Data on farmers' determinants of manure and inorganic fertiliser use in the semi-arid Ethiopian Rift Valley

This article contains the data on farmers' determinants of binary choices for manure use (i.e., manure is used or unused) and fertiliser use (i.e., fertiliser is used or unused) at their fields in semi-arid northern Ethiopian Rift Valley. The data includes (i) a schematic diagram that represents local farmers' distinctions of the crop field types in terms of the distance from their houses and soil fertility and (ii) a table that describes a representative farmer's crop sequences and soil fertilisation methods in two consecutive years. Details about the literature review of the previous case studies on farmers' determinants of manure application technique adoption conducted in some parts of sub-Saharan Africa where cattle dung is used for manure are also summarized in a table. A table shows descriptive statistics of the independent variables used in the empirical analyses. Summary statistics of 4 binomial logit models and 4 multinomial logit models are indicated in a table, which represent model fit. Last two tables exhibited in this article show the logit analyses.


a b s t r a c t
This article contains the data on farmers' determinants of binary choices for manure use (i.e., manure is used or unused) and fertiliser use (i.e., fertiliser is used or unused) at their fields in semiarid northern Ethiopian Rift Valley. The data includes (i) a schematic diagram that represents local farmers' distinctions of the crop field types in terms of the distance from their houses and soil fertility and (ii) a table that describes a representative farmer's crop sequences and soil fertilisation methods in two consecutive years. Details about the literature review of the previous case studies on farmers' determinants of manure application technique adoption conducted in some parts of sub-Saharan Africa where cattle dung is used for manure are also summarized in a table. A table shows descriptive statistics of the independent variables used in the empirical analyses. Summary statistics of 4 binomial logit models and 4 multinomial logit models are indicated in a These data can be compared to the similar type of analyses conducted in other areas.

Data
This article includes a schematic diagram representing local farmers' distinctions of the crop fields in terms of the distance from their houses and soil fertility (Fig. 1), table that summarizes the literature review of the previous case studies on determinants of manure application techniques conducted in other parts of sub-Saharan Africa where cattle dung is used for manure (Table 1), table  describing a representative farmer's crop sequences and soil fertilisation methods in 2011 and 2012  (Table 2). Table 3 shows descriptive statistics of the independent variables used in the empirical analyses. The logit analysis data include the summary statistics of four binomial logit models and four multinomial logit models, which represent model fit (Table 4), variable coefficients of the two binomial logit models, which are selected as appropriate models in Table 4 (Table 5), and average marginal effects of the three multinomial logit models, which are selected as appropriate models in Table 4 (Table 6).    [18]. Thus main food and cash crops in each of the study areas were not described here.

Study area
Adama and Boset districts in Oromia region, Ethiopia, are classified into five agroecological subzones (tef zone, maize zone, semi-pastoral zone, sorghum and tef zone, and wheat and tef zone) [1]. Subsistence crop (sorghum, maize, and barley) and cash crop (tef, wheat, haricot bean, and vegetables) fields are mixed in all zones. The two districts are categorised into mid-altitude dry (MD) subarea and mid-altitude moist (MM) sub-area in terms of major maize growing areas in Ethiopia [2].

Sample
The following two-step procedures were used to select sample plots: in the first step, we set a goal to select 150 households from each maize growing sub-area. The target numbers of households were Table 4 Summary statistics of the four binomial logit models (A1, A2, A3, and A4) and four multinomial logit models (B1, B2, B3, and B4). LL, log-likelihood; χ 2 , Wald χ 2 ; Df, the degrees of freedom; R 2 , pseudo R 2 ; %, % correctly predicted; SSR, the sum of squared residuals.

Table 5
Variable coefficients of the two binomial logit models (models A1 and A2). Numbers in parentheses are standard error. In both models A1 and A2, variable fertiliser was omitted due to collinearity with the dependent variable. The analyses of models A3 and A4 were for reference. * Po 0.1. ** Po 0.05. *** Po 0.01. equally split by the number of sub-zones in each sub-area: four and two sub-zones in the MD and MM sub-areas, respectively, and were allocated to each sub-zone. Semi-structured questionnaires were prepared for interviewing randomly selected household heads in November and December 2012. After eliminating questionnaires with invalid data, we had data from 146 and 173 household heads living in the MD and MM sub-areas, respectively. It was found these 146 and 173 household heads had 313 continuous maize cropping fields (CMCFs; 151 for MD and 162 for MM sub-areas) and 302 other than maize cropping fields (OCMFs; 131 for MD and 171 for MM sub-areas). In the second step, we randomly selected 262 CMCFs (131 for MD and 131 for MM sub-areas) and 262 OCMFs (131 for MD and 131 for MM sub-areas) from these 313 CMCFs and 302 OCMFs to match the numbers of the plot data between CMCFs and OCMFs and between MD and MM sub-areas. The total number of the plot data became 524 (262 CMCFs þ262 OCMFs).

Empirical models
An preliminary field survey conducted in 2011 showed that the CMCFs (n¼ 262) had three fertilisation options: (i) no fertilisation (n ¼39), (ii) manure application (n ¼220), and (iii) fertiliser use (n ¼3), while the OCMFs (n ¼262) had three fertilisation options: (i) no fertilisation (n ¼4), (ii) fertiliser use (n ¼ 153), and (iii) both compost and fertiliser use (n ¼105). The following two empirical exercises were conducted by using different econometric models: (i) To analyse farmers' determining factors in binary manure use options (dependent variable, manure: 1¼ used, 0 ¼not used), two binomial logit models were formulated for CMCF and OCMF subdatasets (model A1 and model A2, respectively). Another two binomial logit models were created with and without variable crop (main cropping system to which the sample plot belonged; 1 ¼CMCFs, 0 ¼OCMFs) for the pooled dataset (model A3 and model A4, respectively); and (ii) To assess the farmers' determinants of four fertilisation options (dependent variable, fertilisation: 1 ¼no fertilisation, 2¼ manure application, 3¼ fertiliser use, 4 ¼both manure and fertiliser use), two multinomial logit models were built for the CMCF and OCMF subdatasets (model B1 and model B2, respectively). Another two multinomial logit models were formulated with variable crop (model B3) and without variable crop (model B4) for the pooled dataset. Hausman test or Small-Hsiao test [3] was conducted to verify the independence of irrelevant alternatives (IIA) hypothesis.
Independent variables selected in this study were based upon the literature on technology adoption studies of manure/fertiliser use (Tables 1 and 3). To select appropriate models for further analyses, indicators of the optimum model selection for logit models [4] including the log-likelihood, McFadden's pseudo-R 2 , Akaike's information criterion (AIC), Bayesian information criterion (BIC), and the % correctly estimated values were examined. To test the validity of the subsampling method, the sum of squared residuals obtained from the pooled dataset and subdatasets to test the equality of coefficients were compared between the models [5]. Stata 13.0 (StataCorp LP) was used for the empirical calculations.

Logit analyses
Summary statistics (Table 4) showed the indicators of the optimum model selection for logit models and the sum of squared residuals (SSR) obtained from the four binomial logit models (A1, A2, A3, and A4) and four multinomial logit models (B1, B2, B3, and B4). The binomial logit and multinomial logit analyses were shown in Tables 5 and 6, respectively.
Shigeki Kano (Osaka Prefecture University) for providing technical suggestions. Part of this study was financially supported by the Ministry of Agriculture, Forestry and Fisheries, Japan.