Examining complementary relationships among climate change adaptation practices of rice producers in Chitwan, Nepal

ABSTRACT

This study examines how the complementary relationships among multiple climate change adaptation strategies (improved varieties, irrigation practices and direct-seeded rice, referred to as ‘DSR’) are influenced by access to Extension services, training related to farming practices and weather-related information (referred to as ‘institutional factors’). This study hypothesizes the influence of institutional factors and the resulting effects on their joint adoption using survey data collected from 359 rice producers in Chitwan district of Nepal in 2019. The results indicate: (1) adoption of all the three climate change adaptation practices are complementary to each other; (2) all three institutional factors enhance the likelihood of the adoption of variety, access to Extension and training enhances the likelihood of the adoption of improved irrigation, and information enhances the likelihood of the adoption of DSR and (3) access to Extension and training enhance the complementarities among all three pairs of adaptation practices (variety and irrigation; variety and DSR and irrigation and DSR), while information enhances the complementarity between variety and DSR only. The impact of complementary relationships among multiple climate change adaptation strategies influenced by the institutional factors is particularly valuable for policy makers and government agencies to allocate and prioritize resources.

KEYWORDS:

• Climate change adaptation
• complementarity
• institutional factors
• Extension

Introduction

Climate change, which is defined as the long-term variation of the climate that persists for an extended period of time, is one of the greatest challenges to agricultural production and food security in many regions around the world (Intergovernmental Panel on Climate Change [IPCC], 2007, 2019, 2021; Richardson et al., 2018). The frequency of the occurrence of extreme weather events such as drought, intense rainfall, floods and storms are increasing over time and their impacts on natural and human systems are observed worldwide (IPCC, 2019, 2021). For example, climate change is found to be the major determinant of the change in soil erosion (Borrelli et al., 2021), and imposes a risk to the livelihood and food security of rural populations, especially in developing countries (Adger et al., 2003; Cline, 2007; IPCC, 2019; Kabir et al., 2021; Wheeler & Braun, 2013). The vulnerability to food insecurity is projected to increase under all emission scenarios, particularly in parts of Sub-Saharan Africa and South Asia (Richardson et al., 2018).

Likewise, producers in Nepal are faced with warming temperatures and precipitation changes over the years. The mean annual temperature in Nepal is projected to increase by 1.8°C by 2030 and 2.8°C by 2060 (McSweeney et al., 2010). According to a Food and Agriculture Organization (FAO) report, while more intense rainfall is expected during the rainy season, annual rainfall in Nepal is projected to decrease in the range of 20 mm to 100 mm by 2050 (FAO, 2014). The report also indicates about 90% of crop losses in Nepal were associated with weather-related events. In Chitwan, one of Nepal’s rice-producing districts, between 1987 and 2000, producers were faced with extreme precipitation events (>300 mm in a day) and record extreme temperatures of 43°C for maximum and 1.6°C for minimum (Gurung & Bhandari, 2009). Rice producers experienced a decline in rice yield due to weather variability in the south-central plains or Terai region of Nepal (Gumma et al., 2011; Karna, 2014; Malla, 2008). Producers in Chitwan district have altered rice farming practices to protect yield as a response to climate change (Gahatraj et al., 2018)

Due to limited adaptive capacity, subsistence farming and resource-poor agricultural producers in developing countries are highly vulnerable to the impacts of climate change (Kurukulasuriya & Rosenthal, 2013; Morton, 2007). Climate change adaptation, which refers to adjustment in natural, socio-economic, or human systems in response to actual or expected climatic stimuli or their effects (IPCC, 2007), is one of the key responses to moderate harm or exploit beneficial opportunities (Adger et al., 2003). In particular, developing and disseminating appropriate climate change adaptation measures are of utmost importance in mitigating the risks and vulnerability of agricultural production to climate change. Climate change adaptation on a large scale, combined with appropriate mitigation efforts, helps to reduce the negative impacts of climate change, enhance crop production and improve food security (Loboguerrero et al., 2019; Richardson et al., 2018).

Literature on climate change adaptation practices can be broadly categorized as studies on macro level (i.e. global, national, or regional) impacts and studies on the individual farm or household-level decisions. Studies dealing with the macro-level climate change adaptation focus on simulation of yield impacts of climate change with and without adaptation. For example, Challinor et al. (2014) suggest that climate change adaptation enhances yield by 7–15% depending on crop type across temperate and tropical regions. Most of the literature on the farm and households centre on producers’ perceptions of climate change and determinants of climate change adaptation practices (Deressa et al., 2009; Gadédjisso-Tossou, 2015; Gbetibouo, 2009; Tazeze et al., 2012; Thinda et al., 2020; Tilahun & Bedemo, 2014). These studies focus on how producers adopt one of various climate change adaptation practices at a time (e.g. changing cropping patterns, using improved crop varieties, diversifying crops, and adopting soil and water conservation practices) without accounting for potential interdependence among practices.

Despite the focus on a single practice at a time in the literature, agricultural producers often face the decision to adopt multiple practices together. For example, an agricultural producer may consider adopting multiple practices such as adopting drought-tolerant crop varieties, soil conservation practices, adjustment of time of planting and improved irrigation practices simultaneously as climate change adaptation practices (Khanal et al., 2018). Thus, it is more realistic to assume that an agricultural producer faces the decision to adopt a combination of practices to better manage available resources and mitigate climate-induced risk at the same time rather than one practice at a time. Therefore, there is a need to examine the joint adoption of multiple climate change adaptation practices and their interdependence at the farm and household level.

A right mix of multiple climate change adaptation practices is crucial to improve farm productivity and to reduce the negative impacts of climate change (Aryal et al., 2018; Kurukulasuriya & Rosenthal, 2013; Teklewold et al., 2013). Analysing the adoption behaviour of producers by incorporating the interdependence among climate change adaptation practices can be helpful in identifying the factors that enable or limit joint adoption of multiple climate change adaptation practices. Nhemachena and Hassan (2007), Nhemachena et al. (2014) and Mulwa et al. (2017) identified the nature of relationships (complementary or supplementary) among multiple climate change adaptation practices and their determinants. Furthermore, it is important to understand how the complementary relationships among multiple climate change adaptation practices are influenced by access to Extension services, training related to farming practices and access to information about improved farming practices and reliable weather information (referred to as ‘institutional factors’). Despite the importance, studies that examine the influence of the institutional factors and their relationships with multiple climate change adaptation practices are absent indicating a gap in literature. The lack of literature is an impediment for governments or agencies that are often faced with the decision of allocating limited resources to improve climate resiliency of agricultural production practices.

The objective of this study is to examine how the complementary relationships among multiple climate change adaptation practices are influenced by institutional factors. As a case study, we develop an empirical model that hypothesizes the impact of institutional factors on adoption of drought-tolerant and short-duration rice varieties (referred to as ‘variety’), investment in improved irrigation (referred to as ‘irrigation’) and adoption of direct-seeded rice practice (referred to as ‘DSR’), and the resulting effects on their joint adoption using survey data collected from 359 rice producers in Chitwan district of Nepal in 2019. Results will help local government agencies prioritize the investment on the institutional factors among Extension services, training and information that can potentially enhance the uptake of climate change adaptation practices by identifying the complementary relationships.

The rest of paper is organized as follows. The next section describes the methods: survey design, data and regression models. The following section discusses the findings and the last section summarizes and draws conclusions.

Methods

A producer is more likely to adopt climate change adaptation practices if the expected benefit of adoption is greater than non-adoption. Let $U_j$ and $U_0$ represent the expected benefit for producers with and without adoption respectively, then a producer decides to adopt adaptation practices if the net benefit of adoption (latent variable, $U_j$ –$U_0>0)$ is higher than non-adoption. The latent variable is determined by observed household, socio-economic, institutional and farm characteristics, as well as unobserved factors (error terms). We extend this basic conceptual framework to examine complementary relationships among multiple climate change adaptation practices.

To test the hypothesis, we employ a multivariate probit model where the dependent variables represent binary choices of climate change adaptation practices by rice producers: variety, irrigation and DSR. Multinominal logit and multivariate probit are candidate models with multiple binary dependent variables (Gadédjisso-Tossou, 2015; Gbetibouo, 2009). A major limitation of the multinominal logit model is that it assumes that the binary choices are mutually exclusive. We acknowledge that a rice producer can adopt more than one practice simultaneously and hence did not use a multinomial logit model.

To capture the potential complementarities among the three climate change adaptation practices, following Lin et al. (2005), we apply the multivariate probit model, which can be expressed as follows (observation subscripts suppressed): $y_i=1\mathrm{i}\mathrm{f}{X}^{\mathrm{\prime }}{\text{β}}_i+{ϵ}_i>0$ (1) $y_i=0\mathrm{i}\mathrm{f}{X}^{\mathrm{\prime }}{\text{β}}_i+{ϵ}_i\le 0,i=1,2,3$(1)

where $y_i$ denotes binary variables representing three adaptation practices$y_1=$ variety (V), $y_2=$ irrigation (I) and $y_3=$ DSR (D). ${\mathrm{X}}^{\mathrm{\prime }}$ is a vector of explanatory variables, ${\text{β}}_i$are the unknown parameters and ${ϵ}_i$ are random error components. The random error terms are distributed in a multivariate normal distribution with zero conditional mean and unitary variance;${ϵ}_i\sim N(0,\mathrm{\Omega }$), where$,\mathrm{\Omega }$ is a n × n covariance matrix as follows: (2) $\mathrm{\Omega }=\left[\begin{array}{ccc}1& {\rho }_{VI}& {\rho }_{VD}\\ {\rho }_{IV}& 1& {\rho }_{ID}\\ {\rho }_{DV}& {\rho }_{DI}& 1\end{array}\right]$(2)

The off-diagonal elements in the covariance matrix represent the correlation (unobserved) among the error terms of the three adaptation practices (i.e. variety, irrigation and DSR). The unknown parameters (${\text{β}}_i$) are estimated using a simulated maximum likelihood approach in the multivariate probit model (Chib & Greenberg, 1998). It should be noted that unlike linear models, the marginal effect or partial derivative for a given explanatory variable is non-linear and not equal to ${\text{β}}_i$. Thus, we estimate the marginal effects of explanatory variables to interpret in terms of likelihood of adoption of climate change adaptation practices as follows: (3) ${\displaystyle \frac{\mathrm{\partial }{P}_i}{\mathrm{\partial }{X}_i}=\varphi \left(X^{\mathrm{\prime }}\text{β}\right){\text{β}}_i,i=1,2,3}$(3)

where $P_i$ indicates the likelihood of adopting a climate change adaptation practice $i$ (1 = variety, 2 = irrigation and 3 = DSR), $\varphi (.)$ is the normal cumulative density function and $\text{β}$ are the model parameters (Hassan, 1996).

The three climate change adaptation practices; variety, irrigation and DSR were selected as the dependent variables in the equations based on the National Adaptation Program of Action report (Ministry of Environment, 2010) and previous studies (Biggs et al., 2013; IRRI, 2018; Khanal et al., 2018; Kunreuther et al., 2014; Manandhar et al., 2011; Ministry of Environment, 2010; Ministry of Finance, 2014, 2018; Piya et al., 2013). The explanatory variables X_i include institutional factors, household and socio-economic factors, farm characteristics and perception about climate change (see Table 1).

Table 1. Climate change perceptions of rice producers in Chitwan District in the past 10 years (2009–2018).

Climate change perceptions	% of respondents who perceived climate change in the past 10 years (2009–2018)
Overall change in climate patterns	76.6%
High summer temperature	86.9%
High winter temperature	61.9%
Intensity of rainfall	80.8%
Floods	29.1%
Dry spells	71.4%
Weather unpredictability	99.6%
Hailstorms	97.8%
Late onset of monsoon rain	86.2%

Source: Regmi (2020).

Institutional factors

We chose three binary variables: access to Extension, training and information as institutional factors based on an extensive literature review. Agrawal (2008) discussed the role of local institutions as Extension service, capacity building and information dissemination; Gbetibouo (2009) includes agricultural Extension and information as institutional factors; Addisu et al., 2016 used Extension and climate information as institutional characteristics; Below et al. (2012) and Aryal et al. (2018) used access to agricultural Extension services as a proxy of institutional service. These three institutional factors are key determinants of adoption of climate change adaptation practices (Abdulai & Huffman, 2014; Deressa et al., 2009; Piya et al., 2013). Specifically, agricultural producers can receive information about improved farming practices and its benefits from Extension workers, which facilitates producers’ decisions for adoptions (Deressa et al., 2009; Mulwa et al., 2017). Likewise, having access to agricultural training programme helps producers gain in-depth knowledge about new farming techniques and to adopt latest technology (Aryal et al., 2018). Similarly, access to weather information can help producers to adjust farm operations, and access to information about improved farming practices can also help agricultural producers to select suitable climate change adaptation practices (Mulwa et al., 2017).

The role of three institutional factors in the local context (see Table 2) is described below based on communication with local agricultural Extension agents (Chhetri, 2019; Dhital, 2020):

• Extension: Extension agents are responsible for executing the field level Extension activities with individual producers. They implement the governmental plans and programmes (of the Ministry of Agricultural Development) for the agricultural development. During the face-to-face field/farm visit, Extension worker provides technical service related to the improved production technologies, effective plant protection measures and assist producers in making better farm management decisions. Besides technical services, producers can get information on availability of subsidies on seeds, fertilizers, irrigation, farm tools and machinery through Extension contact. Furthermore, Extension agents organize agricultural fairs, farmers field day and farm visits to motivate producers to implement improved farming technologies. Extension services are offered by Extension agents during field visits to rice farms or through the Extension offices at district headquarters. Proximity to the district office plays a role in producers availing Extension services and Extension agents’ ability to travel to farms on field visits. The timeframe for farm visit of Extension office visit typically is 1-day (Singh, 1997).

• Training: Both the government and non-government organizations can provide climate change adaptation training to a group of producers (typically 20–25). Producers are trained on topics including climate change, its impacts, implications and methods to reduce the negative impacts of climate change in training sessions. The trainings also impart knowledge on better utilization of local resources, new management practices and improved technologies in the context of climate change. Trainings are offered to producers as demonstrations or in-service workshops in adoption of new varieties, technology and management practices. These are typically provided in 2–3 d workshops at a central location to encourage producer participation. In certain occasions, a scholarship or stipend is provided to producers to encourage participation (Singh, 1997).

• Information: Information is a mass-contact method offered to producers in the form of factsheets, brochures, publications, booklets, progress reports, Television/radio agricultural programmes, weather forecast announcements, extreme weather events and advisories. Literacy and access to print or digital media affects producers’ access to information. Most of the publications are available for producers at the Extension office (Singh, 1997) or online for producers who have internet accessibilities.

  Table 2. Comparison of institutional factors – Extension, training and information (adapted from Singh, 1997).

  	Extension	Training	Information
  Purpose and content	Needs assessment, identification of an agricultural problem or seeking information	Transfer of agricultural technology, new methods, innovative solutions, skills and techniques	Easy to understand agricultural and weather information that can be rapidly distributed to a wider audience
  Number of contacts	Individual contact method or face-to-face with the producer	Group contact method with a group of producers (20–25 or more)	Mass contact method to provide information quickly
  Instructor/Author	Extension agent	Extension agent or subject-matter expert	Extension agent or Agricultural Graduate Officer
  Location	Farm or Extension office	Extension office, Agricultural Knowledge Center, University, farm, market or a centralized location where producers can meet.	Information mailed to households; transmitted via Radio, Television; or posted on internet
  Methods of instruction	Farm visits, Extension office visits and phone calls	Method or result demonstrations, in-service trainings, workshops, conferences, field trips and discussion meetings.	Factsheets, bulletins, brochures, publications and progress reports. Television/Radio or internet can provide timely information on extreme weather events or advisories.
  Timeframe	1 d	2–3 days	Immediate or less than 24 h
  Cost	Medium	High	Low

The local Extension agents avoid duplication of efforts among the three institutional factors to minimize costs and there is no multicollinearity issue among these variables. While local agrovets services may provide services similar to the three institutional factors, we did not collect any information on local agrovets service which is beyond the scope of this study.

Household and socio-economic factors

As household and socio-economic characteristics, we include gender of the producer who is also the head of the household, dependency ratio (ratio of dependent family members under the age of 16 years or above 60 years to the total economically active members aged 16–60), out-migration (to Gulf and Middle East countries for work), producers’ education years, producers’ years of experience in rice farming, annual household income and membership status of producers in cooperatives or producers’ organizations. Gender of the head of the household is one of the important factors influencing household decision to adopt climate change adaptation practices (Aryal et al., 2018; Mulwa et al., 2017). We use two variables: dependency ratio and out-migration as a proxy for household labour availability. A dependency ratio in the household measures the household labour availability. Among many farm households in Chitwan, younger family members emigrate to work in Middle-East and Gulf countries leaving parents and other family members behind. The out-migration of family members for work may potentially reduce the household labour availability for some producers (Gartaula et al., 2012). Household labour availability determines the manpower to implement farm practices that influence the decision to adopt climate change adaptation practices (Ali & Erenstein, 2017; Nhemachena et al., 2014). Literature also indicates that a family with higher number of economically active family members has a greater propensity to adopt additional labour-intensive farming practices (Deressa et al., 2009). Years of education and farming experience are commonly used as determinants of climate change adaptation practices in literature (Ali & Erenstein, 2017; Khanal et al., 2018; Mulwa et al., 2017). A higher educational attainment of the producer who is the head of the household implies better access to information on improved farming practices, and, thus, greater likelihood of adoption of climate change adaptation practices (Deressa et al., 2009). Furthermore, annual household income and membership status of producers in cooperatives or organizations are included because they are expected to have a positive influence on adoption of climate change adaptation practices (Ali & Erenstein, 2017; Hassan & Nhemachena, 2008; Mulwa et al., 2017; Nhemachena et al., 2014).

Farm characteristics

The farm characteristics, such as number of plots under rice cultivation and rice area are important factors influencing the adoption of climate change adaptation practices (Mulwa et al., 2017; Nhemachena et al., 2014). Thus, we include total rice area and number of rice plots under cultivation by allowing a single producer with multiple plots of land. We hypothesize that the producers who sold rice in the market in addition to self-consumption are more likely to adopt climate change adaptation practices.

Perceptions about climate change

We include producers’ perception about changes in local weather patterns over time. We collect producers’ perception about changes in local weather patterns over past 10 years in terms of temperature, rainfall, dry periods, floods, hailstorms, unpredictability of weather, groundwater table, onset and end of monsoon. Producers are more likely to adopt climate change adaptation practices if they noticed changes in local weather patterns (Khanal et al., 2018; Nhemachena et al., 2014; Nhemachena & Hassan, 2007). In accordance to literature, we hypothesize that if producers perceive the changes in local weather patterns, they are more likely to adopt multiple climate change adaptation practices.

Model estimation issues: multicollinearity and endogeneity

The three equations of the multivariate probit model are estimated by using simulated maximum likelihood estimation developed by Cappellari and Jenkins (2003) using mvprobit command in STATA 16 (Statacorp, 2019). Multicollinearity of explanatory variables are tested by calculating a conditional index value (Belsley, 1991) and variance inflation factor (VIF). A conditional index value represents the collinearity of combinations of variables in data. When the value of the conditional index is less than 30, there is no evidence of multicollinearity (Belsley, 1991). We did not find evidence of multicollinearity because the condition index value is 9.48, which is not above the threshold level of 30 and the mean VIF is 4.4. We do not have a serious issue of multicollinearity when the VIF value is less than threshold of 10 (Marquardt, 1970; Mason et al., 1989). A correlation matrix of institutional factors, the variance inflation factors and conditional index values are presented in Appendix 2.

We test how hypothetical changes of institutional factors influence the complementarities of multiple adaptation practices using the post-estimation prediction. To implement the test, first, we estimate the predicted likelihood of climate change adaptation practices based on the results of the three equations of the multivariate probit model (referred to as ‘baseline predicted likelihood’), then we predict the likelihood of adoption of multiple climate change adaptation practices by hypothetically increasing the rice producers with access to Extension, training and information (or increasing institutional factors) through randomly switching their binary values from zeros to ones by 10% for each of the institutional factors using randomtag command in STATA 16 (Statacorp, 2019). Then, we test the complementarity relationship by testing how correlated the predicted increase in the probabilities across the three climate change adaptation practices as a result of switching the values of the binary variables, representing access to Extension, training and information from zeros to ones by 10% each. We also conduct sensitivity tests by hypothetically increasing access to institutional factors by 20%, 30% and 40%.

We address the potential endogeneity of binary institutional factors. Our model has a binary dependent variable (adoption decision) and the potential endogenous variables (Extension, Training and Information) are also binary. The multivariate probit model is a joint probability model which accounts interrelationship of adoption decisions (dependent variable) and potential endogeneity issue in such probability model (multivariate model having binary Y, binary X) are relatively less emphasized in the literature. Finding three distinct instrumental variables (IVs) are very challenging in model estimation. There are many studies in the literature that emphasize the difficulty in finding relevant IVs and model estimation in limited dependent variable models (Rivers & Voung, 1988; Clarke & Windmeijer, 2012; Bollen, 2012; Shardell & Ferrucci, 2016). We used three instrumental variables:

1. Ethnicity/social status of the respondent is used as an instrument for the access to Extension following Devkota (2005). Each survey participant indicated their ethnicity – Brahmin/Chettri, Janajati or Dalit. Brahmin/Chhetri have higher social status and resources to access Extension services (Devkota, 2005) and ability to travel to Extension office. However, ethnicity does not influence climate change adaptation practices.

2. We use producers risk attitudes as an instrument for climate change adaptation training. Risk averse producers are expected to participate in training programmes (Kunreuther et al., 2014; Nisengwe, 2016) than risk neutral or risk-taking producers.

3. The number of literate family members in the household is used as an instrument for access to agricultural and weather information. Having more educated household members can improve access to agricultural and weather information (Dulal et al., 2010) through multiple sources.

We use two approaches to address potential endogeneity problem. First, we used Durbin-Wu-Hausman test of endogeneity in a two-stage framework (Table 3). We failed to reject the null hypothesis for variety (p-value = 0.3110), irrigation (p-value = 0.1545) and DSR (p value = 0.1492) indicating the variables are exogenous and affirming that endogeneity is not an issue in our model.

Table 3. Durbin–Wu–Hausman test of endogeneity for variety, irrigation and DSR.

	Variety	Irrigation	DSR
Wu–Hausman statistics	1.19671	1.76036	1.78805
p-value	p = 0.3110	p = 0.1545	p = 0.1492

In the second approach, following Rivers and Voung (1988) and Aryal et al., (2018), we first specified the potential endogenous variables as a function of instruments and control variables. Then, we use residuals of the first stage regression as an additional control variable in the model to acquire consistent estimates. If the residuals of the first stage regression are not significant, we do not have an issue of endogeneity (Aryal et al., 2018). The residuals of all three variables are not significant in the model and hence we report coefficients without residual terms. A detailed description of the instrumental variables and results from the first stage regressions for potential endogenous variables are provided in Appendix 3.

Since endogeneity is not an issue, we do not believe there are any problems with reverse causality because reverse causality causes endogeneity problem (Cingolani & Crombrugghe, 2012; Feng & Wu, 2018).

Study area

The study area for this research is Chitwan district, which is located in the south-central Plains (Terai) region of Nepal in South-Asia (see Figure 1). We chose Chitwan district as the study area because rice producers in Chitwan district faced high variation in temperature and precipitation (Gurung & Bhandari, 2009) in the recent years. Upon consultation with the District Agricultural Extension Office (Agricultural Knowledge Center, Chitwan), three villages in Chitwan were recommended as a study area due to a relatively higher proportion of rice producers. The meteorological station in Rampur located in Chitwan district provided temperature and precipitation records for the study. The proximity of student enumerators from Agricultural and Forestry University in Chitwan solidified the study area. While it could be informative to add other study areas, however, given our limited budget, we could not survey producers outside of Chitwan district.

Figure 1. Study area of three villages in Chitwan district, which is located in the south-central Plains (Terai) region of Nepal.

According to the District Agricultural Development Office in Chitwan (DADO, 2017), rice is cultivated in 33,900 hectares, which represents 77% of the district’s total arable land. Approximately 67% of households are involved in agricultural activities. The average household size in Chitwan district was 4.38 members with a mean household income in Nepali Rupees (NR) of 169,070 (equivalent to US Dollar, $1,406). 76.98% of the households were literate and 33.69% of the households were female-headed. The average farm size is 0.46 hectares (1.13 acres) which was lower than national average of 0.68 hectares (1.68 acres). The average rice yield was 3.4 metric tons per hectare, which was lower than national average of 3.76 metric tons per hectare. More than half of the producers (55.72%) were members of farmer’s group/co-operatives (Central Bureau of Statistics, 2018; DADO, 2017).

Survey design

All the data for the multivariate probit model was collected from a field survey of rice producers in Chitwan district, Nepal in June, 2019. First, out of seven administrative units in Chitwan district, the largest administrative unit, Bharatpur was selected, which consists of a total of 69,035 households out of which 45,991 were farming households. Second, in consultation with district agricultural Extension office in Chitwan, three villages where rice is a major crop – Patihani, Jagatpur and Sharadanagar were selected as the target survey area. Third, out of the total number of farming households in three villages, 4,090, the sample size was estimated to be at least 352 at the 95% confidence interval, following the sample size determinization method developed by Krejcie and Morgan (1970).

Survey data

Four undergraduate students of Agriculture and Forestry University, who were proficient in Nepali language and familiar with the landscape and customs of Chitwan were hired as enumerators. They were trained and provided directions to fill the questionnaire before conducting the field survey. The enumerators started at the periphery of the village, selected farming households and progressed towards the centre of the village. Each survey participant was read a consent statement by the enumerator following Institutional Review Board procedures describing their involvement in the study, protecting their privacy and ensuring confidentiality. Upon providing consent that the respondent was 18 years and older, understood the information and agreed to participate in the survey, individual producers answered the survey in a face-to-face interaction with enumerators. A total of 383 producers were contacted, 359 producers (93.7%) answered the survey and 24 producers declined to participate in the survey (6.3%). Surveys were conducted at the producers’ house and at local gathering spots in the three villages. Each survey took approximately 15 min to complete. Summary statistics of the variables used in the multivariate probit model are presented in Table 4.

Table 4. Descriptions of variables used in the multivariate probit model.

Variable	Descriptions	Mean (n = 359)	Standard deviation
A. Dependent variables
Variety	1, if adoption of drought tolerant and short-duration rice varieties, 0 otherwise	0.64	0.47
Irrigation	1, if producer invested in improved irrigation, 0 otherwise	0.60	0.48
Direct-seeded rice (DSR)	1, if adoption of the practice, 0 otherwise	0.15	0.36
B. Explanatory variables
a. Household and socio-economic characteristics
Gender	1, if head of the household is female, 0 otherwise	0.20	0.40
Dependency ratio	Household dependency ratio (Ratio of number of dependent family members to number of economically active family members aged 16–60 years)	0.49	0.65
Out-migration	1, if any of household member migrated to another country for employment, 0 otherwise	0.45	0.49
Education	Producers’ formal education years	7.92	4.12
Farming years	Producers’ years of farming experience	26.01	10.81
Income	Annual household Income (USD)	2090.9	1472.01
Membership status	1, if any of household member is a member of farmers’ organizations (farmers’ group or cooperatives), 0 otherwise	0.68	0.46
b. Farm characteristics
Rice plots	Number of plots under rice cultivation	1.65	0.79
Rice sold	1, if sold rice in the market in addition to self-consumption, 0 otherwise	0.78	0.41
Rice area	Hectares of rice area	0.46	0.30
c. Climate-related factors
Climate	1, if perceived changes in local weather patterns over the past 10 years, 0 otherwise	0.76	0.42
d. Institutional factors
Extension	1, if producer contacted Extension agents in the past year, 0 otherwise	0.54	0.50
Training	1, if producer received agricultural training in the past year, 0 otherwise	0.44	0.48
Information	1, if producer received agricultural and weather information through TV/FM radio, weather applications on mobile phone, text-messages and publications, 0 otherwise	0.54	0.49

Triangulation analysis

The relationship among local climate trend data, producers’ perceptions, and survey responses are illustrated in the conceptual model of triangulation in Figure 2. Rice producers in Chitwan observed climate trends and noted changes in local climate trends over the past several decades, in particular the past 10 years. Through the survey, among many parameters, producers expressed their perceptions on the trends (increased, decreased, or no change) of average temperature, rainfall, dry periods, floods, hailstorms and monsoon onset. The trend analysis of temperature and rainfall data (2009–2018) of Rampur Chitwan supports producers’ perceptions of increasing temperature trends and variability in rainfall patterns. Piya et al. (2012) investigate local communities’ (Chepang) perceptions of climate change in Nepal's rural mid hills. They compared the community responses to the recorded temperature and rainfall trends to see if the perceived changes corresponded to the local climate trends. The survey data suggests that the farm practices were adversely affected by climate change in terms of delay in transplanting rice, reduced irrigation water availability, crop losses due to flooding, increased incidence of diseases/pests, more weeds and grain loss at harvest which in turn reduced the rice yield and net income from rice production. Through the survey, producers also provided their preferences on climate change adaptation practices including – varietal management (drought-tolerant, short-duration, pest-tolerant varieties), investment in improved irrigation practices (new canal construction, change in frequency of irrigation, installing solar water pumps and back-up generators) and use of DSR among other practices. Access to institutional factors such as Extension, training and information increases the likelihood of adopting improved farm practices. We explore complementary relationships among the three climate change adaptation practices that are influenced by the three institutional factors.

Figure 2. Conceptual model of triangulation and complementarity between climate change adaptations and institutional factors.

Complementarity

In this study, we examine how the complementary relationships among multiple climate change adaptation practices are influenced by institutional factors (Figure 2). Among the adaptation practices, we explore adoption of drought-tolerant and short-duration rice varieties, investment in improved irrigation practices such as new canals, irrigation timing as well as solar pumps and adoption of DSR practice. Included in the institution factors are Extension farm visit by agricultural agents, office visits by producers, climate-smart and adaptation training, accessing weather, technology and information sources. The socio-economic explanatory variables that influenced varietal, improved irrigation and DSR include gender, dependency ration, out-migration, number years of education, farming years of experience, income and membership of producers’ organizations. The farm characteristics include number of rice plots, amount of rice sold and area of rice under cultivation. The climate-related factors capture rice producers’ perceptions of climate change and variability.

Results and discussion

This section is organized as follows: first, the key findings of the temperature and precipitation trend analysis from Rampur are presented followed by rice producers’ perceptions to climate change and the strategies adapted as a response to climate change. Next, the multivariate probit results describe the complementarity relationship between climate change adaptations and institutional factors. The limitations of the study are noted and the reasons for producers not taking up adaptation practices are identified.

Weather trends, climate change and climate change adaptation

Between 1987 and 2016, Nepal experienced a trend of warming temperature and declining precipitation (Figures 3 and 4). Trend analysis of ten years of climatic data (2009–2018) of Rampur (Chitwan) indicated increases of 0.019°C/year and 0.003°C/year in maximum and minimum temperature, respectively (Figure 5). Following Piya et al. (2012), a linear trend analysis of maximum and minimum temperature during June–October, which is a rice growing season, between the period of 2009 and 2018 indicated that both maximum and minimum temperatures increased by 0.0255 and 0.0257°C/year at Rampur (Chitwan), respectively (Figure 6). Between 2009 and 2018, a linear trend analysis of precipitation at Rampur, Chitwan indicated a decline of 42.2 mm (or 1.66 in.) in annual rainfall and 34.6 mm (or 1.36 in.) in seasonal rainfall during June–October (Figure 7).

Figure 3. Average annual temperature trend of Nepal (1987–2016). Source: World Bank Group (2020).

Figure 4. Average annual precipitation trend of Nepal (1987–2016). Source: World Bank Group (2020).

Figure 5. Average annual temperature at Rampur, Chitwan (2009–2018). Source: Department of Hydrology and Meteorology, 2019, Kathmandu, Nepal.

Figure 6. Maximum, minimum and average seasonal temperature (June–October) at Rampur, Chitwan (2009–2018). Source: Department of Hydrology and Meteorology, 2019, Kathmandu, Nepal.

Figure 7. Annual and seasonal rainfall (June–October) at Rampur, Chitwan (2009–2018). Source: Department of Hydrology and Meteorology, 2019, Kathmandu, Nepal.

76.6% of rice producers in Chitwan perceived changes in local weather patterns in the past ten years. While 86.9% of producers noted higher summer temperatures, only 61.9% noticed higher temperatures in winter. More than three-fourths of rice producers indicated high intense rainfall and a late onset of monsoon rain but less than a third of producers were affected by floods. Almost all the producers indicated hailstorms and unpredictable weather in the past ten years (see Table 1).

Rice producers in the study area were found to be growing drought-tolerant and short-duration rice varieties. Out of 359, 233 (or 64.9%) producers adopted varietal management. During the household survey, producers reported that they replaced traditional, long-duration rice varieties: Aapjhutte, Anadi, Dhudhraj, Gurdi and Mansara. They also replaced Mansuli variety due to higher disease infestation. Improved varieties including Hardinath, Sukkhadhan-2, Sukkhadhan-4 and Ramdhan as well as hybrid varieties such as Garima, Gorakhanath and Loknath were the most popular varieties cultivated in the study area. Out of 359, 218 (or 60.7%) producers invested in improved water infrastructure and water conservation practices. Producers invested in new irrigation infrastructure like constructing new wells, installing electric motors to pump water, cleaning and maintenance of existing wells and canal construction for better water flow to fields. In some areas, producers also conserved water in lowlands (wetlands also called Ghol in Nepali). Although upland (tadhi in Nepali) may be an important factor in climate change adaptation, our sample did not have any producers growing rice in upland, all the producers cultivated rice in irrigated lowlands. Ridges (earthen structure) were constructed as barriers and the runoff water was re-utilized by pumping it back for irrigation. 56 of 359 (or 15.6%) producers adopted DSR practice. Producers may adopt DSR practice irrespective of the availability of water for land preparation and not depend on the onset of monsoon.

A total of 158 out of 359 (or 44%) producers received training related to climate change adaptation practices. Although local governmental organizations and private institutions organized training related to farming practices, more than half of producers did not have access to such training programmes. Access to Extension services was assessed by ascertaining if Extension agents contacted producers. More than half of producers (54%) had availed agricultural Extension services. The proximity of Extension office to producers may have influenced the participation of producers in training and producers’ access to Extension services. More than half of producers (54%) also had access to technical farming information and weather-related information. Access to technical farming information (including price and market information) and weather update and/or forecasts was assessed by whether producers have access to TV, FM radio, agricultural and weather applications on mobile phones and receiving a text message about farming information and weather updates on a mobile phone.

Information is found to be accurate by 73% of producers and 99% of the producers find it to be useful in adoption of climate change strategies. Over 95% of producers prefer television and FM radio for timely information, more than two-thirds of producers prefer Extension publications as a reliable source of information. Less than one-third of producers prefer mobile applications for weather information and less than 9% of producers rely on internet for weather information.

Multivariate probit results

The Wald test of the three equations of the multivariate probit model (Wald chi², (42) = 315.30; Prob >${\chi }^2$ = 0.000) is highly significant and rejects the null hypothesis that all regression coefficients are jointly equal to zero, and showing that the model fits the data well. We test the complementary relationship among multiple climate change adaptation practices by using pair-wise correlation coefficients across the residuals of the three equations of the multivariate probit model. The likelihood ratio test (${\chi }^2$ (3)  = 29.38, Prob > ${\chi }^2$ = 0.000) rejects the null hypothesis that the covariance of the residuals of the multivariate probit model equations is equal to zero. Results indicate that all three pairs of residuals were positively correlated (0.617 for variety and irrigation; 0.241 for variety and DSR and 0.085 for irrigation and DSR), suggesting that the three climate change adaptation practices are complementary to each other. The variety and the irrigation practices have the highest correlation, indicating that producers tend to jointly adopt both practices, ceteris paribus. This result has important policy implications that among variety and irrigation, enhancement of adoption of either practice would increase the adoption of the other practice because these practices tend to be adopted together.

The parameter estimates of the three equations of the multivariate probit model are presented in Table 5. The results indicate that producers are 15.58% (P < 0.01), 13.14% (P < 0.05) and 9.37% (P < 0.10) more likely to adopt variety if they have access to Extension, training and information, respectively. Imparting knowledge and skills through education, training and information influence producers to use improved rice varieties. These findings align with previous studies where access to Extension services was found to increase the likelihood of adoption of climate change adaptation practices (Deressa et al., 2009; Mulwa et al., 2017; Nhemachena & Hassan, 2007). Access to Extension and training significantly increases the likelihood of adoption of improved irrigation practices by 9.37% (P < 0.05) and 8.77% (P < 0.10), respectively. Gaining more knowledge and awareness through education, training and learning from farming activities encouraged producers to invest in improved irrigation practices resulting in higher rice yield. This result confirms the finding of Aryal et al. (2018), who reported that access to agricultural training enhanced the likelihood of adoption of climate-change adaptation practices. Similarly, having access to information related to weather and improved farming technique significantly increases the likelihood of adoption of DSR practice by 10.35%, which is consistent with the earlier findings of Mulwa et al. (2017) and Aryal et al. (2018). Both of these studies indicated that availability of weather information and information on improved farming practices significantly enhance the adoption of multiple climate change adaptation practices.

Table 5. Empirical results of the multivariate probit model.

Variable	Variety		Improved irrigation		Direct-seeded rice
	Coeff.	M.E.	Coeff.	M.E.	Coeff.	M.E.
a. Household and socio-economic characteristics
Gender	−0.118	−2.69%	−0.251	−6.12%	−0.307	−4.87%
	(0.284)		(0.192)		(0.282)
Dependency ratio	−0.068	−1.57%	−0.147	−3.57%	0.236*	3.74%*
	(0.147)		(0.107)		(0.127)
Out-migration	−0.326*	−7.43%*	−0.104	−2.54%	0.187	2.98%
	(0.190)		(0.171)		(0.186)
Education year	0.163***	3.71%***	0.113***	2.75%***	0.036	0.57%
	(0.033)		(0.027)		(0.031)
Farming year	0.013	0.31%	0.020**	0.49%**	−0.022**	−0.35%**
	(0.009)		(0.009)		(0.009)
Income	0.000	0.02%	0.001**	0.03%**	0.001*	0.02%*
	(0.001)		(0.000)		(0.000)
Membership	−0.392	8.92%	0.619***	15.05%***	0.265	4.20%
	(0.252)		(0.240)		(0.356)
b. Farm characteristics
Rice plot	0.223*	5.09%*	0.023	0.57%	−0.048	−0.76%
	(0.124)		(0.140)		(0.131)
Rice sold	0.325	7.39%	0.450*	10.94%*	−0.386	−6.12%
	(0.237)		(0.245)		(0.304)
Rice area	−0.708	−16.11%	−0.388	−9.44%	0.704**	10.16%**
	(0.433)		(0.332)		(0.334)
c. Climate-related factors
Climate	0.669**	15.20%**	0.699***	17.00%***	0.629*	9.96%*
	(0.272)		(0.243)		(0.418)
d. Institutional factors
Extension	0.685***	15.58%***	0.385**	9.37%**	0.383	6.08%
	(0.206)		(0.190)		(0.233)
Training	0.578**	13.14%**	0.360*	8.77%*	0.167	2.64%
	(0.235)		(0.199)		(0.204)
Information	0.412*	9.37%*	−0.052	−1.27%	0.653**	10.35%**
	(0.224)		(0.260)		(0.284)
Constant	2.915***		−2.713***		−2.733***
	(0.478)		(0.450)		(0.503)

Log-likelihood ratio ${\chi }^2$(24) = 319.010; Prob >= 0.000.

Number of observations = 359.

Log-likelihood = −330.690.

***, ** and * refer to the significance at 1%, 5% and 10% levels, respectively, numbers in parenthesis indicate the robust standard error, Coeff. denotes parameter coefficients and M.E. denotes marginal effects.

As compared with the baseline predicted likelihood, a 10% increase in the rice producers with access to Extension results in the highest increment in the likelihood of adoption of variety (1.38%) followed by training (1.16%) and information (0.84%) (see Table 6). Similarly, a 10% increasein the rice producers with access to Extension results in the highest increment in the likelihood of the adoption of irrigation (0.89%), followed by training (0.88%). In addition, a 10% increase in the rice producers with training results in the highest increments in the likelihood of adoption of DSR (1.24%), followed by information (0.61%) and Extension (0.50%).

Table 6. Predicted likelihood of adoption of multiple climate change adaptation practices by hypothetically increasing institutional factors and pairwise correlation coefficients between the predicted likelihood.

	Predicted likelihood of adoption of multiple climate change adaptation practices			Pairwise correlation coefficients between the predicted likelihood
	Variety	Irrigation	DSR	Variety & Irrigation	Variety & DSR	Irrigation & DSR
Baseline	63.93%	60.45%	15.73%	–	–	–
10% increase in Extension service	65.31%	61.34%	16.23%	0.955***	0.529***	0.631***
10% increase in training related to farming practices	65.09%	61.33%	16.97%	0.942***	0.469***	0.656***
10% increase in farming practices and weather-related information	64.77%	60.33%	16.34%	–	0.602***	–

Baseline refers to the predicted likelihood of climate change adaptation practices based on the results of the three equations of the multivariate probit model.

***, ** and * refers to the significance at 1%, 5% and 10% levels, respectively.

The pairwise correlation coefficients between the predicted likelihood of adoption of multiple climate change adaptation practices across variety and irrigation; variety and DSR and irrigation and DSR are 0.955, 0.529 and 0.631, respectively in the case of increasing access to Extension (see Table 6). Similarly, the pairwise correlation coefficients between the predicted likelihood of adoption of multiple climate change adaptation practices across variety and irrigation; variety and DSR and irrigation and DSR are 0.942, 0.469 and 0.656 respectively in the case of increasing training. The pairwise correlation coefficients between the predicted likelihood of adoption of multiple climate change adaptation practices across variety and DSR is 0.602 in the case of increasing information. These findings suggest that effective Extension and farmer’s training can enhance the complementarities among all three pairs of climate change adaptation practices whereas an increase in information enhances the complementarity between variety and DSR. Results of sensitivity test using 20%, 30% and 40% increase of each of three institutional factors follow similar patterns in terms of magnitude and sign of the institutional factors as in the case of 10% increase (see Appendix 1 for details).

The effects of other variables, including socio-economic, farm characteristics and climate-related variables, on the climate change adaptation practices are also presented in Table 5. 45.9% of households had at least one family member migrate to another country for employment. Households with members who out-migrated are 7.43% less likely to adopt variety as a climate change adaptation practice. This might be due to the shortage of household labour in the age group of 16–60 years of age who typically out-migrate for employment.

The average years of farming experience among rice producers in Chitwan was 26 years. Contrary to our expectations, farming experience is less likely (less by 0.35%) to influence DSR adoption. This might be because farmers with more years of experience are relatively older, and it is likely that older producers are reluctant to change and adopt newer climate change adaptation practices. The direction of influence of dependency ratio and rice area is positive in case of adoption of DSR practice, which is realistic that DSR practices require less labour compared to the practice of transplanted rice. International Rice Research Institute (2018) also reported that DSR is relatively less labour-intensive practice than conventional transplanting method. Furthermore, Membership of producers’ organizations enhances the likelihood of adoption of improved irrigation. 68.5% of producers are members of a farmer’s group or a co-operative. Membership of producers’ organizations enhance the likelihood of adoption of improved irrigation by 15.05%. Being a member of a producer’s organization helps to pool available individual resources, transfer technology to peers and influence other producers in the adoption of climate change adaptation practices. Producers having a greater number of rice plots are more likely to adopt variety. Furthermore, rice selling households are more likely to adopt improved irrigation. If producers perceived variability in local weather patterns over the years, they are more likely to adopt varietal selection, invest in improved irrigation practices and adoption of DSR practice. Consistent with existing literature (Deressa et al., 2009; Khanal et al., 2018; Mulwa et al., 2017), results reveal that an increase in the number of years of education of the head of the household, the likelihood of adoption would be enhanced. These results highlight the benefits of better access to education to enhance the uptake of multiple adaptation practices.

A limitation of this study is that we do not directly measure awareness of climate change adaptation strategies. However, survey data indicates the reason producers do not adopt a particular adaptation practice. The survey responses indicate that among non-adopters, 41% of producers did not adopt DSR due to lack of information about the practice. Similarly, 41.1% of producers did not adopt variety due to lack of information while 59.6% of producers did not adopt improved irrigation practices because of lack of funds.

Summary and conclusions

This paper examines how the complementary relationships among multiple climate change adaptation practices are influenced by access to Extension services, training related to farming practices, farming practices and weather-related information (referred to as ‘institutional factors’). Our study examines how the complementary relationships among multiple climate change adaptation strategies (improved varieties, irrigation practices and DSR) are influenced by access to Extension services, training related to farming practices and weather-related information (referred to as ‘institutional factors’). At a macro-level, our study provides valuable information to policymakers on prioritizing limited resources to support and improve specific services to rice producers – Extension, training and information that are effective in the implementation of climate change adaptation strategies.

As a case study, we develop an empirical model that hypothesizes the institutional factors on the adoption of variety, irrigation and DSR, and the resulting effects on their joint adoption using survey data collected from 359 rice producers in the Chitwan district of Nepal in 2019. The results reveal that: (1) adoption of all the three climate change adaptation practices is mutually complementary; (2) all three institutional factors enhance the likelihood of the adoption of variety, access to Extension and training enhances the likelihood of the adoption of improved irrigation and information enhances the likelihood of the adoption of DSR and (3) access to Extension and training enhance the complementarities among all three pairs of adaptation practices (variety and irrigation; variety and DSR; and irrigation and DSR), while information enhances the complementarity between variety and DSR only.

On the basis of the findings summarized above, local government agencies can prioritize the investment on the institutional factors that help rice producer’s adoption of multiple climate change adaptation practices. The result (1) suggests that the three climate change adaptation practices need to be adopted in an organized manner so that the adoption can be done more effectively. For example, when rice producers adopt variety and irrigation practices jointly, they can coordinate their efforts, minimize redundancies and achieve efficient outcomes. The result (2) suggests that it is equally worthy to allocate resources to improve the likelihood of the adoption of the three climate change adaptation practices through Extension agents’ work on technical services, climate change adaptation training programme by the government and non-government organizations and information in the form of factsheets, brochures, publications, booklets and progress reports. The result (3) suggests that investment on improving access to Extension and training is potentially more efficient than improving information accessibility because the former two institutional factors trigger the complementarities among all three pairs of climate change adaptation practices, while the latter institutional factor prompts the complementarity between variety and DSR only. The evaluation of the relative degree of influence of complementary adaptation measures is beyond the scope of this study and can be explored in future studies.

Given the results, improving access to Extension and training is what local government agencies can focus for the uptake of climate change adaptation practices. For example, enhancing the scope of the Community Agricultural Extension Service Center (CAESC) as envisioned by Agricultural Development Strategy (ADS) of Nepal may be the primary target area for investment that aims to improve rice producers’ adoption of climate change adaptation practices. Furthermore, along with regular Extension programmes, local government agencies can develop and implement producers’ capacity-building programme. For example, farmer field school approach, which is a group-based learning approach that teaches how to experiment and solve problems independently (Dhamankar & Wongtschowski, 2014) may help producers to improve awareness of climate change and adoption of multiple climate change adaptation practices.

The policy recommendation comes with a caveat because our framework overlooks potential costs associated with improving institutional factors. Though exact cost estimates for improving each of the institutional factors were not available at the time of our survey, we acknowledge that the cost of improving Extension and training programme is generally higher than the cost of improving information in our study area (Dhital, 2020). For example, improving information through publishing brochures, factsheets on improved agricultural practices, varieties, weather, price and marketing information is relatively cheaper than developing Extension or training programme. Such information is critical in that improving access to Extension and training to take advantage of their resulting complementary relationships among multiple climate change adaptation practices needs to be carefully evaluated after accounting for the comparisons of the costs associated with them.

To provide more comprehensive recommendations to local government agencies for prioritizing the investment in the institutional factors, future research could explore the complementary relationships among multiple climate change adaptation practices triggered by institutional factors that also consider their costs. Such efforts require cost information that can be incorporated in our three equations of the multivariate probit model. It may be challenging to normalize the cost data that fits in the model because the unit costs for improving access to Extension, training and information are not standardized. Once the normalized cost data that fits in the model is acquired, the variables that capture the interactions between institutional factors and their per unit costs can be included in the three equations, which allows to sort out how the complementary relationships among multiple climate change adaptation practices are influenced by the institutional factors condition upon their costs. This study could also be extended to evaluate how producers would adopt DSR, improved varieties and irrigation practices if governmental funds were provided to support these climate change adaptation strategies.

Acknowledgements

We acknowledge the support and cooperation of rice producers in Chitwan, Nepal.

Disclosure statement

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

Additional information

Funding

We are grateful to National Institute for Mathematical and Biological Synthesis (NIMBioS), University of Tennessee, Knoxville for providing financial support for travel and to conduct the field survey in Chitwan, Nepal.

Notes on contributors

Hari P. Regmi

Hari P. Regmi is a PhD candidate at the Department of Agricultural Economics, Purdue University, West Lafayette, Indiana, USA. His research areas include agricultural policy; forecast evaluation related to agricultural policy; and climate change adaptation.

Sreedhar Upendram

Dr. Sreedhar Upendram is an Assistant Professor in the Department of Agricultural and Resource Economics at the University of Tennessee. His primary areas of research are community economics, economic impact analysis, rural development and natural resource economics. He serves as the statewide Community Economic Development Specialist for the University of Tennessee Extension.

Seong-Hoon Cho

Dr. Seong-Hoon Cho is a Professor at the Department of Agricultural &Resource Economics, University of Tennessee Institute of Agriculture.His main research focuses on promoting sustainability of agriculturaland forested lands by examining economic issues related to land use,ecosystem service provision, and conservation science usingmicroeconomics and econometrics.

Christopher D. Clark

Dr. Christopher Clark is currently serving as Professor and Head of the Department of Agricultural & Resource Economics at the University of Tennessee. His research interests include a variety of topics in environmental and natural resource economics. He teaches graduate and undergraduate courses in natural resource economics and an undergraduate course in agricultural and environmental law. He earned a Ph.D. in Economics from Vanderbilt University and a J.D. from Tulane University.

Appendices

Appendix 1

Results of sensitivity test using 20%, 30% and 40% increase of each of institutional factors.

Table

	Predicted likelihood of adoption of multiple climate change adaptation practices			Pairwise correlation coefficients between the predicted likelihood
	Variety	Irrigation	DSR	Variety & Irrigation	Variety & DSR	Irrigation & DSR
A. 20% Increase case
Baseline	63.93%	60.45%	15.73%
20% increase in Extension service	66.83%	62.26%	16.66%	0.958***	0.521***	0.64***
20% increase in training related to farming practices	66.31%	62.19%	16.21%	0.950***	0.467***	0.604***
20% increase in farming practices and weather-related information	65.59%	60.23%	16.92%		0.595***
B. 30% increase case
Baseline	63.93%	60.45%	15.73%
30% increase in Extension service	68.16%	63.07%	17.10%	0.948***	0.531***	0.658***
30% increase in training related to farming practices	67.48%	63.04%	16.46%	0.937***	0.448***	0.585***
30% increase in farming practices and weather-related information	66.40%	60.10%	17.50%		0.571***
C. 40% increase case
Baseline	63.93%	60.45%	15.73%
40% increase in Extension service	69.54%	63.98%	17.50%	0.947***	0.483***	0.609***
40% increase in training related to farming practices	68.50%	63.80%	16.70%	0.934***	0.394***	0.532***
40% increase in farming practices and weather-related information	67.30%	60.00%	18.20%		0.536***

Baseline refers to the predicted likelihood of climate change adaptation practices based on the results of the multivariate probit model.

***, ** and * refers to the significance at 1%, 5% and 10% levels, respectively.

Appendix 2

Correlation between institutional factors and multicollinearity issues

To verify if the three institutional factors (Extension, Training and Information) are positively correlated and thus cause multicollinearity, which compromises the reliability of model estimates, we first estimate the correlation coefficients among Extension (referred as ‘EXT’ in the table), Training and Information below.

Table A1. Correlation matrix of institutional factors.

	Extension	Training	Information
Extension	1
Training	0.486	1
Information	0.489	0.448	1

In general, an absolute correlation coefficient greater than 0.7 among two or more independent variables indicates the presence of multicollinearity, which we do not find among the three institutional variables. To reassure the absence of the multicollinearity, we calculated a conditional index value following Belsley (1991) and variance inflation factor (VIF). We did not find evidence of multicollinearity because condition index value is 9.48, which is not above the threshold level 30, and the mean VIF is 4.4 as shown in the table below. The most common rule of thumb is that VIFs exceeding 10 are signs of serious multicollinearity needing correction (Marquardt, 1970; Mason et al., 1989).

Table A2. Variation inflation factors for variables.

Variables	VIF	1/VIF
Gender	1.42	0.702261
Dependency ratio	1.69	0.592206
Out-migration	1.89	0.529518
Education	7.19	0.139004
Farming years	5.44	0.183856
Income	3.75	0.266948
Membership	6.33	0.158096
Rice plots	5.8	0.172548
Rice sold	6.14	0.162924
Rice area	4.64	0.215589
Climate	6.69	0.14948
Extension	3.41	0.293179
Training	2.7	0.37052
Information	4.65	0.215014
Mean VIF	4.41

Table A3. Conditional index values for variables.

Condition number using scaled variables = 9.48
Condition indexes and variance-decomposition proportions
SN	Condition index	Gender	Dependancy ratio	Out-migration	Educationyears	Farming year	Income	Rice plot	Rice sold	Rice area	Membership	Climate	Extension	Training	Information
1	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0
2	2.74	0.25	0.11	0.02	0	0	0	0	0	0	0	0	0.01	0.02	0.01
3	4.01	0.14	0.09	0.41	0	0	0.01	0	0	0.01	0	0	0.05	0.08	0.01
4	4.08	0.44	0.42	0.08	0	0	0	0	0	0.02	0	0	0	0.03	0
5	4.48	0.07	0.21	0.34	0	0	0.02	0.02	0.01	0.07	0	0	0.02	0.14	0
6	5.46	0.02	0.07	0.01	0.03	0.06	0	0.04	0	0.06	0.06	0.03	0.02	0.27	0.12
7	6.09	0	0.01	0.01	0.01	0	0.32	0	0.02	0.01	0.03	0.01	0.49	0.21	0
8	6.27	0	0.03	0.04	0	0	0.56	0	0.02	0.04	0	0.02	0.33	0.12	0
9	7.11	0	0.05	0.06	0.01	0.43	0.02	0.01	0	0.22	0	0.14	0.01	0.09	0.08
10	7.6	0.01	0	0.01	0	0.05	0.03	0.27	0.24	0.01	0.09	0	0.01	0.01	0.41
11	7.99	0.01	0	0	0.23	0.12	0	0.11	0.28	0	0.03	0.03	0.02	0.01	0.33
12	8.55	0.04	0	0.01	0.01	0.01	0.02	0.39	0.14	0.49	0.15	0.15	0.01	0	0.02
13	9	0.02	0	0	0.08	0.23	0.01	0.04	0.28	0.05	0.6	0.09	0	0.02	0
14	9.48	0	0	0	0.63	0.09	0.01	0.1	0	0	0.03	0.52	0.03	0	0.01

Appendix 3

Instrumental variables and testing for endogeneity

Rivers and Voung (1988) discuss potential endogeneity issues in the limited dependent variable model. Aryal et al. (2018) follow Rivers and Voung (1988) approach to deal with potential endogeneity issues in the multivariate probit model. We follow the similar approach in our study. Following Rivers and Voung (1988) and Aryal et al. (2018), we first specified the potential endogenous variables (Extension, Training and Information) as a function of instruments and controls variables. Then we use residuals of the first stage regression as an additional control variable in the model to acquire consistent estimates. The first stage regression results indicate that all three instruments are significant in the model (see output screenshots), hence we believe that our instruments are reliable and not weak. If the residuals of the first stage regression are not significant, we do not have an issue of endogeneity (Aryal et al., 2018). The residuals of all three potential endogenous variables are not significant in the model and hence we report coefficients without residual terms.

First stage regression results for potential endogenous variables are presented below.

Table A4. Extension: instrument is ethnicity.

Source	SS	df	MS	Number of obs	=	359
				F(14, 344)	=	8.56
Model	23.0146839	14	1.643906	Prob > F	=	0
Residual	66.066096	344	0.192053	R−squared	=	0.2584
				Adj R−squared	=	0.2282
Total	89.0807799	358	0.248829	Root MSE	=	0.43824
	Coefficient	Std. err.	t	P > t	[95% conf. interval]
Gender	−0.102277	0.063947	−1.6	0.111	−0.2280524	0.023498
Dependency ratio	0.0152228	0.037261	0.41	0.683	−0.0580655	0.088511
Out-migration	−0.0914561	0.048255	−1.9	0.059	−0.1863681	0.003456
Education	0.0173144	0.007397	2.34	0.02	0.0027647	0.031864
Farming years	0.0050328	0.002198	2.29	0.023	0.0007105	0.009355
Income	0.0003402	0.000158	2.15	0.032	0.0000296	0.000651
Membership	0.2791628	0.065422	4.27	0	0.1504861	0.407839
Rice plots	0.008055	0.032028	0.25	0.802	−0.0549405	0.071051
Rice sold	0.0739542	0.06612	1.12	0.264	−0.0560958	0.204004
Rice area	−0.1321851	0.090535	−1.46	0.145	−0.3102576	0.045887
Climate	0.0215655	0.068764	0.31	0.754	−0.1136852	0.156816
Ethnicity	0.0897398	0.051273	1.75	0.081	−0.0111072	0.190587
Risk averse	0.2795078	0.151108	1.85	0.065	−0.0177038	0.576719
Literate members	0.0014804	0.017701	0.08	0.933	−0.0333354	0.036296
Constant	−0.2983857	0.182842	−1.63	0.104	−0.6580147	0.061243

Table A5. Training: Instrument is risk-averse producers.

Source	SS	df	MS	Number of obs	=	3 59
				F(14, 344)	=	8 .03
Model	21.785618	14	1.556116	Prob > F	=	0
Residual	66.6767776	344	0.193828	R−squared	=	0 .2463
				Adj R−squared	=	0 .2156
Total	88.4623955	358	0.247102	Root MSE	=	0 .44026
	Coefficient	Std. err.	t	P > t	[95% conf. interval]
Gender	−0.0701564	0.064241	−1.09	0.276	−0.19651	0.056199
Dependency ratio	−0.0245381	0.037433	−0.66	0.513	−0.09816	0.049088
Out-migration	−0.0350346	0.048478	−0.72	0.47	−0.13038	0.060315
Education	0.0113768	0.007432	1.53	0.127	−0.00324	0.025994
Farming years	0.0077157	0.002208	3.49	0.001	0.003374	0.012058
Income	0.0001923	0.000159	1.21	0.226	−0.00012	0.000504
Membership	0.3006126	0.065723	4.57	0	0.171343	0.429883
Rice plots	0.0144904	0.032176	0.45	0.653	−0.0488	0.077776
Rice sold	0.0490551	0.066425	0.74	0.461	−0.08159	0.179705
Rice area	−0.2217237	0.090953	−2.44	0.015	−0.40062	−0.04283
Climate	0.1083801	0.069081	1.57	0.118	−0.02749	0.244254
Ethnicity	0.0047764	0.051509	0.09	0.926	−0.09654	0.106088
Risk averse	0.2536526	0.151805	1.67	0.096	−0.04493	0.552235
Literate members	0.0085106	0.017783	0.48	0.633	−0.02647	0.043487
Constant	−0.3819974	0.183685	−2.08	0.038	−0.74328	−0.02071

Table A6. Information: instrument is the literacy rate in the family.

Source	SS	df	MS	Number of obs	=	3 59
				F(14, 344)	=	2 4.34
Model	44.2866983	14	3.163336	Prob > F	=	0
Residual	44.7049451	344	0.129956	R-squared	=	0 .4977
				Adj R-squared	=	0 .4772
Total	88.9916435	358	0.24858	Root MSE	=	0 .36049
	Coefficient	Std. err.	t	P > t	[95% conf. interval]
Gender	−0.0840612	0.052602	−1.6	0.111	−0.18752	0.019402
Dependency ratio	−0.053047	0.030651	−1.73	0.084	−0.11333	0.00724
Out-migration	−0.0828835	0.039695	−2.09	0.038	−0.16096	−0.00481
Education	0.0174543	0.006085	2.87	0.004	0.005486	0.029423
Farming years	−0.0002814	0.001808	−0.16	0.876	−0.00384	0.003274
Income	0.0002853	0.00013	2.2	0.029	2.98E−05	0.000541
Membership	0.3471369	0.053816	6.45	0	0.241288	0.452986
Rice plots	0.021845	0.026346	0.83	0.408	−0.02998	0.073665
Rice sold	0.0789426	0.05439	1.45	0.148	−0.02804	0.185922
Rice area	−0.1338541	0.074474	−1.8	0.073	−0.28034	0.012628
Climate	0.195289	0.056565	3.45	0.001	0.084032	0.306546
Ethnicity	0.0887778	0.042177	2.1	0.036	0.005821	0.171735
Risk averse	0.0170403	0.124301	0.14	0.891	−0.22745	0.261526
Literate members	0.0349844	0.014561	2.4	0.017	0.006345	0.063624
Constant	−0.1916093	0.150406	−1.27	0.204	−0.48744	0.104221