Agro-food quality certification, agricultural organizations, and farm performance: evidence from vegetable farmers in China

ABSTRACT

Agro-food quality certification has been gaining greater importance, because it contributes to both agricultural sustainability and food quality. However, small-scale farmers in developing countries have difficulties in obtaining certification and the economic benefits acquired from certification are unclear. This paper investigates the effects of agro-food quality certification on the farm performance which is indicated by profit, yield and price. The impacts of agricultural organizations, farmer cooperatives and agricultural companies in particular, on the adoption of certification by farmers are explored. We also investigate the effects of organizations on the farm performance of certified farmers. A dataset of 410 vegetable farmers in China is applied, and the propensity score matching (PSM) method is employed in the empirical analysis to address selection bias and reverse causality. Our study finds that, overall, the adoption of certification does not significantly influence farm performance. Cooperative membership positively influences the adoption of certification and the farm performance of certified farmers, while contract farming does not generate additional benefits for certified farmers.

KEYWORDS:

• Agro-food quality certification
• agricultural sustainability
• farm performance
• farmer cooperative
• agricultural company

1. Introduction

The intensive use of chemical fertilizers and pesticides in agriculture causes severe negative impacts on environmental sustainability, which eventually harms human health (Andersson & Isgren, 2021; Chatzimichael et al., 2022; Pretty, 2018). Moreover, as people’s income and awareness of environmental protection and healthy diets are growing, consumer demand for environmentally sound, safe, and high-quality foods is correspondingly increasing (Ding & Veeman, 2019; Narrod et al., 2009; Yin et al., 2019; Zhou et al., 2022). In this context, quality certification for agro-food has gained greater importance and has been promoted worldwide. It not only increases product quality (Handschuch et al., 2013; Jena et al., 2012), but also contributes to environmentally sustainable development. The latter is reflected in various environmentally friendly farming practices, e.g. the adoption of organic fertilizers (Blackman & Naranjo, 2012), water conservation (Haggar et al., 2017), integrated pest management (Meemken, 2021), and the improvement of eco-efficiency (Godoy-Durán et al., 2017), carbon stocks (Haggar et al., 2017), and alleviating forest degradation (Takahashi & Todo, 2017), etc.

Despite the importance of agro-food quality certification, the economic benefits that farmers acquire from certification are unclear. A few studies find that the adoption of certification increases farmers’ benefits because of higher prices or higher yields of certified products (e.g. Barham & Weber, 2012; Brako et al., 2021; Jouzi et al., 2017; Liu et al., 2020; Tran & Goto, 2019). However, poor performance of certification adoption is also reported, especially for organic certified production (e.g. Akoyi & Maertens, 2018; Arhin et al., 2022; Gather & Wollni, 2022; Ibanez & Blackman, 2016; Jena et al., 2017).

It is challenging for farmers to obtain certification because they have limited financial, technical, human, physical capital, and access to a high-end market (Asfaw et al., 2009, 2010; Guo and Zhang, 2013). The return on investment in agro-food quality certification is a long-term process and may be insignificant, which also disincentivizes farmers from adopting certifications (Sanders & Xiao, 2010; Snider et al., 2017). The situation in which farmers have difficulties or lack incentives to adopt agro-food quality certification is especially evident in developing countries, and China is no exception. Small-scale farmers are still dominating in agricultural production in China. There are 230 million famer households in China and approximately 90% of them have less than 10 mu of arable farm land.¹

Both empirical observations and academic studies show that agricultural organizations, i.e. farmer cooperatives and agricultural companies, play an important role in the adoption of certification by farmers. This is reflected in two ways. First, organizations provide support for farmers to adopt agro-food quality certification (Asfaw et al., 2010; Handschuch et al., 2013; Oelofse et al., 2010). Second, agricultural organizations can help farmers to gain economic benefits from certification (Jena et al., 2012; Ssebunya et al., 2019). However, very limited quantitative research has been conducted to systematically investigate the role of organizations in the farm performance of certified farmers.

This study investigates the effect of quality certification on farm performance and the role of agricultural organizations in agro-food quality certification in terms of the adoption of certification and the farm performance of certified farmers. Specifically, we focus on three main types of certifications in China, i.e. the certification of pollution-free agricultural products (unique to China), green food certification (unique to China) and organic product certification. We focus on vegetable industry because vegetable production is featured by the intensive use of agrochemicals such as fertilizers and pesticides and, more importantly, certification adoption is more common in the vegetable industry (Li & Lu, 2020).

The following research questions were addressed. First, does vegetable quality certification affect farm performance, i.e. profit, yield, and price? Second, does farmers’ membership of farmer cooperatives or participation in contract farming increase the adoption of vegetable quality certification? Third, how do cooperative membership and contract farming influence the farm performance of certified farmers?

The remainder of this paper is organized as follows. Section 2 summarizes the literature and the contribution of the current study. Section 3 introduces the methodology of the study, including the data, empirical models, and descriptive analysis of variables. Section 4 reports the results and Section 5 is discussions for the results. We conclude in Section 6.

2. Literature review and marginal contributions

2.1. Adoption of agro-food quality certification by farmers

Many studies examine the factors that influence farmers’ adoption of quality certification. The main influencing factors are farmers’ demographic, psycho-behavioural and psychosocial characteristics (e.g. gender, age, education, year of operation, positive attitude, normative and moral obligations, risk preferences), household features (e.g. household size, labour size, farm size, access to non-farm income), production management features (e.g. supply chain coordination, land circulation) and external environment (e.g. market distance, available services, region) (Handschuch et al., 2013; Jena et al., 2012; Jena & Grote, 2017; Li & Guo, 2019; Li & Lu, 2020; Mohan, 2020; Sapbamrer & Thammachai, 2021; Tran & Goto, 2019; Van Ho et al., 2019).

The results reported by previous studies regarding the effects of these factors on farmers’ certification decisions are mixed. A few studies find that young farmers, farmers with high education levels or those with more experience in production are more likely to adopt certification (e.g. Handschuch et al., 2013; Jena et al., 2012; Sapbamrer & Thammachai, 2021). Van Ho et al. (2019) and Li and Lu (2020), however, find that experience in production negatively influences farmers’ adoption of certification. Jena and Grote (2017) find that farmers’ education levels have a U relationship with their certification decisions. Household size, farm size and distance to the market are found to have positive effects on farmers’ certification adoption (Handschuch et al., 2013; Li & Lu, 2020; Van Ho et al., 2019).

The mechanism of certification adoption is also discussed by some scholars. Cui et al. (2021) emphasize that social services help to address the household resource endowment constraints and extend product outlets, which therefore promote the adoption of quality certification by farmers. Wu et al. (2022) find that agricultural informatization promotes the adoption of certification by family farms through releasing technical constraints and extending product outlets.

The results regarding the effect of agro-food quality certification on the economic performance are mixed. Most studies find that the adoption of quality certification has positive influences on farm output or farmers’ economic welfare (e.g. Bacon, 2005; Barham & Weber, 2012; Brako et al., 2021; Handschuch et al., 2013; Jena & Grote, 2017; Jouzi et al., 2017; Liu et al., 2020; Qiao et al., 2016; Tran & Goto, 2019; Wang et al., 2021). For example, the research by Qiao et al. (2016) and Tran and Goto (2019), based on tea farmers, report that quality certification positively affects product prices and sale income or profits. Barham and Weber (2012) and Brako et al. (2021) find increased yields and income resulted from certification adoption.

Some studies, alternatively, argue that quality certification generates no or little economic benefit for farmers (Akoyi & Maertens, 2018; Gather & Wollni, 2022; Ibanez & Blackman, 2016; Jena et al., 2012; Jena et al., 2017; Ruben & Fort, 2012; Weber, 2011). The production of certified products allows only limited or zero use of agrochemicals to meet environmental sustainability and product high-quality requirements. Therefore, certified products tend to have lower yields and sequentially cause profit loss (Ibanez & Blackman, 2016; Oya et al., 2018; Ruben & Fort, 2012).

The effects of agro-food quality certification on social and environmental outcomes have received much attention as well. These are reflected in farmer education, employment opportunities, community development, sustainable development of agriculture, and environmental sustainability (Blackman & Naranjo, 2012; Haggar et al., 2017; Jouzi et al., 2017; Liu et al., 2020; Meemken, 2021; Qiao et al., 2016; Takahashi & Todo, 2017).

2.2. The role of agricultural organizations in farmers’ agro-food quality certification

There is a rich body of literature about the effects of agricultural organizations, either farmer groups or contracting farming with companies, on farmers’ adoption of quality certification (Asfaw et al., 2010; Handschuch et al., 2013; Liang et al., 2021). Organizations facilitate the adoption of certification by farmers in multiple ways. First, agricultural organizations help farmers save the costs of certification through collective action (Asfaw et al., 2010; Handschuch et al., 2013). Second, organizations provide farmers with technical assistance and relevant training to help them meet certification requirements (Asfaw et al., 2010; Oelofse et al., 2010; Sanders & Xiao, 2010). Third, organizations can provide market information and a guaranteed market at good prices for farmers’ certified products (Oelofse et al., 2010; Sanders & Xiao, 2010; Wollni & Zeller, 2007).

The role of agricultural organizations in influencing farmers’ benefits obtained from certification is mentioned by a few scholars (Jena et al., 2012; Snider et al., 2017; Ssebunya et al., 2019). Jena et al. (2012) find that cooperatives with good production and organizational capacities can help farmers gain the benefits from certification. Ssebunya et al. (2019) report that the duration of farmer group membership, as a mediator, contributes to the effects of certification on farm incomes. However, quantitative studies on the role of organizations in the farm performance of certified farmers are hardly seen.

2.3. Contribution of the current study

This study makes three marginal contributions to the literature. First, it adds to the literature by estimating the effect of quality certification on farm performance in vegetable industry. Most current studies on the welfare effects of agro-food quality certification focus on coffee industry and small-scale farmers in the context of major coffee production countries in Latin America and Africa (e.g. Akoyi & Maertens, 2018; Dietz et al., 2020; Dragusanu et al., 2022; Gather & Wollni, 2022; Ibanez & Blackman, 2016; Jena et al., 2017; Jena & Grote, 2022; Ssebunya et al., 2019; Vellema et al., 2015), while the vegetable industry is paid little attention. Our empirical results show that vegetable quality certification has a negative, but not statistically significant, impact on farm performance. This may be associated with the characteristics of the farmers. Detailed discussions are provided in Section 5.

Second, this research extensively explores the roles of farmer cooperatives and agricultural companies in promoting the adoption of agro-food quality certification and gaining the benefits of certification by farmers. Although many studies report that organizations positively influence farmers’ adoption of certification (e.g. Asfaw et al., 2010; Handschuch et al., 2013; Liang et al., 2021), the effects of organizations on the economic benefits of certified farmers have hardly been analysed. We find that farmers’ participation in cooperatives has a positive impact on the profits and yields of certified vegetable farmers, whereas contract farming with a company does not significantly influence the farm performance of certified farmers.

Third, methodologically, this study distinguishes the effect of certification and that of agricultural organizations on the farm performance by using propensity score matching (PSM) methods. The results of most previous studies for the effects of certification on farmers’ economic benefits may be biased because they are conflated with the organization’s effects on farmers’ economic benefits (Sellare et al., 2020). We address the potential biased results problem by multiple estimations based on PSM.

3. Materials and methods

This section introduces the sampling strategy and data collection, establishes the empirical models, and provides a descriptive analysis of the variables.

3.1. Data

The data used in this study were collected through a survey of vegetable farmers in Hebei and Zhejiang provinces in China from June to August 2015. Stratified random sampling method was used to choose the sample. First, two provinces, Hebei and Zhejiang, were selected. Hebei is one of the largest vegetable production provinces in China. The yield of vegetables in Hebei province in 2014 was 81.26 million tons, accounting for 10.7% of the total vegetable output in China and ranked second among all the provinces.² Comparatively, Zhejiang Province is characterized by a large population, shortage of land, limited natural resources, and an advanced economy. It mainly concentrates on the production of high-value farm products such as fruits and vegetables (Liang & Hendrikse, 2013).

Second, 13 main vegetable production counties in these two provinces were selected based on vegetable production intensity. Third, 20 vegetable cooperatives and 10 vegetable companies were randomly selected from each province within the 13 counties.³ Fourth, three members of a cooperative (or three farmers contracted with a company) and four individual farmers from the village where the cooperative (company) was located were randomly selected, resulting in a total of 420 farmers.

Face-to-face interviews were conducted with sampled farmers based on a structured questionnaire. Information on farm performance (e.g. the profit and yield), vegetable production, and demographic characteristics of farmers were collected. After the exclusion of 10 invalid questionnaires because of the lack of key information or inconsistent answers by the same respondent, a dataset comprising 40 cooperatives, 20 companies, and 410 farmers was established.

3.2. Model

3.2.1. Impact of agro-food quality certification on farm performance

The effect of vegetable quality certification on farm performance is estimated using the following model: (1) $Per{f}_i={\alpha }_0+{\alpha }_{1}Cer{t}_i+{\mathit{\delta }}_1\mathit{C}{\mathit{V}}_{\mathit{i}}+{ϵ}_i$(1) where $Per{f}_i$ represents farm performance of farmer $i$, i.e. profit, yield, and price; $Cer{t}_i$ represents vegetable quality certification status of farmer $i$, indicated by a dummy variable; $\mathit{C}{\mathit{V}}_{\mathit{i}}$ is a vector of covariates associated with farmers’ demographic features, household features, production management characteristics, and the external environment; ${\alpha }_0$, ${\alpha }_1$ and ${\mathit{\delta }}_1$ are parameters to be estimated; ${ϵ}_i$ is the error term.

Estimating the regression model directly may lead to biased results due to endogeneity problems, because farmers’ adoption of certification is a self-selection process (Asfaw et al., 2009). Specifically, farmers’ certification decisions may be influenced by a set of factors that simultaneously affect farm performance. It is possible that farmers with larger farm sizes have a higher tendency to adopt certification and earn higher profits in production. In this situation, the effects of certification on farm performance are confounded by the effects of observed covariates (Blackman & Naranjo, 2012). There is also the issue of reverse causality between farm performance and farmers’ adoption of certification, due to farmers’ limitations of financial and technical capital which are generally required by certification (Asfaw et al., 2009; Guo & Zhang, 2013). Therefore,${\alpha }_1$, which represents the magnitude of the effect of certification on farm performance, would be biased if we conducted a multiple regression analysis.

To address the endogeneity problems, we estimate the model using PSM method, which is commonly used to minimize selection bias and reverse causality (e.g. Brako et al., 2021; Ji et al., 2019; Li & Lu, 2020; Liu et al., 2020; Tran & Goto, 2019; Verhofstadt & Maertens, 2014). Two groups of samples are distinguished by the matching. The treated group refers to farmers with certification, while the control group is farmers who do not adopt certification. PSM matches observations that are from treated group and control group respectively based on propensity scores which summarize the observable characteristics of the units (Rosenbaum & Rubin, 1983). The farm performance between the treated observations and their constructed counterfactuals using the matched control units are then compared. Specifically, it involves two steps. First, the propensity score is calculated by the following equation: (2) $P{S}_i=P{r}_i\left(Cert=1|\mathit{C}V\right)$(2) where $P{S}_i$ denotes the propensity score of farmer $i$ predicted by the probit model, i.e. the probability of farmer’s certification adoption conditional on a series of covariates; $Cert$ represents a farmer’s quality certification adoption (i.e. treatment variable), $Cert=1$ represents that the farmer adopted certification; $\mathit{C}V$ is a vector of observed covariates that influence a farmer’s decision to adopt certification.

Second, we calculate the average treatment effects on the treated group (ATT), that is, the average of the differences in farm performance between the treated units and their constructed counterfactuals. The equation is as follows:   $ATT=E\left(Y_i\left(1\right)-Y_i\left(0\right)|Cert=1\right)=E\left\{E[Y_i\right(1)$  (3) $\left|Cert=1,P{r}_i\left(Cert=1|\mathit{C}V\right)\right]-E\left[Y_i\left(0\right)|Cert=0,P{r}_i\left(Cert=1|\mathit{C}V\right)\right]|Cert=1\}$(3) where $Y_i\left(1\right)$ and $Y_i\left(0\right)$ represent the farm performance of farmers who adopted certification and those who did not.

To obtain robust results, three methods are applied to match the treated and control groups. Nearest-neighbour matching (k = 4) requires that each unit in the treated group is matched to four units in the control group with the closest propensity scores, and the counterfactual outcome is the weighted average of the outcomes of these four units. Kernel matching constructs a counterfactual outcome using the weighted average of the outcomes of all units in the control group. Radius matching specifies a caliper, that is, the maximum distance between the propensity scores of the units in the treated and control groups. Those with absolute values of propensity score differences within the given caliper can be matched, and the counterfactual outcome is the weighted average of the outcomes of the matched control units. The caliper used in our study was set at 0.05.

3.2.2. The role of agricultural organizations in agro-food quality certification

We then examine the role of organizations, i.e. cooperatives and companies, in the adoption of certification by farmers. The following model is established: (4) $Cer{t}_i={\text{β}}_0+{\text{β}}_{1}Or{g}_i+{\mathit{\delta }}_2{\mathit{C}V}_{\mathit{i}}^{\mathrm{\prime }}+{ϵ}_i$(4) where $Or{g}_i$ represents the organization participation of farmer $i$, i.e. individual farmer, cooperative membership or contract farming; ${\mathit{C}V}_{\mathit{i}}^{\mathrm{\prime }}$ is a vector of covariates; ${\text{β}}_0$, ${\text{β}}_1$ and ${\mathit{\delta }}_2$ are parameters to be estimated; ${ϵ}_i$ is the error term.

To further investigate whether cooperative membership or participation in contract farming with companies influences the farm performance of certified farmers, we identify the sub-sample of all certified farmers. This sub-sample is divided into three groups: certified farmers with cooperative membership, certified farmers involved in contract farming, and individual certified farmers. The model for estimating the effect of participation in organizations on the farm performance of certified farmers is formulated as follows: (5) $Per{f}_i^{\mathrm{\prime }}={\gamma }_0+{\gamma }_{1}Or{g}_i+{\mathit{\delta }}_3{\mathit{C}V}_{\mathit{i}}^{\mathrm{\prime }}+{ϵ}_i$(5) where $Per{f}_i^{\mathrm{\prime }}$ is the farm performance of certified farmers; ${\gamma }_0$, ${\gamma }_1$ and ${\mathit{\delta }}_3$ are parameters to be estimated; ${ϵ}_i$ is the error term. PSM method is used again to estimate Equations (4)(4) $Cer{t}_i={\text{β}}_0+{\text{β}}_{1}Or{g}_i+{\mathit{\delta }}_2{\mathit{C}V}_{\mathit{i}}^{\mathrm{\prime }}+{ϵ}_i$(4) and (5)(5) $Per{f}_i^{\mathrm{\prime }}={\gamma }_0+{\gamma }_{1}Or{g}_i+{\mathit{\delta }}_3{\mathit{C}V}_{\mathit{i}}^{\mathrm{\prime }}+{ϵ}_i$(5) , in order to address the endogeneity problems caused by the self-selection problem and the reverse causality problem.

The relationships between quality certification, organizations, and farm performance, which are specified in the above models, are depicted in Figure 1.

Figure 1. Empirical strategies.

3.3. Variables and descriptive statistics

The definitions and descriptive statistics for each variable are provided in Table 1. Farm performance is indicated by profit, yield, and price per mu. The price here is a comprehensive indicator of product quality, market behaviours, and organization involvement. Farmers in the sample grow different vegetable varieties, are heterogeneous in marketing capabilities and other market behaviours, and are involved in alternative organizations. The mean values of the three variables were 5364.36 yuan, 5007.85 kg and 2.43 yuan/kg, respectively.

Table 1. Definitions and descriptive statistics of variables.

Variable	Definition	Mean	Std. Dev	Min	Max
Profit	Profit per mu (yuan)	5364.36	6279.48	−55,850	59,589.5
Yield	Yield per mu (kg)	5007.85	3616.45	175	22500
Price	Average selling price (yuan/kg)	2.43	1.80	0	17.78
Veg. quality certification	1 = The farmer has obtained vegetable certification, 0 = otherwise	0.29	0.45	0	1
Cooperative membership	1 = The farmer is a member of a cooperative, 0= An individual farmer	0.32	0.47	0	1
Contract farming	1 = the farmer is engaged in contract farming with a company, 0 = An individual farmer	0.14	0.35	0	1
Age	Age of farmers (years)	48.99	10.50	25	75
Education	1 = The farmer’s educational level is junior high school and above, 0 = otherwise	0.65	0.48	0	1
Family size	Population of the household	4.36	1.47	2	11
% Veg. revenue	Percentage of vegetable revenue to total household income	0.66	0.32	0	1
Veg. production years	Years of a farmer engaged in vegetable production	15.63	10.73	1	50
Veg. land area	Natural logarithm of cultivated vegetable land area three years ago (mu)	1.26	3.93	−11.51	7.31
% Greenhouse land area	Percentage of greenhouse land area to vegetable land area	0.49	0.48	0	1
% Leafy veg.	Percentage of leafy vegetable production area among that of all the vegetables	0.16	0.31	0	1
% Fruit veg.	Percentage of fruit vegetable production area among that of all the vegetables	0.72	0.38	0	1
Market distance	Distance to the nearest selling market (kilometres)	13.13	86.90	0	1500
Location	1 represents Zhejiang province, 0 represents Hebei province	0.51	0.50	0	1

Agro-food quality certification is indicated by a binary variable, which refers to the adoption of pollution-free agricultural products, green food certification, and/or organic product certification.⁴ It takes the value of 1 if farmers adopt any one or more of the three certifications, and 0 otherwise. We do not distinguish between different types of certifications because the adoption rate of each type of certification is low. Specifically, only about 29% of the surveyed farmers adopted certification, while 71% did not. The distinction between the three types of certification would make the sample size of farmers with each type of certification too small. Among the farmers with certification, nearly 80% adopted pollution-free certification, while about 22% and 7% adopted green food certification and organic product certification, respectively.

Cooperative membership (contract farming) is a binary variable. It takes the value of 1 if a farmer is a member of a cooperative (participates in contract farming with a company), and 0 if a farmer is an individual farmer. Among the surveyed farmers, 118 farmers are cooperative members, and 42 farmers are engaged in contracting farming, accounting for 28.78% and 10.24% of all the sampled farmers respectively. The proportions of certified farmers among cooperative members and contracting farmers are 53% and 43%, respectively, while there are only about 15% of certified farmers among individual farmers.

Various potentially relevant covariates are included in the analyses, such as the farmers’ age, education, and production years. A few points are worth noting. Vegetable land area refers to the cultivated vegetable land area three years ago rather than last year, to avoid the possibility that the cultivated vegetable land area in the current year is endogenous from farmers’ choices of participation in organizations (Li & Guo, 2017). The mean value of the natural log of vegetable land area is approximately 1.26 mu, which reflects the small scale of farms. We also control for the category of vegetables.

4. Results

The results of empirical analyses are reported in this section.

4.1. Impact of agro-food quality certification on farm performance

Before presenting the ATT for the effects of certification on farm performance, the first stage of probit model is estimated. Three matching methods, i.e. nearest neighbour matching (k = 4), kernel matching, and radius matching, are used. The results are basically same for all the three methods; therefore, only the results for the nearest neighbour matching are reported since they are representative for all three methods. The results are displayed in Table 2. Both cooperative membership and contract farming are positively associated with farmers’ adoption of certification.

Table 2. The first stage of PSM estimation for the effects of certification on farm performance.

Variables	Veg. quality certification	S. E.
Cooperative membership	1.063***	0.157
Contract farming	0.897***	0.235
Covariates	Yes
Constant	−0.517	0.680
Observations	410

Notes: *, ** and *** represent significance level 10%, 5% and 1%, respectively.

The quality of the matching is checked using common support and balancing tests. The result of the common support test is displayed in Figure 2, and good-quality matching is confirmed. The distribution of predicted propensity scores for the treated and control groups overlaps, showing a very large common support for propensity scores between the treated and control groups. In particular, nearly all observations in the treated group are matched with those in the control group, except for one observation that is off support.

Figure 2. Common support test for Model (1).

The results of the balancing test are shown in Table 3. Covariates are balanced between farmers who adopt certifications and those who do not after matching. Specifically, the biases of most covariates are sharply reduced, and their absolute values are less than 10% after matching. Meanwhile, the mean differences in almost all covariates between the treated and control groups are statistically insignificant after matching. Overall, we can conclude that the matching process is of high-quality.

Table 3. Balancing test for Model (1).

Variables	Unmatched	% bias	% |bias| reduction	p > |t|
	Matched
Cooperative membership	U	75.2		0.000
	M	1.0	98.7	0.948
Contract farming	U	22.4		0.030
	M	8.0	64.0	0.575
Age	U	−32.7		0.004
	M	1.3	96.0	0.914
Education	U	34.0		0.003
	M	2.8	91.7	0.821
Family size	U	−3.1		0.785
	M	5.6	−82.0	0.636
% Veg. revenue	U	17.9		0.110
	M	−12.2	31.9	0.316
Veg. production years	U	−25.4		0.025
	M	9.3	63.5	0.407
Veg. land area	U	25.5		0.019
	M	4.5	82.3	0.695
% Greenhouse land area	U	11.4		0.298
	M	−23.5	−107.0	0.068
% Leafy veg.	U	−0.4		0.968
	M	2.1	−384.6	0.872
% Fruit veg.	U	5.6		0.608
	M	−7.4	−32.7	0.566
Market distance	U	14.9		0.088
	M	3.1	79.4	0.605
Location	U	−10.4		0.342
	M	21.1	−102.7	0.103

Next, the ATT for the impact of certification on farm performance is calculated, and the results are shown in Table 4. The adoption of certification has a negative, but not statistically significant, impact on farmers’ profits, vegetable yield, and prices.

Table 4. PSM results for the effects of certification on farm performance.

Outcome variables	Certified farmers	Non-certified farmers	ATT (S.E.)
Profit per mu
Before matching	5476.49	5319.59	156.90 (687.53)
Nearest neighbour matching (k = 4)	5515.08	6926.01	−1410.93 (1095.31)
Kernel matching	5515.08	6996.54	−1481.46 (966.91)
Radius matching	5515.08	6995.77	−1480.69 (948.82)
Yield per mu
Before matching	5300.10	4891.16	408.94 (395.47)
Nearest neighbour matching (k = 4)	5298.37	5988.26	−689.89 (575.18)
Kernel matching	5298.37	5869.50	−571.13 (461.22)
Radius matching	5298.37	5857.82	−559.45 (477.91)
Price
Before matching	2.39	2.44	−0.05 (0.20)
Nearest neighbour matching (k = 4)	2.41	2.56	−0.15 (0.33)
Kernel matching	2.41	2.57	−0.16 (0.28)
Radius matching	2.41	2.57	−0.16 (0.27)

Note: Standard errors are in parentheses and obtained by bootstrap command.

4.2. The role of agricultural organizations in the adoption of certification

Considering potential endogeneity problems, we use the PSM method to estimate the effects of farmers’ participation in cooperative or contract farming on their adoption of certification. The results of the first stage of PSM estimation, common support test, and balancing test for Model (2) applying nearest neighbour matching are displayed in Table A1, Figure A1, and Table A2 respectively in the Appendix. The matching is of good quality.

The estimation results of ATT for the effects of cooperative membership and contract farming on farmers’ adoption of certification are reported in Table 5. The results indicate that both cooperative membership and contract farming significantly promote farmers’ adoption of certifications. Specifically, the probabilities of certification increase by approximately 34% for cooperative members and 30% for farmers who have contract farming with agricultural companies, respectively.

Table 5. PSM results for the effects of organizations on farmers’ adoption of certification.

Matching method	Cooperative membership	Individual farmers	ATT (S.E.)
Before matching	0.53	0.15	0.38*** (0.05)
Nearest neighbour matching (k = 4)	0.52	0.20	0.32*** (0.07)
Kernel matching	0.52	0.17	0.35*** (0.06)
Radius matching	0.52	0.16	0.36*** (0.05)
Matching method	Contract farming	Individual farmers	ATT (S.E.)
Before matching	0.43	0.15	0.28*** (0.06)
Nearest neighbour matching (k = 4)	0.44	0.12	0.32*** (0.10)
Kernel matching	0.44	0.18	0.26*** (0.10)
Radius matching	0.45	0.14	0.31*** (0.10)

Notes: *, ** and *** represent significance level 10%, 5% and 1%, respectively; Standard errors are in parentheses and obtained by bootstrap command.

Next, we go further to investigate whether cooperatives and companies play a role in helping farmers obtain benefits from certification and the PSM method is used to estimate the effects. To do that, we use a sub-sample of 117 certified farmers and divide them into three groups: 62 cooperative members, 18 farmers who have contract farming with companies, and 37 individual farmers. The results of the first stage of PSM estimation, common support test, and balancing test for Model (3) applying nearest neighbour matching are displayed in Table A3, Figure A2, and Table A4 respectively in the Appendix. The matching is of good quality.

The estimated ATT for the effects of cooperative membership and contract farming on the farm performance of certified farmers are presented in Table 6. The results show that cooperative membership has a positive impact on the profit and yield per mu of certified farmers, yet does not exhibit a significant influence on the price. Specifically, certified farmers who are members of cooperatives obtain profits of around 3597.85–3883.69 Yuan per mu higher averagely than those of individual certified farmers. This gap for the yield is 2541.71–2685.87 kg per mu on average. However, certified farmers’ participation in contracting farming with companies does not significantly influence the profit, yield, or price.

Table 6. PSM results for the effects of organizations on farm performance of certified farmers.

	ATT (S.E.)
Outcome variables	Cooperative membership	Contract farming
Profit per mu
Before matching	1651.02 (1463.75)	−2304.80** (1026.35)
Nearest neighbour matching (k = 4)	3597.85** (1764.73)	123.47 (1683.21)
Kernel matching	3883.69** (1738.56)	−185.74 (2279.37)
Radius matching	3879.25** (1696.62)	−1019.17 (2233.98)
Yield per mu
Before matching	1019.73 (832.29)	−1502.72* (867.62)
Nearest neighbour matching (k = 4)	2685.87*** (888.34)	1170.12 (1258.75)
Kernel matching	2567.41*** (996.23)	2061.68 (1425.62)
Radius matching	2541.71** (1000.68)	1959.50 (1687.64)
Price
Before matching	−0.36 (0.35)	−0.67* (0.38)
Nearest neighbour matching (k = 4)	−0.54 (0.40)	−0.62 (0.59)
Kernel matching	−0.58 (0.42)	−1.20 (0.81)
Radius matching	−0.50 (0.42)	−1.31 (0.86)

Notes: *, ** and *** represent significance level 10%, 5% and 1%, respectively; Standard errors are in parentheses and obtained by bootstrap command.

4.3. Robustness of results to hidden bias

The key assumption of the PSM method is that the process of farmers’ selection into one treatment depends only on observable characteristics. Nonetheless, there may be unobservable characteristics, e.g. farmers’ management abilities, risk preferences, and awareness of environmental protection, that simultaneously affect farmers’ decisions and outcome variables. In this situation, a hidden bias might arise, which causes the association between a treatment and an outcome variable estimated based on observed factors may not be a causal effect (Aakvik, 2001). To check for the robustness of our estimation results, sensitivity analyses are conducted by calculating Rosenbaum bounds to test the sensitivity of the empirical results to hidden bias (Rosenbaum, 2002).

Two parts of the results need to be checked for the robustness. One is the results about the effects of participation in organizations on farmers’ adoption of certification. Considering that the outcome variable is certification adoption, which is a binary variable, we adopt the command developed by Becker and Caliendo (2007) which applies to binary outcome variables and is suited for nearest neighbour matching without replacement. The other is the results regarding the effects of cooperative membership on farm performance of certified farmers. Given that the outcome variables, i.e. profit and yield, are continuous variables, the commonly used command developed by DiPrete and Gangl (2004) is adopted, since it is mostly applicable to continuous-outcome variables and suited for the nearest neighbour matching with one neighbour.

The estimates are more robust if they are less sensitive to hidden biases. Gamma (≥1) represents the magnitude of the effect of unobservable factors on the farmers’ selection process, i.e. the magnitude of the hidden bias. The value of Γ equals 1 if these factors have no influence on farmers’ decisions; that is, there is no hidden bias. As Γ increases, the magnitude of the hidden bias increases. As Aakvik (2001) indicates, if a large hidden bias does not alter our estimates of treatment effects, we can conclude that our results are less sensitive to hidden bias and are more reliable. Specifically, a larger Γ can be observed when the bounds become insignificant, that is, the estimated treatment effect is no longer significant. The range of Γ is typically set from 1 to 2 (e.g. Aakvik, 2001; Liu et al., 2020).

Table 7 displays the sensitivity analysis results for the PSM estimates of the effect of organizations on the certification adoption and that of cooperative membership on the profits and yields of certified farmers. The estimated effect of cooperative membership on the adoption of certification is very robust because the estimate of ATT is still significant at 1% significance level when Γ = 2 for both bounds under the assumption of over- or under- estimation of the treatment effect. The estimated effect of contract farming on the certification adoption is less encouraging. The estimate of ATT is just not significant at 10% significance level when Γ = 1.6 for the bound under the assumption of overestimation of the treatment effect. Therefore, the effect of contract farming on the adoption of certification should be explained cautiously.

Table 7. Sensitivity analysis results for Model (2) and Model (3).

	Certification adoption				Profit per mu	Yield per mu
	Cooperative membership		Contract farming		Cooperative membership	Cooperative membership
Γ	p_mh+	p_mh-	p_mh+	p_mh-	sig+	sig+
1	0.000	0.000	0.017	0.017	0.000	0.000
1.1	0.000	0.000	0.026	0.010	0.001	0.000
1.2	0.000	0.000	0.040	0.006	0.003	0.000
1.3	0.000	0.000	0.056	0.004	0.006	0.001
1.4	0.000	0.000	0.075	0.002	0.011	0.001
1.5	0.000	0.000	0.098	0.001	0.018	0.002
1.6	0.000	0.000	0.122	0.001	0.028	0.004
1.7	0.000	0.000	0.149	0.001	0.042	0.006
1.8	0.000	0.000	0.178	0.000	0.059	0.009
1.9	0.001	0.000	0.208	0.000	0.079	0.013
2	0.001	0.000	0.239	0.000	0.102	0.019

Notes: Gamma refers to log odds of differential assignment due to unobserved factors; sig+ refers to upper bound significance level; p_mh+ refers to significance level (assumption: overestimation of treatment effect), p_mh– refers to significance level (underestimation of treatment effect).

Since the effects of cooperative membership on farm performance of certified farmers are positive, only upper bounds are displayed. The estimated effect of cooperative membership on the profits of certified farmers is robust because the estimate of ATT is still significant at the 10% significance level when Γ = 1.9. Likewise, the estimated effect of cooperative membership on yields is also very robust because the estimate of ATT is still significant at the 5% significance level when Γ = 2. Overall, our results can stand to a large extent.

5. Discussion

The results regarding the effects of certification adoption on farm performance are consistent with some previous studies conducted in an international scenario (e.g. Gather & Wollni, 2022; Ibanez & Blackman, 2016; Ruben & Fort, 2012). These studies also report that certification for coffee does not bring farmers extra gains in production.

The main reason for the insignificant effect of certification on the yields may be that the gap in the production process and management practices between certified and non-certified farmers is small. The data show that among certified farmers, most farmers (almost 80%) adopt pollution-free certification, while only about 22% and 7% of farmers adopt green food certification and organic product certification, respectively. Since pollution-free certification has fewer limits on chemical use during production than green food or organic products do, the difference in yields between certified and non-certified farmers is small.

Theoretically, certified products are expected to obtain high selling prices because of the high quality, which is an important incentive for farmers to adopt the certification. Some studies report that certification positively affects the profits due to the price premium (e.g. Qiao et al., 2016; Tran & Goto, 2019). However, our results are inconsistent with these findings since no significant price premium is found for certified products. One reason for this may be that most farmers in China are smallholders and price-takers with weak bargaining power (Huang & Liang, 2018). They face challenges in obtaining premium prices for certified products. In addition, the increase in consumers’ willingness to pay for certified agro-food is still limited in China because of consumers’ insufficient knowledge and distrust of food quality standards (Scott et al., 2014).

Our results about the effects of cooperative membership on farm performance of certified farmers are consistent with most previous studies reporting that cooperative membership has a significantly positive impact on farmers’ profits and product yields (e.g. Abdul-Rahaman & Abdulai, 2018; Liang et al., 2021; Ma et al., 2022; Ma & Abdulai, 2016). Farmer cooperatives generally provide assistance with techniques for farmers to boost their yields (Barham et al., 2011). The results that contract farming does not significantly influence farm performance of certified farmers indirectly indicate that companies provide less technical training or support for farmers than cooperatives.

The effects of cooperative membership or contract farming on farm prices obtained by farmers vary in the literature. Some studies find a positive effect (e.g. Dubbert, 2019; Liang & Wang, 2020; Serra & Davidson, 2021; Mishra, Kumar et al., 2022), while others report no significant or negative impacts (e.g. Chagwiza et al., 2016; Mishra, Mayorga et al., 2022). Our results show that certified farmers with cooperative membership or involved in contract farming do not gain any extra price premium, compared with individual certified farmers. A possible reason is that most farmers in China are price takers, and the price premium, if there is any, is reaped by cooperatives or companies.

6. Conclusions

Agro-food quality certification has essential significance in agricultural sustainability and food quality. However, the adoption of certification faces challenges because small-scale farmers have difficulties in obtaining certification and the economic benefits acquired from certification are unclear. This study investigates the effects of agro-food quality certification on farm performance and further explores the role of agricultural organizations in farmers’ certification adoption and the farm performance of certified farmers. A dataset comprising 410 vegetable farmers from China is used and PSM estimation method is applied to address selection bias and reverse causality issues.

A few important findings are highlighted. First, the adoption of vegetable quality certification by vegetable farmers does not lead to significant improvements in farm performance, i.e. unit profit, unit yield, or price.

Second, agricultural organizations play an important role in farmers’ adoption of quality certifications. Cooperative membership and contract farming with companies increase farmers’ adoption of certification by approximately 34% and 30%, respectively, compared with individual farmers. However, the effect of contract farming on vegetable certification needs to be explained with caution because the estimated results are sensitive to unobservable factors and there might be overestimation of the treatment effect.

Third, cooperative membership positively influences the profit and yield per mu of certified farmers, while contract farming with agricultural companies does not have a significant impact on farm performance. Specifically, cooperative membership increases the profit of certified farmers by around 3597.85–3883.69 yuan per mu and the yield by 2541.71–2685.87 kg per mu. However, no significant effect on the price is found.

The results provide implications for farmers of all the product categories beyond vegetables and for farmers in other regions of the world as well. That is because the challenges of certification are quite similar for all the agro-food, i.e. farmers’ limitations of financial, technical, human, and/or physical capital, and the unclear economic benefits acquired from the certification. These findings can provide direction for policymakers.

First, efforts need to be made to increase the price premium of certified products by enhancing the public’s understanding of and trust in the agro-food quality certification. The importance of environmental friendly features and quality safety of products should be fully conveyed to consumers through multiple channels, e.g. by the government and the medium. Second, it is necessary for the government to provide support for the development of agricultural organizations to give full play to their role in the adoption of agro-food quality certification. For example, financial support for farmer cooperatives targeting agro-food quality certification can benefit the farmers involved and facilitate the overall development of quality certification.

There are a few limitations of our study. First, the heterogeneity of the effects across different types of certification on farm performance is not analysed due to data limits. Second, the outlets of products, e.g. the destinations of products and the types of clients, can also be important factors influencing the performance and product certification of farmers. However, they are not included in the model due to data limits. Third, the mechanisms how farmer cooperatives influence the farm performance of certified farmers can be explored in future research.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

This work was supported by the National Planning Office for Philosophy and Social Science [grant number 21BGL179] and the National Natural Science Foundation of China [grant number 72161147001].

Notes

1 Data source: http://www.gov.cn/xinwen/2019-03/01/content_5369755.htm; 1 mu = 1/15 hectare.

2 Data source: China Rural Statistical Yearbook (2015), China Statistics Press.

3 The number of selected vegetable cooperatives is the double of that of vegetable companies, according to the relatively size of populations of the two types of organizations.

4 As of September 2019, the number of certificates issued by these three certifications were 68,835, 40,077 and 73,934, respectively. Data source: China Academy for Rural Development-Qiyan China Agribusiness Database (CCAD).

Appendix

Figure A1. Common support test: nearest neighbour matching (k = 4).

Figure A2. Common support test: nearest neighbour matching (k = 4).

Table A1. The first stage of PSM estimation for the effects of cooperative membership and contract farming on farmers’ adoption of certification.

Variables	Cooperative membership	Contract farming
Age	−0.003	0.015
	(0.009)	(0.013)
Education	0.410**	0.395*
	(0.171)	(0.240)
Family size	−0.026	−0.167**
	(0.050)	(0.076)
% Veg. revenue	0.079	0.638*
	(0.238)	(0.345)
Veg. production years	−0.000	−0.021*
	(0.008)	(0.011)
Veg. land area	0.052**	0.035
	(0.022)	(0.027)
% Greenhouse land area	0.428	−1.385***
	(0.261)	(0.364)
% Leafy veg.	−0.396	−0.282
	(0.344)	(0.428)
% Fruit veg.	−0.341	−0.491
	(0.291)	(0.376)
Market distance	0.002	−0.007
	(0.002)	(0.006)
Location	0.283	−0.569*
	(0.255)	(0.327)
Constant	−0.686	−0.284
	(0.663)	(0.951)
Observations	368	292

Notes: Table A1 reports the results of the probit regression; standard errors in parentheses; *, **, and *** represent significance at the 10%, 5%, and 1% levels, respectively.

Table A2. Balancing test for the effects of participation in specific organizations on farmers’ adoption of certification: nearest neighbour matching (k = 4).

	Unmatched	% bias		% |bias| reduction		p > |t|
Variables	Matched	Cooperative membership	Contract faming	Cooperative membership	Contract farming	Cooperative membership	Contract farming
Age	U	−19.5	29.1			0.093	0.111
	M	−7.4	−23.8	61.9	18.5	0.560	0.235
Education	U	33.3	8.8			0.004	0.600
	M	1.4	8.7	95.8	0.4	0.910	0.696
Family size	U	−6.1	−45.5			0.599	0.016
	M	−0.4	1.4	92.6	97.0	0.971	0.944
% Veg. revenue	U	17.0	1.1			0.133	0.945
	M	1.8	−3.6	89.5	−220.0	0.890	0.870
Veg. production years	U	−3.5	−15.1			0.759	0.397
	M	−1.6	−28.1	53.0	−85.5	0.896	0.213
Veg. land area	U	28.6	28.9			0.012	0.059
	M	0.1	3.6	99.5	87.5	0.991	0.854
% Greenhouse land area	U	11.4	−70.9			0.309	0.000
	M	1.2	3.6	89.8	94.9	0.929	0.850
% Leafy veg.	U	−3.0	9.6			0.789	0.555
	M	−2.2	−6.2	25.2	34.9	0.869	0.794
% Fruit veg.	U	−3.4	−29.0			0.757	0.078
	M	1.3	3.3	62.0	88.8	0.924	0.884
Market distance	U	19.1	−14.8			0.035	0.480
	M	−3.2	5.9	83.4	59.8	0.318	0.492
Location	U	2.4	31.4			0.830	0.064
	M	−1.7	−4.9	27.5	84.2	0.896	0.820

Table A3. The first stage of PSM estimation for the effects of participation in specific organizations on farm performance indicators of certified farmers.

Variables	Cooperative membership	Contract farming
Age	−0.005	0.032
	(0.016)	(0.036)
Education	1.075***	1.423*
	(0.356)	(0.745)
Family size	0.011	−0.529*
	(0.106)	(0.309)
% Veg. revenue	−0.250	0.571
	(0.530)	(1.103)
Veg. production years	0.030*	0.104*
	(0.018)	(0.061)
Veg. land area	−0.012	−0.045
	(0.044)	(0.074)
% Greenhouse land area	0.876	−3.874***
	(0.790)	(1.381)
% Leafy veg.	0.245	2.822*
	(0.713)	(1.544)
% Fruit veg.	0.392	4.329***
	(0.655)	(1.611)
Market distance	0.001	−0.001
	(0.003)	(0.024)
Location	1.813**	−1.233
	(0.810)	(1.086)
Constant	−2.110	−4.182
	(1.445)	(2.685)
Observations	99	55

Notes: Table A3 reports the results of the probit regression; standard errors in parentheses; *, **, and *** represent significance levels of 10%, 5%, and 1%, respectively.

Table A4. Balancing test for the effects of participation in specific organizations on farm performance indicators of certified farmers: nearest neighbour matching (k = 4).

	Unmatched	% bias		% |bias| reduction		p > |t|
Variables	Matched	Cooperative membership	Contract faming	Cooperative membership	Contract farming	Cooperative membership	Contract farming
Age	U	5.6	47.4			0.790	0.116
	M	−4.5	18.8	19.3	60.4	0.830	0.603
Education	U	49.9	33.9			0.014	0.255
	M	−1.1	0.0	97.9	100.0	0.951	1.000
Family size	U	−7.4	−45.4			0.723	0.146
	M	−7.9	11.9	−6.2	73.9	0.681	0.732
% Veg. revenue	U	−6.7	−2.3			0.750	0.935
	M	−5.8	13.5	13.8	−490.5	0.756	0.722
Veg. production years	U	29.9	32.4			0.167	0.262
	M	24.0	−8.3	19.7	74.5	0.230	0.852
Veg. land area	U	30.3	34.5			0.142	0.209
	M	−0.1	−22.4	99.7	35.2	0.996	0.549
% Greenhouse land area	U	−48.6	−154.2			0.024	0.000
	M	0.1	−8.8	99.8	94.3	0.997	0.834
% Leafy veg.	U	5.1	15.1			0.806	0.609
	M	5.0	−25.6	2.5	−69.9	0.793	0.603
% Fruit veg.	U	−26.3	−2.2			0.214	0.942
	M	−15.6	18.1	40.6	−725.8	0.434	0.694
Market distance	U	21.9	−17.6			0.347	0.597
	M	1.5	6.1	93.1	65.2	0.675	0.399
Location	U	65.0	91.9			0.003	0.002
	M	−6.9	0.0	89.4	100.0	0.739	1.000