Research design and methods of data collection
The research design was based on multistage sampling procedure. In the first stage, the whole sub-basin constituting fifteen districts was grouped into three strata (Kolla, Woyina Dega, and Dega agroecological zones) based on their agroecological characteristics including the rainfall, soil, and topography. Kolla refers to an area with an altitude ranging between 500 and 1500 m asl, with mean annual temperature between 20 and 28 °C and annual rainfall between 600 and 900 mm. Woyina Dega refers to an altitude ranging between 1500 and 2300 m asl, with mean annual temperature 16 and 20 °C and annual rainfall above 900 mm; Dega refers to an altitude between 2300 and 3200 m asl, with mean annual temperature between 6 and 16 °C and mean annual rainfall above 900 mm [34]. Then, two districts were randomly selected from Kolla and Dega agroecological zones. Similarly, two districts were also selected from Woyina Dega agroecology using simple random sampling technique. Two districts from Woyina Dega were taken to maintain proportionality as the Woyina Dega agroecology covers larger area of the study site. In the second stage, only Peasant Associations (PAs) found in the sub-basin in each sampled districts were listed in consultation with agricultural experts in the area. This is mainly to exclude PAs which are not part of the sub-basin in that particular district. Then, four PAs were randomly selected from each selected districts. Finally, 450 sample respondents were selected from 16 PAs using random sampling technique on the basis of probability proportional to size (PPS). The sampling frame was the list of households which was obtained from the PAs administration. Households for Focussed Group Discussions (FGDs) were also drawn from each identified district, and a member of the group was identified with the help of development agents working in the area.
Both quantitative and qualitative methods of data collection were used to obtain information from the selected respondents. Quantitative data were gathered using semi-structured questionnaire. Qualitative data were obtained from FGDs to complement the information obtained through a semi-structured questionnaire in order to have a better understanding of adaptation strategies used by these farmers and barriers to adopting adaptation options. Besides data on lists of adaptation measures, this survey addressed perceived barriers to those adaptation options. Questions were also posed to investigate factors that constrain/facilitate adaptation measures to change in mean temperature and rainfall over the last two or three decades in the study area. Mean monthly temperature and precipitation from 1991 to 2016 were obtained from Ethiopian metrological station found in each sampled district.
Methods of data analysis
In order to analyze and present the data collected from sampled households, descriptive statistics (frequency, mean, maximum, minimum, and standard deviation), inferential tests (Chi-square test and one-way ANOVA test), and econometric model were used. Qualitative categorical types of data were analyzed using frequency and Chi-square test, while quantitative continuous types of variables were analyzed using one-way ANOVA, minimum, maximum, mean, and standard deviation. Qualitative information was organized and constructed coherently and analyzed based on theoretical and conceptual frameworks. After computing the descriptive statistics and inferential tests, a multinomial logistic regression model was used to identify determinants of household’s adoption of adaptation options where the dependent variable was found to be multioutcome. The data analysis was conducted using Statistical Package for Social Sciences (SPSS) version 20 and Stata 12.
Multinomial logit model specification
Probit and logit models are the two most popular functional forms used in adoption modeling. These models have got desirable statistical properties as the probabilities are bounded between 0 and 1. Apparently, adoption models could be grouped into two broad categories based on the number choices or options available to an economic agent [35]. A choice decision by farmers is ‘inherently a multivariate decision.’ Attempting bivariate modeling excludes useful economic information contained in the interdependent and simultaneous choice decisions [36]. Since farmers decision on the use of adaptation options involves multiple response in which the dependent variable is discrete, it is more appropriate to treat factors which are supposed to determine farmers’ decision on the use of adaptation options as a multiple choice decision. Based on this argument, the appropriate econometric model would be either multinomial logit or multinomial probit regression model. Regarding estimation, both of them estimate the effect of explanatory variables on dependent variable involving multiple choices with unordered response categories [35]. However, multinomial probit is rarely used in empirical studies due to estimation difficulty imposed by the need to solve multiple integrations related to multivariate normal distribution [37]. Moreover, a multinomial logit model is selected not only because of the computational ease but also it exhibits a superior ability to predict livelihood diversification and picking up the differences between the livelihood strategies of rural households [38]. In this study, therefore, a multinomial logit model was employed. This model makes it possible to analyze factors influencing households’ choices of adaptation strategies in the context of multiple choices.
The decision of whether to use any adaptation option or not could fall under the general framework of utility maximization [39]. Following Greene [35], suppose for the ith respondent faced with j choices, we specify the utility choice j as:
$$U_{ij} = Zi_{j} \beta + \varepsilon_{ij}$$
(1)
If the respondent makes choice j in particular, then we assume that U
ij
is the maximum among the j utilities. So the statistical model is derived by the probability that choice j is made, which is:
$${\text{Prob}}\left( {U_{ij} > U_{ik} } \right)\,{\text{for all other}}\,K \ne j$$
(2)
where U
ij
is the utility to the ith respondent from adaptation strategy j, U
ik
the utility to the ith respondent from adaptation strategy k.
If the household maximizes its utility defined over income realizations, then the household’s choice is simply an optimal allocation of its asset endowment to choose adaptation strategy that maximizes its utility [40]. Thus, the ith household’s decision can, therefore, be modeled as maximizing the expected utility by choosing the jth adaptation strategy among J discrete adaptation strategies, i.e.,
$$\max_{j} = E(U_{ij} ) = f_{j} (x_{i} ) + \varepsilon_{ij} ;\quad j = 0 \ldots J$$
(3)
In general, for an outcome variable with J categories, let the jth adaptation strategy that the ith household chooses to maximize its utility could take the value 1 if the ith household chooses jth adaptation strategy and 0 otherwise. The probability that a household with characteristics x chooses adaptation strategy j, P
ij
is modeled as:
$$P_{ij} = \frac{{\exp (X^{\prime}_{i} \beta_{j} )}}{{\sum\nolimits_{j = 0}^{J} {\exp (X^{\prime}\beta_{j} )} }},\,J = 0 \ldots \,3$$
(4)
With the requirement that \(\sum\nolimits_{j = 0}^{J} {P_{ij} = 1}\) for any i, where P
ij = probability representing the ith respondent’s chance of falling into category j; X = predictors of response probabilities
\(\beta_{j} =\) Covariate effects specific to jth response category with the first category as the reference.
Appropriate normalization that removes an indeterminacy in the model is to assume that \(\beta_{1} = 0\) (this arise because probabilities sum to 1, so only J parameter vectors are needed to determine the J + 1 probabilities) [41], so that \(\exp (X_{i} \beta_{1} ) = 1\), implying that generalized Eq. (4) above is equivalent to
$$\Pr (y_{i} = j/X_{i} ) = P_{ij} = \frac{{\exp (X_{i} \beta_{j} )}}{{1 + \sum\nolimits_{j = 1}^{J} {\exp (X^{\prime}_{i} \beta_{j} )} }},\quad {\text{for}}\,j = 0,2 \ldots J\quad {\text{and}}$$
$$\Pr (y_{i} = 1/X_{i} ) = P_{i1} = \frac{1}{{1 + \sum\nolimits_{j = 1}^{J} {\exp (X^{\prime}_{i} \beta_{j} )} }},$$
(5)
where y = A polytomous outcome variable with categories coded from 0… J.
Note The probability of P
i1
is derived from the constraint that the J probabilities sum to 1. That is, \(p_{i1} = 1 - \sum {p_{ij} }.\).
The multinomial logistic models crucially depend on the independence of irrelevant alternatives (IIA) assumption in order to obtain unbiased and consistent parameter estimates. This assumption requires the likely of the household’s using a certain adaptation options need to be independent of other alternative adaptation options used by the same household’s. Hausman test was used to test the validity of the IIA assumption.
The estimated cofficients of multinomial logit model provide only the direction of effect of independent variables on dependent variables, but estimate neither represent the actual magnitude of change nor probabilities [42]. Thus, the Stata version 12 was used to generate the parameter estimates (marginal effect). The marginal effects measure the expected change in the probability of a particular choice being made with respect to a unit change in an independent variable [35].
Definition of variables and hypothesis
The dependent variable in this study is the choice of an adaptation option that is listed in Fig. 2. The potential explanatory variables, which were hypothesized to influence farmers’ use of adaptation options in response to climate variability and change and considered in the analysis, are often classified as personal, physical, socioeconomic, institutional, and climate factors [43, 44]. These variables include age of the household head, gender of the household head, education status, and family size, membership in the social group, access to extension service, access to credit service, and climate warning system, agroecology, livestock ownership, the occurrence of drought and flood, and land cultivated. Table 1 presents the description, definition and unit of measurment for both dependent and independent variables.
Family size
The empirical adoption literature shows that household size has mixed impacts on farmers’ adoption of agricultural technologies. Larger family size positively influences farmers to take up labor-intensive adaptation measures like soil and water conservation (SWC) and irrigation that demand labor which is a critical problem in a peak period of production and livestock rearing [45, 46]. Alternatively, a large family might be forced to divert part of its labor force into non-farm activities to generate more income and reduce consumption demands [42]. We hypothesize that SWC and small-scale irrigation are more labor intensive and hence we expect family size to have a positive influence on the adoption of such adaptation measures. Similarly, this variable is expected to have a positive effect on the use of diversified livelihood options.
Age of the household head
The influence of age on adoption of SWC is unclear. Some studies found a positive relationship between age and conservation investment. This indicates that the likelihood of adoption of conservation practices is more among older farmers than among the younger ones, perhaps because older farmers could adopt SWC because they have more experience that helps them to perceive erosion problems [43, 47]. Conversely, older farmers could be less willing to bear the risk of investing in SWC due to their shorter planning horizons [45, 46]. In this study, we hypothesize that age of the household head has both positive and negative impacts on adaptation measures. Empirical studies by Arega et al. [48] and Gebreyesus [49] showed that age of the household head negatively related to farmers decision to diversify to non-farm and off-farm activities. Thus, age is hypothesized to influence the decision to diversify livelihood options.
Gender of the household head
It is a dummy variable that takes 1 if the household head is male, 0 otherwise. Many previous literature on adaptation, the influence of gender on adoption of adaptation measures are mixed. Female farmers have been found to be more likely to adopt natural resource management and conservation practices [45, 50]. However, some studies found that male household heads had a better opportunity to take an adaptation measure than female household by involving on agronomic practices (such as crop diversification and use of drought-tolerant crop species) and by adopting SWC measures and irrigation to their farm [51, 52]. For instance, Asfaw and Admassie [51] noted that male-headed households are often considered to be more likely to get information about new technologies and take risky businesses than female-headed households. We hypothesize that female- and male-headed households differ significantly in their ability to adapt to climate change because of major differences between them in terms of access to assets, education, and other critical services such as credit, technology, and input supply.
Farm size
It is a continuous variable defined as the years of schooling attained by the household heads. In most of the adoption studies, it has been shown that education is an important factor that positively influences adoption decisions [42, 46, 53,54,55]. These studies have shown that better education and more farming experience increase farmers’ ability to get and use of information and improve awareness of potential benefits and willingness to participate in local natural resource management and conservation activities. Educated and experienced farmers are expected to have more knowledge and information about climate change and agronomic practices that they can use in response [56]. We expect that improved knowledge and farming experience will positively influence farmers’ decisions to take up adaptation measures.
Farm size
Empirical adoption studies have found mixed effects of farm size on adoption of SWC. For example, a study on soil conservation measures and irrigation in Ethiopia found that farmers with larger farms were found to have more land to allocate for constructing soil bunds, stone bunds, check dams, and improved cutoff drains and motivate to use irrigation [47, 53, 57, 58]. Similarly, Gbetibouo [59] revealed that farm size is positively correlated with the probability of choosing irrigation as an adaptation measure. On the other hand, Nyangena [60] found that farmers with a small area of land were more likely to invest in soil conservation than those with a large area. It is also supported by the study conducted by [32]. According to their argument, the need for specific adaptation option (i.e., SWC measures) to climate variability and change is dictated by characteristics of the plot than the size of the farm. This means that it is not the size of the farm. Empirical studies have shown that the area of land owned by the household has a negative correlation with the likelihood of diversifying to non-farm and off-farm activities [48, 61]. Therefore, farm size to have a positive role in the decision to use irrigation is hypothesized. On the other hand, farm size was expected to negatively affect the use of different livelihood diversification options.
Access to agricultural extension services
Extension services are an important source of information on agronomic practices as well as on climate. Extension education is found to be an important factor motivating an increased intensity of the use of specific SWC practices and irrigation use because access to extension services and information help farmers to have better understanding of the land degradation problem and soil conservation practices and hence may perceive SWC practices to be profitable [42, 43, 46, 53]. Thus, extension service is hypothesized to be promoting decision to use SWC practices, agronomic practices, and irrigation use. This study postulates that the availability of better climate and agricultural information helps farmers make comparative decisions among alternative crop management practices and hence choose the ones that enable them to cope better with changes in climate [62].
Access to credit
Several studies have shown that access to credit is an important determinant enhancing the adoption of various technologies [42, 52, 59, 62]. Deressa et al. [32] and Gbetibouo [59] reported that farmers with more financial and other resources at their disposal are able to make use of all their available information to invest on the use of irrigation, use of agricultural inputs, use of drought-tolerant crop species, use of SWC, and take up livelihood diversification in response to changing climatic and other conditions. Credit provision has the advantage to solve financial constraints to meet their need to change their practices to suit the forecasted climate change. Thus, it is hypothesized that access to credit has a positive effect on the use of irrigation, use of SWC measures, use of drought-tolerant crop species, and use of non-farm and off-farm activities.
Market access
Distance to the nearest market is used to proxy for availability of input and output markets. It is another important factor affecting adoption of agricultural technologies. The households located further away from markets are found to adopt lesser adaptation practices [30, 56, 63, 64]. Input markets allow farmers to acquire the inputs they need such as different seed varieties, fertilizers, and irrigation technologies. At the other end, access to output markets provides farmers with positive incentives to produce cash crops that can help improve their resource base and hence their ability to respond to changes in climate [30]. Maddison [56] observed that long distances to markets decreased the probability of farm adaptation in Africa and that markets provide an important platform for farmers to gather and share information. Lapar and Pandely [65] found that in the Philippines access to markets significantly affected farmers’ use of conservation technologies. Piya et al. [66] showed that in Nepal distance to markets negatively and significantly affected the use of SWC technologies. It is expected that the households located further away from the road are less likely to adopt livelihood diversification strategies, varietal selection, and the construction of tanks, but more likely to depend on traditional coping strategies.
Livestock ownership in TLU
Previous studies have shown mixed evidence about the relationship between livestock ownership and farmers’ decision in relation to SWC investment [42, 47, 55, 58]. Amsalu and De Graaff [47] showed that livestock ownership has negative influence to adopt stone terrace. On the contrary, more specialization in livestock negatively influences the use of SWC by reducing the economic impact of soil erosion. Hence, the effect of the size of livestock holding on conservation decision is difficult to hypothesize a prior. Livestock holding negatively influences household’s choice of non-farm and off-farm activities that means the farmer with lower livestock holding would be obliged to diversify livelihoods into off- and non-farm in order to meet needs [67, 68]. Therefore, it is hypothesized to have a negative relationship with diversifying livelihood options.
Access to weather information
Smallholder farmers require different types of climate information during each stage of the agricultural production process in order to adapt to climate variability and change. Major climate change information includes early warning signals, weather forecasts, pest attacks, input management, cultivation practices, pest and disease management, and prices [69,70,71]. Nhemachena and Hassan [72] reported that better access to weather information has a positive influence on the decision to invest in SWC measures, use of irrigation, use of drought-tolerant crop varieties, and diversify livelihood options in response to climate change problem. In the same way, Deressa et al. [52] found that access to information increases the likelihood of using SWC measures and different crop varieties to adapt climate change. The effect of access to weather information on the decision to use SWC measures, irrigation, drought-tolerant crop varieties, and livelihood diversification is expected to be positive.
Membership in a social group
Membership to social groups or organizations enables farmers to acquire information on proper agronomic practices, credits, and productive inputs as well as attend training and workshops at which stakeholders meet and exchange ideas. Self-help grouping and formation of cooperatives are a more reliable and pragmatic means of achieving social capital and ensuring dissemination and adoption of innovative technology [73, 74]. Tafa et al. [44] found that being a member of a social group increased the probability of adapting climate variability and change using conservation agriculture, drought-tolerant varieties, and irrigation. Thus, membership in social groups positive effect on adoption of adaptation options in response to climate change impact is hypothesized.
Agroecological setting
In Ethiopia, Kolla agroecology (lowland) is characterized by relatively hotter and drier climate, whereas Weyina Dega (middle land) and Dega agroecology (highland) are wetter and cooler [32]. Evidence revealed that farmers living in different agroecological settings have their own choice of adaptation methods [32, 75, 76]. For instance, Deressa et al. [32] observed that farming in the Kolla zone significantly increases the probability of SWC practices, compared to farming in Weyina Dega. However, farming in Kolla significantly reduces the probability of using different crop varieties, planting trees, and irrigation as compared to farming in weyna Dega. Hence, agroecology was hypothesized to have a positive or negative effect on household’s adoption decision on climate change adaptation options.