-
This face-to-face survey study was carried out in Ilam Province, which is situated on the western part of Iran at a latitude of 33°38′N and longitude of 46°26′ E (Fig. 2). The research sites comprise the Townships of Sirwan and Chardawol, with a population of 41 469 and 37 981, respectively, accounting for 16.56% and 28.81% of the total rural population. 79.54% of the population of Sirwan are farmers, 15.65% of cattle breeders, the rest are employed in other services (4.81%). In addition, 54.15% of the population of Chardawol are farmers, 38.75% are livestock farmers, the remainder of the population have other service jobs (7.7%) (Statistical Yearbook of Ilam Province, 2016). Two townships are located in a stretch of the central Zagros Mountains, the areas where have been severely affected by declining rainfall and therefore decreased local ware resources (Seymareh River) due to frequent droughts. Before the occurrence of droughts, prior to 2001, these regions have been the center of producing a specific cultivar of rice, as locally spoken as Anbarboo. It is also, according to the information of the Agricultural Jahad Organization, in the western part of Iran, which includes parts of Kurdistan, Kermanshah, Hamedan, Lorestan and Ilam provinces, rice is planted only in Ilam Province and especially in Chardavol and Sirvan cities. Along with climate change and droughts in recent years and the drying up of agricultural water resources and reduced rice cultivation, rice farmers living in the villages of the study area have suffered socio-economic losses, including the loss of the main job, the disappearance of the main source of household income, the decrease in the level of household income, the return of households to false jobs such as peddlers along the streets of large cities and migration to larger cities such as Tehran.
-
The research population includes all rural householders in the townships of Sirwan and Chardawol (N = 19 856). Using Bartlet et al.’s (2001) approach, 200 farm householders were determined and surveyed under the multi-stage stratified random sampling. By stratifying the research area based on the characteristics like the typology of cultivation (i.e., rice, grain, and vegetables), the topological structure of area (i.e., highland or close to rivers), and irrigation method (i.e., well water or rivers), the householders of 31 villages were surveyed in the rural districts of Karazan, Lowmar, Zangwan, and Roudbar.
Using exploratory field research and archival data analysis, we developed a four-page A4-format questionnaire consisting of sections containing questions derived from the conceptual frameworks. The experts of the DOA delivered their corrective suggestions on bias wording, thus approved the validity of the indices of the questionnaire. By trialing with 30 farmers in a small-scale pilot survey, the internal consistency of indicator variables with the respective latent variables in the questionnaire (i.e., reliability) was achieved using Cronbach’s Alpha method (Thompson, 2003). The results include
$ {\alpha }_{BCC}=0.91 $ ,$ {\alpha }_{PRCC}=0.71 $ ,$ {\alpha }_{AES}=0.80 $ ,$ {\alpha }_{AC}=0.83 $ ,$ {\alpha }_{MSA}=0.79 $ ,$ {\alpha }_{CES}=0.92 $ ,$ {\alpha }_{PEC}=0.82 $ , and$ {\alpha }_{PDsI}= $ $ 0.72 $ . These variables are also abbreviated in Table 1.Level Dimension Predictor variables Abbreviated References Household-level Perceived climate change Belief in CC BCC Li et al. (2017); Shi-Yan et al. (2018); Khanal et al. (2018) Perceived risk of CC PRCC Dang et al. (2014a) Demographic factors Age AGE Ofuoku (2011); Yegbemey et al. (2013) Gender Gender Zamasiya et al. (2017); Shikuku et al. (2017) Marital status MS Jamshidi et al. (2014) Female heads FH Tenge et al. (2004) Household size SH Jamshidi (2015) Educational attainment EA Maddison (2006); Yegbemey et al. (2013) Economic factors Household income from farming HIF Jamshidi (2015) Household sources of income HSI Jamshidi (2015) Secondary job SJ Ghambarali et al. (2012); Khosravipour et al. (2013) Access to credit AC Shahidur et al. (2002); Yegbemey et al. (2013) Labor sources LS Jamshidi (2015) Farming experience FE Adesina and Baidu-Forson (1995); Maddison (2007); Ghambarali et al. (2012); Khosravipour et al. (2013) Rights on land RL Yegbemey (2013) Social factors Membership in social associations MSA Jamshidi et al. (2014) Participation in extension classes PEC Jamshidi et al. (2014) Farm-level Farm size FS Jamshidi (2015) Land slope LSE Jamshidi (2015) Agricultural water resource AWR Jamshidi (2015) Distance to markets for agricultural inputs and products DMAIs/DMAOs Maddison (2006); Nhemachena and Hassan (2007); Deressa et al. (2009); Yegbemey et al. (2013), Dang et al. (2014a) Institutional-level Access to extension service AES Jamshidi et al. (2014) Institutional arrangements on land IAL Yegbemey et al. (2013) Contact with extension CES Jamshidi et al. (2014) Perceived incentives/ disincentives PIs/PDs Jamshidi (2015); Shi-Yan et al. (2018) Note: Adaptation to CC was operationalized in terms of the different strategies of adaptation documented by Bradshaw et al. (2004); Kurukulasuriya and Mendelsohn (2006); Maddison (2006); Nhemachena and Hassan (2007) Table 1. The variables of predictor
-
The predictor variables that were examined in the study are listed in Table 1, some of which are factual in nature, whereas, the remainder of which are attitudinal, measured on a 5-point Likert scale, (1) strongly disagree to (5) strongly agree. Also, the criterion variable of the study was established based on the literature review, which contains a vast variety of strategies that farmers use to adapt to CC like changing the irrigation methods, application of drought-resistant varieties, changing the time of planting, and the usage of the livestock breeds adaptable to CC (Yegbemey et al., 2013).
The Linear Discriminant Analysis (LDA) is a multivariate classification procedure used to model the variation of a criterion variable or dependent variable based on its association with one or more predictors or independent variables. The data analysis was carried out at the interface of SPSS. To achieve a realistic interpretation of the data, the statistics of U and Wilks’ Lambda were utilized, which helped us to compare the equality of means of the respective groups.
-
As demonstrated by the findings, the study participants made use of the five types of adaptation strategies as follows: 1) crop diversification and crop management strategies (12.0%) (e.g., different crops and rotation); 2) agricultural water management strategies (4.5%) (e.g., rainwater storage and harvesting the groundwater to use during the dry season by pumping water from the river); 3) adjusting farm management strategies (6.5%) (e.g., making adjustment to the cultivation calendar and the use of insurance); 4) diversification strategies (8.0%) (e.g., raising the livestock, planting the crops before the outbreak of drought, and usage of drought-resistant crops), and 5) diversification in activities, for instance, fishery and mining (5.0%).
-
In this section, we deliver four major impediments as farmers adapt to CC, including: 1) water shortage (20.0%), 2) the lack of alternative jobs (13.5%), 3) the dispersed lands owned by the same farmers (13.0%), and 4) the shortage of a permanent source of income (11.0%).
-
At the first stage of the analysis (i.e., preliminary data exploration), we examined the measure of Kolmogorov-Smirnov test to probe into normality (KST) of data distribution, therefore, resulting in 0.52 for KST at the P-value more than 0.5, acceptable measure was achieved (Kolmogorov-Smirnov = 0.54, P-value = 0.92). Using the backward stepwise method, the discriminant function was developed in 11 steps to minimize the measurement of the Wilk’s Lambda at each step. The result reveals that 11 variables are the major differentiating drivers (F > 3.84), by which the research sample would be divided into two groups of adapters and non-adapters (G1 and G2). The rest of the variables were disregarded owing to having the respective value of the F statistic lower than the input value (i.e., F < 2.71) (Table 2).
Step Entered variables Wilks’ Lambda Value df-1 df-2 df-3 F exact Value df-1 df-2 P-value 1 Farming experience 0.589 1 1 198 138.3 1 198 0.001 2 Farm size 0.435 2 1 198 127.8 2 197 0.001 3 Earned farming income / % 0.372 3 1 198 110.4 3 196 0.001 4 Education 0.335 4 1 198 96.6 4 195 0.001 5 Rights on land 0.305 5 1 198 88.2 5 194 0.001 6 Age 0.285 6 1 198 80.9 6 193 0.001 7 Access to extension services 0.269 7 1 198 74.6 7 192 0.001 8 Belief in CC 0.256 8 1 198 69.3 8 191 0.001 9 Perceived risk of CC 0.242 9 1 198 66 9 190 0.001 10 Subsidiary jobs 0.235 10 1 198 61.5 10 189 0.001 11 Access to credit 0.229 11 1 198 57.3 11 188 0.001 Table 2. Results of the discriminant analysis with the backward stepwise selection method
As resulted by the classification analysis model, the resulting measures of canonical correlation analysis (CCA) and Wilks’ Lambda test indicated that the canonical correlation coefficient between the variables of two groups and the discriminant score was 0.878 (
$ {R}_{cononical} $ = 0.878), there is therefore a strong relationship between the two groups and discriminant score. When canonical correlation coefficient is high, it indicates the relevance of the derived function in dividing farmers into target groups, in this case, adaptors and non-adaptors. The eigenvalue is another measure relating to the estimation of the discriminating function, which explains the variation of the criterion variable. According to Marey-Pérez and Rodríguez-Vicente (2011), a large value for this measure indicates that the function has a convincing power to account for the variation (Eigenvalue = 0.787 or 78.7%). The Wilks’ Lambda test shows that 22.9% of the total variance in discriminative scores is not explained by the differences between adapters and non-adaptors.For a better interpretation of Wilks’ Lambda test results, it is recommended to use the converted Chi-square. The result unveils that the level of critical significance for the chi-square test (i.e., transformed value of the Lambda statistic) is lower than 0.05 (χ2 = 283.345, df = 11; P ≤ 0.001), thus disclosing that there is a highly significant difference between the group centroids, more clearly, little overlap between the adapters and non-adapters (Table 3).
Statistics Measure Eigenvalue 0.787 Percentage of variance 100% Cumulative percentage 100% Canonical correlation 0.878 Wilks’ Lambda 0.229 Chi-square 283.345 df 11 P-value 0.0001 Table 3. Canonical correlation and Wilks’ Lambda in the discriminant model
We also assessed the relative impact of different factors that contribute to the discriminant scores of the differentiation of adapters and non-adapters, hereby the standardized coefficients of the canonical discriminant function were used (Table 4). Therefore, variables such as farm experience, farm size and land rights had the most notable share in the discriminant function. Moreover, beliefs in CC, the perceived risks of CC, educational attainment, household income from farming, access to extension, subsidiary jobs, age, and access to credit were priorities, respectively. The discriminant analysis predicts the likelihood of positioning a farmer in groups of farmers that adapt to CCs and others that do not. This is done by placing the individual values for each variable into the discriminating function. The average discriminating score for the group of farmers who adapt CC is 2.54 and –1.309 for another group. By subtracting these two numbers from one another, the average value of the total function of the two groups gave 1.23. Therefore, if the value is less than 1.23, it is predicted that a farmer is not able to adapt to CC. In the opposite vein, the measures more than 1.23 indicate that a farmer falls into adapting group, being able to adapt to CC.
Predictor variables Coefficient Correlation Farming experience –0.454 –0.456 Farm size 0.387 0.409 Rights on land –0.322 –0.34 Belief in CC 0.297 0.181 Perceived risk of CC 0.266 0.263 Education 0.259 0.365 Household income from farming / % –0.235 –0.412 Access to extension service 0.229 0.178 Subsidiary jobs 0.219 0.329 Age –0.208 –0.344 Access to credit 0.181 0.141 Table 4. The standardized coefficients of the canonical discriminant function
The values of βi coefficients correspond to the estimated measures for each of the groups. XP is the variable P. Fisher classification functions are calculated to complete the LDA and then weightings for each predictor are obtained for each group (i.e., adopters and non-adaptors). Equation 1 is the discriminant function of adapters.
$$ Y={\beta }_{0}+{\beta }_{1}{X}_{1}+{\beta }_{2}{X}_{2}+\ldots +{\beta }_{P}{X}_{P}$$ (1) $$ \begin{split} Y= &-72.62-{0.569X}_{AGE}+{2.39X}_{FE}+{0.157X}_{HIF}+\\ &{4.16X}_{FS}+{2.49X}_{EA}+{1.52X}_{SJ}+{6.2X}_{BCC}+\\ &{6.2X}_{PRCC}+{0.026X}_{RL}+{3.49X}_{AES}+{2.93X}_{CES} \end{split} $$ (1) -
As shown in Table 5, using cross-validated procedure, the result of classification of the samples indicate that 64 observations (94.1%) of 68 participants in the adapter-group have been accurately predicted to be embedded in the group of the farmers who adapt to CC. Furthermore, 4 observations (5.9%) were mistakenly incorporated in group 2 (i.e., non-adapters). Moreover, 130 observations (98.5%) out of 132 observations of group 2 (i.e., non-adapters) were accurately placed in the target group, and 2 observations (1.5%) have been mistakenly placed in group 1. Accordingly, 95% of the total observations have been accurately assigned in the respective groups.
Group Predicted group membership Total Group 1 Group 2 Original Count G1 64 4 68 G2 2 130 132 Percentage / %
G1 94.1 5.9 100 G2 1.5 98.5 100 Cross-validated Count G1 63 5 68 G2 5 127 132 Percentage / % G1 92.6 7.4(%) 100 G2 3.8 96.2 100 Note: G1: Adapters to CC, G2: Non-adapters to CC Table 5. The matrix of classification from discriminant analysis
-
To realize the real impact of the predictor variables in differentiating the two groups of adapters and non-adapters as more precisely as possible, we calculated eta-squared (
$ {\eta }^{2} $ ) that is a measure of effect size for LDA, consistent with the procedure delivered by Vacha-Haase and Thompson (2004). The eta-squared values of 0.01, 0.06, and 0.14 are interpreted as small, moderate, and large, respectively (Cohen, 1992). The corresponding effect size was calculated using the following formula (Eq. 2) ( Hahs-Vaughn, 2016):$$\mathrm{P}\mathrm{a}\mathrm{r}\mathrm{t}\mathrm{i}\mathrm{a}\mathrm{l}\;{\eta }^{2}=1-{\lambda }^{1/3}$$ (2) where
$ {\eta }^{2} $ is eta squared for effect size,$ \mathrm{\lambda } $ is a measure for Wilks’ Lambda, which is equal to 0.22 in this study. Therefore, the value of 0.38 is calculated for$ \mathrm{P}\mathrm{a}\mathrm{r}\mathrm{t}\mathrm{i}\mathrm{a}\mathrm{l}\;{\eta }^{2} $ , which indicates a powerful effect size of the LDA.
Determinants of Adaptation to Climate Change: A Case Study of Rice Farmers in Western Province, Iran
doi: 10.1007/s11769-021-1246-0
- Received Date: 2020-09-09
- Accepted Date: 2021-01-10
- Publish Date: 2022-01-01
-
Key words:
- climate change (CC) /
- drought /
- adaptation management /
- rice farmers /
- Iran
Abstract: The decisions made by agricultural households to adjust to climate change (CC) in Iran are not well known. This study is intended to investigate the influence of perceptions and socioeconomic, institutional features on farmers’ adaptation decisions about CC, which constitute the hypothetical statements of the study. We undertook a survey of 200 farm householders from 31 villages of Ilam Province, situated in the western Iran, as randomly selected. The result discloses that the proposed discriminant model matches the dataset well, with a strong effect size of partial eta-squared
Citation: | JAMSHIDI Alireza, JAMSHIDI Masomeh, ABADI Bijan, 2022. Determinants of Adaptation to Climate Change: A Case Study of Rice Farmers in Western Province, Iran. Chinese Geographical Science, 32(1): 110−126 doi: 10.1007/s11769-021-1246-0 |