Can Information from the Internet Improve Grain Technical Efficiency? New Evidence from Rice Production in China

Abstract

The Internet has become an important channel through which farmers obtain technical information regarding agricultural production. While previous studies have examined the effect of information from the Internet on technical efficiency in cash-crop production, there is little knowledge about how information from the Internet affects technical efficiency in grain production. This study aims to provide new evidence for the effect of information from the Internet on technical efficiency in rice production using the random survey data of 1122 rice farmers from the Yangtze River Basin in China. A stochastic frontier production function is employed to estimate technical efficiency in rice production, and the endogenous switching regression model is utilized to address the potential self-selectivity bias. The results show that 13.6% of surveyed rice farmers obtain information regarding agricultural production from the Internet. After addressing the self-selectivity bias, information from the Internet is found to increase rice farmers’ technical efficiency by 6.657 percentage points using the endogenous switching regression model. Meanwhile, information from the Internet exerts greater positive effects on technical efficiency in rice production among farmers with larger farm size and less farming experience as well as those in the Guizhou and Hubei provinces.

1. Introduction

Since the second half of the 20th century, the Internet has become one of the most influential information and communications technologies. In particular, China has experienced an impressive increase in rural netizens over the past two decades. By the end of 2021, there were 284 million rural netizens in China [1]. Previous studies showed that Chinese farmers increasingly seek and obtain information on fertilization, crop pest control, and natural condition forecasting from the Internet [2,3].

Previous studies so far have empirically investigated the effect of the adoption of ICTs especially the Internet. For example, many studies analyzed the effect on farmers’ production behavior. Kaila and Tarp suggested that Internet access could increase total agricultural output by 6.8% in Viet Nam, and improvement of fertilizer use efficiency could manifest a positive effect [4]. In Kenya, Ogutu et al. showed that ICT-based information services led farmers to increase the usage of seed, fertilizer, and land [5]. Hou et al. provided evidence that computer usage enlarged farm size and reduced labor input in China [6]. Deng et al. showed that Internet adoption reduced the probability of crop abandonment by 43.2% [7]. Liu et al. found that farmers’ access to agricultural information from the Internet significantly promoted land transfer in China [8]. Zhao et al. found that Internet use was conducive to reducing pesticides in vegetable production in China [9]. Yuan et al. also provided evidence that Internet use could significantly reduce chemical fertilizer use through increasing the farmers’ human capital [10]. In addition, Gao et al. indicated that farmers’ usage of WeChat official accounts and Apps regarding agricultural extension could promote new technology adoption [11]. Ma and Wang, and Yan et al. also found that Internet adoption could promote farmers to adopt technologies for sustainable agriculture [12,13]. Another group of studies attached attention to the effect of Internet adoption on market participation. For example, Zanello found that market information by telephone promoted farmers in Ghana to participate in the market [14]. Shimamoto et al. reported that mobile phone use improved Cambodian farmers’ access to market information, which further raised the rice selling price [15]. In Ethiopia, Tadesse and Bahiigwa provided evidence that farmers’ usage of the mobile phone could increase the prices they received from the market [16]. In China, Ma et al. concluded that farmers’ willingness for E-commerce adoption increased by 20% through using information from the Internet [17].

Given the central role of technical efficiency in increasing agricultural productivity, a better understanding of how information from the Internet influences agricultural technical efficiency can provide effective policy implications. A few studies focused on this issue, and generally concluded that Internet adoption can exert a positive effect on technical efficiency [3,18,19]. It should be noted that previous studies focused on cash-crop production, such as vegetables, apple and banana and that there exists a large difference in the complexity of technologies used in the production of grain and cash crops. Compared with the technologies for cash-crop production, those for grain production are much less complex. Moreover, the farmers’ demands for technologies in grain and cash-crop production are also greatly different. In grain production, farmers always adopt technologies to increase the productivity. However, both productivity-oriented and quality-oriented technologies are needed in cash-crop production. In addition, there may also exist heterogeneity in the farmers’ ability to utilize the information from the Internet between grain and cash-crop production, given the coexistence of both correct and incorrect information on the Internet. Thus, it is important to investigate the effect of information from the Internet on grain production, given the increasingly important role of grain production in promoting food security in China and even the whole world. However, little is known about how information from the Internet affects technical efficiency in grain production. Moreover, the heterogeneity in the effects of information from the Internet on grain technical efficiency across different groups of farmers also remains unclear.

To fill the research gaps, this study attempts to provide new evidence as to whether information from the Internet improves grain technical efficiency in rice production in China, and investigate the heterogeneity in the effects of the information from the Internet on technical efficiency. Overall, this study contributes to the literature in three main aspects. First, this study uncovers how information from the Internet affects technical efficiency in grain production rather than in the production of other cash crops, which would produce more policy implications for enhancing food security in the context of Internet development. Second, this study also explores the heterogeneity in the effects of information from the Internet on technical efficiency by farm size and farming experience as well as across regions, which provides more knowledge about the relationship between the information from the Internet and the technical efficiency in grain production for different farmers. Third, the self-selectivity issue regarding the farmers’ decision to obtain information from the Internet is addressed, which provides unbiased estimation results.

The other sections of this study proceed as follows. Section 2 describes the background of rice production in China, and provides a theoretical analysis on the relationship between information from the Internet and technical efficiency in rice production. Section 3 introduces the empirical methods and materials, including the stochastic frontier production function, endogenous switching regression model, and data collection and descriptive analysis, followed by the results and discussion of the empirical analysis in Section 4. We conclude the study with policy implications in Section 5.

2. Background and Theoretical Analysis

2.1. Rice Production in China

This study focuses on rice production, since rice is one of the most important staple grain crops in China. It should be noted that China is among the largest rice producers in the world. Although China’s rice yield is higher than the global average level, there is still room for improvement in technical efficiency [20,21,22]. In 2020, the total area of rice sown in China was 30.1 million hectares (ha), ranking second to India, while the total rice output in China was 211.9 million tonnes, ranking it first in the world [23]. From 2000 to 2020, rice yield in China has increased by 12.69% (Figure 1), which is the major reason for the growth of total rice output given the decline in rice sown area.

However, rice technical efficiency in China has declined during the past decades. As shown in Figure 1, while labor input in rice production in China has decreased by 66.64% from 2000 to 2020, capital input in real terms has increased by 38.49%, higher than the growth rate of the rice yield. More importantly, rice production cost per kilogram (kg) in real terms has increased by 34.69%, 22 percentage points higher than the growth rate of rice yield (Figure 1). In comparison, the annual growth rates of rice yield and production cost per kg of rice during the period 2000–2020 were 0.60% and 1.50%, respectively (Figure 2). It illustrates that the increase in rice yield in China may be largely attributed to the increase in factor input rather than improvement of technical efficiency.

2.2. Theoretical Analysis

In the literature, technical efficiency is defined as the ratio of actual output to maximum output given the available technology [3,19]. In this study, we consider that information from the Internet affects technical efficiency in rice production in three ways. First, information from the Internet has a diffusion effect, which indicates that information from the Internet can extend the farmers’ access to improved technologies and better production factors. Given the dominant role of smallholders in China, information asymmetry increases the cost of farmers’ access in traditional ways to improved technologies and better production factors [26]. In this context, farmers can obtain information from the Internet with a relatively lower information cost [27,28,29], which is conducive to improving technical efficiency.

Second, information from the Internet can also increase technical efficiency through its learning effect. In this study, the learning effect refers to the premise that information from the Internet can facilitate farmers’ knowledge acquisition of improved technologies through online learning [30]. Traditionally, farmers in China always learn agricultural knowledge from public agricultural extension agents [31]. However, there is great capacity for expanding the coverage and improving the effectiveness of frontline public agricultural extension services in rural China [32]. For farmers in remote villages, in particular, their access to the resources of agricultural knowledge learning is much poorer. However, those farmers can conveniently obtain online learning materials.

Third, the timing effect is another way in which information from the Internet affects technical efficiency in rice production. In general, the timing effect refers to the fact that information from the Internet can reduce the time that farmers would spend on decision making. Note that timely information plays a decisive role in farmers’ decision making. However, an information delay would happen when farmers seek information from traditional sources. For example, public agricultural extension and training activities are often organized on a regular or irregular basis rather than as a timely response to the farmers’ demand for technical information [33,34]. In comparison, farmers can obtain information in time through the Internet [27,28,30]. According to the analysis above, the first hypothesis in this study is as follows:

Hypothesis 1.

Information from the Internet may increase technical efficiency in rice production.

It should be noted that both correct and incorrect information coexist on various Internet platforms. While an increasing number of individuals and organizations upload technical information regarding agricultural production to online platforms [3], most information on the Internet is afforded to specific planting seasons or given regions. Meanwhile, the correctness of information on the Internet is subject to the uploaders’ knowledge and capability. In this context, farmers have to verify the correctness of information on the Internet based on their own knowledge endowment [35]. Moreover, farmers with different farm sizes and different farming experience as well as being in different regions are much more likely to have different abilities for information identification and application [36]. Thus, the second hypothesis is as follows:

Hypothesis 2.

Information from the Internet may exert heterogeneous effects on technical efficiency by farm size and farming experience as well as across regions.

3. Methods and Materials

In this study, our empirical analysis was conducted in three steps. First, a stochastic frontier production function was used to calculate technical efficiency in rice production. Second, the endogenous switching regression model was used to examine the determinants of farmers’ decision to obtain information from the Internet, and the effect and heterogeneity of information from the Internet and other factors on technical efficiency in rice production. Meanwhile, the validation of instrumental variable was also verified in this step. Third, both robustness checks and heterogeneity analysis were conducted.

3.1. Stochastic Frontier Production Function

Technical efficiency was estimated using a stochastic frontier production function rather than data envelopment analysis, because the latter cannot account for stochastic factors, such as extreme weather events. The stochastic frontier production function was developed as:

$$Q_i=Q({\mathit{X}}_i;\mathit{\alpha })\cdot \exp (v_i-u_i),$$

(1)

where i indicates the i-th farmer; Q_i indicates rice yield; X_i indicates a group of inputs in rice production, such as fertilizer, pesticide, machinery, and labor; α indicates the coefficients to be estimated; v_i indicates a random error term; and u_i indicates a non-negative inefficiency term.

Farmers’ technical efficiency in rice production can be calculated as:

$$Ef{f}_i=\frac{Q_i}{Q({\mathit{X}}_i;\mathit{\alpha })\cdot \exp (v_i)}\times 100\%=\exp (-u_i)\times 100\%,$$

(2)

where Eff_i indicates the i-th farmer’s technical efficiency. Note that farmers’ technical efficiency ranges from zero to 100%.

3.2. Endogenous Switching Regression Model

Note that farmers’ decision to obtain information from the Internet may be self-selected, resulting in a potential self-selectivity bias. To address the self-selectivity bias, the endogenous switching regression model consisting of a treatment equation and two outcome equations was widely used in the literature [3,37,38,39].

In this study, a random utility framework was utilized to analyze the farmers’ decision to obtain information from the Internet. We assume that D_i^* indicates the difference in the utility between obtaining information from the Internet and not obtaining information from the Internet. A farmer would obtain information from the Internet if D_i^* > 0, and not obtain information from the Internet when D_i^* ≤ 0. Hence, the farmers’ decision to obtain information from the Internet is modeled as:

$$D_i^{\ast }={\mathit{T}}_i\mathit{\beta }+{\omega }_i,D_i=\left\{\begin{array}{c}1\mathrm{if}{D}_i^{\ast }>0\\ 0\mathrm{if}{D}_i^{\ast }\le 0\end{array}\right.,$$

(3)

where D_i is a dummy variable that is equal to one if a farmer obtains information from the Internet and zero otherwise; T_i are factors influencing the farmers’ decision to obtain information from the Internet; β are coefficients to be estimated; and ω_i is a random error term with a zero mean.

In addition to information from the Internet, there are other factors affecting technical efficiency. Thus, two outcome equations were developed as:

$$Ef{f}_{1i}={\mathit{Z}}_i{\mathit{\delta }}_1+{\upsilon }_{1i}\mathrm{if}{D}_i=1,$$

(4)

$$Ef{f}_{0i}={\mathit{Z}}_i{\mathit{\delta }}_0+{\upsilon }_{0i}\mathrm{if}{D}_i=0,$$

(5)

where 1 and 0 indicate farmers obtaining and not obtaining information from the Internet, respectively; Eff_1i and Eff_0i indicate technical efficiency in rice production among farmers obtaining and not obtaining information from the Internet, respectively; Z_i indicates exogenous factors affecting technical efficiency; δ₁ and δ₀ are coefficients to be estimated; and υ_1i and υ_0i are random error terms with zero means.

Given the presence of the self-selectivity bias, the expected average technical efficiency in the actual and counterfactual situations among farmers obtaining information from the Internet are:

$$E(Ef{f}_{1i}|D_i=1)={\mathit{Z}}_i{\mathit{\delta }}_1+{\sigma }_{\omega \upsilon 1}{\lambda }_{1i},$$

(6)

$$E(Ef{f}_{0i}|D_i=1)={\mathit{Z}}_i{\mathit{\delta }}_0+{\sigma }_{\omega \upsilon 0}{\lambda }_{1i},$$

(7)

where σ_ωυ1 is the covariance of ω_i and υ_1i; σ_ωυ0 is the covariance of ω_i and υ_0i; and λ_1i = φ(T_iβ)/Φ(T_iβ) indicates the inverse Mills ratio. Note that φ(•) indicates the standard normal probability density, and Φ(•) indicates the cumulative distribution function of the standard normal distribution.

The difference in the expected average technical efficiency in rice production in the actual and counterfactual situations among farmers obtaining information from the Internet, referred to as the average treatment effect on the treated (ATT), is calculated as:

$$\mathrm{ATT}=E(Ef{f}_{1i}|D_i=1)-E(Ef{f}_{0i}|D_i=1)={\mathit{Z}}_i({\mathit{\delta }}_1-{\mathit{\delta }}_0)+{\lambda }_{1i}({\sigma }_{\omega \upsilon 1}-{\sigma }_{\omega \upsilon 0}),$$

(8)

Let σ_ω, σ_υ1 and σ_υ0 denote the standard deviations of ω_i, υ_1i and υ_0i, respectively. Thus, ρ_ωυ1 = σ_ωυ1/(σ_ωσ_υ1) indicates the correlation coefficient between ω_i and υ_1i, and ρ_ωυ0 = σ_ωυ0/(σ_ωσ_υ0) indicates the correlation coefficient between ω_i and υ_0i. The significant ρ_ωυ1 and ρ_ωυ0 indicate the presence of self-selectivity bias.

To estimate the endogenous switching regression model, at least one instrumental variable is required. The instrumental variable should be included in T_i but not included in Z_i. Meanwhile, it should be correlated with the farmers’ decision to obtain information from the Internet, but not correlated with technical efficiency except through obtaining information from the Internet. We employ the full information maximum likelihood method to estimate the endogenous switching regression model [3,39,40].

3.3. Data and Descriptive Analysis

In this study, data were collected from a random survey of rice farmers located in the Guizhou, Hubei, Jiangsu, and Zhejiang provinces from the Yangtze River Basin, which produces more than half of the total rice output in China [41]. The survey was conducted in October and November, 2016, after the harvest of rice. Rice farmers were randomly selected in each province using a multi-stage sampling procedure. First, four representative counties were randomly selected in each province after all counties were sorted by per capital rural income. Second, the random sampling approach was then used to select two townships in each county and two villages in each township. Third, about 20 rice farmers were randomly selected from a household list in each village. After excluding farmers failing to provide complete information, a total of 1122 rice farmers remained in the final sample.

We collected a wide range of information using a structured questionnaire. The first set of data collected in the survey included farmers’ individual and household characteristics, such as gender, age, education, and village-leader status. The second set of data contained farm characteristics including rice farm size, adoption of hybrid variety, and utilization of seeding method. The third set of data included detailed information of inputs and output in rice production, such as fertilizer, pesticide, machinery, labor, as well as rice yield. In addition, we also collected data about whether farmers obtained information regarding rice production from the Internet.

The dependent variable, technical efficiency in rice production, was estimated using the aforementioned stochastic frontier production function. Note that the inputs used in the stochastic frontier production function included chemical and organic fertilizers, pesticide, agricultural machinery expenditure, and manual labor per hectare, and the output was rice yield. The independent variable of interest was a dummy variable equal to one if a farmer obtained information regarding rice production from the Internet platforms (e.g., WeChat official accounts and Apps) using a mobile phone or computer, and zero otherwise.

In terms of the instrumental variable, we define a dummy variable indicating whether a farmer’s neighbors obtained information regarding rice production from the Internet. As aforementioned, as Smith et al. pointed out, there may exist a peer effect for farmers’ technology adoption and usage of the Internet [42]. Thus, it is reasonable to assume that a farmer is more likely to obtain information regarding rice production from the Internet if his or her neighbors do so. Thus, this instrumental variable is highly correlated with this farmer’s decision to obtain information regarding rice production from the Internet. In addition, the neighbors’ obtaining information from the Internet would not directly affect this farmer’s technical efficiency in rice production, but only correlate with farmers’ technical efficiency through obtaining information from the Internet. Thus, the dummy variable indicating whether a farmer’s neighbors in the same village obtained information from the Internet would be a valid instrumental variable [7,19,27].

Table 1. Definition and descriptive statistics of variables.

Variable	Definition or Description	Mean	SD
Yield	Quantity of rice yield (tonnes/ha)	8.008	1.703
Chemical fertilizer	Quantity of chemical fertilizer (tonnes/ha)	0.522	0.351
Organic fertilizer	Quantity of organic fertilizer (tonnes/ha)	0.637	3.886
Pesticide	Quantity of pesticide (kg/ha)	16.435	18.410
Machinery	Agricultural machinery expenditure (1000 yuan/ha)	0.201	0.472
Labor	Total hours of manual labor (1000 h/ha)	0.442	0.410
Information from the Internet	1 if farmer obtains information regarding rice production from the Internet, 0 otherwise	0.136	0.343
Male	1 if farmer is male, 0 otherwise	0.904	0.295
Age	Age of farmer (years)	57.000	9.680
Education	Formal schooling of farmer (years)	6.622	3.281
Village leader	1 if farmer is a village leader, 0 otherwise	0.099	0.299
Farm size	Total rice sown area (ha)	1.944	10.953
Hybrid	1 if farmer adopts hybrid variety, 0 otherwise	0.485	0.500
Direct seeding	1 if farmer adopts direct seeding, 0 otherwise	0.468	0.499
Instrumental variable	1 if farmer’s neighbors in the same village obtain information regarding rice production from the Internet, 0 otherwise	0.818	0.386

Notes: Authors’ survey.

summarizes the definitions and descriptive statistics of the main variables used in this study.

Table 2. Differences between farmers obtaining and not obtaining information from the Internet.

Variable	Farmers Obtaining Information from the Internet (n = 153)	Farmers not Obtaining Information from the Internet (n = 969)	Mean Differences
Yield	8.430 (1.651)	7.942 (1.703)	0.488 ***
Chemical fertilizer	0.496 (0.299)	0.526 (0.358)	−0.029
Organic fertilizer	0.715 (5.550)	0.624 (3.556)	0.091
Pesticide	11.807 (12.085)	17.165 (19.121)	−5.356 ***
Machinery	0.274 (0.572)	0.189 (0.454)	0.085 **
Labor	0.286 (0.307)	0.466 (0.419)	−0.181 ***
Male	0.902 (0.298)	0.904 (0.295)	0.002
Age	49.556 (9.232)	58.175 (9.218)	−8.620 ***
Education	9.056 (2.619)	6.238 (3.210)	2.818 ***
Village leader	0.242 (0.430)	0.076 (0.266)	0.165 ***
Farm size	7.846 (27.433)	1.012 (3.794)	6.834 ***
Hybrid	0.431 (0.497)	0.493 (0.500)	−0.062
Direct seeding	0.412 (0.494)	0.477 (0.500)	−0.065

Notes: Standard deviations are in parentheses. ** p < 0.05, *** p < 0.01.

presents the mean differences of the main variables between farmers obtaining and those not obtaining information from the Internet. It shows that there indeed exist differences in many variables between these two groups of rice farmers. Compared with those not obtaining information from the Internet, farmers obtaining information from the Internet attain a significantly higher rice yield while using fewer pesticides, more agricultural machinery expenditure, and lower levels of manual labor in rice production. However, there is no significant differences in the usage of chemical and organic fertilizers between the two groups of farmers. In terms of farmers’ individual and planting characteristics, farmers obtaining information from the Internet were relatively younger and better educated, and had a larger rice farm size. These significant differences may indicate the presence of self-selectivity bias.

4. Results and Discussion

4.1. Estimates of Technical Efficiency

In terms of the stochastic frontier production function, we estimated both Cobb–Douglas and translog specifications (

Table 3. Estimation results of the stochastic frontier production function.

Variables	Coefficient	Robust Standard Error
(1) Frontier production equation
Ln(chemical fertilizer)	0.018 ***	0.006
Ln(organic fertilizer)	0.006 **	0.002
Ln(pesticide)	−0.000	0.006
Ln(machinery)	0.003 **	0.002
Ln(labor)	−0.015 *	0.009
Hybrid	0.057 ***	0.022
Provincial effects	Yes
Constant	9.106 ***	0.075
(2) Inefficiency equation
Information from the Internet	−0.180	0.168
Male	−0.205	0.186
Age	−0.008	0.007
Education	−0.041 **	0.019
Village leader	−0.114	0.183
Ln(Farm size)	−0.046	0.049
Hybrid	0.003	0.199
Direct seeding	0.480 ***	0.129
Provincial effects	Yes
Constant	−1.070 **	0.465
Log-likelihood	215.244 (Prob > χ2 = 0.582)
n	1122

Notes: * p < 0.10, ** p < 0.05, *** p < 0.01.

). To determine the better specification, we conducted a likelihood ratio test. The χ² statistic was equal to 215.244, and but not significant (Prob > χ² = 0.582). It suggests that the Cobb–Douglas specification of stochastic frontier production function is nested in its translog specification. In the frontier production equation, each 1% increase in chemical fertilizer, organic fertilizer, and machinery expenditure would increase rice yield by 0.018, 0.006, and 0.003%, respectively. However, there was a significantly negative effect of manual labor input on rice yield. This finding was not consistent with traditional economic theory but consistent with several previous studies [43,44]. It might be economically explained in two ways. First, there might exist a collinearity of labor input and other inputs, which makes it difficult to distinguish the yield impact of labor from that of other inputs [45]. Second, while a large number of farmers migrate to urban areas, there is also a large probability of an agricultural surplus labor force, given the rapid technological progress in agriculture, and thus, the marginal output of labor input might tend to be zero or even negative [43,46]. The average rice yield of hybrid variety is 5.7% higher than that of non-hybrid varieties. In the inefficiency equation, information from the Internet seems not to affect technical inefficiency in rice production. However, this result fails to address the potential self-selectivity bias.

The average technical efficiency in rice production using the estimation results of the stochastic frontier production function are presented in

Table 4. Comparison of technical efficiency between farmers obtaining and not obtaining information from the Internet.

Group	Mean	SD
Total	81.191	11.255
Farmers obtaining information from the Internet	83.656	10.027
Farmers not obtaining information from the Internet	80.801	11.393
Difference	2.855 ***

Notes: *** p < 0.01.

. On average, the technical efficiency among all 1122 farmers was 81.191, which may be a considerable level as previously calculated [20,21,22]. By comparison, technical efficiency among farmers obtaining information from the Internet was significantly different from that among those not obtaining information from the Internet. Specifically, the average technical efficiency among farmers obtaining information from the Internet was 83.656, 3.53% [(83.656 − 80.801)/80.801 × 100%] higher than that among those not obtaining information from the Internet. These findings indicate that there is considerable room for improvement in technical efficiency in rice production.

4.2. Determinants of Obtaining Information from the Internet

Table 5. Estimation results of the endogenous switching regression model.

Variables	Information from the Internet	Technical Efficiency
		Farmers Obtaining Information from the Internet	Farmers Not Obtaining Information from the Internet
Male	−0.305	6.952 ***	1.085
	(0.202)	(2.332)	(1.365)
Age	−0.035 ***	0.032	0.078 *
	(0.007)	(0.086)	(0.041)
Education	0.121 ***	0.660 **	0.264 **
	(0.021)	(0.342)	(0.113)
Village leader	0.545 ***	0.559	−0.006
	(0.143)	(1.792)	(1.223)
Ln(Farm size)	0.196 ***	0.661	−0.385
	(0.039)	(0.458)	(0.301)
Hybrid	−0.262	0.688	−0.588
	(0.168)	(3.443)	(0.970)
Direct seeding	−0.205	−3.064	−4.289 ***
	(0.133)	(1.995)	(0.812)
Provincial effects	Yes	Yes	Yes
Instrumental variable	0.621 ***
	(0.200)
Constant	−0.219	59.231 ***	67.067 ***
	(0.454)	(7.555)	(3.039)
ρωυ1		0.212 **
		(0.097)
ρωυ0			−0.271 ***
			(0.069)
Indep. eqs. (χ2)	19.540 ***
n	1122	153	969

Notes: Robust standard errors are in parentheses. * p < 0.10, ** p < 0.05, *** p < 0.01.

presents the estimation results of the endogenous switching regression model. The χ² statistic for the Wald test of independent equations was significant, which indicated the presence of dependence between the treatment and the two outcome equations. In addition, both the correlation coefficient ρ_ωυ1 and ρ_ωυ0 were significant, which confirmed that there existed a self-selectivity bias arising from both observed and unobserved factors. It implies that using the endogenous switching regression model was appropriate and necessary.

We verified the validity of the instrumental variable using the falsification test [37,47,48]. The results of the falsification test are summarized in

Table 6. Estimation results of falsification test on the instrumental variable.

Variable	Information from the Internet	Technical Efficiency
Instrumental variable	0.562 *** (0.203)	1.159 (0.883)
Male	−0.310 (0.200)	0.883 (1.359)
Age	−0.034 *** (0.007)	0.053 (0.041)
Education	0.122 *** (0.021)	0.328 *** (0.114)
Village leader	0.534 *** (0.145)	0.522 (1.201)
Ln(Farm size)	0.196 *** (0.039)	−0.237 (0.293)
Hybrid	−0.306 * (0.165)	−0.742 (0.966)
Direct seeding	−0.204 (0.134)	−4.483 *** (0.815)
Provincial effects	Yes	Yes
Constant	−0.111 (0.447)	68.345 *** (3.028)
n	1122	969

Notes: The first column is estimated using Probit model, while the second column is estimated using Tobit model. Robust standard errors are in parentheses. * p < 0.10, *** p < 0.01.

. They showed that the instrumental variable had a significantly positive relationship with the farmers’ decision to obtain information from the Internet, while it had no significant relationship with technical efficiency among farmers not obtaining information from the Internet. Thus, it is a valid instrumental variable.

The statistical significance and magnitude of the estimated coefficients of the variables associated with the farmers’ decision to obtain information from the Internet in the endogenous switching regression model are presented in Table 5. The coefficient of farmers’ age was significantly negative, indicating that older farmers have a lower probability of obtaining information from the Internet, which is consistent with previous studies [49,50,51]. This finding is reasonable because older farmers may have poorer ability in using the Internet.

A higher level of education encourages farmers to obtain information from the Internet (Table 5), since the coefficient of farmers’ years of formal schooling is positive and significant at the 1% level. This is also reasonable because better educated farmers would be more skilled at using the Internet and learning new things from online platforms [42,52,53].

Village leaders had a higher probability of obtaining information from the Internet since the coefficient of village-leader variable was significantly positive (Table 5). As Zhu et al. pointed out, an important task that village leaders in rural China undertake is to spread the technical information they receive from public agricultural extension agents to farmers in their villages, which would largely encourage them to use the Internet [3].

In addition, farmers with a larger farm size have a higher probability of obtaining information from the Internet (Table 5). When farmers can increase profit through obtaining information from the Internet, they would decide to do so [53]. In the context of imperfect public agricultural extension services, farmers with larger farm size are more likely to obtain information from the Internet, because they may face a higher level of profit-reduction risks [3,51].

4.3. Effects of Information from the Internet and Other Factors on Technical Efficiency

The endogenous switching regression model provides evidence that information from the Internet can benefit rice farmers through increasing technical efficiency in rice production. Using the estimation results of the endogenous switching regression model, we were able to calculate the expected values of technical efficiency among the farmers obtaining information from the Internet in the actual and counterfactual situations (

Table 7. ATT of information from the Internet on technical efficiency among farmers.

Group	Technical Efficiency (%)		ATT
	Obtain Information from the Internet	Do Not Obtain Information from the Internet
Farmers’ obtaining information from the Internet	83.659	77.003	6.657 ***

Notes: ATT denotes the average treatment effect on farmers obtaining information from the Internet. *** p < 0.01.

), which further enabled us to calculate the treatment effects of information from the Internet on technical efficiency in rice production.

Table 7 shows that information from the Internet can increase the technical efficiency in rice production among farmers obtaining information from the Internet. Specifically, technical efficiency among farmers obtaining information from the Internet was 6.657 percentage points higher than the level among those not obtaining information from the Internet. It indicates that information from the Internet exerted a positive effect on technical efficiency in rice production. While differences exist in the complexity of and demand for technologies between rice production and other cash-crop production, the estimation results in this study provide evidence that information from the Internet not only improves technical efficiency in the production of vegetable, banana and apple as previous documented [3,18,19], but also benefits rice farmers through increasing technical efficiency. This result is reasonable in the context of China. It should be noted that there is a large gap between the provision of agricultural extension services and the farmers’ demand for information about agricultural technologies in China [22,31,32,33,34]. In recent years, an increasing number of farmers have obtained information regarding rice production from the Internet [22,31,32,33,34]. In the absence of timely and accurate agricultural extension services, information from the Internet can empower farmers in their agricultural production. Our findings in this study provide new evidence for a positive role of information from the Internet in agricultural production, in addition to previous studies [3,18,19].

Technical efficiency in rice production is also affected by other factors (Table 5). The estimation results of the endogenous switching regression model provide richer information about the effects of other factors on technical efficiency in rice production. For example, the technical efficiency of male farmers obtaining information from the Internet was 6.952 percentage points higher than that of female farmers obtaining information from the Internet. Each extra year on the farmers’ age increased the technical efficiency by 0.078 percentage points among farmers not obtaining information from the Internet. While better education failed to increase technical efficiency among farmers obtaining information from the Internet, each extra year in formal schooling increased the technical efficiency among farmers not obtaining information from the Internet by 0.264 percentage points. Direct seeding reduced technical efficiency among farmers not obtaining information from the Internet.

4.4. Robustness Check

We attempted to check the robustness of the results analyzed above using the treatment-effect model. It should be noted that the treatment-effect model consists of a treatment equation and an outcome equation, and it has also been widely used to address the self-selectivity bias in the literature [3,39]. The major difference between the treatment-effect model and the endogenous switching regression model is that there is only one outcome equation in the treatment-effect model but two outcome equations in the endogenous switching regression model. According to the estimation results of the treatment-effect model presented in

Table 8. Estimation results of the treatment-effect model.

Variable	Information from the Internet	Technical Efficiency
Information from the Internet		6.250 ***
		(2.117)
Male	−0.283	2.046
	(0.204)	(1.272)
Age	−0.034 ***	0.085 **
	(0.007)	(0.039)
Education	0.123 ***	0.264 **
	(0.021)	(0.110)
Village leader	0.537 ***	−0.482
	(0.143)	(1.060)
Ln(Farm size)	0.200 ***	−0.209
	(0.039)	(0.263)
Hybrid	−0.254	−0.285 *
	(0.170)	(0.936)
Direct seeding	−0.190	−4.088 ***
	(0.133)	(0.756)
Provincial effects	Yes	Yes
Instrumental variable	0.539 ***
	(0.202)
Constant	−0.206	65.825 ***
	(0.457)	(2.955)
ρωυ	−0.249 ***
	(0.092)
Indep. eqs. (χ2)	6.702 ***
n	1122	1122

Notes: Robust standard errors are in parentheses. * p < 0.10, ** p < 0.05, *** p < 0.01.

, we found that the χ² statistic for Wald test of independent equations was significant, which once again indicates the presence of dependence between the treatment and outcome equations. The correlation coefficient (ρ_ωυ) was significantly negative, further indicating that farmers with below-average technical efficiency tended to obtain information regarding rice production from the Internet, which confirmed the presence of self-selectivity bias arising from observed and unobserved factors. The positive coefficient of information from the Internet demonstrated that information from the Internet increased the technical efficiency in rice production by 6.250 percentage points among the surveyed rice farmers. This was in line with our expectation and findings in previous studies [3,51].

4.5. Heterogeneity Analysis

We further analyzed the heterogeneity in the effects of information from the Internet on technical efficiency in rice production among the surveyed farmers by farm size and farming experience as well as across regions using the endogenous switching regression model (

Table 9. ATT of information from the Internet on technical efficiency among farmers by groups.

Groups of Farmers Obtaining Information from the Internet	Technical Efficiency (%)		ATT
	Obtain Information from the Internet	Do Not Obtain Information from the Internet
(1) By farm size
Farm size > 1 ha	85.135	74.789	10.346 ***
Farm size ≤ 1 ha	82.684	78.300	4.384 ***
(2) By farming experience
Farming experience > 35 years	84.056	76.477	7.579 ***
Farming experience ≤ 35 years	82.264	72.820	9.445 ***
(3) Across regions
Guizhou & Hubei	81.401	69.138	12.263 ***
Jiangsu & Zhejiang	85.562	82.155	3.407 ***

Notes: ATT denotes the average treatment effect on farmers obtaining information from the Internet. *** p < 0.01.

). Information from the Internet can increase technical efficiency in rice production among farmers with a farm size larger than 1 ha by 10.346 percentage points, while increasing the technical efficiency among those with a farm size not larger than 1 ha by 4.384 percentage points. In other words, relative to those with a larger farm size, farmers with a smaller farm size benefit less from information from the Internet. Moreover, farmers with more than 35 years of farming experience increased their technical efficiency in rice production by 7.579 percentage points, while those with not more than 35 years of farming experience increased their technical efficiency by 9.455 percentage points. This implies that farmers with less farming experience might benefit relatively more when they obtain information regarding rice production from the Internet. The magnitude of the positive effect of information from the Internet on technical efficiency among farmers in Guizhou and Hubei was much larger than that in Jiangsu and Zhejiang. Compared with Guizhou and Hubei which are two inland provinces, the rural information infrastructure and agricultural extension services are better in Jiangsu and Zhejiang which are two coastal provinces [54]. As there is greater room for improvement in technical efficiency, farmers in Guizhou and Hubei would to some extent benefit more from obtaining information from the Internet.

5. Conclusions and Policy Implications

This study examines the effect and heterogeneity of information from the Internet on technical efficiency in rice production using the random survey data of 1122 rice farmers in China. We employed a stochastic frontier production function to estimate technical efficiency in rice production. The endogenous switching regression model was utilized to examine the effect and the heterogeneity of information from the Internet through addressing the self-selectivity bias arising from both observed and unobserved factors. The treatment-effect model was used to conduct a robustness check.

The results show that the technical efficiency among the surveyed rice farmers is overall around 80%, and 13.6% of surveyed farmers obtain information from the Internet. This study shows that while information from the Internet overall increases technical efficiency in rice production, it also exerts heterogeneous effects on technical efficiency in rice production among farmers by farm size and farming experience as well as across regions. After addressing the self-selectivity bias, information from the Internet can increase rice farmers’ technical efficiency by 6.657 percentage points. It should be noted that the findings in this study were also robust when an alternative model specification was used. The heterogeneity analysis shows that information from the Internet induces a greater increase in technical efficiency among farmers with a larger farm size and less farming experience. In addition, rice farmers in Guizhou and Hubei provinces benefit more through obtaining information from the Internet than those in Jiangsu and Zhejiang provinces.

The findings in this study have several important policy implications. First, an effective Internet plus agricultural extension system should be developed, given the positive role of information from the Internet on grain production. In China, there has been impressive progress in the popularity of the Internet [1]. However, the quality of information on the Internet is uncertain. In this context, farmers with poor ability for information filtering are more likely to obtain incorrect information from the Internet. An effective Internet plus agricultural extension system can provide farmers with more correct information from the agricultural extension system. More importantly, farmers would have more access to correct information from the agricultural extension system with a lower information cost. Second, more targeted measures regarding the development of Internet should be taken to meet the specific needs of different types of farmers since there exist heterogeneous effects of information from the Internet. For example, efforts should be made to encourage smallholder farmers to obtain more information regarding agricultural production from the Internet, given that farmers with smaller farm size are less likely to obtain information from the Internet. Thus, more technical training on how to use the Internet should be conducted for smallholder farmers to reduce the technical barriers and information cost. Meanwhile, similar technical training should also be conducted for farmers with less farming experience to improve the positive effect of information from the Internet on technical efficiency. It should be noted that Internet development in inland provinces lags behind that in coastal provinces, but farmers in inland provinces (i.e., Guizhou and Hubei) were found to benefit more from obtaining information from the Internet. Hence, both Internet infrastructure and information platforms should be enhanced.

It should be acknowledged that there exist some drawbacks in this study. Our analysis fails to describe the details of information that farmers obtain from the Internet, which makes it difficult to examine the accurate effect of different types of information from the Internet on technical efficiency. Given the coexistence of correct and incorrect information on the Internet, more effort should be made to fill this gap. In addition, our analysis was based on a cross-sectional dataset, which prevented us from exploring the dynamic effect of information from the Internet on technical efficiency.