Impact of Farmer Field School on Crop Income, Agroecology, and Farmer’s Behavior in Farming: A Case Study on Cumilla District in Bangladesh

Abstract

The Farmer Field School is a season-long training for farmers involving participatory activities and interactive learning with the doctrine of integrated pest management and agroecosystem analysis. It has become a popular education and extension approach worldwide. This study tried to evaluate the FFS as a vehicle for sustainable agriculture which has economic viability, ecological soundness, and social acceptability. The study aimed to analyze the impact of the FFS on crop income, agroecology, and farmers’ behavior in farming. The empirical models, such as propensity score matching, Mahalanobis distance matching, and difference in differences, were applied for estimating the impact of FFS on crop income, more specifically, real income from brinjal. The environmental impact quotient was used to assess the agroecological impact of using pesticide, and a graded response model was used to investigate farmer behavioral changes in farming. The treatment effect based on the empirical models has shown a positive, significant effect on crop income. The findings also revealed that FFS farmers had a lower agroecological impact from pesticide use, and their behavior in farming practices was improved. Therefore, FFS was demonstrated to be a key strategy in strengthening agricultural extension services, which will contribute to promoting sustainable agriculture.

1. Introduction

Non-formal agricultural education has played a vital role in the development of a sustainable agriculture sector in many developing countries where the emergence of farmer field school (FFS) was the influence for this method. The FFS approach, as pioneered by the Food and Agriculture Organization (FAO), is a way to introduce farmers to discovery-based learning for addressing the pest management issues. Rola et al. [1] identified FFS as a season-long training of farmers involving participatory activities, hands-on analysis, and decision making. It has been developed as a ‘bottom-up’ extension approach based on experiential and reflective learning to strengthen farmers’ problem-solving capabilities by highly qualified facilitators working with farming communities, as judged by Larsen and Lilleor in 2014 [2]. According to the FFS guidance document [3], FFSs were mostly constructed for smallholder farmers who are resource-poor and often have limited access to education, information, extension services, and market access. The approach mandated to fill the gap in local knowledge, conduct holistic research on the agroecosystem, and increase awareness and understanding of phenomena that are not obvious or easily observable. The approach is very much practical, and farmers are taught directly in the field and the physical place is normally close to the field under a tree or a small shelter. FFS learning was designed based on the theory of adult learning, theory of constructivism, and theory of experiential learning (Figure A1). Integrated pest management (IPM) and agroecosystem analysis (AESA) are the core activities of FFS, other activities are designed to support it [3] and IPM is treated as an economic threshold concept of FFS.

FAO has imputed a lot of effort in the incubation, development, and spread of FFS from the outset. FFS began in Southeast Asia and quickly spread to other parts of Asia in the early 1990s, Africa in the mid-1990s, and then the rest of the world [3]. The approach was schemed to respond to the lack of awareness among the Asian farmers relating to agricultural ecology, especially the relationship between insect pests and their natural enemies [4]. The first IPM-FFSs were started in 1989 in Indonesia to reduce farmer reliance on pesticides in rice [5]. It was originally implemented to address the challenge of ecological heterogeneity and integrated pest management that would allow farmers to reduce pesticide use, improve crop management, and secure better profit margins [6]. The agro-ecological content and experiential methodology of the FFS approach have influenced extension programs in various countries [4]. Now, the program has been actively forwarded by the FAO in the context of measures designed to enhance food security, farmer income, climate change adaptation, and agricultural sustainability [7]. The FAO continues to support FFS in the different regions, through expertise, networking, and funding [8].

Bangladesh is one of the most densely populated, smallholder farms, and intensive agricultural countries, with approximately 87% of rural inhabitants’ income derived from agricultural activities [9]. Agriculture is the largest sector of employment in Bangladesh. 40.6% of the total working population is involved in the agriculture sector, contributing 14.23% of the national gross domestic product [10,11]. Even so, the farming community in Bangladesh has increasingly been threatened by population pressure on the use of arable land and natural resources; to produce more to meet the increasing demand of an ever-growing population with low per capita cultivable land of 0.05 hectare [11]. Hence, farmers use more chemical inputs, such as fertilizers to produce more and pesticides to safeguard crops against harmful insects, pests, and diseases. Only 4% of farmers are formally trained in pesticide use and over 47% of farmers overuse pesticides in Bangladesh [12]. The irrational use of pesticides pollutes the ecosystem through contaminating soil, groundwater, and surface water [13]. That is to say, the overuse of chemical inputs has a negative impact on soil, health, and the environment and leads to decreased agricultural income due to raised costs. To tackle this challenge, it is required to design an effective program that goes beyond the dissemination of the concept of IPM and AESA among farmers, helps them to get organized, and empowers the farming community in problem-solving. The FAO-induced FFS could be able to satisfy these needs which require investment to aware and educate farmers for good agricultural practices.

The agricultural extension system in Bangladesh has a long history of evolution which has taken different shapes over time [14]. The traditional training and visit extension approach which is “top-down” in nature was failing most of the developing countries to address the issue of overusing pesticides [14], and Bangladesh is not an exception. Afterwards, there was a need for a more participatory “bottom-up” extension approach considering the ecological aspect with a principle of integrated pest management, and the emergence of FFS met the situation demand policy [14,15]. In Bangladesh, the first Farmer Field Schools were organized in the early 1990s, assisted by the FAO intercountry program for rice [16]. It has now been conducted in different agricultural crops; FFS on vegetables is one of them. Most of these vegetable FFSs focused on brinjal (eggplant) [17]. Brinjal is the second most important vegetable in Bangladesh in terms of both production area and yield [18], and it is a primary source of cash income for farmers. Moreover, insect and pest attack is one of the most significant hurdles to large-scale brinjal cultivation in Bangladesh [19]. Raza et al. [19] claimed that brinjal is attacked by 17 species of insects and six types of different diseases in Bangladesh, and farmers sprayed insecticides more than 40 times in a single cropping season. Thus, this study tried to evaluate the brinjal-FFS in the context of ensuring sustainability in agriculture.

The Plant Protection Wing of the Directorate of Agriculture Extension (DAE) is responsible for the implementation of FFS activities in Bangladesh. Bangladesh’s government is committed to supporting the education and betterment of its farmers through a field-level educational program aimed at the empowerment of farmers and local communities [20]. Investing in farmer education is treated as a necessary complement to research and extension services and as a strategy in accumulating wealth from agriculture [21]. Moreover, returns on investments in FFS cannot be appraised until the pesticide-use externalities have been considered and properly quantified [22].

Some previous studies focused on the economic aspect of the FFS program found that the FFS participants have significantly more knowledge about IPM practices; they have the potential to improve production and productivity [23,24]. Another study by Larsen and Lilleor [2] claimed that the FFS mechanism triggers to agricultural production may have led to strong positive effects on food security, but no effect on poverty. Some studies highlighted the environmental aspect and identified that intensification of agricultural production has raised concerns about environmental facts [25] and without sustainable management of variable agroecosystems—considering the major consequences in terms of declining soil quality, soil erosion, pollution of surface and groundwater, and loss of biodiversity—no agricultural development program would be successful [21]. Studies on the social aspect asserted that FFS can led to an entry point to establishing a link between farmer education, empowerment, and a pathway toward increased well-being [26]. A recent study argued that the FFS remains relevant at the field level, helping farmers to adapt their farming practices and livelihood situation to changing circumstances, contribute to the role in rural development [8]. When the FFS program was scaled up in Bangladesh, a lot of focus was placed on evaluating the impact of the new horizon of extension policy and how farmers’ behavior changed over time in relation to IPM practice.

The reviewed papers indicate the substantial impact of FFS in terms of farm productivity and production, not to differentiate trained individuals with untrained in terms of agricultural income. Some papers showed that increased knowledge of IPM led to sustainable management of agri-environment but did not find and dissociate the effect of overusing pesticides on agroecology, nor did they study the changes of farmer behavior in farming towards the pathway of social well-being. Therefore, there remains a need to balance the economy, ecology, and improvement of behavior in farming practice. Above all, sustainable agriculture requires the consolidation of economic viability, ecological soundness, and the farmer’s behavioral improvement. Therefore, the impact analysis of all the domains of sustainability aims to justify the success of FFS program. Moreover, no study has been done to determine the impact of brinjal-FFS on the domains of sustainable agriculture in the selected area of Bangladesh.

Therefore, this study aims to do so as a case study on Cumilla district examining whether the FFS program, more specifically, brinjal-FFS has played a significant role in generating more crop income, maintaining agroecology, and changing farmers’ behavior in farming responding to the sustainable agriculture, and therefore, tried to find the answer to the following research questions:

Does the FFS program make a difference to crop income? How are other aspects of FFS beneficial towards sustainable agriculture?

To answer these identified research questions, this study set the general objective to find the impact of FFS on crop income, agroecology, and farmer behavior in farming. The specific objectives are to analyze the difference in crop income, to assess the difference of agroecological effect of using pesticides, and to identify the behavioral changes of farmers in farming by employing a causal inference technique with a thorough robustness check.

2. Materials and Methods

2.1. Study Site

This study took place in 3 upazilas (sub-districts) of Cumilla district in Bangladesh as a case study (Figure 1). Cumilla is situated between the capital city Dhaka and the port city Chattagram and is considered as one of the hot spots for vegetable production and sales. Cumilla has a wide coverage of FFS, especially since agriculture is the main occupation of most households (49.15%) [11]. This study examined the FFS on brinjal because it is the most vulnerable crop to pests and produces fruit almost year-round. Moreover, FFS on brinjal was conducted in the selected three sub-districts of Cumilla district during the treatment year 2019. The survey took place from October to November 2020.

2.2. Sampling

Sub-districts, namely, Burichong, Chandina, and Daudkandi from Cumilla district, were purposively selected. The target population were the farmers who were cultivating brinjal in the selected sub-districts. A list of farmers who participated in FFS on brinjal in the treatment year was collected from the selected sub-districts. As there are some criteria for participating in the FFS program, treatment is nonrandomly assigned before participation. We strategically collected the list of common vegetable farmers (as there was no separate list of brinjal farmers) and from them brinjal farmers were identified by consulting the sub-assistant agriculture officer who usually works at the village level, from the villages where FFS on brinjal was done. Then, we constructed a sampling frame including FFS participants (75) and non-FFS brinjal farmers (636) from the sub-districts. Finally, 150 respondents were selected from the sampling frame for the interview where FFS participants were treated as treatment group (48) and non-FFS farmers were treated as the control group (86); 16 respondents were excluded as they only responded partially.

This study tried to gather data from a significant number of respondents with nearly the same farming characteristics, more specifically, brinjal farmers. Yet due to COVID-19 pandemic, this goal was not achieved. Although there have some other alternatives, it was not feasible due to lack of infrastructure, as well as a detailed questionnaire being very time-consuming.

2.3. Data Collection

As a participatory method, the data were collected via face-to-face interview by questionnaires, with semi-structured questions containing the FFS program as a treatment variable where agricultural income, field use environmental impact quotient (FEIQ) value, and farmers’ behavior in farming as the outcome variables with different covariates, through an intensive survey. It is worth mentioning that the data before intervention were gathered on a recall basis. The survey was approved by the research ethics committee of IDEC, Hiroshima University.

2.4. Data Analysis

This study largely used case study leading quantitative analysis. Hollweck [27] proclaimed that a case study strategy was adopted when the researcher has no control over the conditions, contextual factors, and outcomes of an intervention. This study also attempted to outline the qualitative traits of farmer behavior in farming.

Determination of net crop income:

To determine net crop income from brinjal, the total value of brinjal production (Y) is multiplied by the average selling price (P) from which the total production cost ($\mathrm{P}\mathrm{i}\mathrm{X}\mathrm{i}$) is deducted. The total production cost is the summation of input quantity (x_i) multiplied by the unit cost (p_i).

$$\mathsf{\pi }=\mathrm{P} \times \mathrm{Y}-{\displaystyle \sum }\mathrm{PiXi}$$

Calculation of real value:

For consistent comparisons of input costs and income before and after the treatment, the nominal value was divided by the January–December 2018 Consumer Price Index (CPI) of Bangladesh [28] to obtain the real value:

$$\mathrm{Real} \mathrm{value}=\frac{\mathrm{Nominal} \mathrm{Value}}{\mathrm{CPI}}\times {100}_{ }$$

Benefit–Cost Ratio (BCR):

This study calculates the benefit–cost ratio that determine the financial feasibility of the FFS program in future. To obtain the benefit–cost ratio, the following formula is used:

$$\mathrm{BCR} (\mathrm{Discounted})=\frac{\mathrm{Benifits} \left(\mathrm{Value} \mathrm{of} \mathrm{TPP}\right)}{\mathrm{Costs} \left(\mathrm{Total} \mathrm{Production} \mathrm{Cost}\right)}$$

Two-sample t-test:

To compare the mean value and test the statistical significance of the socio-demographic characteristics, brinjal production-related information, and EIQ data, two-sample t-tests (two-tailed) were used. This test is essential for satisfying the matching principles, as well.

Variable Selection:

This study selected the covariates that are time-invariant in nature. Gender and education are absolutely time-invariant; age, farming experience are also treated as time-invariant due to their proportional change nature and the time variation of marital status, household size, farm hours, and other training is negligible. These covariates were selected according to the previous study of [24,29] and field reality. The selected covariates are also expected to affect the outcome variable of crop income and field use EIQ. The definition of the selected variables is shown in the

Table 1. Variable Descriptions.

Variables	Definition
Outcome Variables:
Crop Income Field Use EIQ Farmer Behavior	Real value of Crop Income from Brinjal (BDT) Value of field use Environmental Impact Quotient of pesticides Changes of farmer’s behavioral skills in farming
Treatment Variable:
FFS Participation	1 for participation in brinjal FFS, 0 for nonparticipation
Covariates:
Age	Age of farmers (years)
Gender	0 for female, 1 for male
Marital status	0 for single, 1 for married
Household size	Number of family members
Education	Years of formal education
Farming experience	Length of years of farming
Farm hours	Average time spent in farm per day (hours)
Other training	1 for having training, 0 for no training

.

2.5. Empirical Models

2.5.1. Propensity Score Matching (PSM)

The matching approach is applied to predict causal treatment effects in a multifaceted field of study. The central role of the propensity score in observational studies is related to causal effects [30]. Ideally, the impact of an intervention would be measured with the participants by observing the outcome with and without the intervention [24]. However, we did not have both with and without FFS intervention farming data from the same farmers. Besides, to evaluate treatment effects, only comparing the mean of two separate groups is not advisable as there remains a difference in the outcome between the participant and non-participant groups even in the absence of treatment. Propensity score matching is a possible solution to the problem. Austin [31] proclaimed that an observational or nonrandomized study can be designed and analyzed using the propensity score to imitate some of the characteristics of a randomized controlled trial. Moreover, matching is likely to reduce the underlying selection bias with two assumptions—conditional independence assumption and common support assumption [32]. This study fulfills the necessary conditions to assess the impact of FFS program in terms of selecting covariates and ensure their sufficient balancing. Rosenbaum and Rubin [30] established the propensity score as the probability of treatment assignment based on observed baseline covariates. The propensity score p(x) of participation can be represented as follows:

Propensity Score (PS), p(x) = P (p = 1lx)

(1)

p is the treatment variable and x represent a set of time-invariant covariates (Table 1). In this study, the logit model is used to estimate the propensity score with balancing tests of covariates. This study performed the nearest-neighbor matching, Kernel matching, and Radius caliper (0.05) matching algorithms for checking the conformity of the results. Imbens [33] asserted that propensity score matching allows the researcher to estimate treatment effects for an individual, and the average treatment effect on treated (ATET). Even then, after matching the propensity score between treated and control groups is satisfactory with a set of covariates, we can compare the treatment effect on the outcome variable, using the following equations adopted from the study of Sanglestsawai et al. [34]:

ATET^PSM = [(y₁|p = 1, p(x)) − (y₀|p = 1, p(x))]

(2)

where p = participation in FFS (p = 1 if participated in FFS, and p = 0 if did not participate in FFS); y₁ = net income of the participant, y₀ = net income of the non-participant.

PSM may provide inconsistent estimates due to misspecification bias; the Inverse probability weighted regression adjustment (IPWRA) estimates are used to resolve such issues [29]. Holmes [35] recommends applying weighted regression when working with a small sample size. Moreover, IPWRA has a double robust feature, which has a coherent impact on outcome. Even so, PSM may yield biased results due to some unobservable factors; Mahalanobis distance matching (MDM) is the effective measure to reduce the probable biasness in PSM as the distance normalized by the variance–covariance matrix of the covariates [36]. King and Neilsen [37] critique the non-robust estimates of PSM paradox which could be addressed by employing a potentially more robust approach such as MDM, which matches directly the covariate space, while PSM does not. The mathematic expression of Mahalanobis’ distance is adopted from the study of King et al. [38]:

$$\mathrm{M}({\mathrm{X}}_{\mathrm{i}}, {\mathrm{X}}_{\mathrm{j}})=\sqrt{{\left(\mathrm{X}\mathrm{i} - \mathrm{X}\mathrm{j} \right)}^{\prime }{S}^{-1} \left(\mathrm{X}\mathrm{i} - \mathrm{X}\mathrm{j} \right)}$$

(3)

Therefore, this study endeavored to justify the robustness of the treatment effect and reduce the bias from unobservable factors and small sample size by applying IPWRA and Mahalanobis distance matching.

2.5.2. Difference in Differences (DID)

The Difference in Differences (DID) measure the impact of a program or treatment by considering the interaction of the treatment and time variables. Larsen and Lilleor [2] argued that there might be potential bias in the cross-sectional comparison for determining impact (ATET) through PSM due to unbalanced observable; the DID estimation does not suffer such kind of biasness. This method accounts for unobservable time and group characteristics that can confound the treatment’s effect on the outcome [39]. The concept of DID is adopted from Cunningham [40], as in the following equation:

$${\delta }_{DID}=(\mathsf{\Delta }\mathrm{y}|\mathrm{FFS} - \mathsf{\Delta }\mathrm{y}\left|\mathrm{non}-\mathrm{FFS}\right)=[y_1 |\mathrm{FFS} - y_0\left|\mathrm{FFS}\right] - [y_1 |\mathrm{non}-\mathrm{FFS} - y_0|\mathrm{non}-\mathrm{FFS}]$$

(4)

where $y_1$ = crop income after treatment, $y_0$ = crop income before treatment.

The current study intends to evaluate the impact of FFS program as a treatment on the outcome of crop income from brinjal. This study construct two groups from an indifferent sampling frame indexed by treatment status where non-FFS indicates the individuals who do not receive treatment, i.e., the control group, and FFS indicates individuals who do receive treatment, i.e., the treatment group. We collected data from individuals in two periods, where 0 indicates a period before the treatment group receives treatment, i.e., before FFS (2018), and 1 indicates a period after the treatment group receives treatment, i.e., after FFS (2020). This study used two points of data instead of multiple points to reduce the serial correlation problem highlighted by Fredriksson and Oliveira [41], having considered the parallel trends assumption and no spillover effects. In effect, FFS has no relation to agricultural income during the before-treatment period.

The DID delivers a non-experimental model for estimating the ATET by comparing the difference in outcome means between the control and treatment groups across time. Fredriksson and Oliveira [41] argued that matching of covariates may be a way to achieve a robust DID result. Therefore, this study adopted the potential treatment outcome regression model with matching of covariates to estimate the causal effect of the FFS program on crop income from Lang and Donald [42].

$${\mathrm{y}}_{\mathrm{i}}={\mathsf{\gamma }}_{\mathrm{g}}+{\mathsf{\gamma }}_{\mathrm{t}}+{\displaystyle \sum }\beta i.\mathrm{Z} gt +\mathsf{\delta }{\mathrm{D}}_{\mathrm{gt}}+{\mathsf{\epsilon }}_{\mathrm{it}}$$

(5)

Here, y = crop income, g = group, t = time, Z = covariates, D = treatment occurs at the group and time levels, and ε = error term.

In fact, this study tried to employ DID as a more robust technique, compared to corresponding methods for estimating economic causal inference of FFS.

2.5.3. Environmental Impact Quotient (EIQ)

The Integrated Pest Management (IPM) program of Cornell University developed a model, responding to the environmental impact of using pesticides, called the environmental impact quotient (EIQ) of pesticides. This model reduces the environmental impact information to a single value [43]. In Asia, the pesticide risk indicator model EIQ has been used in the assessment of environmental impact of Farmer Field Schools [5]. By using the field use EIQ value, this study tried to incorporate agroecological effects along with the efficacy of using pesticides through the intervention of the FFS program. The following equation based online EIQ calculator was used to find the field use EIQ value of pesticides [43]:

Field use EIQ = EIQ × % of active ingredient × use rate = (Farmworker EI + Consumer EI + Ecological EI)/3

(6)

The EIQ Value for Active Ingredient was found in the New York State IPM EIQ Database, the FEIQ value for pesticide is calculated by considering the average effect of consumer, farmworker, and ecological components [44]. To compare the FEIQ value and its components between two groups, a t-test was performed.

2.5.4. Graded Response Model (GRM)

The graded response model (GRM) developed by Samejima [45] is applied in the analysis of data collected from a Likert-type attitude scale. The model runs in a two-step process; the first step estimates the probability of a certain skill that an individual has chosen from the given scale by the following equation, as adopted from Aune et al. [46]:

$${\mathrm{P}*}_{\mathrm{m}} (\theta )=\frac{1}{1+e^{-1.7a\left(\theta -\mathrm{bm}\right) }}$$

(7)

Here, a is the discrimination parameter of skills (slope); b_m is a set of threshold parameter; e is the natural log;$\theta$ is the latent trait, in this study—behavioral skill; P*_m ($\theta )$ is the probability of responding in a scale; P*_(m+1) ($\theta )$ is the probability of the next responding scale.

The second step estimates the probability of the subtraction that an individual responds in each scale and the next one is defined as the following equation:

$${\mathrm{P}}_{\mathrm{m}} (\theta )={\mathrm{P}*}_{\mathrm{m}} (\theta ) -{\mathrm{P}*}_{(\mathrm{m}+1)} (\theta )$$

(8)

2.6. Data Analysis Programs

Data were tabulated using Microsoft Excel Worksheet. The STATA¹⁷ program was applied for descriptive statistics, obtaining the result of ATET, and behavioral change. The online EIQ calculator was used to calculate the overall field use EIQ value, as well as the EIQ values of its components.

3. Results and Discussion

3.1. Socio-Demographic Information of FFS and Non-FFS Farmers

Table 2. Summary Statistics of socio-demographic characteristics.

Variables	FFS (n = 48)			Non-FFS (n = 86)			t-Stat.	p-Value
	Mean (SD)	Min.	Max.	Mean (SD)	Min.	Max.
Gender	0.75 (0.44)	0	1	0.89 (0.32)	0	1	2.24	0.026 **
Marital Status	0.96 (0.20)	0	1	0.92 (0.30)	0	1	−0.877	0.382
Age	45.00 (12.85)	22	70	44.73 (12.85)	19	70	−0.114	0.909
Household size	5.79 (1.87)	3	12	6.31 (2.10)	4	12	1.433	0.154
Education	7.35 (3.72)	0	15	5.74 (3.94)	0	15	−2.314	0.022 **
Farming Experience	19.31 (12.35)	3	50	18.76 (12.53)	2	50	−0.248	0.805
Daily Farm hours	6.19 (1.92)	3	10	6.40 (2.51)	2	12	0.498	0.619
Other Training	0.375 (0.46)	0	1	0.291 (0.49)	0	1	−0.998	0.319

SD = Standard Deviation; Min. = Minimum; Max. = Maximum; ** = Significant at 5% level.

presents and compares the socio-demographic characteristics, including age, gender, marital status, education, household size, farming experience, average daily farm hours, and other training. The result shows that out of 134 farmers, most of them are male and married in both groups. The number of female respondents of FFS is greater than of non-FFS farmers, but the overall number of female respondents is lower due to the local social customs. The selected characteristics showed no significant difference among the brinjal farmers except for gender and education, due to some requirement of FFS participation, although they usually do not strictly resemble field reality. The FFS farmers are more educated and experienced in farming than the non-FFS farmers. Moreover, non-FFS farmers have invested more time in farming activities, but they have less training than FFS farmers. Therefore, the groups would be statistically fit for comparison in propensity score matching after conducting the balance property test (

Table A1. Balancing property of FFS vs. non-FFS farmers.

	Mean		Bias Reduction (%)	p-Value
	Treated	Control
Before matching:
Gender	0.75	0.89		0.026 **
Marital Status	0.96	0.919		0.382
Age	45.00	44.73		0.910
HH Size	5.79	6.31		0.154
Education	7.35	5.74		0.022 **
Farming Experience	19.31	18.76		0.805
Farm Hours	6.19	6.39		0.619
Other Training	0.375	0.291		0.320
After matching:
Gender	0.75	0.81	57.0	0.464
Marital Status	0.96	0.917	−4.9	0.404
Age	45.00	44.67	−24.6	0.898
HH Size	5.79	5.81	96.0	0.955
Education	7.35	7.79	72.8	0.550
Farming Experience	19.31	19.52	62.6	0.937
Farm Hours	6.19	5.67	−150.6	0.248
Other Training	0.375	0.417	50.6	0.680

** = significant at 5% level.

) and propensity score matching quality test (

Table A2. Propensity score matching quality test.

Items	Before Matching	After Matching
Pseudo R2	0.079	0.031
p-value	0.085	0.842
Mean Standardize Bias (%)	19.6	10.3

).

3.2. Brinjal Production Information at before FFS

Total production cost consists of the land preparation cost, seedling cost, fertilizer cost, labor cost, irrigation cost, pesticide cost is treated as variable costs, and fixed costs include land use cost and interest on cash capital. The real value was used for better comparison.

Before FFS, fertilizer, irrigation, pesticide, and fixed cost showed that both groups used nearly the same input costs except for land preparation, seedling, and labor cost, which showed significant difference (

Table 3. Summary Statistics of Brinjal Production information in real value (before FFS).

Variables (Unit: BDT)	FFS (n = 48)	Non-FFS (n = 86)	Diff.	S.E.	t-Stat.	p-Value
	Mean (SD)	Mean (SD)
Land Preparation Cost	4664.56 (212.35)	4793.20 (477.87)	−128.64	72.76	−1.77	0.079 *
Fertilizer Cost	11,967.52 (444.45)	11,934.52 (531.95)	33.00	90.54	0.36	0.716
Seedling Cost	5301.62 (418.19)	5019.47 (562.98)	282.15	92.99	3.03	0.003 ***
Labor Cost	21,785.61 (430.59)	21,560.18 (796.85)	225.43	124.16	1.82	0.072 *
Irrigation Cost	2677.28 (172.00)	2628.87 (270.22)	48.41	43.22	1.12	0.265
Pesticide Cost	10,336.00 (352.14)	10,355.26 (253.95)	−19.26	52.74	−0.36	0.715
Fixed Cost	6515.33 (315.19)	6439.42 (484.98)	75.90	77.88	0.97	0.331
Total Production Cost	63,247.92 (1654.06)	62,730.94 (3002.72)	516.98	469.14	1.10	0.272
TPP (Kg)	12,220.83 (810.82)	12,098.84 (828.39)	121.99	148.13	0.82	0.412
Value of TPP	105,501.2 (3042.02)	104,644.7 (3803.76)	856.47	639.84	1.39	0.183
Crop Income	42,253.25 (1899.61)	41,906.82 (1958.35)	346.43	349.34	0.99	0.323
BCR	1.668 (0.027)	1.669 (0.041)	−0.001	0.0067	−0.19	0.848

*** = Significant at 1% level, * = Significant at 10% level ($1 = Bangladeshi Taka (BDT) 85).

). However, there was no significant difference in total production cost, total physical production, and crop income of FFS and non-FFS farmers. The results indicate that the farming practice of using inputs and gaining return between the groups was mostly the same before the treatment. The BCR of FFS and non-FFS farmers are 1.668 and 1.669 times, respectively, and no significant difference was revealed.

3.3. Brinjal Production Information at after FFS

Before FFS, the groups were indifferent to using inputs for brinjal production, especially for pesticides and fertilizer, which were the main concern of this study in addition to the agricultural income. After FFS, the result showed that the FFS participants use a significantly lower cost of inputs than that of non-FFS farmers. The focal point is that the reduction of using fertilizer and pesticides is a principle of IPM practice which ultimately reduces the total cost and eventually translates to increasing income. Interestingly, the average total production of FFS farmers is lower but the average real crop income of them is significantly higher than that of non-FFS farmers (

Table 4. Summary Statistics of Brinjal Production information in real value (After FFS).

Variables (Unit: BDT)	FFS (n = 48) Mean (SD)	Non-FFS (n = 86) Mean (SD)	Diff.	S.E.	t-Stat.	p-Value
Land Preparation Cost	4797.76 (228.15)	5205.11 (380.46)	−407.34	60.23	−6.763	0.00 ***
Fertilizer Cost	8086.45 (652.57)	9541.62 (474.89)	−1455.16	98.16	−14.82	0.00 ***
Seedling Cost	5227.98 (365.58)	5811.88 (278.60)	−583.89	56.28	−10.37	0.00 ***
Labor Cost	15,311.64 (530.52)	21,532.94 (303.81)	−6221.29	71.99	−86.42	0.00 ***
Irrigation Cost	2687.21 (151.79)	2976.13 (291.34)	−288.91	45.17	−6.39	0.00 ***
Pesticide Cost	5261.90 (380.99)	8386.72 (487.08)	−3124.82	98.18	−31.82	0.00 ***
Fixed Cost	6254.72 (335.27)	7115.92 (252.83)	−861.20	73.01	−11.79	0.00 ***
Total Production Cost	47,627.68 (1625.23)	60,570.31 (1962.33)	−12,942.63	333.19	−38.84	0.00 ***
TPP (Kg)	9563.54 (380.61)	10,094.19 (520.48)	−530.64	85.66	−6.19	0.00 ***
Value of TPP	94,257.58 (2964.90)	103,049.70 (3544.43)	−8792.16	603.49	−14.57	0.00 ***
Crop Income	46,629.90 (2268.59)	42,479.43 (2501.63)	4150.47	436.23	9.51	0.00 ***
BCR	1.98 (0.055)	1.70 (0.041)	0.28	0.008	33.08	0.00 ***

*** = Significant at 1% level ($1 = Bangladeshi Taka (BDT) 85).

). A study in Cambodia demonstrated that the FFS approach allows for the efficient use of farm inputs and is expected to be successful in improving rice production sustainability [47]. Two other studies from the Philippines showed that onion FFS reduced pesticide usage, reduced pesticide expense, and increased income while maintaining the same yield [34,48]. BCR of FFS and non-FFS farmers are 1.98 and 1.70 times, respectively, and there is a significant difference in BCR between the two groups after FFS. The increased BCR of FFS participants indicate the economic viability of the FFS program. Moreover, the crop income of FFS farmers increased by 10.36% compared to before FFS and the crop income of non-FFS farmers also increased by 1.37%; however, this may have been caused by unobservable factors.

3.4. Results for Economic Domain

Effects of FFS Program on Farmers’ Income from Brinjal in Matching Estimation

To estimate the impact of the FFS program on the crop income from brinjal production, propensity score matching was used after the balance property test of the sociodemographic characteristics and propensity score matching quality test between the groups. The average treatment effect on treated (ATET) using nearest-neighbor matching, kernel matching, and radius caliper matching have shown a positive and significant effect on income from brinjal farming by BDT 4885.46, BDT 4399.05, and BDT 4266.45, respectively (

Table 5. Average Treatment Effect on Treated (ATET) between FFS and non-FFS farmers.

Crop Income (BDT/Acre)	Nearest Neighbor (1) Matching		Kernel Matching		Radius Caliper (0.05) Matching
	Coef.	t-Stat.	Coef.	t-Stat.	Coef.	t-Stat.
ATET FFS Training (1 vs. 0)	4885.46 (563.03)	8.68 ***	4399.05 (484.92)	9.07 ***	4266.45 (477.78)	8.93 ***

*** = Significant at 1% level.

). Whereas, the MDM model showed a positive significant effect on brinjal farming income by BDT 4191.32 (

Table 6. MDM Estimation for ATET on Net Income.

Crop Income (BDT/Acre)	Coef.	Std. Err.	t-Stat.
ATET FFS Training (1 vs. 0)	4191.32	526.85	7.96 ***

*** = Significant at 1% level.

), which verified the robustness of PSM results. The IPWRA findings also verify that the FFS program significantly increases the brinjal farming income by BDT 4300.44 (

Table 7. IPWRA Estimates for Net Income.

Crop Income (BDT/Acre)	Coef.	Std. Err.	z	P >|z|
ATE FFS Training (1 vs. 0)	4300.44	446.25	9.64	0.000 ***

*** = Significant at 1% level.

), corroborating the findings of Moahid et al. [29]. Increased income was also reported in the study from the Philippines by using a PSM model, with which they found that IPM-FFS farmers profited more than non-FFS farmers [34]. Despite reduced pesticide use, a study on IPM in rice in Thailand found no significant impact of the FFS on gross margin [49]. As the MDM and IPWRA aim to reduce the bias of PSM results, this study highlights the results of MDM and IPWRA.

A sufficient overlap between the treatment and control group is expected in the process of matching. The propensity score graph (Figure 2) showed that there is some overlap in the range of propensity scores between treatment and control groups. Moreover, the distribution of propensity scores between treated and control groups have some dissimilarities (Table 1). The balancing property test (Table A1) was conducted to match the distribution of covariates among control and treatment groups. Additionally, the smaller sample size also reduced the chance of overlapping. Nonetheless, after the matching of propensity score between control and treated observations, the graph shows homogenous distributions of propensity score among the brinjal farmers.

3.5. Effects of FFS Program on Farmers’ Income from Brinjal in Difference in Differences Estimations

To test the conformity of matching results, this study applied the DID approach where time and treatment effect were envisaged with covariates. The average treatment effect on treated (ATET) in DID estimation has shown a significant positive change in crop income from brinjal by BDT 3809.91 (

Table 8. Difference in Differences (DID) effect on net income.

Crop Income (BDT)	Coef.	Robust Std. Err.	t	p > |t|
ATET FFS training (1 vs. 0)	3809.91	543.54	7.01	0.000 ***

*** = Significant at 1% level.

). A study on cotton FFS found that FFS caused significant increases in household income for cotton farmers in China, India, and Pakistan by lowering the cost of insecticide [50], which supports the result of this study. Another study conducted by Feder et al. in 2003 [51] used a DID model and found that the impact of the FFS program did not provide evidence of increased yield or lower pesticide usage for either trained or control group farmers in Indonesia. The ATET in DID estimation is lower than that of matching results, which implies that there may have selection bias in matching methods. Thus, DID have imparted a more robust outcome by comparing the results of matching techniques, removing the constraints in the matching models. Therefore, the ATET in DID estimation was more prominent in this study as an economic impact of FFS.

Results for Agroecological Domain

Table 9. Pesticide utilization information.

Pesticides Name	No. of Farmers		Average Use Rate/Acre
	FFS	Non-FFS	FFS	Non-FFS
Cypermethrin/Rolethrin	45 (93%)	61 (71%)	1.8 Kg	2.5 Kg
Cartap	12 (25%)	-	0.8 L	-
Ridomil	30 (62%)	42 (49%)	5.0 Kg	7.0 Kg
Indofil	1 (2%)	8 (10%)	2.0 Kg	3.0 Kg
Voliam Flex	03 (6%)	27 (31%)	160 mL	200 mL
Tundra	03 (6%)	-	3.0 L	-
Success/Tracer	03 (6%)	07 (8%)	250 mL	250 mL
Thiovit	-	10 (12%)	-	4.0 kg

Source: Field Survey, 2020.

presents the frequency and the average use of pesticides among the farmers. It was found that non-FFS farmers used pesticides more frequently than FFS farmers. Similar results were obtained in the study by Sharma et al. [52]. According to the WHO hazard class, Ridomil is slightly hazardous and other pesticides are moderately hazardous [52]. The field reality is that the FFS participants are more concerned by using hazardous pesticides. Based on the pesticide use data, this study calculates and compares the FEIQ value in

Table 10. Mean Value of Field use EIQ and its component values.

Value of EIQ	FFS	Non-FFS	|Diff.|	t-Stat	p-Value
Field use EIQ	213.26 (44.35)	255.28 (59.95)	42.02	4.24 ***	0.000
Consumers	65.89 (24.89)	76.71 (24.89)	10.82	1.65 *	0.099
Farmworkers	92.93 (25.11)	110.65 (36.07)	17.72	3.02 ***	0.003
Ecological	480.89 (91.39)	565.55 (137.60)	84.66	3.81 ***	0.002

SD in parentheses; *** = Significant at 1% level; * = Significant at 10% level.

. It was also found that both groups use different traps (Figure 3), but FFS farmers have more practice in using traps instead of using pesticides.

3.6. Effect of FFS Program on Agroecology in FEIQ Estimation of Pesticides

As the mean scores of field use EIQ of non-FFS farmers were higher than that of FFS farmers are, it is implied that non-FFS farmers have had a greater environmental impact than FFS farmers. Even then, we can observe that the average EIQ values for consumers, farmworkers, and ecological components among the non-FFS farmers were 76.71, 110.65, and 565.55 respectively, while among FFS farmers, the components were valued at 65.89, 92.93, and 480.89 (Table 10), respectively. Therefore, it is obvious that non-FFS farmers have had a significant impact on agroecology by applying pesticides in brinjal production. Mwungu et al. [53] found the environmental component of the EIQ was high among both the IPM and non-IPM farmers, but there was a significant difference in the EIQ field use between the two categories of farmers. Ahamad et al. [54] found that improvement in the consumer, ecology, and farmer environment quotients is the plausible outcome of the FFS training.

3.7. Effects of FFS Program on the Value of Field use EIQ in Matching Estimation

Table 11. Average Treatment Effect on Treated (ATET) of FEIQ value for using pesticides.

FEIQ Value	Nearest Neighbor Matching		Kernel Matching		Radius Matching
	Coef.	t-Stat.	Coef.	t-Stat.	Coef.	t-Stat.
ATET FFS Training (1 vs. 0)	−54.95 (13.28)	−4.14 ***	−59.94 (10.70)	−5.60 ***	−60.36 (10.51)	−5.74 ***

*** = Significant at 1% level.

presents the average treatment effect on treated (ATET) applying nearest-neighbor matching, kernel matching, and radius caliper matching of the value of field use EIQ of using pesticides. These matching models reveal that the FFS training significantly reduced the FEIQ value. The IPWRA findings, used for checking the robustness, marked the FFS program’s significant decrease of the FEIQ value by 55.17 (

Table 12. IPWRA Estimates for FEIQ value.

FEIQ Value	Coef.	Std. Err.	z	p > |z|
ATE FFS Training (1 vs. 0)	−55.17	9.27	−5.95	0.000 ***

*** = Significant at 1% level.

), which conform the consistency of the result found in the above three matching scores. The FEIQ results indicate the agroecological impact of FFS which is inversed to agricultural income. As we could not locate any previous study checking the effect of using pesticide on agroecology through PSM, this study attempts to do so to provide novel findings.

Results for the Behavioral Domain

This study tried to explore the changes in farmers’ farming practice, one of the crucial parts of social behavior, in terms of different behavioral skills. Most of the respondents were agreed with the quick decision, leadership, farmer-to-farmer extension, IPM knowledge, and community network skills, but most of them were neutral towards agroecosystem knowledge (

Table 13. Degree of response to different behavioral skills of FFS farmers.

Skills	Strongly Agree (%)	Agree (%)	Neither Agrees nor Disagree (%)	Disagree (%)	Strongly Disagree (%)	Total (%)
Quick Decision	27%	48%	17%	8%	0%	100%
Leadership	21%	48%	19%	10%	3%	100%
F-to-F extension	19%	43%	25%	13%	0%	100%
IPM Knowledge	19%	54%	27%	0%	0%	100%
Agroecosystem	4%	31%	33%	29%	3%	100%
Community Network	17%	46%	21%	10%	6%	100%

Source: Field Survey, 2020.

). The response rate implies that the FFS participants have improved their farming skills. Muhammad et al. [55] identified that the different aspects of social well-being (decision making, confident building, leadership quality, resource management) of the farming community had improved in the project area because of the FFS program.

The development of skills through different methods of learning led to a positive change in behavior according to the logic behind the response in the Likert scale. Most of the respondents were rational in assessing themselves as they became expert in managing agroecology by changing their traditional practice in farming. These behavioral changes satisfied the concepts of learning theories of the FFS program in the sense that FFS participants were mostly skilled enough to construct their own understanding together. Talibo [56], as cited by Hansen and Duveskog, claims that in line with best practice, inputs, complexity, and yields, FFS enables the farmer to contrast the new practices to their own. This process is grounded in constructivism learning theory. This study indicated that the FFS farmers were able to build up their personal (quick decision, leadership ability), technical (IPM knowledge, agroecosystem), and social (F-to-F extension, community network) constructivism (Table A2). Moreover, they improved their knowledge on IPM and agroecosystem through experiential learning, i.e., learning by doing. The FFS farmers gained more knowledge in pest and nutrient management and actively exercised interpersonal networks to share knowledge among themselves, but very little with other farmers [1], joint decision making, and group capacity building based on learning outcomes [6]. The best change in the behavioral aspect is that the knowledge can be translated into practice in farming. Field realities also support findings on the behavioral aspects.

3.8. Effect of FFS Program on Farmer’s Behavior in Farming under Graded Response Model

The item thresholds parameter ranges from −6.20 (b₁ item community network) to 2.28 (b₄ item IPM knowledge) and the item discrimination parameter ranges from 0.65 to 3.78 (

Table 14. Graded Response Model (GRM) parameter estimates for the behavioral change.

Items (Skills)	a (s.e.)	B1 (s.e.)	b2 (s.e.)	b3 (s.e.)	B4 (s.e.)	p-Value
Quick Decision	0.69 (0.37)	-	−3.70 (1.93)	−1.58 (0.87)	1.74 (0.94)	0.061 *
IPM Knowledge	0.69 (0.36)	-	-	−1.64 (0.87)	2.28 (1.18)	0.056 *
F-to-F Extension	0.86 (0.40)	-	−2.59 (1.11)	−0.73 (0.46)	1.94 (0.88)	0.030 **
Community Network	0.65 (0.35)	−6.20 (3.43)	−3.22 (1.67)	−1.16 (0.73)	1.59 (0.97)	0.061 *
Leadership	1.06 (0.41)	−4.06 (1.66)	−2.26 (0.81)	−1.04 (0.44)	1.48 (0.58)	0.010 **
Agroecosystem	3.78 (1.81)	−2.11 (0.61)	−0.59 (0.23)	−0.34 (0.21)	1.98 (0.46)	0.247

** = Significant at 5% level; * = Significant at 10% level.

). The item with the lowest discriminative ability was quick decision and that with the highest discriminative ability was agroecosystem. Analysis of the behavioral items with the GRM provides evidence that the skills provide a greater level (b₄) of information in behavioral change where quick decision, leadership, community network, F-to-F extension, and IPM knowledge have a significant impact. Therefore, it may presage that the FFS program has improved the farming behavior of FFS farmers in brinjal cultivation, consequently introducing them to sustainable agriculture.

4. Conclusions and Policy Implication

4.1. Summary of Results and Conclusion

FFS acts as a vehicle to improve the knowledge level of farmers in the doctrine of IPM and AESA. This study showed that FFS can empower farmers so that they can decide on resource allocation properly, and the more efficiently use of inputs resulted in lowering overall production costs, which leads to higher income. The findings of this study have shown that there is a reduction of brinjal production after FFS, which differed from the results of the study of Cai et al. [57], where they found that FFS has a positive impact on the yield of tomato in Beijing. Higher BCR of FFS-trained farmers implies the future feasibility of the intervention.

The most significant findings of this study are that the PSM, MDM, and DID model has shown a positive change of the crop income from the selected vegetable and significantly reduced the FEIQ value, which has led to the robustness of impact analysis. Behavioral change, another domain of sustainability, also showed a positive indication as per GRM parameter. Therefore, the effects of FFS on improvements in farmer’s social, agroecological, and economic livelihood are desirably sustainable in nature [58]. Furthermore, apart from the objective of assessing the impact of FFS on crop income, agroecology, and farmer’s behavior in farming, more academic interest and debate may have emerged to motivate future study.

The summary suggests that the FFS program could serve as a key strategy to widen the agricultural extension services by targeting sustainable agriculture where all the domains would be positively impacted. Therefore, the marginal contribution of this study is to improve impact analysis of FFS by addressing the gap of previous studies and all the pillars of sustainability.

Thus, it can be concluded that the FFS program will make a positive change in agricultural practice and has the potential to change farmers’ livelihoods by increasing their income. It can also contribute to promoting sustainable agriculture, as indicated by SDG-2 [59], by orienting farmers to use fewer chemical inputs that positively impact on crop income, agroecology, and farmers’ behavior in farming.

4.2. Policy Implications

Agricultural extension and farmer education initiatives are important policy tools for policymakers aiming to boost agricultural income while also protecting the environment [60]. This study has provided the evidence that FFS-trained farmers practice reducing the use of chemical inputs and improving crop income, which is justification for policymakers to devote more resources to this program and possibly expand it to cover more crops.

FFS curricula should be developed in a flexible manner that allows each FFS to be customized for diverse target groups and local conditions. Furthermore, Extension officials should be well trained in using a participatory approach to effectively deliver the message of the FFS program. An action plan might be followed by the government on how these would be ensured.

FFS intervention should be planned with the goal of maximizing the potential to build on previous efforts and create more significant change by focusing more on agroecology and sustainable farming practices.

4.3. Limitations of the Study

In this study, the major limitation is that the treatment was not randomly assigned; rather, it was taken ex post facto and we cannot make measurements with and without treatment effects from the same individuals. The small sample size is another limitation of this study due to the COVID-19 pandemic, the comprehensive questionnaire, and not being able to conduct an online survey or telephone interview due to insufficient facilities. This study tried to mitigate the above-mentioned limitations by applying different methods of empirical analysis, as per suggestions in the literature.