Jukuri, open repository of the Natural Resources Institute Finland (Luke) All material supplied via Jukuri is protected by copyright and other intellectual property rights. Duplication or sale, in electronic or print form, of any part of the repository collections is prohibited. Making electronic or print copies of the material is permitted only for your own personal use or for educational purposes. For other purposes, this article may be used in accordance with the publisher’s terms. There may be differences between this version and the publisher’s version. You are advised to cite the publisher’s version. This is an electronic reprint of the original article. This reprint may differ from the original in pagination and typographic detail. Author(s): Shijia Kang, Amer Ait Sidhoum, Fabian Frick, Johannes Sauer & Shaofeng Zheng Title: The Impact of Information and Communication Technology on the Technical Efficiency of Smallholder Vegetable Farms in Shandong of China Year: 2023 Version: Accepted manuscript Copyright: The Author(s) 2023 Rights: CC BY 4.0 Rights url: http://creativecommons.org/licenses/by/4.0/ Please cite the original version: Shijia Kang and others, The Impact of Information and Communication Technology on the Technical Efficiency of Smallholder Vegetable Farms in Shandong of China, Q Open, 2023;, qoad017, https://doi.org/10.1093/qopen/qoad017 © The Author(s) 2023. Published by Oxford University Press on behalf of The European Agricultural and Applied Economics Publications Foundation. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/ licenses/by/4.0/), which permits unrestricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited. O R IG IN A L U N E D IT E D M A N U S C R IP T Title page Title The Impact of Information and Communication Technology on the Technical Efficiency of Smallholder Vegetable Farms in Shandong of China Author Names Shijia Kang, Amer Ait Sidhoum, Fabian Frick, Johannes Sauer and Shaofeng Zheng Author Affiliations 1. Shijia Kang (corresponding author, shijia.kang@tum.de), Technical University of Munich, Agricultural Production and Resource Economics, Freising, Germany 2. Amer Ait Sidhoum, Natural Resources Institute Finland (Luke), Business Economics, Helsinki, Finland 3. Fabian Frick, Technical University of Munich, Agricultural Production and Resource Economics, Freising, Germany 4. Johannes Sauer, Technical University of Munich, Agricultural Production and Resource Economics, Freising, Germany 5. Shaofeng Zheng, Northwest A&F University, College of Economics and Management, Yangling, China Acknowledgements The authors wish to thank Stefan Wimmer (ETH Zürich) and Maria Vrachioli (Technical University of Munich) for their valuable insights and thank William H. Greene (New York University) and Hung-pin Lai (National Chung Cheng University) for their advice on the use of the LIMDEP (software) and for providing code and data. We would also like to thank the participants in the AAEA annual meeting and several other seminars in Freising for all their insightful comments and suggestions on the earlier version of the paper. We also thank three anonymous reviewers and the Editor in Chief for their helpful and constructive comments on an earlier draft. Any mistakes are, however, ours. Competing Interests All the authors declare none. Funding Statement This research was funded by the National Natural Science Foundation of China (Grant No. 71773093) and China Scholarship Council (File No. 201906300061). The name of the project is Influence Mechanism of information and Communication Technology Application on Farmers' Market Participation and Performance - based on the perspective of transaction cost. Data Availability The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request. D ow nloaded from https://academ ic.oup.com /qopen/advance-article/doi/10.1093/qopen/qoad017/7161147 by Finnish Forest R esearch Institute / Library user on 17 M ay 2023 O R IG IN A L U N E D IT E D M A N U S C R IP T Abstract Farmers have started to adopt Information and Communication Technology (ICT), which has considerable potential to impact farm performance. This study uses data from a 2018 survey of 763 vegetable smallholder farms in China to estimate the impact of ICT on technical efficiency (TE). We adopt propensity score matching to create a balanced sample of ICT users and non- users, and a stochastic frontier model with sample selection correction to compare the two groups’ TE. After accounting for self-selection bias from both observables and unobservables, the study finds a positive effect of ICT use on TE. On average, the TE score of ICT users is 0.64, whereas ICT non-users have a lower score of 0.57. A quantile regression analysis further reveals a heterogeneous impact of ICT on TE, with the largest effects among less efficient farms. These results suggest that vegetable farmers’ performance could be fostered by the widespread use of ICT. Keywords: Information and Communication Technology (ICT), Technical efficiency, Vegetable production, Selection bias, Propensity score matching. JEL codes: D24, Q12, Q16 D ow nloaded from https://academ ic.oup.com /qopen/advance-article/doi/10.1093/qopen/qoad017/7161147 by Finnish Forest R esearch Institute / Library user on 17 M ay 2023 O R IG IN A L U N E D IT E D M A N U S C R IP T 1. Introduction China, as the foremost global producer of vegetables cultivating over half of the world’s total output, has for a long time prioritized agricultural development and the increase of farm productivity (Food and Agriculture Organization, “FAO”, 2023). China's agricultural production is characterized by small-scale family farms less than 2 hectares in size which account for 80% of vegetable farms (National Bureau of Statistics of China, “NBSC”, 2017). One of the major challenges faced by farmers is the lack of timely and effective information for farm management (Feather and Amacher, 1994). This is especially obvious for vegetable producers in remote villages since issues such as information distortion make it difficult to ensure effectiveness and efficiency in resource use (Zheng et al., 2021). Therefore, understanding how to broaden and optimize the information channels for small farmers to improve agricultural productivity is crucial. Since the 1990s, Information and Communication Technology (ICT), represented by all devices that allow data to be digitized, stored, and transmitted, has gradually gained popularity and profoundly changed the way farmers obtain information (Lio and Liu, 2006). The Chinese government attempted to seize this opportunity by promoting rural ICT infrastructure construction and adoption to narrow the urban-rural Digital Divide.1 These activities, such as mobile phone trainings and the "Internet +” 2 demonstration project for farmers, have been organized and implemented with increasing intensity in rural areas since 2015. According to the 50th Statistical Report on China’s Internet Development released by China Internet Network 1 The digital divide refers to the information gap between various entities in the global digitalization process. 2 "Internet+" refers to the integration of the Internet and other innovative information technologies into traditional industries in China. D ow nloaded from https://academ ic.oup.com /qopen/advance-article/doi/10.1093/qopen/qoad017/7161147 by Finnish Forest R esearch Institute / Library user on 17 M ay 2023 O R IG IN A L U N E D IT E D M A N U S C R IP T Information Center (2022), with the implementation of the Digital Rural Development strategy, the scale of Chinese rural netizens continues to expand. At the end of 2022, 82.2% of farm households had access to a mobile phone, and internet penetration in rural areas of China reached 58.8%. The number of rural ICT users in China amounted to 293 million or 27.9% of the national total. The salient role of ICT in improving farmers’ productivity in developing countries has been confirmed by several studies in recent years (Azhar et al., 2017; Mwalupaso et al., 2019; Zheng et al., 2021). However, previous research concentrated on arable crops such as cereals, or fruits. For example, based on a survey carried out by the Punjab Economic Research Institute in 2017, Azhar et al. (2017) illustrated the potential role of ICT in enhancing agricultural productivity by using a sample of more than 500 crop farms in India. Mwalupaso et al. (2019) reported a positive association between mobile phone use and technical efficiency (TE) for a sample of 200 corn farms in Zambia. Zheng et al. (2021) analyzed the positive impact of internet use on the TE of smallholder farms in China by using data on banana production. However, neglecting that farmers' decision to adopt ICT could be affected by various personal and demographic factors while investigating a causal link between ICT use and efficiency may lead to inaccurate results due to self-selection bias3. Previous studies addressed this problem by implementing propensity score matching (PSM), an instrumental variable (IV) approach, or endogenous switching regression (ESR) models (Issahaku et al., 2018; Deng et al., 2019; Twumasi et al., 2021; Zhu et al., 2021; Ma and Zheng, 2022). For example, Issahaku et al. (2018) used PSM and concluded 3 Self-selection bias occurs when adopters differ from non-adopters due to unobserved factors that are also relevant for the outcome variable (TE in our case). Farmers who use ICT may have different characteristics, attitudes, or experiences compared to those who do not use it. This can result in a biased estimate for the effect of adoption on the outcome. D ow nloaded from https://academ ic.oup.com /qopen/advance-article/doi/10.1093/qopen/qoad017/7161147 by Finnish Forest R esearch Institute / Library user on 17 M ay 2023 O R IG IN A L U N E D IT E D M A N U S C R IP T that mobile phone use significantly improves agricultural productivity in Ghana. Zhu et al. (2021) applied stochastic frontier (SF) analysis and the ESR model to investigate how the TE of apple farmers in China is improved by internet use. Ma and Zheng (2022) used an IV approach to examine the effects of smartphone use on rural development in China. In this study, we examine the relationship between ICT use and farm TE, based on survey data for 763 vegetable farmers in Shandong province of China. The Chinese vegetable sector is an interesting sector to investigate for at least three reasons. First, vegetables as a group are the most widely cultivated and economically essential crop in Chinese agriculture. According to FAO (2023), the vegetable planting area in China has reached 20 million hectares and vegetable output was over 600 million metric tons in 2021, accounting for 52% of the global production volume of vegetables. Considering the ever-increasing global demand, it is crucial for Chinese vegetable farmers to improve production efficiency and maximize profits, thereby meeting the growing need and maintaining its position as a leading producer and supplier of vegetables in the global market (Gale and Hu, 2012). Second, vegetable production is characterized by intensive farming with a low fallow ratio, greater use of inputs such as capital and labor, and higher yields per unit of land due to the shorter growing season and multi-harvest (Stringer et al., 2009; Silva Dias, 2010). More frequent pest and disease management as well as more careful attention to soil fertility and water management are needed during production compared to other crops (Watson et al., 2002; Dinham, 2003). Therefore, seeking out optimal production practices can help farmers grow vegetables with higher TE and guarantee their well-being. Third, ICT use might prove particularly effective in improving TE in vegetable production due to special marketing conditions. Vegetables have very limited shelf-life, and adapting vegetable production and sales timing has become increasingly critical (Ruben et al., 2007). Vegetable producers must meet D ow nloaded from https://academ ic.oup.com /qopen/advance-article/doi/10.1093/qopen/qoad017/7161147 by Finnish Forest R esearch Institute / Library user on 17 M ay 2023 O R IG IN A L U N E D IT E D M A N U S C R IP T higher requirements in terms of rapid access to market information and broadening distribution channels (Richards and Rickard, 2020). Given the increasing prominence of ICT and the urgency to improve the productivity of vegetable production, our study can shed light on whether ICT adoption affects the TE of smallholder vegetable farms in China. The contributions of our article to the empirical literature are as follows. First, instead of using indicators similar to previous studies such as general ICT investment or ownership (Becchetti et al., 2003; Ruddock, 2006), we make use of a dataset that allows defining ICT adoption more accurately as using ICT specifically to find information for purposes of vegetable production4. In this way, we can make sure that farmers’ ICT use is closely linked to vegetable production. Second, potential economic benefits and poverty alleviation impacts associated with vegetable production are well recognized (Schreinemachers et al., 2017), but research on vegetable production is, in general, scarce. This study contributes to closing this gap. Third, by estimating quantile treatment effects (QTE), policymakers are offered novel insights into the heterogeneous nature of the effect of ICT on TE, which can facilitate the design of tailored and effective solutions that cater to the diverse needs of different vegetable farmers subgroups. The remainder of this paper is organized as follows: The following section briefly reviews the conceptual framework of our study. The subsequent sections describe the data collection and summary statistics as well as the empirical strategy. Section 5 shows and discusses the estimated results. The final section concludes and discusses policy implications. 4 Farmers invest in ICT for a variety of purposes, including contacting friends and family for social networks, playing games for entertainment, messaging or calling others for problem-solving, and obtaining information online to enhance production (Lwoga, 2010; Zheng and Lu, 2021). Adopting ICT investment or ownership as a treatment variable cannot reflect the impact of ICT on farmers’ production performance, which would be misleading. D ow nloaded from https://academ ic.oup.com /qopen/advance-article/doi/10.1093/qopen/qoad017/7161147 by Finnish Forest R esearch Institute / Library user on 17 M ay 2023 O R IG IN A L U N E D IT E D M A N U S C R IP T 2. Conceptual Framework In the following, we elaborate on key concepts in the productivity framework to clarify the possible pathways regarding how ICT adoption can affect the TE of farms. We start out with a simple representation of the production frontier, which represents the maximum output attainable from each input level. Productivity can be measured by the ratio of (aggregated) output over (aggregated) input (Coelli, 1995). Therefore, the farms on the frontier are technically efficient and those below are not because a greater amount of output can be achieved with the same input level or inputs can be saved without compromising the level of output. Thus, attaining elevated TE necessitates either augmenting the output with the current inputs or diminishing the inputs with the prevailing output. ICT is a potential TE driver for several reasons: First, ICT can help farmers make better decisions and guide farmers to apply appropriate farming practices. Challenges faced by vegetable farmers in developing countries include a lack of training and skills, limited access to inputs, and inadequate agricultural extension services. ICT improves the transfer of knowledge through exchanging direct, timely, and worldwide information and ideas among farmers, experts, and other organizations to address these problems (Hobbs, 1996; Bozoǧlu and Ceyhan, 2007; Aker, 2008). Also, farmers might have easier access to guidance and training from authoritative experts. Secondly, the internet provides information on products and services and facilitates access to agricultural inputs of higher quality or lower price (Zhu et al., 2021). Farmers are no longer limited to the few options they had in the past and can keep up with factor markets by accessing the newest market information. Third, ICT can effectively connect farmers with suppliers and customers by facilitating D ow nloaded from https://academ ic.oup.com /qopen/advance-article/doi/10.1093/qopen/qoad017/7161147 by Finnish Forest R esearch Institute / Library user on 17 M ay 2023 O R IG IN A L U N E D IT E D M A N U S C R IP T communication and thus helps rural households to distribute labor and capital in a more efficient manner (Zanello and Srinivasan, 2014; Hou et al., 2019). Timely access to market and price information enables farmers to capture market dynamics and then adjust production strategies against possible risks and losses as soon as possible (Wellman et al., 1996; Wellman et al., 2010; Galperin and Viecens, 2017). 3. Data and Descriptive Statistics 3.1 Data collection Our empirical analysis relies on existing data from a farm-level survey that was conducted in Shouguang City during 2018 in the context of a project concerned with the influence of ICT adoption on farmers’ performance. Shouguang, the “hometown” of Chinese vegetables, is located in the coastal plain area of northern central Shandong Province (Figure 1), which ranked first among the provinces in vegetable production with a total production of 82 million tons, accounting for 12% of the national vegetable supply in 2018 (NBSC, 2019). The vegetable industry in Shouguang has experienced rapid growth, which can be attributed to several factors. Firstly, the region benefits from favorable geographical conditions, including a suitable climate and fertile soil. Secondly, a mature market environment has emerged in the area, with established supply chains and distribution networks supporting the vegetable industry. Additionally, the local government has played a key role in supporting the growth of the vegetable industry through various policy measures and investment initiatives, such as stimulating R&D, organizing trade exhibitions, and supporting land transfers as well as standardization and quality management systems (Ministry of Agriculture and Rural Affairs of China, 2022). As the largest vegetable distribution center in China, Shouguang vegetables are now even exported to various countries across the globe. To take production capacity to a higher level, information technology has been gradually enhanced in Shouguang. The construction of mobile communication and wireless D ow nloaded from https://academ ic.oup.com /qopen/advance-article/doi/10.1093/qopen/qoad017/7161147 by Finnish Forest R esearch Institute / Library user on 17 M ay 2023 O R IG IN A L U N E D IT E D M A N U S C R IP T paging systems has been hastened, while the prevalence of broadband networks is on the rise. In addition, government-backed websites (e.g., http://sg.vegnet.com.cn) have been set up to release timely information on vegetable prices, wholesaling, supply, and demand, which are updated daily. Various mobile applications (e.g., Shouguang Vegetables) are also available, providing farmers with the latest news on vegetable farming and the opportunity to interact with leading experts in the field online. Therefore, exploring the effects of ICT on Shouguang vegetable producers' TE could provide policymakers to undertake further policy actions to help rural farmers benefit from ICT use in agricultural production. Probability proportional sampling and random sampling methods were employed to select vegetable farmers in Shouguang. First, six towns were randomly chosen which include Gucheng, Hualong, Luocheng, Sunjiaji, Tianliu, and Wenjia. Second, 4 to 7 villages were randomly selected in each town. Finally, face-to-face in-depth interviews with around 15 to 25 vegetable farmers randomly selected in each of the selected villages were conducted. All the interviewers were from the research team of Northwest A & F University and have been trained for data collection purposes. A structured questionnaire was used to collect detailed information on the household head’s characteristics (e.g., age, gender, and education), demographic and socioeconomic characteristics (e.g., household size, dependency ratio, and farm size), output and inputs (e.g., total income, expenditures on seeds and fertilizers) of vegetable production, and farmers’ status of ICT adoption. In total, 796 household observations were obtained. For our study, after checking for inconsistencies and excluding invalid questionnaires that had missing data, a total of 763 observations could be retained for our empirical analysis. D ow nloaded from https://academ ic.oup.com /qopen/advance-article/doi/10.1093/qopen/qoad017/7161147 by Finnish Forest R esearch Institute / Library user on 17 M ay 2023 O R IG IN A L U N E D IT E D M A N U S C R IP T 3.2 Descriptive statistics and variable selection Table 1 contains summary information on the variables used in the empirical analysis. The treatment variables ICT use (respondents indicating whether they use ICT to find information related to vegetable production) is used to categorize the smallholder farmers into the treatment groups of ICT users (ICT) and the control group of ICT non-users (NICT). Output is farmers’ vegetable sales income in 2018. Input variables refer to the factors used in production. Labor measures the costs of household labor and hired labor (see Appendix A.1 for the details). Land is the total size of vegetable production (unit: mu5). Fertilizer and Pesticide are the expenditures on fertilizer and pesticides. Expenses for irrigation, seeds, machinery, and mulch are consolidated into Other. For determinants of ICT use, previous studies pointed out household characteristics, local conditions, and geographic attributes (Issahaku et al., 2018; Mwalupaso et al., 2019; Zheng et al., 2021; Zhu et al., 2021). We choose the householder’s gender, age, education (number of school years), experience (number of years working as a farmer), certificate (having the official professional farmer certificate), family burden ratio (the number of family members without own income divided by the number of members with an income), market (distance to closest farmers’ market), government (distance to closest government administration), car (having a car), cooperative (participating in a cooperative), training (ICT training in the village), acquaintances (number of frequent acquaintances), information literacy (ability to acquire and process information, see Appendix A.2 for the definition of this variable), social capital (frequency and quality of social contacts, see Appendix A.3), and five locational dummy variables (Gucheng/Hualong/Luocheng/Sunjiaji/Tianliu) as a set of relevant covariates. 5 1 mu ≈ 0.0667 hectares. D ow nloaded from https://academ ic.oup.com /qopen/advance-article/doi/10.1093/qopen/qoad017/7161147 by Finnish Forest R esearch Institute / Library user on 17 M ay 2023 O R IG IN A L U N E D IT E D M A N U S C R IP T The selection of variables is grounded on theoretical economic reasoning and prior empirical findings, while the inputs and control variables undergo variance inflation factor tests to ensure they do not suffer from multicollinearity issues (see Appendices B) (Abate et al., 2014; Aldosari et al., 2019; Key and Mcbride, 2014; Smith et al., 2004). Table 1 also presents the mean and standard deviation of pooled treatment (ICT) and control (NICT) groups. There are 369 observations in the NICT group and 394 in the ICT group. Compared with the NICT group, farmers who use ICT are younger and higher educated. The ICT users are more likely to have an official professional farmer certificate and participate in the cooperative. ICT users have a higher potential to own a car. More of the farmers who live in a village with training use ICT. Farmers’ information literacy and social capital positively impact their potential in using ICT. In addition, the set of variables that are significantly higher in ICT users than in non-users contains family burden ratio and distance to local government. The significant differences in these variables demonstrate a potential self-selectivity problem when estimating the ICT effect. 4. Empirical Approaches To study the effect of ICT on TE, we follow a multi-step procedure that gradually corrects potential bias due to observables and unobservables. First, we present the results of the SF model on the original (unmatched) sample, which likely suffers from selection bias. We then use PSM to form a balanced sample of ICT users and non-users to address bias from observed characteristics. Then, Greene’s (2010) sample selection model is applied to the matched sample to correct for possible bias arising from unobserved factors. We compare the TE scores of ICT users and non-users resulting from different combinations of these correction procedures, with the most reliable results being those of the sample selection SF model on the matched sample. For brevity’s sake, the robustness checks relying on different matching routines, functional forms D ow nloaded from https://academ ic.oup.com /qopen/advance-article/doi/10.1093/qopen/qoad017/7161147 by Finnish Forest R esearch Institute / Library user on 17 M ay 2023 O R IG IN A L U N E D IT E D M A N U S C R IP T for frontier estimation, and an ESR model are reported in Appendix E. 4.1 Stochastic frontier model: TE estimation TE measures an individual’s ability to maximize output from given inputs and can be estimated in different ways such as data envelopment analysis and the SF model. SF model, a parametric approach, has the advantage of smaller sensitivity to measurement errors by involving a symmetric random variable to allow for statistical noise, while a one-sided random variable is included to allow for inefficiency (Bauer, 1990; Battese, 1992; Førsund et al., 1980). The general form is as follows: 𝑌௜ = 𝑓(𝑋௜; ) ∗ exp(𝑽௜ − 𝑼௜) #(1) where  is a vector of parameters to be estimated. 𝑌௜ is the output of the ith individual. The set of 𝑋௜ is the independent input variable. 𝑉௜~𝑖𝑖𝑑(0, 𝜎௩ଶ) represents omitted variables, functional form errors, and measurement errors, and 𝑈௜~𝑖𝑖𝑑(0, 𝜎௨ଶ) denotes a non-negative random variable capturing the inefficiency effect. A Translog (transcendental logarithmic) SF model6, as a flexible functional form with the ability to more accurately model production processes, is used in our research to approximate the production technology as follows: 6 The initial application of SF was based on Cobb-Douglas (CD) production function which has the restrictive properties of assuming constant output elasticities and elasticities of input substitution equal to unity (Wilson et al., 1998). However, when determining the form of a farmer’s agricultural production function, the substitution elasticities between various inputs are not known in advance. However, the Translog production function does not entail as many restrictions as in the CD form, and allows for more complex relationships among the inputs (Addai et al., 2014). Additionally, we conducted a likelihood ratio test to compare the performance of the Translog SF model and the CD form and the statistic for testing H0 (CD form) against H1 (Translog form) is 53.54, indicating that the Translog SF model provides a better fit for our analysis. D ow nloaded from https://academ ic.oup.com /qopen/advance-article/doi/10.1093/qopen/qoad017/7161147 by Finnish Forest R esearch Institute / Library user on 17 M ay 2023 O R IG IN A L U N E D IT E D M A N U S C R IP T 𝑙𝑛𝑌௜ = 𝛽଴ + ෍ 𝛽௝ ௃ ௝ୀଵ 𝑙𝑛𝑋௜௝ + 1 2 ෍ ෍ 𝛽௝௠𝑙𝑛(𝑋௜௝)𝑙𝑛(𝑋௜௠) ெ ௠ୀଵ ௃ ௝ୀଵ + 𝑣௜ − 𝑢௜#(2) TE is defined as the ratio of observed output to the corresponding stochastic frontier output (Battese, 1992; Jondrow et al., 1982) and can be calculated as: 𝑇𝐸௜ = 𝑌௜ 𝑓(𝑋௜;) ∗ exp(𝑽௜) = exp(−𝑼௜) #(3) 4.2 Propensity score matching: observed bias correction This study aims to measure the average impact of ICT on small farm households’ TE. Merely comparing the gaps between TE scores of users and non-users in a direct and simple way, without considering the differences in the initial conditions of the two groups of farmers, may not reflect the effect of ICT. In 1974, Rubin proposed a counterfactual framework named the Rubin Causal Model (Rubin, 1974). The counterfactual is the potential result or state of affairs that will occur if the cause does not exist (Shadish et al., 2002). Our concern is how the farmers' TE would have been impacted if those who did not use ICT had utilized it. However, as this scenario was never observed, PSM proposed by (Rosenbaum and Rubin, 1983a) is applied to create a control group with similar observed characteristics as the treatment group, thereby creating a counterfactual outcome. According to the Rubin Causal Model, this study divides sample households into a treatment group (ICT) and a control group (NICT). We use i to indicate the individual farmer and 𝐷௜ to indicate whether farmer i adopts ICT or not. For the next step, a probit regression is used to estimate farmers’ propensity scores (P- score), which is defined as the conditional probability 𝑝(z௜) that an individual is predicted to adopt ICT given observed characteristics z௜. The covariates that we selected for matching ICT users and non-users include the householder's gender, age, education, experience, certificate, D ow nloaded from https://academ ic.oup.com /qopen/advance-article/doi/10.1093/qopen/qoad017/7161147 by Finnish Forest R esearch Institute / Library user on 17 M ay 2023 O R IG IN A L U N E D IT E D M A N U S C R IP T distance to market, distance to government, family burden ratio, cooperative participation, training, acquaintances, information literacy, social capital, and locational variables. The propensity score is thus estimated as follows: 𝑝(z௜) ≡ 𝑝(𝐷௜ = 1|z = z௜)#(4) Then, based on the calculated P-score, each ICT user is matched with a similar non-user. Various matching algorithms are tested to evaluate the matching quality in terms of how much selection bias is reduced. In this study, we test the performance of nearest neighbor matching, radius matching, and kernel matching, and find that all three alternative methods lead to very similar results for the estimated bias. Considering the trade-off between sample size and matching quality, the best result is obtained using Gaussian kernel matching. After matching, the standardized bias (S) is used to check if the distribution of the relevant variables is balanced in both the control and treatment groups. After conditioning the propensity score, there should be no big differences between the covariates. The expression for S is: 𝑆 = |𝑧୍̅ େ୘ − 𝑧୒୍̅େ୘| ඨ𝑠௭,୍େ୘ ଶ − 𝑠௭,୒୍େ୘ ଶ 2 #(5) where 𝑧୍̅ େ୘, 𝑧୒୍̅େ୘, s௭,ூ஼்ଶ and s௭,ேூ஼்ଶ represent the mean and variance of the covariate of both two groups respectively. Generally, the standardized bias should not exceed 10% (Rosenbaum and Rubin, 1983b). 4.3 Selection-corrected SF model: unobserved bias correction PSM can eliminate the self-selection bias based on the unconfoundedness assumption, which states a strong identifying assumption that all variables influencing the adoption decision and outcome variables should be included. The standard regression techniques result in biased D ow nloaded from https://academ ic.oup.com /qopen/advance-article/doi/10.1093/qopen/qoad017/7161147 by Finnish Forest R esearch Institute / Library user on 17 M ay 2023 O R IG IN A L U N E D IT E D M A N U S C R IP T inconsistent estimators if the correlation between the unobservable factors affecting outcome and those affecting the selection process has not been considered (Greene, 2010; Lai, 2015; Bravo- Ureta et al., 2021; Vrachioli et al., 2021). Thus, the selection-corrected SF model is applied to eliminate selection bias from unobservable factors. The setting of the SF model with sample selection consists of two equations. Sample selection: 𝐷௜ = ℎ௜𝛾 + 𝑒௜ , 𝑤𝑖𝑡ℎ 𝐷௜ = ൜ 1, 𝑖𝑓 𝐷௜ > 0 0, 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒 #(6) SF model: 𝑌௜ = ൜ 𝑓(𝑋௜; 1) ∗ 𝑒𝑥𝑝(𝒗ଵ௜ − 𝒖ଵ௜) , 𝑖𝑓 𝐷௜ = 1 𝑓(𝑋௜; 2) ∗ 𝑒𝑥𝑝(𝒗ଶ௜ − 𝒖ଶ௜) , 𝑖𝑓 𝐷௜ = 0 #(7) where ℎ௜ represents a vector of the individual factors which may affect farmers’ choice of using ICT to obtain vegetable production information, 𝛾 is the corresponding coefficient vector, and 𝑒௜ is the normalized error. Given the sample selection, the output 𝑌௜ is observed in two possible group production technologies, corresponding to the vectors of ଵ and ଶ to be estimated. 𝒗ଵ௜ , 𝒖ଵ௜, 𝒗ଶ௜, and 𝒖ଶ௜ follow the same definition as in equation (1). To derive the likelihood function of equations (6) and (7), the three symmetric errors are imposed to be independent of the vectors of explanatory variables, and assumed to be a sequence of i.i.d. bivariate normal random vectors such that 1 1 2 1 2 2 2 1 1 1 12 1 2 2 2 12 1 2 22 0 1 0 , 0 v i v v i v v v v v v vi e v v                                                 (8) where ଵଶ denotes the correlation coefficient between 𝑉ଵ௜ and 𝑉ଶ௜ ; ଵ denotes the correlation D ow nloaded from https://academ ic.oup.com /qopen/advance-article/doi/10.1093/qopen/qoad017/7161147 by Finnish Forest R esearch Institute / Library user on 17 M ay 2023 O R IG IN A L U N E D IT E D M A N U S C R IP T coefficient between 𝑉ଵ௜ and e୧; ଶ denotes the correlation coefficient between 𝑉ଶ௜ and e୧. We use a two-step approach to estimate the equation system (Greene, 2010; Lai, 2015). In the first step, we estimate the selection equation given in (6) by the probit model. Let ଵ and ଶ be the first step, maximum likelihood estimator. Then, we estimate the frontier model (7) with the given ଵ and ଶ for the sample selection model. If ଵ or ଶ is nonzero, then the unobserved factor 𝑣ଵ௜ or 𝑣ଶ௜ affecting the 𝑦௜ is correlated with the unobserved factor 𝑒௜ affecting the selection process. Otherwise, the endogenous self-selection bias from unobservable factors could be ignored. 5. Results and Discussions 5.1 Results of the stochastic frontier analysis: Unmatched sample The estimated results of the conventional SF model and selection-corrected SF model for ICT users and non-users using the whole sample are shown in Table 2. ଵ is statistically significant at the 1% level, which suggests the SF model with correcting selection should be considered. The likelihood ratio (LR) tests7 in both regimes also reject the conventional SF model. Before estimation, all variables are divided by their geometric mean, so the first-order coefficients can be interpreted as output elasticities evaluated at the sample mean (Orea, 2002; Alvarez and Arias, 2004). For ICT users, the output elasticity with respect to fertilizer is 0.30, which indicates that a 1% increase in fertilizer use will lead to a 0.3% increase in output. Land contributes the most to farm production with an output elasticity of 0.49 for farmers using ICT. The finding aligns with Bozoǧlu and Ceyhan (2007)’s survey of vegetable farms in Turkey as well as Dong et al. (2019)’s research on a household-level survey of greenhouse-grown vegetable 7 𝐻଴: ଵ=0. The LR statistic is (-2)*[(-316.95)-(-512.46)] =391.02 ~ 𝜒 ଶ(1) 𝐻଴: ଶ=0. The LR statistic is (-2)*[(-312.30)-(-517.95)] =411.3 ~ 𝜒 ଶ(1) D ow nloaded from https://academ ic.oup.com /qopen/advance-article/doi/10.1093/qopen/qoad017/7161147 by Finnish Forest R esearch Institute / Library user on 17 M ay 2023 O R IG IN A L U N E D IT E D M A N U S C R IP T planting in northern China. The impact of other intermediate inputs and labor on output is lower, at 0.20 and 0.08, respectively. Additional pesticide use is not found to improve ICT users’ output, which is in agreement with the results of previous studies on vegetable production by Padmajani et al. (2014) in Sri Lanka and Yang et al. (2019) in China. One possible explanation is that farmers' excessive use of pesticides to mitigate the risk of crop loss caused by pests and diseases can lead to lower yields in vegetable production. In the case of ICT non-users, land size has the highest elasticity of 0.30, while the contributions of fertilizer and other inputs to output are roughly 0.26 and 0.23, respectively. The output elasticity of pesticide and labor is comparatively lower at around 0.1%. Our estimations are similar to the studies of Dong et al. (2019) and Zheng et al. (2021). The sums of all partial production elasticities for both ICT users and non-users are around 1, indicating the constant returns to scale, which is consistent with the findings of Udoh (2005) and (Shrestha et al., 2016). The average TE scores obtained by the conventional and selection-corrected SF model for the unmatched sample are shown in Table 6. The average TE scores in the conventional SF model are 0.62 for ICT users and 0.57 for ICT non-users. In the selection-corrected SF model, slightly higher TE scores are estimated. If we control for the unobservable bias, the TE score of ICT non-users increases by 0.03 while the TE score of ICT users increases by 0.01. The TE scores obtained from two-group frontiers in the selection-corrected SF present the positive effect of ICT use on vegetable farmers’ TE when considering unobserved bias. 5.2 Results of propensity score matching We apply PSM to match ICT users and non-users to address observed bias. After estimating the P-scores and matching ICT users and non-users, there are 739 observations left. As shown in Appendix C, the standardized biases of the covariables are significantly reduced after matching, D ow nloaded from https://academ ic.oup.com /qopen/advance-article/doi/10.1093/qopen/qoad017/7161147 by Finnish Forest R esearch Institute / Library user on 17 M ay 2023 O R IG IN A L U N E D IT E D M A N U S C R IP T and their absolute values are all reduced to less than 10%, indicating that the matching is effective. The coefficients and marginal effects of the probit model in Table 3 show how the factors impact farmers' choice of using ICT to obtain information. Gender has a significant and positive impact on using ICT, which suggests that male farmers are more likely to apply ICT to obtain information than females. Age has a significant and negative impact on the decision to use ICT, which presents that older farmers are not inclined to use ICT. This is consistent with the view that older farmers are more likely to have poorer ICT skills. Experience increases the likelihood of using ICT for agricultural information because experienced farmers are better informed about technology adoption (Okello et al., 2012; Paustian and Theuvsen, 2017). Households who live closer to the government or have ICT training in the village are more inclined to use ICT because they are often more susceptible to favorable policies designed to strengthen agriculture and benefit farmers, particularly in terms of innovative technologies (Kiiza and Pederson, 2012). Farmers who participate in agricultural cooperatives are more likely to use ICT because agricultural cooperatives often promote ICT adoption and transmit information via ICT (Abdul- Rahaman and Abdulai, 2018). Having a higher information literacy score plays a vital role in ICT adoption and its benefit from the resources available, which is in line with the previous literature (Aker, 2011; Zanello and Maassen, 2014). It should be noted that both owning a certificate and cooperative membership have the potential to affect TE, which may bias the results. We address this endogeneity issue by using the two-stage control function model suggested by Wooldridge (2015), with further elaboration and the first-stage estimates provided in Appendix D. Table 3 presents the coefficients of the generalized residuals predicted from the first stage of the control function for the variables of certificate and cooperative. The results indicate that the coefficients D ow nloaded from https://academ ic.oup.com /qopen/advance-article/doi/10.1093/qopen/qoad017/7161147 by Finnish Forest R esearch Institute / Library user on 17 M ay 2023 O R IG IN A L U N E D IT E D M A N U S C R IP T of both residuals are statistically significant, suggesting that certificate and cooperative are endogenous in the ICT decision model. 5.3 Results of the stochastic frontier analysis: Matched sample Table 4 shows the parameter estimation results of the conventional and selection-corrected SF models for the matched sample. Compared to the unmatched sample, the output elasticities and the return to scale in both models do not change much. The estimation of the selection-corrected SF models reveals that the coefficient of the sample-selection-bias variable ଵ is statistically different from zero for the ICT group, which is consistent with the unmatched sample. This result again suggests the presence of selection bias when estimating the conventional SF for ICT users. Moreover, we cannot find evidence strongly supporting the significant statistic of ଶ which implies the selection bias in the SF of ICT non-users. As shown in Table 5, after matching, the average TE scores of ICT users are 0.63 in conventional SF and 0.64 in selection-corrected SF, respectively, while the scores of ICT non- users are 0.57 in both conventional SF and selection-corrected SF. Our analysis shows that the difference in mean TE between ICT users and non-users is higher in the matched sample than in the unmatched sample, with an increase from 0.03 to 0.07. This indicates that neglecting selection bias resulting from both observed and unobserved factors can lead to an underestimation of the mean TE difference between ICT users and non-users. This finding aligns with previous studies by Zheng et al. (2021) and Zhu et al. (2021). In terms of the mean score of TE, there are a few noteworthy points to consider. Firstly, our findings align with previous studies conducted by Liang et al. (2019) and Dong et al. (2019) in China, with an average score of around 0.62. However, when compared to neighboring countries that are major vegetable producers worldwide, such as Vietnam where farmers have an average D ow nloaded from https://academ ic.oup.com /qopen/advance-article/doi/10.1093/qopen/qoad017/7161147 by Finnish Forest R esearch Institute / Library user on 17 M ay 2023 O R IG IN A L U N E D IT E D M A N U S C R IP T efficiency score of 0.75 (Nguyen et al., 2021), or India where the score is 0.77 (Murthy et al., 2009), Chinese vegetable farmers appear to have lower TE scores. One possible explanation for this gap could be the land tenure system in China, which may not be as supportive of efficient vegetable farming practices as in other countries. Land use or transfer restrictions limit producers' ability to invest in land resources and improve their TE (Krusekopf, 2002). Moreover, as the largest producer and consumer of vegetables globally, China faces significant environmental challenges, such as soil pollution and water scarcity (Khan et al., 2009). These challenges make it difficult for vegetable farmers to operate efficiently, thereby lowering their TE. Secondly, vegetable farmers tend to have lower efficiency scores than other crop farmers in China. For instance, the TE scores for apple farms and rice farming were found to be 0.83 and 0.9, respectively (Ma et al., 2018; Wang et al., 2020). The labor-intensive nature of vegetable farming work is one of the reasons for this discrepancy, as it involves tasks like hand-weeding, multi-harvesting, and diverse pest control (Stringer et al., 2009; Silva Dias, 2010). Moreover, vegetables are more susceptible to environmental factors such as changes in temperature, water availability, and soil quality than some other crops (Tripathi et al., 2016). Additionally, the lack of institutional and socio-economic support, such as extension services and cooperative assistance, can hinder the improvement of farmers’ TE in China (Zheng et al., 2021). These findings suggest that the effects of efficiency might be country- or crop-specific and thus need further empirical evidence. It is also worth noting that different productivity analysis models and variable settings can influence the level of the efficiency score (Madau, 2015). 5.4 Quantile treatment effect of ICT use on TE To design appropriate agricultural development policies, it is important to understand the D ow nloaded from https://academ ic.oup.com /qopen/advance-article/doi/10.1093/qopen/qoad017/7161147 by Finnish Forest R esearch Institute / Library user on 17 M ay 2023 O R IG IN A L U N E D IT E D M A N U S C R IP T heterogeneity of the effect of ICT on TE. To achieve this, we utilize the residualized quantile regression (RQR) model proposed by Borgen et al. (2021), which offers a flexible approach for estimating the treatment effects across the distribution of outcomes. The RQR model is estimated through a two-step approach in our study. First, the treatment variable (ICT) is regressed on control variables using ordinary least squares, which allows us to decompose the variance of the treatment variable into two components: a portion that can be explained by the observed control variables and a residual component that is orthogonal to the controls. In the second step, the outcome variable (TE) is regressed on the residualized treatment variable using the method of minimum absolute deviation. At last, QTE can be computed from observed data while adjusting for selection bias by comparing quantiles (𝜏) of the outcome distribution for individuals with different treatment values as 𝑄𝑇𝐸ఛ = 𝑄ூ஼்ఛ − 𝑄ேூ஼்ఛ . The results can be found in Table 6. The coefficients except for the quantile at 90th exhibit a positive and significant relationship between ICT use and TE, which is in line with our previous results. Specifically, from the 10th to 25th percentile, the coefficient for the ICT treatment has a slight increase. The largest effect is observed at the 25th percentile, with a coefficient of 0.12, suggesting that the ICT treatment has a more pronounced effect on the lower quantiles of the distribution (Zheng et al., 2021). These findings suggest that adopting ICT is likely to be more effective for less efficient farms to begin with, as they have more room for improvement. The effect is not significantly different from zero at the 90th percentile, indicating that ICT use is less effective for the most efficient farms, potentially because they have already optimized their production processes to a greater extent with the help of other information channels. 6. Conclusion and Policy Implication This study uses data from a survey of 763 small vegetable farms in China to estimate the impact D ow nloaded from https://academ ic.oup.com /qopen/advance-article/doi/10.1093/qopen/qoad017/7161147 by Finnish Forest R esearch Institute / Library user on 17 M ay 2023 O R IG IN A L U N E D IT E D M A N U S C R IP T of farmers’ ICT use on vegetable production performance. Propensity score matching and selection-corrected stochastic frontier model are combined to correct for selection bias from both observed and unobserved factors. The results reveal that the difference in TE between ICT users and non-users is statistically and economically significant after addressing both observed and unobserved biases, which highlights the potential benefits brought by ICT use to vegetable production in China. Further empirical analysis conducted by the quantile treatment effect model reveals that the impact of ICT on TE is heterogeneous, with the largest effects observed among the less efficient farms, and decreasing as we move towards the median and becoming statistically insignificant for the most efficient farms. The findings have several important policy implications: First, our findings indicate the considerable potential of ICT use to improve TE in the sector, which suggests increasing government subsidies to improve the penetration rate of ICT and further promote the modernization of rural areas, especially the investment in broadband infrastructure. Second, ICT use appears to be particularly effective for low-TE farms, indicating that those farms should be targeted with greater focus. Third, results from the propensity score model suggest that effective ways to foster ICT adoption are the provision of training and enhancing information literacy in general. Farmers might still lack the ability to make effective use of information acquired by ICT, which can be found in our survey that only 16% of farmers have ICT training in their villages. Thus, the government should encourage and guide farmers to use ICT to obtain agricultural information by providing ICT-related training, stimulate the provision of agricultural information by, e.g., extension services, and cultivate their information awareness and literacy. Especially for the less technical efficient and less-educated farmers, their level of using the ICT should be improved so that they can access and apply fundamental information and benefit from D ow nloaded from https://academ ic.oup.com /qopen/advance-article/doi/10.1093/qopen/qoad017/7161147 by Finnish Forest R esearch Institute / Library user on 17 M ay 2023 O R IG IN A L U N E D IT E D M A N U S C R IP T it. Fourth, as a more ancillary result, the department of agricultural information services should provide effective guidance and regulation which will lead to an optimum match between information supply and demand thereby facilitating farmers’ utilization of information resources via ICT. Limitations and outlook of our study include the fact that our survey sample is limited to a cross-sectional data set. Panel data that might be obtained in future research will allow more rigorous treatment of endogeneity due to time-invariant unobservables and additionally better reflect the impact of ICT on productivity change. As a future outlook, new ICT developments might bring about new implications for farm performance. The Central Government of China proposed to accelerate the planning and construction of the fifth-generation communication technology (5G) in rural areas, establish the agricultural big data system, and promote the in- depth integration of new-generation information technology with agricultural production and operation. Therefore, the advent of the 5G era is not only a new opportunity for agricultural development but also a new chapter to study new ICT developments’ implications for farm performance. D ow nloaded from https://academ ic.oup.com /qopen/advance-article/doi/10.1093/qopen/qoad017/7161147 by Finnish Forest R esearch Institute / Library user on 17 M ay 2023 O R IG IN A L U N E D IT E D M A N U S C R IP T References Abate G. T., Francesconi G. N. and Getnet K. (2014) ‘Impact of Agricultural Cooperatives on Smallholders’ Technical Efficiency: Empirical Evidence from Ethiopia’, Annals of Public and Cooperative Economics, 85: 257–286. Abdul-Rahaman A. and Abdulai A. (2018) ‘Do Farmer Groups Impact on Farm Yield and Efficiency of Smallholder Farmers? Evidence from Rice Farmers in Northern Ghana’, Food Policy, 81: 95–105. Addai K. N., Owusu V. and Danso-Abbeam G. (2014) ‘Effects of Farmer-Based-Organization on the Technical Efficiency of Maize Farmers across Various Agro-Ecological Zones of Ghana’, Journal of Economics and Development Studies, 3: 149–172. Aker J. C. (2008) ‘Does Digital Divide or Provide? The Impact of Cell Phones on Grain Markets in Niger’, SSRN Electronic Journal. Aker J. C. (2011) ‘Dial “A” for Agriculture: A Review of Information and Communication Technologies for Agricultural Extension in Developing Countries’, Agricultural Economics, 42: 631–647. Aldosari F., Al Shunaifi M. S., Ullah M. A., Muddassir M. and Noor M. A. (2019) ‘Farmers’ Perceptions Regarding the Use of Information and Communication Technology (ICT) in Khyber Pakhtunkhwa, Northern Pakistan’, Journal of the Saudi Society of Agricultural Sciences, 18: 211–217. Alvarez A. and Arias C. (2004) ‘Technical Efficiency and Farm Size: A Conditional Analysis’, Agricultural Economics, 30: 241–250. Azhar A., Tahir M. A., Ata K. and Adil S. (2017) ‘Enhancing Agriculture Efficiency : The Differential Impact of Communication Technologies and Other Modernisation Techniques’, Papers and proceedings: 433–445. Battese G. E. (1992) ‘Frontier Production Functions and Technical Efficiency: A Survey of Empirical Applications in Agricultural Economics’, Agricultural Economics, 7: 185–208. Bauer P. W. (1990) ‘Recent Developments in the Econometric Estimation of Frontiers’, Journal of Econometrics, 46: 39–56. D ow nloaded from https://academ ic.oup.com /qopen/advance-article/doi/10.1093/qopen/qoad017/7161147 by Finnish Forest R esearch Institute / Library user on 17 M ay 2023 O R IG IN A L U N E D IT E D M A N U S C R IP T Becchetti L., Londono Bedoya D. A. and Paganetto L. (2003) ‘ICT Investment, Productivity and Efficiency: Evidence at Firm Level Using a Stochastic Frontier Approach’, Journal of Productivity Analysis, 20: 143–167. Borgen N. T., Haupt A. and Wiborg Ø. (2021) ‘A New Framework for Estimation of Unconditional Quantile Treatment Effects: The Residualized Quantile Regression (RQR) Model’, SocArXiv: 1–61. Bozoǧlu M. and Ceyhan V. (2007) ‘Measuring the Technical Efficiency and Exploring the Inefficiency Determinants of Vegetable Farms in Samsun Province, Turkey’, Agricultural Systems, 94: 649–656. Bravo-Ureta B. E., González-Flores M., Greene W. and Solís D. (2021) ‘Technology and Technical Efficiency Change: Evidence from a Difference in Differences Selectivity Corrected Stochastic Production Frontier Model’, American Journal of Agricultural Economics, 103: 362–385. China Internet Network Information Center. (2022) The 50th Statistical Report on China’s Internet Development, http://www.stats.gov.cn/english/pdf/202010/P020201012324332948860.pdf. Coelli T. J. (1995) ‘Recent Developments in Frontier Modelling and Efficiency Measurement’, Australian Journal of Agricultural Economics, 39: 219–245. Deng X., Xu D., Zeng M. and Qi Y. (2019) ‘Does Internet Use Help Reduce Rural Cropland Abandonment? Evidence from China’, Land Use Policy, 89: 104243. Dinham B. (2003) ‘Growing Vegetables in Developing Countries for Local Urban Populations and Export Markets: Problems Confronting Small-Scale Producers’, Pest Management Science, 59: 575–582. Dong Y., Mu Y. and Abler D. (2019) ‘Do Farmer Professional Cooperatives Improve Technical Efficiency and Income? Evidence from Small Vegetable Farms in China’, Journal of Agricultural and Applied Economics, 51: 591–605. FAO. (2023). Leading producers of fresh vegetables worldwide in 2021, https://www.statista.com/statistics/264662/top-producers-of-fresh-vegetables-worldwide/. Feather P. M. and Amacher G. S. (1994) ‘Role of Information in the Adoption of Best Management Practices for Water Quality Improvement’, Agricultural Economics, 11: 159–170. D ow nloaded from https://academ ic.oup.com /qopen/advance-article/doi/10.1093/qopen/qoad017/7161147 by Finnish Forest R esearch Institute / Library user on 17 M ay 2023 O R IG IN A L U N E D IT E D M A N U S C R IP T Førsund F. R., Lovell C. A. K. and Schmidt P. (1980) ‘A Survey of Frontier Production Functions and of Their Relationship to Efficiency Measurement’, Journal of Econometrics, 13: 5–25. Gale H. F. and Hu D. (2012) ‘Food Safety Pressures Push Integration in China’s Agricultural Sector’, American Journal of Agricultural Economics, 94: 483–488. Greene W. (2010) ‘A Stochastic Frontier Model with Correction for Sample Selection’, Journal of Productivity Analysis, 34: 15–24. Hobbs J. E. (1996) ‘A Transaction Cost Approach to Supply Chain Management’, Supply Chain Management, 1: 15–27. Hou J., Huo X. and Yin R. (2019) ‘Does Computer Usage Change Farmers’ Production and Consumption? Evidence from China’, China Agricultural Economic Review, 11: 387–410. Issahaku H., Abu B. M. and Nkegbe P. K. (2018) ‘Does the Use of Mobile Phones by Smallholder Maize Farmers Affect Productivity in Ghana?’, Journal of African Business, 19: 302–322. Jondrow J., Knox Lovell C. A., Materov I. S. and Schmidt P. (1982) ‘On the Estimation of Technical Inefficiency in the Stochastic Frontier Production Function Model’, Journal of Econometrics, 19: 233–238. Key N. and Mcbride W. D. (2014) ‘Sub-Therapeutic Antibiotics and the Efficiency of U.S. Hog Farms’, American Journal of Agricultural Economics, 96: 831–850. Khan S., Hanjra M. A. and Mu J. (2009) ‘Water Management and Crop Production for Food Security in China: A Review’, Agricultural Water Management, 96: 349–360. Kiiza B. and Pederson G. (2012) ‘ICT-Based Market Information and Adoption of Agricultural Seed Technologies: Insights from Uganda’, Telecommunications Policy, 36: 253–259. Krusekopf C. C. (2002) ‘Diversity in Land-Tenure Arrangements under the Household Responsibility System in China’, China Economic Review, 13: 297–312. Lai H. pin (2015) ‘Maximum Likelihood Estimation of the Stochastic Frontier Model with Endogenous Switching or Sample Selection’, Journal of Productivity Analysis, 43: 105–117. D ow nloaded from https://academ ic.oup.com /qopen/advance-article/doi/10.1093/qopen/qoad017/7161147 by Finnish Forest R esearch Institute / Library user on 17 M ay 2023 O R IG IN A L U N E D IT E D M A N U S C R IP T Liang Y., Jing X., Wang Y., Shi Y. and Ruan J. (2019) ‘Evaluating Production Process Efficiency of Provincial Greenhouse Vegetables in China Using Data Envelopment Analysis: A Green and Sustainable Perspective’, Processes, 7. Lio M. and Liu M. (2006) ‘ICT and Agricultural Productivity : Evidence from Cross-Country Data’, Agricultural Economics, 34: 221–228. Lwoga E. T. (2010) ‘Bridging the Agricultural Knowledge and Information Divide: The Case of Selected Telecenters and Rural Radio in Tanzania’, The Electronic Journal of Information Systems in Developing Countries, 43: 1–14. Ma W., Renwick A., Yuan P. and Ratna N. (2018) ‘Agricultural Cooperative Membership and Technical Efficiency of Apple Farmers in China: An Analysis Accounting for Selectivity Bias’, Food Policy, 81: 122–132. Ma W. and Zheng H. (2022) ‘Heterogeneous Impacts of Information Technology Adoption on Pesticide and Fertiliser Expenditures: Evidence from Wheat Farmers in China*’, Australian Journal of Agricultural and Resource Economics, 66: 72–92. Madau F. A. (2015) ‘Technical and Scale Efficiency in the Italian Citrus Farming: A Comparison between SFA and DEA Approaches’, Agricultural Economics Review, 16: 15–27. Ministry of Agriculture and Rural Affairs of China. (2022). http://www.shouguang.gov.cn/zwgk/NYJ/202206/t20220617_6070878.htm. Murthy D. S., Sudha M., Hegde M. R. and Dakshinamoorthy V. (2009) ‘Technical Efficiency and Its Determinants in Tomato Production in Karnataka , India : Data Envelopment Analysis ( DEA ) Approach’, Agricultural Economics Research, 22: 215–224. Mwalupaso G. E., Wang S., Rahman S., Alavo E. J. and Tian X. (2019) ‘Agricultural Informatization and Technical Efficiency in Maize Production in Zambia’, Sustainability, 11: 2451. NBSC. (2019) China Statistical Yearbook, http://www.stats.gov.cn/tjsj/ndsj/2019/indexch.htm. NBSC. (2017) China’s Third National Agricultural Census, http://www.stats.gov.cn/english/pdf/202010/P020201012324332948860.pdf. D ow nloaded from https://academ ic.oup.com /qopen/advance-article/doi/10.1093/qopen/qoad017/7161147 by Finnish Forest R esearch Institute / Library user on 17 M ay 2023 O R IG IN A L U N E D IT E D M A N U S C R IP T Nguyen Trung Thanh, Tran V. T., Nguyen Thanh Tung and Grote U. (2021) ‘Farming Efficiency, Cropland Rental Market and Income Effect: Evidence from Panel Data for Rural Central Vietnam’, European Review of Agricultural Economics, 48: 207–248. Okello J. J., Kirui O., Njiraini G. W. and Gitonga Z. (2012) ‘Drivers of Use of Information and Communication Technologies by Farm Households: The Case of Smallholder Farmers in Kenya’, Journal of Agricultural Science, 4: 111–124. Orea L. (2002) ‘Parametric Decomposition of a Generalized Malmquist Productivity Index’, Journal of Productivity Analysis, 18: 5–22. Padmajani M. T., Aheeyar M. M. M. and Bandara M. M. M. (2014) Assessment of Pesticide Usage in Up-Country Vegetable Farming in Sri Lanka, Colombo, Sri Lanka, Hector Kobbekaduwa Agrarian Research and Training Institute. Paustian M. and Theuvsen L. (2017) ‘Adoption of Precision Agriculture Technologies by German Crop Farmers’, Precision Agriculture, 18: 701–716. Richards T. J. and Rickard B. (2020) ‘COVID-19 Impact on Fruit and Vegetable Markets’, Canadian Journal of Agricultural Economics, 68: 189–194. Rosenbaum P. R. and Rubin D. B. (1983a) ‘The Central Role of the Propensity Score in Observational Studies for Causal Effects’, Biometrika, 70: 41–55. Rosenbaum P. R. and Rubin D. B. (1983b) ‘Assessing Sensitivity to an Unobserved Binary Covariate in an Observational Study with Binary Outcome’, Journal of the Royal Statistical Society: Series B (Methodological), 45: 212–218. Ruben R., Boselie D. and Lu H. (2007) ‘Vegetables Procurement by Asian Supermarkets: A Transaction Cost Approach’, Supply Chain Management, 12: 60–68. Rubin D. B. (1974) ‘Estimating Causal Effects of Treatment in Randomized and Nonrandomized Studies’, Journal of Educational Psychology, 66: 688–701. Ruddock L. (2006) ‘ICT in the Construction Sector: Computing the Economic Benefits’, International Journal of D ow nloaded from https://academ ic.oup.com /qopen/advance-article/doi/10.1093/qopen/qoad017/7161147 by Finnish Forest R esearch Institute / Library user on 17 M ay 2023 O R IG IN A L U N E D IT E D M A N U S C R IP T Strategic Property Management, 10: 39–50. Schreinemachers P., Simmons E. B. and Wopereis M. C. S. (2017) ‘Tapping the Economic and Nutritional Power of Vegetables’, Global Food Security, 16: 36–45. Shadish W. R., Cook T. D. and Campbell D. T. (2002) Experimental and Quasi-Experimental Designs for Generalized Causal Inference, Boston, MA, Houghton Mifflin. Shrestha R. B., Huang W. C., Gautam S. and Johnson T. G. (2016) ‘Efficiency of Small Scale Vegetable Farms: Policy Implications for the Rural Poverty Reduction in Nepal’, Agricultural Economics (Czech Republic), 62: 181–195. Silva Dias J. (2010) ‘World Importance, Marketing and Trading of Vegetables’. In XXVIII International Horticultural Congress on Science and Horticulture for People (IHC2010): International Symposium on Horticulture for Development, pp. 153–169. Smith A., Goe W. R., Kenney M. and Paul C. J. M. (2004) ‘Computer and Internet Use by Great Plains Farmers’, Journal of Agricultural and Resource Economics, 29: 481–500. Stringer R., Sang N. and Croppenstedt A. (2009) ‘Producers, Processors, and Procurement Decisions: The Case of Vegetable Supply Chains in China’, World Development, 37: 1773–1780. Tripathi A., Tripathi D. K., Chauhan D. K., Kumar N. and Singh G. S. (2016) ‘Paradigms of Climate Change Impacts on Some Major Food Sources of the World: A Review on Current Knowledge and Future Prospects’, Agriculture, Ecosystems and Environment, 216: 356–373. Twumasi M. A., Jiang Y., Zhou X., Addai B., Darfor K. N., Akaba S. and Fosu P. (2021) ‘Increasing Ghanaian Fish Farms ’ Productivity: Does the Use of the Internet Matter ?’, Marine Policy, 125: 104385. Udoh E. (2005) ‘Technical Inefficiency in Vegetable Farms of Humid Region: An Analysis of Dry Season Farming by Urban Women in South South Zone, Nigeria’, Journal of Agriculture & Social Sciences, 1: 79–85. Vrachioli M., Stefanou S. E. and Tzouvelekas V. (2021) ‘Impact Evaluation of Alternative Irrigation Technology in Crete: Correcting for Selectivity Bias’, Environmental and Resource Economics, 79: 551–574. D ow nloaded from https://academ ic.oup.com /qopen/advance-article/doi/10.1093/qopen/qoad017/7161147 by Finnish Forest R esearch Institute / Library user on 17 M ay 2023 O R IG IN A L U N E D IT E D M A N U S C R IP T Wang J., Etienne X. and Ma Y. (2020) ‘Deregulation, Technical Efficiency and Production Risk in Rice Farming: Evidence from Zhejiang Province, China’, China Agricultural Economic Review, 12: 605–622. Watson C. A., Atkinson D., Gosling P., Jackson L. R. and Rayns F. W. (2002) ‘Managing Soil Fertility in Organic Farming Systems’, Soil Use and Management, 18: 239–247. Wellman B., Haase A. Q., Witte J. and Hampton K. (2010) ‘Does the Internet Increase, Decrease, or Supplement Social Capital? Social Networks, Participation, and Community Commitment’, American behavioral scientist, 45: 436–455. Wellman B., Salaff J., Dimitrova D., Garton L., Gulia M. and Haythornthwaite C. (1996) ‘Computer Networks as Social Networks: Collaborative Work, Telework, and Virtual Community’, Annual Review of Sociology, 22: 213–238. Wilson P., Hadley D., Ramsden S. and Kaltsas I. (1998) ‘Measuring and Explaining Technical Efficiency in UK Potato Production’, Journal of Agricultural Economics, 49: 294–305. Wooldridge J. (2015) ‘Control Function Methods in Applied Econometrics’, The Journal of Human Resources, 50: 420–445. Yang M., Zhao X. and Meng T. (2019) ‘What Are the Driving Factors of Pesticide Overuse in Vegetable Production? Evidence from Chinese Farmers’, China Agricultural Economic Review, 11: 672–687. Zanello G. and Maassen P. (2014) ‘Strengthening Citizen Agency and Accountability Through ICT’, Public Management Review, 13: 363–382. Zanello G. and Srinivasan C. S. (2014) ‘Original Article Information Sources , ICTs and Price Information in Rural Agricultural Markets’, European Journal of Development Research, 26: 815–831. Zheng H., Ma W., Wang F. and Li G. (2021) ‘Does Internet Use Improve Technical Efficiency of Banana Production in China? Evidence from a Selectivity-Corrected Analysis’, Food Policy: 102044. Zheng X. and Lu H. (2021) ‘Does ICT Change Household Decision-Making Power of the Left-behind Women? A Case from China’, Technological Forecasting and Social Change, 166: 120604. D ow nloaded from https://academ ic.oup.com /qopen/advance-article/doi/10.1093/qopen/qoad017/7161147 by Finnish Forest R esearch Institute / Library user on 17 M ay 2023 O R IG IN A L U N E D IT E D M A N U S C R IP T Zhu X., Hu R., Zhang C. and Shi G. (2021) ‘Does Internet Use Improve Technical Efficiency ? Evidence from Apple Production in China’, Technological Forecasting & Social Change, 166: 120662. D ow nloaded from https://academ ic.oup.com /qopen/advance-article/doi/10.1093/qopen/qoad017/7161147 by Finnish Forest R esearch Institute / Library user on 17 M ay 2023 O R IG IN A L U N E D IT E D M A N U S C R IP T Figures Figure 1 Map of the study area. Source: Use ArcGIS Map to draw. Shandong Province D ow nloaded from https://academ ic.oup.com /qopen/advance-article/doi/10.1093/qopen/qoad017/7161147 by Finnish Forest R esearch Institute / Library user on 17 M ay 2023 O R IG IN A L U N E D IT E D M A N U S C R IP T Table 1 Variable Selection and Descriptive Statistics. Variable Interpretation Mean All ICT NICT Diff. Treatment Variable: ICT Does farmer use ICT to obtain vegetable production information? 0.52 0.48 (0.50) (0.50) Outcome Variable: Output Vegetable sales revenue in 2018 (10,000 yuan) 13.23 15.07 11.25 3.82*** (-12.71) (-14.55) (-10.04) 4.06 4.25 3.87 0.38* (-2.71) (-2.78) (-2.63) Input Variables: Labor Household labor costs and hired labor costs (10,000 yuan; Appendix A.1) 6.16 6.16 6.15 0.01 (-4.39) (-4.43) (-4.36) 2.30 2.14 2.47 -0.33** (-2.06) (-2.02) (-2.10) Land The farm size of vegetables (mu) 3.97 4.32 3.61 0.71* (5.30) (5.63) (4.91) Fertilizer The costs of fertilizer (10,000 yuan) 1.83 1.86 1.80 0.06 (-1.79) (-1.96) (-1.60) 0.62 0.60 0.65 -0.05 (-0.54) (-0.59) (-0.49) Pesticide The costs of pesticide (10,000 yuan) 0.51 0.57 0.44 0.13*** (-0.64) (-0.72) (-0.54) 0.18 0.19 0.16 0.03* (-0.26) (-0.30) (-0.21) Other The costs of irrigation, seeds, plastic and machinery (10,000 yuan) 1.15 1.27 1.03 0.24** D ow nloaded from https://academ ic.oup.com /qopen/advance-article/doi/10.1093/qopen/qoad017/7161147 by Finnish Forest R esearch Institute / Library user on 17 M ay 2023 O R IG IN A L U N E D IT E D M A N U S C R IP T (-1.65) (-1.90) (-1.33) 0.36 0.38 0.34 0.04 (-0.44) (-0.54) (-0.29) Control Variables: Gender Gender of household head 0.97 0.98 0.95 0.03*** (0.17) (0.12) (0.22) Age Age of household head 51.85 49.17 54.71 -5.54*** (8.68) (8.39) (8.07) Education How many years of education does household head have? 8.24 8.63 7.83 0.80*** (2.63) (2.41) (2.80) Experience How many years has household head been engaged in vegetable production? 20.695 21.02 20.33 0.69 (10.18) (9.39) (10.96) Certificate Does household head have an official professional farmer certificate? 1 =Yes; 0=No 0.06 0.08 0.04 0.04** (0.24) (0.27) (0.20) Ratio Dependency ratio of those typically not in the labor force and those typically in the labor force 0.75 0.87 0.63 0.24*** (0.67) (0.72) (0.59) Market The distance from household to markets (km) 1.97 2.01 1.92 0.09 (3.23) (3.83) (2.44) Government The distance from household to local governments(km) 6.92 8.11 5.65 2.46*** (7.17) (9.35) (3.15) Cooperative Does household participate in a cooperative? 1=Yes; 0=No 0.10 0.14 0.06 0.08*** (0.30) (0.34) (0.23) Acquaintance The number of frequent acquaintances 20.95 22.73 19.04 3.69*** (16.19) (16.94) (15.14) Car Does household own a private car? 1=Yes; 0=No 0.50 0.56 0.45 0.11*** (0.50) (0.50) (0.50) Training Is there ICT training in the village? 0.16 0.20 0.13 0.07*** (0.37) (0.40) (0.33) Information literacy Ability to acquire and utilize information (Appendix A.2) 53.44 55.64 51.09 4.55*** D ow nloaded from https://academ ic.oup.com /qopen/advance-article/doi/10.1093/qopen/qoad017/7161147 by Finnish Forest R esearch Institute / Library user on 17 M ay 2023 O R IG IN A L U N E D IT E D M A N U S C R IP T (6.48) (5.89) (6.27) Social capital Frequency and quality of social contacts (Appendix A.3) 41.47 42.45 40.42 2.03*** (6.77) (6.98) (6.39) IV: cooperative Is there a cooperative in the village? 0.57 0.65 0.47 0.18*** (0.50) (0.48) (0.50) IV: certificate Share of certificate owner in the village 0.06 0.07 0.05 0.02*** (0.06) (0.06) (0.06) Source: Farm household survey (2018). Notes: Values in bold italics show the per mu data of farm inputs and output in vegetable production; Other values measures per farm. Standard deviation in parentheses. *, **, and *** denote mean difference (t-test) between ICT non-users (NICT) and users (ICT) at the 10%, 5%, and 1% significant levels, respectively. Note: D ow nloaded from https://academ ic.oup.com /qopen/advance-article/doi/10.1093/qopen/qoad017/7161147 by Finnish Forest R esearch Institute / Library user on 17 M ay 2023 O R IG IN A L U N E D IT E D M A N U S C R IP T Table 2 Stochastic Production Frontiers Estimates: Unmatched Sample. Variables Conventional SF Selection-corrected SF ICT NICT ICT NICT Constant 0.268*** 0.482*** 0.352** 0.278 (0.104) (0.107) (0.153) (0.170) Labor 0.059 0.071 0.076 0.090 (0.048) (0.053) (0.054) (0.067) Land 0.538*** 0.363*** 0.488*** 0.305*** (0.054) (0.069) (0.071) (0.084) Pesticide -0.067* 0.093** -0.069 0.111** (0.035) (0.037) (0.043) (0.050) Fertilizer 0.280*** 0.228*** 0.296*** 0.255*** (0.047) (0.042) (0.061) (0.056) Others 0.199*** 0.227*** 0.199*** 0.225*** (0.050) (0.049) (0.059) (0.069) Labor*Labor 0.051 -0.086 0.070 -0.072 (0.104) (0.108) (0.116) (0.164) Land*Land -0.331*** -0.011 -0.318*** 0.008 (0.095) (0.112) (0.107) (0.121) Pesticide*Pesticide -0.025* 0.024 -0.025* 0.014 (0.013) (0.047) (0.013) (0.049) Fertilizer*Fertilizer 0.014 0.032 0.008 0.041 (0.078) (0.054) (0.126) (0.077) Others*Others 0.099 0.057 0.099 0.063 (0.066) (0.050) (0.092) (0.084) Labor*Land 0.079 -0.013 0.082 -0.044 (0.073) (0.081) (0.078) (0.095) Labor* Pesticide -0.068 -0.097** -0.056 -0.100* (0.048) (0.048) (0.059) (0.055) Labor* Fertilizer -0.051 -0.044 -0.032 0.000 (0.065) (0.053) (0.076) (0.075) Labor* Others 0.090 0.175*** 0.074 0.171** (0.068) (0.059) (0.084) (0.070) Land* Pesticide -0.017 -0.085* -0.024 -0.067 (0.039) (0.050) (0.038) (0.051) Land* Fertilizer 0.143** 0.049 0.162* 0.034 (0.070) (0.072) (0.087) (0.088) Land* Others 0.037 0.032 0.017 0.011 (0.069) (0.071) (0.089) (0.090) Pesticide* Fertilizer 0.043 -0.004 0.046 -0.013 (0.045) (0.044) (0.058) (0.051) Pesticide * Others -0.042 0.092** -0.044 0.085 (0.046) (0.046) (0.044) (0.058) D ow nloaded from https://academ ic.oup.com /qopen/advance-article/doi/10.1093/qopen/qoad017/7161147 by Finnish Forest R esearch Institute / Library user on 17 M ay 2023 O R IG IN A L U N E D IT E D M A N U S C R IP T Fertilizer *Others -0.123** -0.174*** -0.123 -0.157*** (0.060) (0.043) (0.088) (0.061) Gucheng 0.341*** 0.203 0.350*** 0.278* (0.125) (0.124) (0.133) (0.154) Hualong 0.204* 0.073 0.242* 0.159 (0.119) (0.119) (0.132) (0.120) Luocheng 0.352*** 0.362*** 0.354*** 0.385*** (0.119) (0.124) (0.120) (0.143) Sunjiaji 0.357*** 0.013 0.320** 0.015 (0.116) (0.129) (0.128) (0.143) Tianliu 0.395*** 0.429*** 0.374*** 0.467*** (0.120) (0.125) (0.133) (0.139)  -0.475*** 0.341 (0.179) (0.228) 𝜎 u 0.652*** 0.800*** 0.616*** 0.734*** (0.113) (0.070) (0.126) (0.120) 𝜎 v 0.380*** 0.325*** 0.416*** 0.420*** (0.061) (0.043) (0.061) (0.068) Log-likelihood -316.947 -312.301 -512.460 -517.948 N 394 369 394 369 Source: Farm household survey (2018). Notes: Asterisks denote significance at the 10% (*), 5% (**), and 1% (***) levels. Numbers in the parentheses are the standard errors. The LR statistic for testing H0 (two-regime model) against H1 (one-regime model) is 42.62, so we adopt the two-regime model. D ow nloaded from https://academ ic.oup.com /qopen/advance-article/doi/10.1093/qopen/qoad017/7161147 by Finnish Forest R esearch Institute / Library user on 17 M ay 2023 O R IG IN A L U N E D IT E D M A N U S C R IP T Table 3 Probit Model Marginal Effects: Unmatched and Matched Sample. Unmatched Matched Coefficient Marginal effect Coefficient Marginal effect Gender 0.677** 0.201** 0.673** 0.207** (0.324) (0.095) (0.322) (0.098) Age -0.051*** -0.015*** -0.051*** -0.016*** (0.008) (0.002) (0.008) (0.002) Education -0.005 -0.001 -0.007 -0.002 (0.022) (0.006) (0.022) (0.007) Certificate 1.173 0.348 1.073 0.330 (0.728) (0.215) (0.774) (0.237) Experience 0.028*** 0.008*** 0.028*** 0.009*** (0.006) (0.002) (0.007) (0.002) Ratio 0.109 0.032 0.112 0.034 (0.084) (0.025) (0.083) (0.026) Government 0.051*** 0.015*** 0.049*** 0.015*** (0.014) (0.004) (0.015) (0.004) Market 0.013 0.004 0.012 0.004 (0.022) (0.006) (0.021) (0.007) Training 0.339** 0.101** 0.322** 0.099** (0.157) (0.046) (0.157) (0.048) Cooperative 1.087** 0.323** 0.971** 0.299** (0.465) (0.137) (0.476) (0.145) Car 0.096 0.029 0.101 0.031 (0.112) (0.033) (0.111) (0.034) Acquaintances -0.002 -0.001 -0.001 -0.000 (0.004) (0.001) (0.004) (0.001) Information literacy 0.071*** 0.021*** 0.071*** 0.022*** (0.010) (0.003) (0.010) (0.003) Socialcapital 0.000 0.000 0.001 0.000 (0.008) (0.003) (0.009) (0.003) Gucheng -0.072 -0.021 -0.098 -0.030 (0.237) (0.070) (0.234) (0.072) Hualong -0.072 -0.021 -0.077 -0.024 (0.210) (0.062) (0.208) (0.064) Luocheng -0.189 -0.056 -0.211 -0.065 (0.226) (0.067) (0.226) (0.069) Sunjiaji 0.127 0.038 0.129 0.040 (0.212) (0.063) (0.207) (0.064) Tianliu 0.542** 0.161** 0.466** 0.143** (0.233) (0.069) (0.228) (0.070) Residual_cooperative -0.549** -0.163** -0.462* -0.142* (0.265) (0.078) (0.270) (0.083) Residual_certificate -0.697* -0.207* -0.626 -0.193 (0.388) (0.115) (0.411) (0.126) Constant -3.021*** -3.027*** (0.785) (0.780) Log-likelihood --398.515 -399.593 D ow nloaded from https://academ ic.oup.com /qopen/advance-article/doi/10.1093/qopen/qoad017/7161147 by Finnish Forest R esearch Institute / Library user on 17 M ay 2023 O R IG IN A L U N E D IT E D M A N U S C R IP T N 763 763 739 739 Source: Farm household survey (2018). Notes: Asterisks denote significance at the 10% (*), 5% (**), and 1% (***) levels. Numbers in the parentheses are the standard errors. D ow nloaded from https://academ ic.oup.com /qopen/advance-article/doi/10.1093/qopen/qoad017/7161147 by Finnish Forest R esearch Institute / Library user on 17 M ay 2023 O R IG IN A L U N E D IT E D M A N U S C R IP T Table 4 Stochastic Production Frontiers Estimates: Matched Sample. Variables Conventional SF Selection-corrected SF ICT NICT ICT NICT Constant 0.227** 0.461*** 0.311* 0.336** (0.107) (0.106) (0.164) (0.156) Labor 0.075 0.093* 0.086 0.115* (0.049) (0.053) (0.056) (0.068) Land 7.513*** 0.371*** 0.469*** 0.329*** (0.055) (0.070) (0.072) (0.087) Pesticide -0.060* 0.107*** -0.063 0.124** (0.036) (0.038) (0.045) (0.051) Fertilizer 0.294*** 0.224*** 0.306*** 0.236*** (0.048) (0.042) (0.062) (0.055) Others 0.187*** 0.202*** 0.190*** 0.193*** (0.051) (0.049) (0.061) (0.072) Labor*Labor 0.080 -0.030 0.092 -0.021 (0.106) (0.110) (0.118) (0.162) Land*Land -0.306*** 0.004 -0.301*** 0.031 (0.097) (0.112) (0.107) (0.117) Pesticide*Pesticide -0.021* 0.052 -0.021 0.038 (0.013) (0.048) (0.014) (0.047) Fertilizer*Fertilizer 0.003 0.052 -0.004 0.055 (0.084) (0.054) (0.128) (0.072) Others*Others 0.089 0.093* 0.089 0.097 (0.067) (0.054) (0.096) (0.091) Labor*Land 0.066 -0.048 0.068 -0.079 (0.074) (0.082) (0.080) (0.094) Labor* Pesticide -0.092* -0.095* -0.079 -0.087 (0.049) (0.050) (0.062) (0.055) Labor* Fertilizer -0.023 -0.088 -0.012 -0.065 (0.067) (0.054) (0.077) (0.070) Labor* Others 0.112 0.253*** 0.094 0.257*** (0.069) (0.065) (0.084) (0.073) Land* Pesticide -0.009 -0.077 -0.011 -0.064 (0.039) (0.050) (0.040) (0.054) Land* Fertilizer 0.152** 0.060 0.168* 0.054 (0.072) (0.074) (0.089) (0.088) Land* Others -0.004 0.003 -0.015 -0.031 (0.071) (0.072) (0.091) (0.095) Pesticide* Fertilizer 0.045 -0.006 0.049 -0.014 (0.046) (0.044) (0.060) (0.051) Pesticide * Others -0.030 0.070 -0.035 0.061 (0.047) (0.048) (0.047) (0.062) Fertilizer *Others -0.130** -0.197*** -0.126 -0.178*** D ow nloaded from https://academ ic.oup.com /qopen/advance-article/doi/10.1093/qopen/qoad017/7161147 by Finnish Forest R esearch Institute / Library user on 17 M ay 2023 O R IG IN A L U N E D IT E D M A N U S C R IP T (0.061) (0.044) (0.091) (0.061) Gucheng 0.367*** 0.228* 0.370*** 0.280* (0.127) (0.122) (0.135) (0.153) Hualong 0.260** 0.099 0.290** 0.172 (0.123) (0.117) (0.136) (0.124) Luocheng 0.368*** 0.355*** 0.367*** 0.368*** (0.121) (0.121) (0.122) (0.142) Sunjiaji 0.313*** -0.011 0.291** -0.020 (0.121) (0.126) (0.131) (0.138) Tianliu 0.410*** 0.427*** 0.387*** 0.442*** (0.123) (0.122) (0.135) (0.137)  -0.430** 0.349 (0.190) (0.275) 𝜎 u 0.618*** 0.820*** 0.581*** 0.821*** (0.139) (0.066) (0.143) (0.093) 𝜎 v 0.397*** 0.304*** 0.430*** 0.359*** (0.071) (0.042) (0.064) (0.066) Log-likelihood -303.130 -303.363 -499.624 -506.804 N 377 362 377 362 Source: Farm household survey (2018). Notes: Asterisks denote significance at the 10% (*), 5% (**), and 1% (***) levels. Numbers in the parentheses are the standard errors. D ow nloaded from https://academ ic.oup.com /qopen/advance-article/doi/10.1093/qopen/qoad017/7161147 by Finnish Forest R esearch Institute / Library user on 17 M ay 2023 O R IG IN A L U N E D IT E D M A N U S C R IP T Table 5 Technical Efficiency: Unmatched and Matched Sample. NICT ICT Diff. Unmatched Conventional SF 0.575 0.616 0.042*** (0.178) (0.155) Selection-corrected SF 0.597 0.631 0.033*** (0.150) (0.145) ESR 0.590 0.648 0.058*** (0.182) (0.143) Matched Conventional SF 0.572 0.629 0.058*** (0.182) (0.144) Selection-corrected SF 0.573 0.643 0.070*** (0.174) (0.135) ESR 0.585 0.658 0.073*** (0.188) (0.135) Source: Farm household survey (2018). Notes: Asterisks denote mean differences (t-test) between ICT non-users and ICT users are significant at the 10% (*), 5% (**), and 1% (***) levels. The numbers in the parentheses are the standard deviations. D ow nloaded from https://academ ic.oup.com /qopen/advance-article/doi/10.1093/qopen/qoad017/7161147 by Finnish Forest R esearch Institute / Library user on 17 M ay 2023 O R IG IN A L U N E D IT E D M A N U S C R IP T Table 6 Quantile treatment effects of ICT on TE. Quantile level Coef. Std. Err 10th 0.096*** 0.028 25th 0.115*** 0.026 50th 0.079*** 0.018 75th 0.044*** 0.010 90th 0.001 0.010 Source: Farm household survey (2018). Notes: Asterisks denote significance at the 10% (*), 5% (**), and 1% (***) levels. Standard errors are based on 1000 bootstrap repetitions in both the first-step residualization and the second-step QTE estimation. D ow nloaded from https://academ ic.oup.com /qopen/advance-article/doi/10.1093/qopen/qoad017/7161147 by Finnish Forest R esearch Institute / Library user on 17 M ay 2023