the threat of opportunistic expropriation by leaseholders (Abdulai and Goetz, Reference Abdulai and Goetz2014). Consequently, without formal contracts established, short-run production gains and soil harvesting may be preferred over long-term investments in soil conservation, resulting in unsustainable practices; on the contrary, with written contracts, LSPUs are able to overcome the hold-up problems by adjusting investments to the point where they fully recover the returns, grounded on the conditions (e.g., contract duration) prescribed beforehand (Jacoby and Mansuri, Reference Jacoby and Mansuri2008; Yoder et al., Reference Yoder, Hossain, Epplin and Doye2008).

An indirect effect of contract type on long-term investment is its impact on access to credit. Due to the lack of qualified collateral, Chinese farmers face widespread financing problems (Feder and Onchan, Reference Feder and Onchan1987; Feder et al., Reference Feder, Lau, Lin and Luo1992; Bandiera, Reference Bandiera2007). LSPUs are more likely to become eligible for borrowing from formal financial institutions if their operated farmland is leased through a formal (long-term) contract. In such cases, land tenure security will most likely also result in larger investments through the indirect effect of improved credit access (Feder and Onchan, Reference Feder and Onchan1987).

3.1 Data collection and research area

Data were collected at the LSPU and plot level through a survey conducted in 40 townships in Jiangsu and Jiangxi Provinces in August and September 2016. Both provinces are important rice production areas in China. While Jiangsu Province is part of China's more developed coastal area, Jiangxi Province, located in Southeast China, is less developed. We consider the chosen two provinces as good representatives of major rice-producing areas in China, or even in other countries with similar socioeconomic characteristics.

In Jiangsu Province, Southern Jiangsu is considered more economically developed, followed by Middle Jiangsu and Northern Jiangsu. In Jiangxi Province, the Nanchang-Jiujiang Zone is most developed followed by the Jingdezhen-Yingtan Zone, Xinyu-Pingxiang Zone and Ganzhou Zone (Peng, Reference Peng2007).Footnote ³ For each province, a multi-stage stratified random sampling method was applied for townships and LSPUs in order to capture a diversity of levels of economic development, land endowment and geographical conditions. The number of countiesFootnote ⁴ in the sample was first determined for each economic zone in proportion to its arable land area after eliminating counties in which arable land area is less than 10 per cent of the total land area; sampled counties were then randomly selected within each economic zone. Next, sampled townships were selected following the same procedure within each county. In each township, 10 to 20 LSPUs were then randomly selected from a list of LSPUs provided by local township governments. In this paper, we consider family farms and professional farmers with at least 30 mu (2 ha),Footnote ⁵ which is 10 times the average area of local cultivated land per household.Footnote ⁶ For each LSPU, we plan to collect information on the two largest plots,Footnote ⁷ one rented directly from small farmers and another indirectly accessed via intermediaries.

We used a structured questionnaire that asked for detailed information on household and employment status, household assets, land endowment, land transfer, farm inputs, investments and production. The resulting dataset has nested information at the plot and LSPU level. For each LSPU, contract-related data were collected at the plot level allowing inferences to be drawn about the relationship between contract type for a specific plot, the investment activities on this plot, and its characteristics such as plot size and soil quality.

In total, 743 plots from 705 LSPUs were surveyed in Jiangsu (JS) and Jiangxi (JX) Provinces. After data cleaning, deleting observations with missing values and ignoring outliers, the final sample for this analysis consists of 699 plots (375 plots in JS and 324 plots in JX) from 664 LSPUs (346 LSPUs in JS and 318 LSPUs in JX). Among these, only 40 LSPUs choose to access land through two channels and thereby information on two plots were collectedFootnote ⁸; for the remaining LSPUs, in which 381 only rent in land directly from smallholders, while 243 only access land with the involvement of intermediary organizations, we only collected information on one plot for each LSPU.

Specifically, the farm size operated by the LSPUs sampled ranges from 30 mu (2 ha) to 2005 mu (133.7 ha) with a mean of 237.45 mu (15.8 ha); in terms of quantiles based on the distribution of farm size, the farm size in the 10–90 per cent quantiles ranges from 53 mu (3.5 ha) to 502 mu (33.5 ha), averaged at 194.16 mu (12.9 ha). Characteristics of the sampled plots in terms of crop composition, farmland tenure security, and investment in soil fertility are given in table 1. A notable finding is that plot sizes in JS are on average six times larger than those in JX. Rice-wheat cropping systems predominated on plots in JS, whereas single rice and early-late rice cropping systems account for 42 and 58 per cent, respectively, of the total cultivated area in JX. Average yields are higher on plots in JX compared to those in JS, which may result from differences in cropping systems between the two provinces. In total, 64.5 per cent of the plots are transferred through a written contract. Furthermore, more plots in JS are transferred on the basis of written contracts compared to those in JX. Straw is applied to 56.4 per cent of the plots; organic fertilizer is applied to 17.3 per cent of the plots. The number of plots on which green manure is applied is as low as 3.6 per cent. Organic fertilizer and straw are more frequently applied in JS and the use of green manure is equally low in both provinces.

Table 1.

Descriptive statistics at plot level

Notes: The χ² and t-test are used to test the differences between the two provinces for the categorical and continuous variables, respectively. *** Significant at 1%. 1 mu = 1/15 ha; US$1 = Yuan6.68 in September 2016.

3.2 Model specification

We assume that decisions on long-term investments in soil-improving measures depend not only on land tenure security but also on household and farm characteristics, plot characteristics and geographic location (Place and Hazell, Reference Place and Hazell1993; Knowler and Bradshaw, Reference Knowler and Bradshaw2007; Ali et al., Reference Ali, Abdulai and Goetz2012). Proxies for these characteristics are used to test the effect of contract type on investments in soil-improving activities at the plot level as follows:

(1)\begin{align} {\rm{Inv}}^{*}&=f( {W,{X^h},{X^p},{X^l}} )\\ {\rm Inv}&=\begin{cases} 1, & {\rm if} \ {\rm In}{{{\rm v}}^{*}}\ge 0 \\ 0, & {\rm otherwise} \end{cases},\notag \end{align}

where Inv^* is a latent variable implying the application of organic fertilizer, green manure and straw on that particular plot in 2015 and Inv refers to the observed binary variable. For organic fertilizer, we consider the application of farmyard waste and commercial organic fertilizer. The bulkiness of farmyard waste makes this application particularly labor-intensive. Green manure refers to the planting of a crop used primarily as a soil improver and nutrient source for subsequent crops (Cherr et al., Reference Cherr, Scholberg and McSorley2006). Fabaceae or Astragalus sinicus are often planted as green manure. These crops are easy to plant and manage and are low on labor demand. In the study area, straw application is done mechanically by crushing the stubble after harvest and the crushed straw is ploughed back into the soil. This makes straw application a capital-demanding investment.

We treat the variable of interest, contract type W, as a binary variable which equals one when the contract is concluded in written form and zero otherwise. As previously discussed, the written nature of contracts is expected to positively affect long-term investments through two combined channels: the assurance effect and the collateralization effect. We exploit the nested data available at the plot and LSPU level to define the control variables, including household and farm characteristics (X^h), plot characteristics (X^p) and location variables (X^l) (see table A1 in the appendix).

The number of houses and cars owned is used as a proxy for households' financial capitalFootnote ⁹ (Ma et al., Reference Ma, Heerink, van Ierland, van den Berg and Shi2013), which is assumed to be positively related to investment. Non-farm income provides LSPUs with additional resources and eases their liquidity constraints, allowing more production factors to be invested on plots. However, as an alternative income source, availability of more non-farm income could result in higher opportunity costs of agricultural activities (Knowler and Bradshaw, Reference Knowler and Bradshaw2007). The direction of the correlation of share of non-farm income to total income and land-related investment decisions therefore has to be determined empirically (Sumner, Reference Sumner1982). Household size may positively affect investments on plots since larger households may need more food; however, they also have more family labor at their disposal for farm activities on certain plots (Feng et al., Reference Feng, Heerink, Ruben and Qu2010; Abdulai and Goetz, Reference Abdulai and Goetz2014). We also control for different household characteristics such as age, education and Communist Party membership, dependency ratio, and female-to-male labor availability in the household. Farm characteristics added to the model include total farm size, agricultural machinery assets, and access to extension services.

To control for plot-level effects, we add plot size to the model, and account for the distance between the plot and LSPU's homestead. Because it is often hypothesized that returns of soil-improving investments are higher on fertile land (Ali et al., Reference Ali, Abdulai and Goetz2012), we also include a perceived soil quality variable. Finally, we add location variables such as the province and the distance from the village to the county government. Generally, more soil-improving practices may be adopted by LSPUs on plots located in remote rural areas since the importance of farm activities is higher for them.

3.3 Empirical strategy

The three types of soil-improving investments (organic fertilizer, green manure and straw application) compete for labor and capital in each LSPU. As a result, the investment decision in one type of soil-improving measure within one LSPU is not independent of investment in the other. This means that instead of estimating the models that test the likelihood of investing in each type separately, considering these investments as a system is more efficient as error terms of individual models may not be independent and need to be assessed as correlated (Vellema et al., Reference Vellema, Buritica Casanova, Gonzalez and D'Haese2015). A multivariate probit model developed by Cappellari and Jenkins (Reference Cappellari and Jenkins2003) is used to account for the joined decision-making process on land investments. In addition, cluster-robust standard errors are needed to account for problems of clustering that might arise on 40 LSPUs on which two plots are sampled.

All explanatory variables in the model are considered exogenous. However, contract type may be potentially endogenous as it may depend on unobservable characteristics and be changed by the LSPU with the investment decision. To capture this potential endogeneity, a two-step control function approach suggested by Wooldridge (Reference Wooldridge2014) is considered. The first step consists of estimating a probit model of the contract type following equation (2):

(2)\begin{align} {W^*} &= f( {{Z^h},{Z^p},{Z^l},IV} ) \\ W &=\begin{cases} 1, & if \ {{W}^{*}}>1 \\ 0, & {\rm otherwise} \end{cases},\notag \end{align}

where Z^h, Z^p and Z^l are respectively household and farm characteristics, plot characteristics and location variables. We add an instrumental variable (IV) indicating whether there is at least one farmland transfer intermediary organization in the county (0 = no; 1 = yes). With the presence of intermediaries, LSPUs are more likely to access plots via these organizations and thus increase their opportunities to sign written contracts given that intermediary organizations not only reduce transaction costs involved in farmland transfer, but also ensure the stability of transaction relationships between small farmers and LSPUs (Shi et al., Reference Shi, Chen, Ma and Lan2018). Moreover, intermediary organizations mainly play a role in land acquisition, i.e., publishing farmland transfer information, proposing land rental price, and providing guidance on contract signing, and therefore should not have direct effects on land investments. The significance of IV in equation (2) indicates the strength of the instrument (Liu et al., Reference Liu, Rommel, Feng and Hanisch2017).

The generalized residual R is obtained from equation (2) and introduced as a regressor in equation (1) as the second step estimation:

(3)\begin{equation} {\rm{Inv}^*} = f( {W,{X^h},{X^p},{X^l},\;R} )\end{equation}

The significance statistics on the coefficient of R are sufficient to test the null hypothesis that contract type is exogenous. To test the validity of the IV, an alternative version of equation (3) is estimated with the IV included, following Abdulai et al. (Reference Abdulai, Owusu and Goetz2011):

(4)\begin{equation} {\rm{Inv}^*} = f( {W,{X^h},{X^p},{X^l},\;R,IV} ). \end{equation}

If the coefficient of IV is not significantly different from zero, the IV can be excluded from equation (3), indicating that the instrument affects land investments only via contractual arrangements.