Climate change and farmers’ perceptions: impact on rubber farming in the upper Mekong region

Abstract

This article examines the impact of farmers’ perceptions of temperature change on implementing environmentally friendly agriculture practices on rubber plantations. Based on the data collected from 611 smallholder rubber farmers in Xishuangbanna Dai Autonomous Prefecture (XSBN) in the upper Mekong region, an endogenous switching probit model and an endogenous treatment effects model are applied to estimate the impacts of farmers’ perceptions of temperature change on implementing environmentally friendly rubber plantations (EFRP) proxied by the intercropping system. While the real annual average temperature in XSBN has been increasing, only 59% of respondents perceived an increasing trend, whereas over 38% perceived no change. Farmers’ perceptions of temperature change appear to hinge on their education and socioeconomic characteristics and the experience of shocks related to regional climate change. Improving farmers’ perceptions of increasing temperature can significantly foster their practice of EFRP. Hence, policies that promote awareness of regional climate change can effectively encourage the implementation of mitigation practices.

Appendices

Appendix A tables and figures

Table 8 Crops intercropped with rubber at the plot level

Table 9 Summary and descriptive statistics of the variables at the plot level

Table 10 Probit regressions for farmers’ perceptions of increasing temperature and the adoption of rubber intercropping at the household level

Fig. 3

The location of XSBN and the distribution of sample townships

Fig. 4

Cumulative distribution of probabilities of adopting rubber intercropping between farmers who perceive increasing temperature and farmers who do not perceive increasing temperature

Appendix B Models

1.1 Theoretical framework

In this study, we focus on smallholder rubber farmers’ decisions to adopt the EFRP which is proxied by rubber intercropping. This decision is assumed to be made by the farmer after the decision of land allocation for rubber farming has been completed. The farmer’s utility from rubber farming is assumed to include two components: (1) the utility derived from the profit of rubber farming and intercropping, which is affected by the weather condition during the crop season, and (2) the utility derived from improved environmental conditions when adopting rubber intercropping under the weather condition during the crop season. Thus, the farmer should determine the proportion of the rubber plantation to allocate for intercropping to maximize the total utility from rubber farming.

Specifically, the expected profit of rubber farming and intercropping (π) is assumed to be determined as follows:

$$ {\displaystyle \begin{array}{c}\uppi =\int \left[\mathbf{y}-c\left(\mathbf{x}\right)\right]f\left(\mathbf{y}\right)d\mathbf{y}\\ {}s.t.\kern0.5em {\sum}_{i=1}^2{x}_i=1\end{array}} $$

(5)

where the vector x = (x₁, x₂), where x₁ and x₂ represent the proportions of rubber plantations allocated for intercropping and monoculture rubber plantations, respectively. L is defined as the planting area of natural rubber available to the farmer. y (y = y(x| L)) is a vector of outputs corresponding to x given the planting area of rubber (L), while c(x) is the cost function corresponding to x. f(y) is the farmer’s subjective probability density function for y, which can be assumed to be solely related to the weather condition (w_t) in the coming crop season (Bai et al. 2015). It is assumed that all smallholder rubber farmers in XSBN face the same market prices of rubber, intercrops, and inputs in the observation year; therefore, the price variables are omitted in the profit function (5).

Moreover, we further assume that the environmental utility of the farmer’s rubber farming depends on the planting area of natural rubber (L), the proportions of intercropping and monoculture rubber plantations (x), and the weather condition (w_t). Thus, the environmental utility can be expressed as

$$ \mathrm{U}\left(\mathrm{E}\right)=h\left(\mathbf{x},L,{w}_t\right) $$

(6)

By combining the profit function (5) and the environmental utility function (6), the farmer’s utility maximization problem can be written as

$$ {\displaystyle \begin{array}{c}\underset{\mathbf{x}}{\max }\ \mathrm{U}=\underset{\mathbf{x}}{\max}\left[\mathrm{U}\left(\uppi \right)+\mathrm{U}\left(\mathrm{E}\right)\right]\\ {}s.t.\kern0.5em {\sum}_{i=1}^2{x}_i=1\end{array}} $$

(7)

where U indicates the total utility from rubber farming, while U(π) denotes the utility from the profit of rubber farming and intercropping.

As farmers do not know the weather condition in the coming crop season (w_t), they make the decision on x based on their predictions of the weather condition. Here, we assume that farmers’ predictions of weather condition in the coming crop season ($\hat{w_t}$) rely on the real weather condition in previous years and their perceptions of weather condition change in previous years. Thus, $\hat{w_t}$ can be expressed as

$$ \hat{w_t}=g\left({w}_{t-1},P\right) $$

(8)

where w_t − 1 denotes the real weather condition in previous years, while P represents farmers’ perceptions of the weather condition change in previous years.

By incorporating the weather condition prediction function (8) into the utility maximization problem (7), the optimal choice of x can be conceptually derived as

$$ {{\mathbf{x}}_t}^{\ast }=z\left(P,L\right) $$

(9)

where the real weather condition in previous years (w_t − 1) is omitted as there is an implicit assumption that all rubber farmers faced the same weather condition in previous years. Given that temperature is a primary measurement of weather condition, the perception of a change in weather condition (P) can, to some extent, be proxied by the perception of temperature change (P′). Then, the optimal proportion of rubber plantations allocated for intercropping can be expressed as

$$ {x_1}^{\ast}=z^{\prime}\left(P^{\prime },L\right) $$

(10)

According to Eq. A6, two hypotheses could be simply derived as follows. First, as x₁^∗ > 0 indicates that the farmer adopts intercropping, the hypothesis (1) is that the adoption of rubber intercropping is affected by the perception of temperature change (P′). Second, Eq. 10 shows the land allocated for rubber intercropping is a function of farmers’ perceptions of temperature changes; thus, we propose the hypothesis (2) that farmers’ perceptions of temperature change (P′) also influence the adoption intensity of rubber intercropping.

1.1.1 Endogenous switching probit model and counterfactual analysis

According to Lokshin and Glinskaya (2009), a farmer’s propensity to perceive increasing temperature can be expressed in a linear form as

$$ {P}_i^{\ast }=\gamma {Z}_i+{\mu}_i $$

(11)

where subscript i represents the farmer. Z_i denotes a vector of independent variables reflecting the socioeconomic characteristics of the respondent, household, land, and located village, while γ is a vector of corresponding parameters to be estimated. μ_i is an error term. Therefore, the observed farmer’s perception of increasing temperature (P_i) can be expressed as

$$ {P}_i=\left\{\begin{array}{c}1\kern1.25em if\kern0.75em {P}_i^{\ast}\ge 0\\ {}0\kern1.5em otherwise\end{array}\right. $$

(12)

where P_i = 1 represents that the farmer perceives the trend in increasing temperature in the local area, while P_i = 0 denotes that the farmer does not perceive this trend.

The propensity of the farmer’s household to adopt rubber intercropping as a mitigation behavior for the farmer’s perception of increasing temperature is expressed as

$$ {A}_{iP}^{\ast }={\beta}_P{X}_i+{v}_{iP} $$

(13)

where the subscript P denotes the two regimes presented in Eq. 12. X_i is a vector of variables regarding the characteristics of the respondent, household, land, and village, while β_Pis a regime-specific vector of the parameters to be estimated; v_iP is a regime-specific error term.

Hence, by combining Eqs. 12 and 13, the observed mitigation behavior regarding rubber intercropping can be written as follows:

$$ {A}_{i1}=\left\{\begin{array}{c}1\kern1.25em if\kern0.75em {A}_{i1}^{\ast}\ge 0\\ {}0\kern1.75em otherwise\end{array}\right.\kern2.75em \left(P=1\right) $$

(14a)

$$ {A}_{i0}=\left\{\begin{array}{c}1\kern1.25em if\kern0.75em {A}_{i0}^{\ast}\ge 0\\ {}0\kern1.75em otherwise\end{array}\right.\kern2.75em \left(P=0\right) $$

(14b)

where Eqs. 14a and 15b indicate whether the farmer adopts rubber intercropping under the conditions of P = 1 and P = 0, respectively.

According to previous studies (Lokshin and Glinskaya 2009), the error terms (μ_i, v_i0, v_i1) from Eqs. 12, 14a, and 14b are assumed to be jointly normally distributed with a zero-mean vector and correlation matrix:

$$ {\Omega}_{\mathrm{m}}=\left(\begin{array}{ccc}1& {\rho}_{\mu 0}& {\rho}_{\mu 1}\\ {}& 1& {\rho}_{01}\\ {}& & 1\end{array}\right) $$

(15)

where the terms ρ_μ0 and ρ_μ1 are the correlations between v_i0, v_i1, and μ and ρ₀₁ is the correlation between v_i0 and v_i1. However, as A_i1 and A_i0 are never observed simultaneously, the joint distribution of (v₀, v₁) is not identified; accordingly, ρ₀₁ cannot be estimated. Hence, following the study by Lokshin and Sajaia (2011), we further assume that γ is estimable only up to a scalar factor (ρ₀₁ = 1); therefore, this model can be identified by nonlinearities in its functional form. Following the study by Lokshin and Glinskaya (2009), we can express the log likelihood functions for the simultaneous system of Eqs. 12, 14a, and 14b as follows:

$$ \ln \left(\upxi \right)=\sum \limits_{P_i\ne 0;{A}_i\ne 0}\ln \left\{{\Phi}_2\left({\beta}_1{X}_i,\gamma {Z}_i,{\rho}_{\mu 1}\right)\right\}\kern5.75em +\sum \limits_{P_i\ne 0;{A}_i=0}\ln \left\{{\Phi}_2\left(-{\beta}_1{X}_i,\gamma {Z}_i,-{\rho}_{\mu 1}\right)\right\}+\sum \limits_{P_i=0;{A}_i\ne 0}\ln \left\{{\Phi}_2\left({\beta}_0{X}_i,-\gamma {Z}_i,-{\rho}_{\mu 0}\right)\right\}+\sum \limits_{P_i=0;{A}_i=0}\ln \left\{{\Phi}_2\left(-{\beta}_0{X}_i,-\gamma {Z}_i,{\rho}_{\mu 0}\right)\right\} $$

(16)

where Φ₂ is the cumulative function of a bivariate normal distribution. Accordingly, function (20) can be estimated by the FIML method. The ESP model takes into account the unobserved variables that could simultaneously affect the farmer’s perception of increasing temperature and the farmer’s decision to adopt rubber intercropping. The application of the FIML method to simultaneously estimate the functions of these two decisions can yield consistent standard errors of the estimates (Lokshin and Sajaia 2011).

The impact of a farmer’s perception of increasing temperature on the adoption of rubber intercropping can be defined as treatment effects, including the effect of treatment on the treated (TT), the effect of the treatment on the untreated (TU), and the treatment effect (TE). Following previous studies (Lokshin and Glinskaya 2009; Lokshin and Sajaia 2011), the formulas of these treatment effects are given as:

$$ \mathrm{TT}(x)=\Pr \left({A}_1=1|P=1,X=x\right)-\Pr \left({A}_0=1|P=1,X=x\right) $$

(17)

$$ \mathrm{TU}(x)=\Pr \left({A}_1=1|P=0,X=x\right)-\Pr \left({A}_0=1|P=0,X=x\right) $$

(18)

$$ \mathrm{TE}(x)=\Pr \left[A=1|X=x\right]-\Pr \left[A=0|X=x\right] $$

(19)

Furthermore, the average treatment effect on the treated (ATT), the average treatment effect on the untreated (ATU), and the average treatment effect (ATE) can be obtained from Eqs. 17, 18, and 19 by averaging TT(x), TU(x), and TE(x) over the sample, respectively. ATT reflects the average difference between the predicted probability of adopting intercropping by a household that perceives increasing temperature and the predicted likelihood of adopting intercropping for the household had they not perceive increasing temperature. ATU is the average expected effect of perceiving increasing temperature on the probability that households with observed characteristic X, which do not perceive increasing temperature, would adopt intercropping. ATE is the average impact of perceiving increasing temperature on the probability that a household randomly drawn from the households with characteristics x would adopt intercropping. Additionally, the ATT, ATU, and ATT for a subgroup of the households are the averages of TT(x), TU(x), and TE(x) for that subgroup (Lokshin and Sajaia 2011).

1.1.2 Endogenous treatment effects model

Following Maddala (1983), the adoption intensity of rubber intercropping can be expressed as a treatment effects model:

$$ {y}_i=\beta {X}_i+{\gamma P}_i+{\upvarepsilon}_i $$

(20)

where the definitions of X_i and P_i are the same as in Eq. 13. β and γ are parameters to be estimated, while ε_i is an error term. Meanwhile, ε_i and μ_i (in Eq. 11) are assumed to be bivariate normal with zero and covariance matrix:

$$ \left(\begin{array}{c}{\varepsilon}_i\\ {}{\mu}_i\end{array}\right)\sim N\left[\left(\begin{array}{c}0\\ {}0\end{array}\right)\left(\begin{array}{cc}{\sigma}_{\varepsilon}^2& \rho {\sigma}_{\varepsilon}\\ {}\rho {\sigma}_{\varepsilon }& 1\end{array}\right)\right] $$

(21)

where ρ is the correlation coefficient between ε_i and μ_i. According to Maddala (1983), the log likelihood for observation i can be written as:

$$ \ln {L}_i=\left\{\begin{array}{c}\mathrm{ln}\Phi \left\{\frac{{\tau Z}_i+\left({y}_i-\beta {X}_i-\delta \right)\rho /\sigma }{\sqrt{1-{\rho}^2}}\right\}-\frac{1}{2}{\left(\frac{y_i-\beta {X}_i-\delta }{\sigma}\right)}^2-\ln \left(\sqrt{2\pi}\sigma \right)\kern3em {P}_i=1\\ {}\mathrm{ln}\Phi \left\{\frac{-{\tau Z}_i-\left({y}_i-\beta {X}_i\right)\rho /\sigma }{\sqrt{1-{\rho}^2}}\right\}-\frac{1}{2}{\left(\frac{y_i-\beta {X}_i}{\sigma}\right)}^2-\ln \left(\sqrt{2\pi}\sigma \right)\kern3.75em {P}_i=0\end{array}\right. $$

(22)

where Φ(∙) is the cumulative distribution function of the standard normal distribution. Thus, Eqs 10 and 12 could be simultaneously estimated by maximum likelihood estimation.

Appendix C Test

Table 11 Falsification test for the validity of the instrumental variable

The estimation results of a falsification test for the validity of the proposed two IVs are reported in Appendix Table 11. The results show that the proposed two instrumental variables significantly affect farmers’ perceptions of increasing temperature. However, for farmers who do not perceive increasing temperature, the proposed two IVs have insignificant impacts on the adoption of rubber intercropping. The proposed two instrumental variables meet the exclusion restriction. Hence, the falsification test empirically confirms the validity of the proposed two instrumental variables to control for the endogeneity of farmers’ perceptions of temperature change in explaining farmers’ implementation of the EFRP model.

Appendix D Unobserved heterogeneity

Fig. 5

Heterogeneities in the effects of perceiving increasing temperature on the adoption of rubber intercropping by unobserved component (95% confidence interval)

The effect of perceiving increasing temperature on the adoption of rubber intercropping by households can vary by observed household characteristics X and unobservables μ (Lokshin and Glinskaya 2009). To account for the unobserved heterogeneity, we can further simulate the MTE:

$$ \mathrm{MTE}\left(x,\mu \right)=\Pr \left({A}_1=1|\ X=x,\mu =\overline{\mu}\right)-\Pr \left({A}_0=1|\ X=x,\mu =\overline{\mu}\right) $$

(23)

The MTE identifies the effect of perceiving increasing temperature on households induced to adopt rubber intercropping because of perceiving increasing temperature (Heckman and Vytlacil 2001; Lokshin and Glinskaya 2009).

Based on the estimation results of ESR, the simulated MTE is 0.342, nearly equal to the ATE, and heterogeneity in the effects of perceiving increasing temperature based on unobserved characteristics is also found (Fig. 5). Following the MTE framework (Lokshin and Glinskaya 2009), Fig. 5 plots the MTE of perceiving increasing temperature on the adoption of rubber intercropping against the normalized values of unobservable component (μ) at the household means for Xs according to Eq. 14a and 14b. The estimates of the MTE for perceiving increasing temperature on the adoption of rubber intercropping are monotonically increasing in μ, indicating that smallholders who are more likely to perceive increasing temperature are also more likely to adopt rubber intercropping. Additionally, the MTEs of perceiving increasing temperature on the adoption of rubber intercropping are not flat, which confirms the presence of unobservable heterogeneity in the impacts of farmers’ perceptions of increasing temperature on farmers’ decisions to adopt rubber intercropping.