The Impacts of Crop Diversity Adoption on Rubber Farmers’ Welfare: Evidence from Thailand

doi:10.21203/rs.3.rs-3233023/v1

Download PDF

Research Article

The Impacts of Crop Diversity Adoption on Rubber Farmers’ Welfare: Evidence from Thailand

https://doi.org/10.21203/rs.3.rs-3233023/v1

This work is licensed under a CC BY 4.0 License

Journal Publication

published 19 Jun, 2024

Read the published version in Agroforestry Systems →

You are reading this latest preprint version

Given the environmental and socioeconomic challenges of monoculture rubber plantation, FAO emphasized the use and protection of crop diversity which also has been proposed for sustainable development of rubber industry in Thailand. But the adoption of such technology is relatively low that makes rubber crop diversity highly controversy in Thailand. With 2,159 observations at rubber household level collected in 2012–2014, present study investigated the impacts of crop diversity adoption on rubber farmers’ net income in Thailand with use of propensity score matching models. The study found that rubber crop diversity adoption had statistically significant negative impacts on the net income of rubber farmers. And the rubber crop diversity farmers got lower net income of [-11,785.29, -10,887.21] baht per person than monoculture rubber farmers. As the rubber crop diversity system usually grants environmental benefits to the local eco-systems which are exactly the positive externalities and non-rewarded in the market, more incentives and supports are necessary to steam up rubber crop diversity adoption and the sustainability of rubber industry in Thailand.

Rubber

crop diversity

welfare evaluation

ATT

Thailand

Natural rubber plantation has long history in Thailand since the first rubber tree was introduced to its southern region in 1899 (Tongkaemkaew et al. 2018). Thailand became the top natural rubber producer in the world since 1991 (Thongyou 2014), and about 1 million households (16% of agricultural households) were involved in rubber plantation (Department of Agricultural Extension 2016). As the largest natural rubber producer and exporter in the world, in 2020, Thailand planted 3.57 million hectares rubber trees, more than one quarter of rubber plantation area in the world, and produced 4.69 million tons of natural rubber, about 32% of global production, and exported 31.5% of global traded natural rubber (OAE 2020). Moreover, rubber plantation became one key poverty elimination program in some poor regions, including the provinces in northeast Thailand.

Given the great importance of rubber industry, its sustainable development is essential for both the economy growth and rubber farmers in Thailand. Monoculture rubber farms with intensive inputs were largely promoted since 1960s to improve latex yield due to the increasing rubber demand (Janchum 2016). And, nowadays, most rubber farmers are highly specialized, especially in south region where is the traditional rubber production base in Thailand. Although monoculture rubber is featured with simple farm management and generally high rubber yields (Kelty 2006; Nichols et al. 2006; Piotto 2008), many studies pointed out that monoculture plantation is an intensive way to exploit natural resources and associated with negative environmental impacts, including deforestation, loss of crop diversity, soil erosion, drought, higher risks of fire and flood, etc. (Baltodano 2000; Jongrungrot 2014; Liu et al. 2018; Somboonsuke et al. 2018). Moreover, as the existence of big fluctuations in global natural rubber markets, rubber farmers are vulnerable to price risks, especially for the farmers who get stuck in poverty. Those largely undermined the rubber farmers’ welfare and rubber industry development in Thailand (Benton et al. 2003; Erskine et al. 2006; Smajgl et al. 2015).

The reducing crops diversity and negative impacts of monoculture system have raised concerns worldwide. The Food and Agriculture Organization gave the important details of crop varieties development and emphasized the use and protection of crop diversity (CropTrust 2015). An intuitive illustration of the difference between the rubber monocultural and crop diversity farms is presented in Fig. 1. Many researchers recognized the benefits of crop diversity agroecosystem adoption, such as mitigating the adverse effects of monoculture and adapting to climate change (Benton et al. 2003; Liu et al. 2018; Nichols et al. 2006). Moreover, some studies also pointed out the possible economic gains from rubber crop diversity plantation (Jongrungrot et al. 2014; Simien and Penot 2011). Tongkaemkaew et al. (2018) revealed that increasing crops diversity based on suitable species can reduce fertilizer inputs and achieve higher yields. Indeed, given the increasingly serious prices risks encountered by monoculture rubber farmers, Thai government issued policies and recommended rubber farmers to adopt rubber crop diversity farms (no more than 15 plants per rai) since 1997. However, rubber crop diversity system is generally more complicated, and crop diversity farmers need to invest extra inputs and spend more time in farm management and then generally have less opportunities to earn off-farm incomes. Therefore, the cost, productivity and welfare would be very different across rubber crop diversity and monocultural groups. In fact, rubber crop diversity usually results in changes of rubber pattern and lower household income, that makes rubber crop diversity plantation less attractive and non-popular in Thailand (Jongrungrot 2014; Somboonsuke et al. 2011).

Farmers are the key driver of rubber crop diversity transformation. The welfare changes followed by crop diversity adoption may directly affect farmers’ expectations, perception, and inputs used in rubber production. Therefore, the impacts of rubber crop diversity adoption on rubber farmers’ welfare are essential for understanding and promoting rubber crop diversity plantation, but there are a very few studies on such topic. Moreover, the existing literature on the relationships between the rubber crop diversity adoption and rubber farmers’ welfares are not consistent (Charernjiratragul et al. 2015; Jongrungrot et al. 2014; Somboonsuke et al. 2011). For instance, some research shows crop diversity adoption improved the rubber farmers income, and provided benefits to local environment (Charernjiratragul et al. 2015; Department Of Agriculture 2015; Jongrungrot et al. 2014), while, Somboonsuke et al. (2011) revealed the commercial (monoculture) rubber plantation was more profitable than the rubber crop diversity system.

In addition, since the rubber farmers make the choice to adopt crop diversity themselves and cannot be randomly assigned to be adopter and non-adopter categories, the impacts of crop diversity adoption must be separated from observed and unobserved socioeconomic factors which simultaneously affect the farmers’ welfare. Failure to deal with farmer self-selection problem will lead biased estimation of welfare impacts. To the best of our knowledge, most present studies estimated the impacts of crop diversity adoption on rubber farmers’ welfare with problematic methods due to the existence of self-selection and endogeneity problems in the process of crop diversity adoption. Therefore, the contribution of this study is to further scrutinize the impacts of rubber crop diversity adoption on the welfare of rubber households with the propensity score matching method (PSM) which generates the condition of random assignment across adopters and non-adopters and reasonably creates the identification of a causal link between the crop diversity choice and welfare outcomes.

The rest of this paper is structured as follows. The second section provides rubber industry and rubber crop diversity system in Thailand. Section three shows the conceptual framework and methodology. The fourth section explains data and descriptive analysis. The estimation results are presented in the fifth section and followed by the discussion and conclusions.

The patterns of rubber crop diversity are quite flexible and highly depended on farmers’ preference, and intercrop species also largely varied across rubber farms and different regions in Thailand. Taking the availability of data and previous studies on crop diversity into consideration, the definition of rubber crop diversity in present study focuses on the variety of crops in rubber farms, and the rubber crop diversity is observed when additional crop species (such as rice and/or corn and/or durian and so on) in rubber plots are intercropped, bordered, or mixed with rubber trees.

The farmers usually make the adoption decision of rubber crop diversity by maximizing their expected utility according to their characteristics and local factors. Following the previous literature, a random utility framework is used to model the adoption of rubber crop diversity (Asfaw et al. 2012). Given a risk-neutral and utility-maximizing rubber farmer, a latent index function to estimate the rubber crop diversity adoption is represented as ${T}_{i}^{\text{*}}$, which denotes the utility difference between the rubber crop diversity (${U}_{ai}$) and non-diversity adoption (${U}_{ni}$). Individual farmer $i$ will adopt rubber crop diversity if the farmer gains utility from diversity adoption is greater than the counterpart $\left({T}_{i}=1|{T}_{i}^{\text{*}}= {U}_{ai}-{U}_{ni}>0\right)$ and vice versa $\left({T}_{i}=0|{T}_{i}^{\text{*}}= {U}_{ai}-{U}_{ni}\le 0\right)$ (Leepromrath et al. 2021). The following equation provides an estimate of the difference in utilities.

${T}_{i}^{\text{*}}=\beta {Z}_{i}+{u}_{i} \text{w}\text{i}\text{t}\text{h} {T}_{i}=\left\{\begin{array}{c}1 if {T}_{i}^{\text{*}}>0\\ 0 otherwise\end{array}\right.$ (Eq. 1)

The ${Z}_{i}$ is a vector of explanatory variables, including characteristics of family, inputs, geographic and climatic conditions of the farm; $\beta$ is a vector of parameters to be estimated; and $u$ is the error term. Moreover, net income is the very common welfare objective in utility optimization process (Kassie et al. 2011; Mendola 2007). As mentioned earlier, the inputs and costs of agricultural production would largely change in the transformation of monoculture into crop diversity rubber farms, so does the net income of rubber farmers. Generally, for household $i$, the treatment effect of rubber crop diversity in a counterfactual framework is equal to:

${TE}_{i}={Y}_{1i}-{Y}_{0i}$ (Eq. 2)

Where ${Y}_{1i}$ and ${Y}_{0i}$ denote net income of household $i$ adopts rubber crop diversity and the outcome that does not adopt crop diversity, respectively. In fact, either ${Y}_{1i}$ or ${Y}_{0i}$ is normally observed, but not both outcomes can be recorded for each rubber farmer at the same time. The net income could be normally observed as follows:

${Y}_{i}={Y}_{1i}\text{*}{T}_{i}-{Y}_{0i}\text{*}\left(1-{T}_{i}\right), {T}_{i}=\text{0,1}$ (Eq. 3)

Commonly, the basic relationships between the net income ${Y}_{i}$ and rubber crop diversity adoption are assumed to be a linear function of a vector of influential variables ${X}_{i}$ and the dummy variable for rubber crop diversity adoption ${T}_{i}$:

${Y}_{i}=\alpha {X}_{i}+\eta {T}_{i}+{\epsilon }_{i}$ (Eq. 4)

Where $\alpha$ is a vector of parameters to be estimated, and ${\epsilon }_{i}$ is an error term. The impact of diversity adoption on the net income ${TE}_{i}$ is measured by the parameter $\eta$. However, $\eta$ can precisely measure the impact of rubber crop diversity only when the adoption is exogenously determined and rubber farmers are randomly assigned to diversity and monocultural groups (Becerril and Abdulai 2009; Kassie et al. 2011). In fact, the adoption of rubber crop diversity is dependent on expected benefits, characteristics of the household and farm, and the factors that affect rubber crop diversity choices may also influence the net income. That implies the rubber crop diversity adopters and nonadopters are not ex-ante randomly assigned and the equivalent distributions of two groups do not hold in this case. Given the relationships between the rubber crop diversity adoption and net income, the correlation between the two error terms ${\epsilon }_{i}$and ${u}_{i}$ is greater than zero and the endogeneity issue in the crop diversity adoption process would create a spurious link between the diversity adoption and net income and lead to a biased estimation of crop diversity impact (Rubin 1974).

The two-step Heckman and instrumental variable (IV) approaches can deal with the bias caused by self-selection. However, the Heckman method depends on the restrictive assumption of normally distributed errors, and it’s not easy to identify reasonable instruments in practice for IV. Moreover, these two methods tend to address the endogeneity problem by imposing distributional and linear functional form assumption which might lead the welfare outcomes extrapolate over regions where no similar adopter and non-adopter observations exist. In addition, those methods have the strong assumption that the coefficients on the control variables are similar for diversity adopters and nonadopters, but the coefficients could vary from each other (Becerril and Abdulai 2009; Kassie et al. 2011).

To reduce the potential selection bias and endogeneity problem, many studies suggested that it is necessary to avoid functional form assumptions and impose a common support condition (Cerulli 2015). The propensity score matching studies the probability of participants who receive a treatment based on observed characteristics to calculate the causal effect with use of the propensity scores and matching algorithm (Li 2013). PSM does not require to specify a specific functional form for the relationship between confounding variables and potential outcomes (Kassie et al. 2011), and with propensity matching each individual adopter and non-adopter based on similar observable characteristics, PSM overrides the exogeneity of covariates to identify the causal effect of interest. Moreover, PSM approach allows a wide set of different matching procedures, thus it enables researchers to compare various estimators and provide robustness to results (Cerulli 2015; Rosenbaum and Rubin 1983). Thus, the propensity score-matching method is adopted to reliably recover the treatment effect. With the conditional independence assumption that, diversity adoption is independent of potential outcomes given set of covariates, the average treatment effect on the crop diversity farmers (ATT) is defined within the region of common support which ensures that farmers with the similar covariates $X$ have a positive probability of being both adopter and nonadopter in propensity score matching (Guo et al. 2020; Rosenbaum and Rubin 1983). The ATT effect can then be estimated as follows:

$${ATT}_{i}=E\left[{Y}_{1i}|{T}_{i}\text{=1}\right]-E\left[{Y}_{0i}|{T}_{i}\text{=1}\right]$$

$$={E}_{p|T=1}\left\{E\left[{Y}_{1i}|{T}_{i}\text{=1,}p\left(X\right)\right]-E\left[{Y}_{0i}|{T}_{i}\text{=1,}p\left(X\right)\right]\right\}$$

$={E}_{p|T=1}\left\{E\left[{Y}_{1i}|{T}_{i}\text{=1,}p\left(X\right)\right]-E\left[{Y}_{0i}|{T}_{i}\text{=0,}p\left(X\right)\right]\right\}$ (Eq. 5)

Where $E\left(\bullet \right)$ is the expectation in the population. ${T}_{i}$ represents the treatment of crop diversity. In practice, both the probit and logit model are commonly used to predict propensity score for each observation. And several methods which were proposed in the previous studies to impose the common support region and match similar adopters and nonadopters, such as Nearest neighbor matching and Kernel-based matching, are also adopted in present study. In addition, as PSM assumes diversity adoption is based on observable variables, there might be systematic differences between the net income and non-diversity rubber farmers even after conditioning on observables due to the existence of unobservable variables that may affect both crop diversity choice and net income. Therefore, Rosenbaum bounds test is commonly employed as the sensitivity analysis that examines how great an effect of unobservable may have on changing inference about the crop diversity.

The household data in present study came from the socioeconomic and labor surveys mainly carried out by the Office of Agricultural Economics (OAE). The annual surveys were based on the agricultural census by National Statistical Office of Thailand and farmer registration information by Department of Agriculture Extension and stratified two-stages random sampling was applied. The survey covered all provinces in Thailand and collected the information such as household characteristics, land and water resources, agricultural inputs and extension services, income and expenditure, agricultural and non-agricultural activities, etc. Given the data availability, the final dataset consisted of 2,159 observations and covered the rubber households from 67 provinces in Thailand from 2012 to 2014. And the data arrangement and analysis were worked out in STATA 16.0 program.

Table 1 presents the definition of variables and their differences across rubber crop diversity and non-diversity groups. There are several kinds of factors included in the vector of observed variables, namely household characteristics, geographic and climatic conditions, agricultural inputs and extension services which may affect rubber farmer’s production decision (Leepromrath et al. 2021). The significant differences were found in age, education, working on-farm, main rubber region, number of lands, farm size, rubber land size, rainy season rainfall, land problem, crop sale channels, organic fertilizer, credit access, young rubber.

Non-adopters are more likely to encounter the lack of access to market. Increasing number of land blocks, particularly in main rubber production region (south) where rubber plantation is common and highly competitive, the chances to adopt monocultural plantation get higher. The higher education level does not necessarily lead to crop diversity system adoption as the results show that crop diversity group has lower education level for the household heads. Meanwhile, crop diversity adopter is more likely to work full time on farm, and the farms tend to adopt crop diversity when the rubber trees are growing at young stage. There are significant differences in farm size and rubber land size, implying that crop diversity adopters have more land areas than their counterparts. The credit access for crop diversity group is better than the non-diversity group that is in line with the existence of extra inputs of crop diversity. The difference in water resources in rainy season between adopters and non-adopters is statically significant, and crop diversity group exhibits higher proportion of adequate water in rainy season. In addition, rubber crop diversity farmers report more land problems and the use of organic fertilizer which are consistent with the expectation that land problems and organic fertilizer in rubber plantation have significant impacts on rubber crop diversity adoption.

Table 1

The difference of variables between crop diversity and non-crop diversity rubber households.
Variables	Description	Diversity	Non-diversity	Difference
		N = 500	N = 1,659
Net income	Average net income per capita of households (baht/person)	68746.85 (55883.51)	70246.72 (56767.97)	-1499.87
Age	Age of household head (years)	55.88 (10.69)	54.67 (10.96)	1.21**
Gender	Gender of household head (1 = male, 0 = otherwise)	0.81 (0.39)	0.82 (0.39)	-0.01
Education¹	Education level of household head	3.81 (1.34)	4.00 (1.42)	-0.19***
Working on-farm	Work status of household head (1 = worked full time on the farm, 0 = otherwise)	0.82 (0.38)	0.78 (0.42)	0.04**
Agricultural training	Dummy variable (1 = household head attended in training programs, 0 = otherwise)	0.23 (0.42)	0.21 (0.41)	0.02
Household size	The number of family members	4.44 (1.62)	4.32 (1.48)	0.12
Cooperative membership	Dummy variable (1 = household head joined cooperative organization; 0 = otherwise)	0.77 (0.42)	0.80 (0.40)	-0.03
Age below 12	Percentage of person below age 12 within households	12.54 (15.76)	12.37 (15.79)	0.17
Age above 65	Percentage of person above age 65 within households	10.12 (19.12)	8.78 (17.78)	1.32
Main rubber region	Dummy variable (1 = farms located in main rubber production region (south), 0 = otherwise)	0.42 (0.49)	0.54 (0.50)	-0.12***
Number of lands	Number of land blocks of households	3.28 (1.75)	3.91 (1.63)	-0.63***
Farm size²	Total area of farm cultivated by households (rai, ln form)	3.32 (0.64)	3.11 (0.71)	0.21***
Rubber land size	Total area of rubber land cultivated (rai, ln form)	2.90 (0.75)	2.35 (0.84)	0.55***
Land right	Dummy variable (1 = household had land right certificate for rubber land, 0 = otherwise)	0.48 (0.50)	0.50 (0.50)	-0.02
Irrigation access	Dummy variable (1 = had access to irrigation, 0 = otherwise)	0.02 (0.15)	0.02 (0.13)	0.00
Rainy season rainfall	Dummy variable (1 = had adequate rainfall in rainy season, 0 = otherwise)	0.97 (0.17)	0.95 (0.22)	0.02**
Land problem	Dummy variable (1 = rubber land bogged in any land problems, i.e., soil salinity, soil acidity, and soil erosion, 0 = otherwise)	0.05 (0.22)	0.02 (0.12)	0.03***
Crop sale channels	Dummy variable (1 = rubber households could sell productions only by themselves, 0 = otherwise, e.g., company, cooperative group)	0.56 (0.50)	0.67 (0.47)	-0.11***
Organic fertilizer	Dummy variable (1 = organic fertilizer was used in rubber plantation, 0 = otherwise)	0.12 (0.33)	0.09 (0.28)	0.03**
Hiring labors	Dummy variable (1 = hired labors for rubber plantation, 0 = otherwise)	0.12 (0.33)	0.15 (0.36)	-0.03*
Credit access	Dummy variable (1 = had credit access for agriculture production, 0 = otherwise)	0.68 (0.47)	0.59 (0.49)	0.09***
Land rent	Dummy variable (1 = household with agricultural land rent, 0 = otherwise)	0.03 (0.18)	0.04 (0.20)	-0.01
Computer	Dummy variable (1 = household had computer, 0 = otherwise)	0.27 (0.44)	0.27 (0.44)	0.00
Young rubber	Dummy variable (1 = rubber plantation was in young stage, 0 = otherwise)	0.37 (0.48)	0.24 (0.43)	0.13***
Note: ¹ The survey records the education level as follows: 1 = illiterate; 2 = can read and write; 3 = primary school (primary 1–5); 4 = primary school (primary 6–7); 5 = junior high school; 6 = high school; 7 = vocational education; 8 = higher education. ² Land area is measured in a unit of Rai in Thailand, 1 rai = 0.16 hectare. Standard errors are provided in parentheses. Standard deviations are provided in parentheses. P < 0.05, P < 0.01 and **P < 0.001.

The distributions of the rubber households’ net income for crop diversity and non-diversity groups are presented in Fig. 2. The distributions of net income for two groups generally follow normal distributions with asymmetric long tails on the right. The difference in net income was − 1,499.87 baht/person, indicating that rubber crop diversity adopters were generally worse off than non-adopters.

4.1 Estimations of propensity scores

The propensity scores of rubber crop diversity adoption were estimated in both probit and logit models and served as indicators to balance the observed distribution of variables across the diversity and non-diversity groups. The results were reported in Table 2 and most variables had the expected signs and statistically significant associations with rubber crop diversity adoption. As the propensity scores only serve as a device to balance the observed distribution of covariates across the groups of adopters and non-adopters, present study did not illustrate the interpretation of the propensity score estimates.

Table 2

Probit and Logit estimates of the propensity to adopt rubber crop diversity.
	Probit		Logit
	Coefficient	S.E.	Coefficient	S.E.
Age	0.001	(0.004)	0.002	(0.007)
Gender	0.002	(0.088)	0.01	(0.155)
Education	-0.022	(0.027)	-0.042	(0.049)
Working on-farm	0.212**	(0.086)	0.412***	(0.155)
Agricultural training	-0.039	(0.085)	-0.056	(0.15)
Household size	0.021	(0.024)	0.039	(0.043)
Cooperative membership	-0.22***	(0.084)	-0.394***	(0.149)
Age below 12	-0.001	(0.002)	-0.001	(0.004)
Age above 65	0.002	(0.002)	0.003	(0.004)
Main rubber region	-0.379***	(0.074)	-0.652***	(0.131)
Number of lands	-0.174***	(0.025)	-0.309***	(0.048)
Farm size	-0.387***	(0.093)	-0.814***	(0.176)
Rubber land size	0.948***	(0.075)	1.781***	(0.148)
Land right	0.189***	(0.07)	0.333***	(0.123)
Irrigation access	0.283	(0.247)	0.521	(0.42)
Rainy season rainfall	0.285	(0.184)	0.505	(0.335)
Land problem	0.782***	(0.191)	1.293***	(0.326)
Crop sale channels	-0.443***	(0.081)	-0.805***	(0.144)
Organic fertilizer	0.24**	(0.107)	0.421**	(0.184)
Hiring labors	-0.215**	(0.099)	-0.383**	(0.176)
Credit access	0.181**	(0.074)	0.32**	(0.131)
Land rent	0.132	(0.182)	0.302	(0.324)
Computer	-0.048	(0.078)	-0.059	(0.136)
Young rubber	0.765***	(0.085)	1.324***	(0.15)
Y2013	0.07	(0.088)	0.133	(0.156)
Y2014	0.047	(0.084)	0.107	(0.148)
_cons	-1.691***	(0.39)	-2.865***	(0.699)
LR chi2	488.37		498.85
Prob > chi2	< 0.01		< 0.01
Log likelihood	-924.24		-919
Pseudo R2	0.209		0.2135
N	2159		2159
Note: P < 0.05, P < 0.01 and **P < 0.001.

The both Nearest neighbor matching (NNM) and Kernel-based matching (KBM) were employed in the evaluation. As the analysis is condition on the propensity score rather than all pre-treatment variables, it necessary to check the matching quality that whether covariates get balanced distribution in the adopters and nonadopters matching process. The common support condition was checked after estimating the propensity score and all the treated and control households in net income model got considerable overlaps in the region of common support. Figure 3 displays the histogram of the estimated propensity score distributions and regions of common support in net income model with probit and logit regressions. Under the common support condition in probit model, there were 477 treated and 1,615 controlled rubber households in the region of common support. The distributions of propensity scores and the regions of common support were highly similar in logit regression.

Besides the visual inspections of density distributions of the estimated propensity scores and the common support conditions, the further matching quality checks were worked out. The t-test and the test for standardized bias (SB) are commonly used to ensure the balance of the strata (Rosenbaum and Rubin 1983). The t-test compares whether the means of covariates significant different between the treated and the matched control groups. The SB approach calculates the mean difference (for each covariate $X$) in the treated and the matched control groups as a percentage of the square root of the average sample variance in two groups before and after matching. Table 3 presents the matching quality indicators of those tests. The reductions in the mean absolute standardized bias between the matched and unmatched were significant. The standardized mean difference for overall covariates was 14.90% before matching while the remaining standardized differences after counterfactual matching reduced to below 5.0%, with substantial total bias reductions of 66–72%.

Moreover, one can expect to get fairly low pseudo-${R}^{2}$ when re-estimate the propensity score on the matched samples, and p-values of the likelihood ratio in the re-estimating functions should be insignificant because there should be no systematic differences in the distribution of covariates between adopters and nonadopters after matching (Sianesi 2004). The pseudo-R² dropped significantly from 0.21 before matching to around zero after matching. The p-values of the likelihood ratio tests in the re-estimating functions became totally insignificant, indicating that systematic differences in the distribution of the variables between adopters and non-adopters were eliminated after matching. In summary, the low pseudo-R², insignificant p-values of the likelihood ratio test, low mean standardized bias after matching, and high total bias reduction suggest that the matching processes of balancing the distributions of the covariates between the diversity and non-diversity groups were fairly successful.

Table 3

PSM quality tests before and after matching.
Matching algorithm	Model type	Sample	Pseudo R²	LR chi2	p > chi2	Mean standardized bias	(Total)% \|bias\| reduction
Nearest neighbor Matching	Probit	Unmatched	0.21	487.72	0.000	14.90
		Matched	0.01	15.86	0.893	4.20	72%
	Logit	Unmatched	0.21	487.72	0.000	14.90
		Matched	0.02	20.32	0.678	5.00	66%
Kernel-based matching	Probit	Unmatched	0.21	487.72	0.000	14.90
		Matched	0.01	18.37	0.785	4.50	70%
	Logit	Unmatched	0.21	487.72	0.000	14.90
		Matched	0.01	14.75	0.928	4.10	72%

4.2 Estimation of average crop diversity effect (ATT)

Table 4 reports the estimation of the average crop diversity adoption effects on net income of rubber farmers based on Nearest neighbor matching and Kernel-based matching in both probit and logit models. The findings were generally consistent in all approaches. The NNM estimates indicated that crop diversity adoption caused statistically significant decrease of net income around 10,965.65 to 11,785.29 baht for rubber farmers. The effect evaluation from KBM estimates were similar in the range from − 11,732.79 to -10,887.21 baht per capita. Given the problem of hidden selection bias may still arise with PSM method when there are unobserved variables that simultaneously affect the diversity adoption and welfare outcomes for rubber households. Present study then checked the hidden bias after matching with the use of a sensitivity analysis on the estimated average treatment effects of diversity adoption (ATT). Rosenbaum bounds test is commonly employed in such sensitivity analysis and the critical levels of gamma Γ, at which the causal inference of significant diversity adoption impact may be uncertain, was listed in Table 4 (Becerril and Abdulai 2009). The gamma Γ was of 1.46–1.67 in all models which implies that if farmers who have the same covariates differ in their odds of rubber crop diversity adoption by an unobservable factor of 46–67%, the significance of the crop diversity adoption effect on net income may be questionable. That would be fairly large for the sensitivity of the hidden bias and the final effect of crop diversity on farmers’ welfare is robust and insensitivity to the unobservable factor.

Table 4

Average treatment effects (ATT) of rubber crop diversity on net income and sensitivity analysis.
Matching algorithm	Model type	ATT	S.E.	T-stat	Critical level of hidden bias (Γ)	Number of treated	Number of controls
Nearest neighbor matching	Probit	-10965.65***	4149.36	2.64	1.46–1.47	477	1,615
Nearest neighbor matching	Logit	-11785.29***	4437.38	2.66	1.53–1.54	478	1,617
Kernel-based matching	Probit	-11732.79***	3802.48	3.09	1.66–1.67	477	1,615
Kernel-based matching	Logit	-10887.21***	3865.86	2.82	1.65–1.66	478	1,617
Note: ATT is Average Treatment Effect on the Treated (rubber crop diversity); P < 0.05, P < 0.01 and **P < 0.001. Source: Own calculation

The results indicated that rubber crop diversity adoption significantly decreased the net income for rubber farmers in Thailand that is consistent with the study by Somboonsuke et al. (2011) which revealed monoculture rubber plantation was more profitable. As is well known, the economy of scale commonly exists in agricultural production. And management requirements of rubber plantation are relatively high, and they become more complicated in rubber crop diversity production which usually demand more inputs and time. Moreover, the research of Jongrungrot et al. (2014) indicated that rubber farmers who work off the farm can commonly get their income improved due to the relatively high salary levels in non-agricultural sectors. As the non-diversity farmers spent more time on off-farm work, Lambert et al. (2006) pointed out that off-farm income and the high value crops were negatively associated with participation in crop diversity programs. Finally, the existence of the constraints and negative impact on net income largely effected the crop diversity perception and fairly explained that most of rubber farmers in Thailand still preferred monoculture rubber plantation system (Somboonsuke et al. 2011).

This study evaluated the impact of crop diversity adoption on net income of rubber farmers in Thailand. Based on the data from the Office of Agricultural Economics in Thailand, propensity score matching models were employed to deal with potential bias of self-selection and endogenous problems. The impacts of rubber crop diversity on farmers’ welfare were estimated with Nearest neighbor matching and Kernel-based matching in both probit and logit probability propensity algorithms. The results indicate that rubber crop diversity adoption not only induce the changes of rubber pattern, but also lead varied inputs and costs which led statistically significant lower net income for the rubber crop diversity farmers. More specifically, the study found that the effect of crop diversity adoption on net income ranged from − 11,785.29 to -10,887.21 baht per person than the counterparts.

Given the increasing challenges of the rubber industry, the sustainable development strategies are essential and urgent for the rubber industry. Rubber crop diversity plantation system is proposed as one solution to the side effects of monoculture rubber plantation and expected to make the agroecosystems more resilient to climate changes. However, the environmental benefits of rubber crop diversity systems are exactly the positive externalities and not fully rewarded by market mechanisms and then the supply of these benefits is usually less than social optimal levels. That is a justification for government involvement in promoting rubber crop diversity adoption. Therefore, given positive externality and negative net income impact of rubber crop diversity, more incentives and supports should be provided in the rubber crop diversity promoting programs to steam up rubber industry sustainable development. And the estimation of the impact of crop diversity on rubber farmers’ net income in this study would be one useful reference for the policy makers when redesigning the rubber sustainable development strategies in Thailand.

Acknowledgments: The authors thank Pichet Chaipanich, the director of Chachoengsao Rubber Research Center, and staffs from the institution for useful information and guideline of policy implications.

Funding: The work was supported by the National Natural Science Foundation of China [NSFC: 72174088,72103090], the earmarked fund for China Agriculture Research System [CARS-28], and “A Project Funded by the Priority Academic Program Development of Jiangsu Higher Education Institutions” [PAPD].

Asfaw S, Shiferaw B, Simtowe F, Lipper L (2012) Impact of modern agricultural technologies on smallholder welfare: Evidence from Tanzania and Ethiopia. Food policy 37: 283–295
Baltodano J (2000) Monoculture forestry: a critique from an ecological perspective. Tree trouble: a compilation of testimonies on the negative impact of large-scale monoculture tree plantations prepared for the 6th COP of the FCCC
Becerril J, Abdulai A (2009) The impact of improved maize varieties on poverty in Mexico: A propensity score-matching approach. World development 38: 1024–1035
Benton TG, Vickery JA, Wilson JD (2003) Farmland biodiversity: is habitat heterogeneity the key? Trends in ecology & evolution 18: 182–188
Cerulli G (2015) Econometric Evaluation of Socio-Economic Programs Theory and Applications. Springer
Charernjiratragul S, Satsu P, Romyen A (2015) Practical Knowledge and Lessons Learned From Driving the Policy on Expanding the Area for the Rubber-based Intercropping Systems Journal of Social Development 17: 35–50
CropTrust (2015) Securing crop diversity for sustainable development. Global Crop Diversity Trust, Bonn, Germany,
Department of Agricultural Extension (2016) Farmers Registration Database. Department of Agricultural Extention, Bangkok, pp 2–6
Department Of Agriculture (2015) The alternatives activities in rubber plantation with rubber-inter crops and rubber-multi crops to generating rubber farmers' income. Department Of Agriculture, Songkhla, Thailand, pp 1–27
Erskine PD, Lamb D, Bristow M (2006) Tree species diversity and ecosystem function: can tropical multi-species plantations generate greater productivity? Forest Ecology and Management 233: 205–210
Guo S, Fraser M, Chen Q (2020) Propensity Score Analysis: Recent Debate and Discussion. Journal of the Society for Social Work & Research 11: 463–482
Janchum N (2016) Price risk management strategies in a natural rubber industry: A case study of rubber business intermediaries in Thailand. Northumbria University, UK,
Jongrungrot V (2014) The Economic Efficiency and Social Security of Smallholder Farming System which Practices Associate Crops in Rubber Plots in Southern Thailand. Tropical Agricultural Resource Management Prince of Songkla University, Songkla, Thailand, p 279
Jongrungrot V, Thungwa S, Snoeck D (2014) Tree-Crop diversification in rubber plantations to diversity sources of income for small-scale rubber farmers in Southern Thailand. Bois et forêts des tropiques: 21–32
Kassie M, Shiferaw B, Muricho G (2011) Agricultural Technology, Crop Income, and Poverty Alleviation in Uganda. World Dev 39: 1784–1795
Kelty MJ (2006) The role of species mixtures in plantation forestry. Forest Ecology and Management 233: 195–204
Lambert D, Sullivan P, Claassen R, Foreman L (2006) Conservation-Compatible Practices and Programs: Who Participates? United States Department of Agriculture, pp 1–43
Leepromrath S, Zhu J, Zhou J, Li T, Zhou D (2021) Rubber crop diversity and its influential factors in Thailand. Journal of Rubber Research 24: 461–473
Li M (2013) Using the propensity score method to estimate causal effects: A review and practical guide. Organizational Research Methods 16: 188–226
Liu CLC, Kuchma O, Krutovsky KV (2018) Mixed-species versus monocultures in plantation forestry: Development, benefits, ecosystem services and perspectives for the future. Global ecology and conservation 15: e00419
Mendola M (2007) Agricultural technology adoption and poverty reduction: A propensity-score matching analysis for rural Bangladesh. Food policy 32: 372–393
Nichols JD, Bristow M, Vanclay JK (2006) Mixed-species plantations: prospects and challenges. Forest Ecology and Management 233: 383–390
OAE (2020) Agricultural Statistics of Thailand. Office of Agricultural Economics, Ministry of Agricultural and Cooperatives, Bangkok, Thailand, pp 115–118
Piotto D (2008) A meta-analysis comparing tree growth in monocultures and mixed plantations. Forest Ecology and management 255: 781–786
Rosenbaum PR, Rubin DB (1983) The central role of the propensity score in observational studies for causal effects. Biometrika 70: 41–55
Rubin DB (1974) Estimating causal effects of treatments in randomized and nonrandomized studies. Journal of Educational Psychology 66: 688–701
Sianesi B (2004) An Evaluation of the Swedish System of Active Labor Market Programs in the 1990s. The Review of Economics and Statistics 86: 133–155
Simien A, Penot E (2011) Current evolution of smallholder rubber-based farming systems in Southern Thailand. Journal of sustainable forestry 30: 247–260
Smajgl A, Xu J, Egan S, Yi Z-F, Ward J, Su Y (2015) Assessing the effectiveness of payments for ecosystem services for diversifying rubber in Yunnan, China. Environmental Modelling & Software 69: 187–195
Somboonsuke B, Kongmanee C, Thongtrai N, Phitthayaphinant P (2018) Potential of para-rubber production in Thailand: Production technology management, problems and suggestions for yield improvement Kasetsart Journal of Social Sciences 39: 645–659
Somboonsuke B, Wetayaprasit P, Chernchom P, Pacheerat K (2011) Diversification of smallholding rubber agroforestry system (SRAS) Thailand. Kasetsart Journal (Social Sciences) 32: 327–339
Thongyou M (2014) Rubber cash crop and changes in livelihoods strategies in a village in Northeastern Thailand. Asian Social Science 10: 239
Tongkaemkaew U, Sukkul J, Sumkhan N, Panklang P, Brauman A, Ismail R (2018) Litterfall, litter decomposition, soil macrofauna, and nutrient content in rubber monoculture and rubber-based agroforestry plantations. Forest and Society 2: 138–149

No competing interests reported.

Download PDF

Journal Publication

published 19 Jun, 2024

Read the published version in Agroforestry Systems →

Editorial decision: Revision requested
16 Feb, 2024
Reviews received at journal
02 Feb, 2024
Reviewers agreed at journal
19 Sep, 2023
Reviewers invited by journal
18 Sep, 2023
Editor assigned by journal
07 Aug, 2023
Submission checks completed at journal
05 Aug, 2023
First submitted to journal
03 Aug, 2023

You are reading this latest preprint version

The Impacts of Crop Diversity Adoption on Rubber Farmers’ Welfare: Evidence from Thailand

Status:

Journal Publication

Version 1

Abstract

Figures

1. Introduction

2. Conceptual framework and welfare evaluation method

3. Data and descriptive analysis

4. Results and discussion

4.1 Estimations of propensity scores

4.2 Estimation of average crop diversity effect (ATT)

5. Conclusions

Declarations

References

Additional Declarations

Status:

Journal Publication

Version 1