2.1 Abiotic stress and climate-smart varieties

In the face of changing climate, drought and submergence are two major abiotic stresses which significantly constrain rice production in India. Out of nearly 20 million hectares (Mha) of rainfed rice in India, about 14 Mha are prone to drought (Arora et al., Reference Arora, Bansal and Ward2015) and the value of rice production lost in drought years has been estimated to be as high as 36 per cent of the total value of rice production in eastern India, costing several hundred million dollars per year (Pandey et al., Reference Pandey, Wang, Bhandari, Johnson, Haefele and Hardy2012). On average, 10–12 Mha of rice area are flood affected every year, causing losses to human life, property, forests and crops (NRAA, 2013). Flash floods can occur at any stage of crop growth and can damage the crop completely. Consequently, rural poverty and food insecurity are persistent in those rainfed and flood-prone rice production areas. About 30 per cent of the 700 million people living in absolute poverty in Asia are from rainfed rice-growing areas of South Asia, and half of them live in India, Bangladesh and Nepal. They live in the rainfed areas that are prone to abiotic stresses (Ismail et al., Reference Ismail, Singh, Singh, Dar and Mackill2013).

Green revolution rice technologies have played a critical role in improving food security and reducing poverty in many developing Asian countries (Evenson and Gollin, Reference Evenson and Gollin2003; Hazell, Reference Hazell2010). In the early stage of the green revolution, many technologies were developed and disseminated to the irrigated and favorable rice environments. But they have largely avoided unfavorable growing environments such as areas affected by droughts, floods, salinity, soil toxicity and nutrient deficiencies, resulting in low and uncertain yields in eastern India (Khush, Reference Khush1990; Samal et al., Reference Samal, Pandey, Kumar and Barah2011).

To counteract flood risk, the IRRI developed SS1, a submergence-tolerant rice variety, through marker-assisted backcrossing of the Sub1 QTL (quantified trait locus) from Indian rice cultivar FR13A into the most popular Indian variety, Swarna (Neeraja et al., Reference Neeraja, Maghirang-Rodriguez, Pamplona, Heuer, Collard and Septiningsih2007; Septiningsih et al., Reference Septiningsih, Pamplona, Sanchez, Neeraja, Vergara, Heuer, Ismail and Mackill2009). SS1 can survive up to 14 days of full submergence and, under normal conditions, studies find no significant differences in agronomic performance, grain yield and grain quality between Swarna and SS1 (Sarkar et al., Reference Sarkar, Reddy, Sharma and Ismail2006; Neeraja et al., Reference Neeraja, Maghirang-Rodriguez, Pamplona, Heuer, Collard and Septiningsih2007). SS1, however, shows a twofold or higher yield advantage over Swarna after submergence for 10 days or more during the vegetative stage (Septiningsih et al., Reference Septiningsih, Pamplona, Sanchez, Neeraja, Vergara, Heuer, Ismail and Mackill2009). SS1 has been distributed in eastern India by IRRI under the Stress-Tolerant Rice for Africa and South Asia Project and its collaborators. Starting in 2010, seed distribution was significantly expanded when the National Food Security Mission included STRV in its eastern India programs.

2.2 Information, technology adoption and economic impacts

The agri-food system has gone through major changes in India, which makes agricultural knowledge and information intensive (Birthal et al., Reference Birthal, Kumar, Negi and Roy2015). Many new technologies were introduced rapidly and knowledge transfer in agriculture is generally not at the expected pace, especially in the case of small and marginal farmers (Raghu et al., Reference Raghu, Manaloor and Nambi2014). Several studies have shown that access to information or extension services significantly increased the probability of adoption of agricultural technologies such as improved varieties, which in turn increased crop yields (Matuschke et al., Reference Matuschke, Mishra and Qaim2007; Shiferaw et al., Reference Shiferaw, Kebede, Kassie and Fisher2015; Ainembabazi et al., Reference Ainembabazi, Asten, Vanlauwe, Ouma, Blomme, Birachi, Nguezet, Mignouna and Manyong2016; Wossen et al., Reference Wossen, Abdoulaye, Alene, Haile, Feleke, Olanrewaju and Manyong2017).

Varma (Reference Varma2018) analyzed the role of access to information in the adoption of System of Rice Intensification (SRI) technology in India, where 96 per cent of SRI adopters received information whereas only 56 per cent of SRI non-adopters received information. The role of information is critical, especially when a technology is new to an area; and a study done by Keil et al. (Reference Keil, D'Souza and McDonald2017) showed that access to extension service was significantly associated with the awareness and adoption of zero tillage technology in the Eastern Indo-Gangetic Plain. Although information on technology plays a vital role, having access alone will not influence adoption behavior, as the quality of the information will also influence behavior, especially in the case of knowledge-intensive technologies.

Yokouchi and Saito (Reference Yokouchi and Saito2016) found that about 20 per cent of farmers stopped growing new rice varieties in Africa due to the lack of appropriate information on the varietal characteristics and farming practices of the respective varieties. In recent years, farmers have expected to obtain information on weather forecasts and advisories for agricultural inputs, agronomic practices, pest management, markets and prices (Aker, Reference Aker2011). Farmers' exposure to risk and uncertainty is often aggravated by the lack of information about weather, inputs, farm management practices or market prices, which adversely affects crop production and income (Mittal, Reference Mittal2012; Mittal and Mehar, Reference Mittal and Mehar2012).

In addition to information related to new technologies, farmers value other supplementary information when making adoption decisions. There is a significant and positive association between market information and the adoption of new rice varieties in Pakistan, and a study highlighted the fact that farmers need up-to-date information regarding inputs, new technology, developed and released improved varieties, prices and new agronomic practices (Chandio and Yuansheng, Reference Chandio and Yuansheng2018). Apart from information access related to crop varieties, access to seasonal climate forecast information influenced planting/harvesting of the crop and the adoption of improved crop varieties in Africa and South Asia (Wood et al., Reference Wood, Jina, Jain, Kristjanson and DeFries2014). Studies by Aker (Reference Aker2011) and Mittal and Mehar (Reference Mittal and Mehar2015) indicated that access to market information had a positive influence on the adoption of new seed technologies, and increased crop productivity and the livelihood of the farmers. The major barriers in adopting drought-tolerant maize in eastern and southern Africa are the unavailability of improved seed, inadequate information and resources, and farmers' perceptions of variety attributes (Fisher et al., Reference Fisher, Abate, Lunduka, Asnake, Alemayehu and Madulu2015).

Besides access to information, the quality of and trust in the information are important criteria in adoption decisions. A study by Ward and Pede (Reference Ward and Pede2015) reveals that farmers' decisions to adopt hybrid rice technology are based on frequent interaction with peers about their learning and experience with those varieties, and they directly link the quality of and trust in the information. Information reach through formal agricultural extension was found to be slow. Strengthening agricultural extension services and improving the skill of extension officers in supplying good-quality information minimizes the risk in adoption due to lack of trust and incomplete information transfer (Beyene and Kassie, Reference Beyene and Kassie2015). Small farmers in various countries have indicated a willingness to pay for extension services that meet their needs (Gautam, Reference Gautam2000; Holloway and Ehui, Reference Holloway and Ehui2001), showing that information quality is central for technology adoption and dissemination.

The impact of information (or extension access) on agricultural productivity, income and household welfare was analyzed across different studies earlier (Owens et al., Reference Owens, Hoddinott and Kinsey2003; Dinar et al., Reference Dinar, Karagiannis and Tzouvelekas2007; Birthal et al., Reference Birthal, Kumar, Negi and Roy2015; Wossen et al., Reference Wossen, Abdoulaye, Alene, Haile, Feleke, Olanrewaju and Manyong2017). Some of the studies estimated direct impacts whereas few of them addressed impact through the technology adoption pathway. A study on agricultural extension programs for smallholder women farmers in Uganda showed that technology adoption leads to improved food security for farmers and better shock-coping methods (Pan et al., Reference Pan, Smith and Sulaiman2018). Birthal et al. (Reference Birthal, Kumar, Negi and Roy2015) estimated an enhancement of net return in farming by 12 per cent through information averaging across different cropping systems. Farmers who adopted stress-tolerant varieties could mitigate yield loss under stress conditions and reported an increase in rice yield of 15.5 per cent (Jie-hong et al., Reference Jie-hong, Li-qun and Yu2018). Khatri-Chhetri et al. (Reference Khatri-Chhetri, Aryal, Sapkota and Khurana2016) found a significant increase in the adoption of various climate-smart practices and technologies in target areas, and this resulted in substantial economic benefits for smallholder farmers.

This article contributes to adoption literature that addresses the role of information, particularly the importance of good quality information in driving technology adoption decisions. Although several studies exist on the role of information in the adoption of technology, few studies depart from equating information access to complete information. Among those few studies, our study simultaneously considered information content (as a quality indicator) along with information access as driving factors in adoption decisions. Second, we evaluated the adoption process in a stage-wise approach, distinguishing between informed choices and non-informed choices. Such informed choices lead to sustainable adoption, which gives policymakers insight into the pathway for better dissemination of technology using information platforms.

The remainder of the article is organized as follows. Section 3 briefly describes the study region and sampling procedure. Section 4 discusses the analytical approach to estimating the adoption of SS1 using a sequential logit framework, and then develops an empirical model to estimate the effect of information on STRV adoption. We present and discuss the main analytical results in section 5, consisting of the performance of submergence-tolerant rice varieties during normal and flood conditions, details on information on STRVs, the sequential adoption model and the projected impact of SS1 under different information scenarios. Section 6 concludes the study, drawing major findings and policy implications.

3.1 Study area

This study was conducted in three eastern Indian states: Assam, Odisha and West Bengal. Eastern India contributes nearly one-third of the country's total rice area and production. Table A1 in the online appendix illustrates the importance of the study area vis-à-vis the Indian rice scenario. The kharif (wet) season in these three states contributes one-fourth of the total rice area and production. As noted from table A1, approximately 30 per cent of the total rice area in these three states experienced flood in the 2015 kharif season.

3.2 Sampling and data

A primary survey was conducted in three eastern Indian states: Assam, Odisha and West Bengal. The flood-affected (pertaining to wet season, 2015) rice areas were identified using remote-sensing information (see figure A1 in the online appendix) and accordingly the list of villages was prepared. A total of 475 villages (155 in Assam and 160 each in Odisha and West Bengal) from 41 districts (19 in Assam, 13 in West Bengal and 9 in Odisha) were selected for this study. A census of all these villages was carried out and, from the census information, a list of all rice farmers was prepared. Ten rice farmers were randomly selected from each village. The final sample comprises 4,744 rice-farming households from three eastern states (see table A2, online appendix).

A comprehensive household questionnaire was developed using the Surveybe computer-assisted personal interview (CAPI) software, and data collection started immediately after the wet season harvest (December 2015 to April 2016). The questionnaire contains different modules pertaining to household and socioeconomic conditions, stress occurrence, SS1 adoption, incremental gain due to SS1 adoption, household spending and so forth.

Census data was collected from 4,744 households belonging to 475 villages (table A2). Because of some technical and logistical issues, a few sampled households/villages had to be excluded from the final analysis. The final sample used for the analysis contained data from 475 villages and 4,698 households. From the sampled households, it is evident that Odisha has a higher level of SS1 adoption (16.6 per cent) than West Bengal (4.2 per cent) and Assam (1.9 per cent).

4.1 Sequential adoption

The adoption decision on climate-smart varieties is a two-stage process: first is the acquisition of information on the variety and its advantages to farmers (access and learning) and second is the person adopting the variety (adoption). The learning stage comprises access to information on STRVs and passing through effective information transitions. Thus, the adoption decision is a sequential process as described in the decision tree (figure 1) and at each stage farmers evaluate future utility and make choices accordingly. For example, in the first stage of access to information, the farmer may be exposed to a climate-smart variety (with a probability of ${\pi _1}$) and, in the learning stage, the farmer makes an effort to collect additional information and thereby increase his/her knowledge on SS1 (${\pi _2}$ proportion able to pass the effective information hurdle). Thus, in learning-stage information, farmers collect information pertaining to the net present value of the potential profit from adopting SS1, which leads to the adoption decision regarding SS1 (probability of ${\pi _3}$). Evaluation of the perceived utility of information on STRVs depends on the individual's capacity and other socioeconomic as well as farm characteristics.

Figure 1.

Sequential adoption decision process with effective information.

The sequential adoption decision process can be modeled using a utility framework (Dimara and Skuras, Reference Dimara and Skuras2003; Shiferaw et al., Reference Shiferaw, Kebede, Kassie and Fisher2015). The first-stage learning process and information acquisition at optimum level $( {{L^O}} )$ is an outcome of an underlying utility maximization problem, which depends on the farmer's socioeconomic and demographic factors and farm characteristics, as well as perceived incidence of risk. The farmer is aware $( {A_{L, i}^\ast } )$ of the STRV if the level of information acquired is above the threshold level $( {{L^T}} )$ (Saha et al., Reference Saha, Love and Schwart1994) That is,

$$A_{L, i}^\ast\equiv {L^O} - {L^T} \gt 0 \equiv \tau {z_i} + {\varepsilon _L},$$

where $\tau$ is a vector of parameters to be estimated and ${\varepsilon _L}$ is the error term related to the first stage of the adoption process (learning and awareness). Since $A_{L, i}^\ast$ is an unobservable latent variable, we observe whether the farmer is aware or not $( {A_{L, i}}) \;$ such that

$${A_{L, i}} = \begin{cases} 1, &\textrm{if}\ A_{L, i}^{\ast} \ge 0,\quad \tau {z_i} \ge - {\varepsilon_L}\\ 0, &\textrm{if}\ A_{L, i}^{\ast} \lt 0, \quad \tau {z_i} \lt - {\varepsilon_L} \end{cases}.$$

In the next stage, the producer evaluates the potential future utility of the stress-tolerant variety. We used the utility framework rather than a profit maximizing one as this climate-smart technology increases the farmer's utility by reducing the risk of abiotic stresses, which depends on climatic fluctuations. Farmer i adopts stress-tolerant variety SS1 if the utility derived from the new variety is more than the utility from the old variety $( {U_{S, i}} \gt {U_{O, i}})$. The adoption decision on the new variety by the farmer $( {A_{S, i}})$ is

$${A_{S, i}} = \begin{cases} 1 &\textrm{if}\ {U_{S, i}} \ge {U_{O, i}},\quad \beta {x_i} \ge - {\epsilon_A}\\ 0 &\textrm{if}\ {U_{S, i}} \lt {U_{O, i}},\quad \beta {x_i} \lt - {\epsilon_A} \end{cases},$$

where $\beta$ is a vector of parameters to be estimated and ${\epsilon _A}$ is the error term related to the second stage of the adoption process (adoption decision). In practice, one would observe only the qualitative variable ${A_i}$, the final adoption of a stress-tolerant variety. But ${A_i}$ involves two hurdles: first, the farmers pass through information screening, and those who pass this hurdle are the potential adopters; and second, those potential adopters evaluate the potential benefits (utility) of STRV technology (new variety) vis-à-vis their own cultivated variety (old variety). Thus, the final adoption of a stress-tolerant variety is a partial observation (Dimara and Skuras, Reference Dimara and Skuras2003), given as

$${A_i} = {A_{L, i}}{A_{S, i}} = \begin{cases} 1 &\textrm{if\ SS1\ is\ adopted}\\ 0 & \textrm{if\ SS1\ is\ not\ adopted} \end{cases}.$$

${A_i} = 1$ will occur when the farmer has an optimum level of information that is more than threshold information $( {A_{L, i}} = 1)$ and the farmer decides to cultivate SS1 $( {A_{S, i}} = 1)$. ${A_i} = 0$ can occur if either the farmer has a below-optimum level of information that is less than threshold information $( {A_{L, i}} = 0)$ or, even if the farmer has acquired more than the threshold level of information and is aware of the STRV $( {A_{L, i}} = 1)$, she/he willingly decides not to adopt SS1 $( {A_{S, i}} = 0)$. The information level that facilitates this adoption decision (either to cultivate or not) when ${A_{L, i}} = 1\;$is often termed effective information (Varma, Reference Varma2018). $\pi = {\pi _1} \times {\pi _2} \times {\pi _3}$ is the probability of people making a decision based on the above-described sequential adoption process passing through different hurdles of information and utility/benefit evaluation.

4.2 Empirical model

The effect of information on STRV adoption is decomposed into weighted sum of effects on odds of passing each stage (access to information $INFO_1^{STRV}$, effective information $INFO_2^{Quality}$ and adoption $ADOP{T_3}$) with a sequential logit model. In this model, information quality is defined by considering both the content of the information and the source from which the person receives that information. A farmer may receive more than one content pertaining to the STRV, but one should be either information on SS1 or its seed, sourced from one or more trusted sources such as progressive farmers with prior STRV cultivation experience (S_p), NGOs involved in STRVs (S_n), institutionalized agricultural extension service providers (S_e), seed dealers on seeds (S_s) or paddy traders on marketing (S_t). That is,

$$INFO_2^{Quality} = \begin{cases} {1} & \textrm{if}\ \textrm{number}\ \textrm{of}\ \textrm{content}\vert {(\textrm{source} = {S_p}\ \textrm{or}\ {S_n}\ \textrm{or}\ {S_e}\ \textrm{or}\ {S_s}\ \textrm{or}\ {S_t}) \gt 1}\\ 0 & \textrm{otherwise} \end{cases}.$$

Thus, depending on information content, the trusted source may vary. For example, a formal agricultural extension service or university is a good source of information on STRV whereas seed dealers are a good source of information on seed availability, paddy traders are a good source for marketing of produce, and so on. The content should be a minimum of two pertaining to STRV and related information such as STRV seed availability, price and mitigating crop damage. All information that does not meet these two criteria is grouped as non-effective information.

Following Buis (Reference Buis2011, Reference Buis2015), the model assumes that the farmer is at risk of passing through each stage and had to pass through all lower stages. For example, if a farmer reaches stage 3 (adoption decision), she/he had to pass through two previous hurdles, $INFO_1^{STRV}$ and $INFO_2^{Quality}$. The model assumes that the person who adopted the STRV was ‘at risk’ of passing through the first two transitions, but the decisions at each stage are assumed to be independent. The sequential logit model is shown below:

\begin{align*}&\textbf{Transition}\ \textbf{1}{:}\ \boldsymbol{\textit{Pr}}({INFO_{1,i}^{STRV} = 1\vert {{x_{1i}},{x_{2,i}}} } ) = {\hat{\pi }_{1i}} \\ &\quad = \boldsymbol{\Lambda }( {{{\hat{\beta }}_{01}} + {{\hat{\beta }}_{11}}\textrm{H}{\textrm{H}_i} + {{\hat{\beta }}_{21}}{\textrm{H}_i} + {{\hat{\beta }}_{31}}\textrm{FAR}{\textrm{M}_i} + {{\hat{\beta }}_{41}}\textrm{SO}{\textrm{C}_i}} )\end{align*}

\begin{align*}&\textbf{Transition}\ \textbf{2}{:}\ \boldsymbol{\textit{Pr}}({INFO_{2, i}^{Quality} = 1\vert {x_{1i}}, \;{x_{2, i}},\ INFO_{1, i}^{STRV} = 1}) = {\hat{\pi }_{2i}} \\ &\quad = \boldsymbol{\Lambda }( {{{\hat{\beta }}_{02}} + {{\hat{\beta }}_{12}}\textrm{H}{\textrm{H}_i} + {{\hat{\beta }}_{22}}{\textrm{H}_i} + {{\hat{\beta }}_{32}}\textrm{FARM}_i + {{\hat{\beta }}_{42}}\textrm{SO}{\textrm{C}_i}} )\ \textrm{if}\ INFO_{1, i}^{STRV} = 1\end{align*}

\begin{align*}&\textbf{Transition}\ \textbf{3}{:}\ \boldsymbol{\textit{Pr}}({ADOP{T_{3, i}} = 1\vert {x_{1i}}, \;{x_{2, i}},\ INFO_{2, i}^{Quality} = 1}) = {\hat{\pi }_{3i}} \\ &\quad = \boldsymbol{\Lambda }( {{{\hat{\beta }}_{03}} + {{\hat{\beta }}_{13}}\textrm{H}{\textrm{H}_i} + {{\hat{\beta }}_{23}}{\textrm{H}_i} + {{\hat{\beta }}_{33}}\textrm{FARM}_i + {{\hat{\beta }}_{43}}\textrm{SO}{\textrm{C}_i}} )\ \textrm{if}\ INFO_{2, i}^{Quality} = 1\end{align*}

The function $\boldsymbol{\Lambda}(.)$ denotes a standard logistic function $\left[{\Lambda (.) = \frac{{exp(.) }}{{1 + exp(.)}}} \right]$. The conditional probability that farmer i passes transition k is ${\hat{\pi }_{ki}}$, and ${\hat{\beta }_{mk}}$ represents the association between the variable m (m includes household head (HH), household (H), farm (FARM) and social (SOC) characteristics of the farmer) and transition probability ${\hat{\pi }_k}$, and ${\hat{\beta }_{0k}}$ is the constant for transition k. The model assumes that everybody is at risk in the first transition, that is, at the stage of the information on STRVs. The sequential logit model also models how the expected outcome differs between these individuals:

$$\frac{{\partial E( {outcome} ) }}{{\partial x}} = \sum \;( {at\;ris{k_k}\;\times \;varianc{e_k}\;\times \;gai{n_k}} ) \;{\beta _k}.$$

In this equation, x is the explanatory variable; $at\;ris{k_k}\;$is the proportion of persons at risk of passing transition k; $varianc{e_k}$ is the variance of the dependent variable for transition k, that is, $P{r_k}( {1 - P{r_k}} )$; $gai{n_k}$ is how much a person can expect to gain from passing transition k; and ${\beta _k}$ is the effect of variable x on the log odds of passing transition k. Thus the total effect is a weighted sum of the effects on each transition, and a transition receives more weight when more people are at risk, and people can expect to gain much from passing that transition. It is important to note that neither does virtually everybody pass nor does virtually everybody fail that transition Buis (Reference Buis2011, Reference Buis2015).