2.1. Study Area The study was conducted in the northeast part of Bangladesh in May 2017. We purposively selected the north and northeast (Jamalpur and Gaibandha) of Bangladesh, since the agricultural produce is diverse and agricultural products are the primary source of income in these regions. Jamalpur is an important market center of the rice, sugarcane, jute, tobacco, and mustard produced in the region, whereas the main crops in Gaibandha district are paddy, wheat, jute, sugarcane, potato, brinjal, mustard seed, chili, onion, garlic, and vegetables. Secondly, the food insecurity and poverty rate is high in these regions. Thirdly, higher illiteracy rates, lack of irrigation facilities, low output price, labor scarcity, lack of proper knowledge about improved varieties, insect pest and diseases management, and weak research extension farmers’ linkages, etc. are other important factors affecting crop growth in these areas [105]. Finally, there is an established relationship between the researchers and the farmers; therefore, it was assumed that access to the relevant populations would be high. Furthermore, Rabi season (starting from November and ending around April) was chosen since during this season the farmers usually grow a variety of crops, including: Wheat, maize, boro rice, potato and sweet potato, mustard, sesame, lentil, brinjal, tomato, carrot, bottle gourd, country bean, chili, onion, garlic, coriander, cumin, sugarcane, tobacco, watermelon, etc.
2.2. Sampling Technique and Questionnaire Design We employed multistage sampling techniques for this study. There are 14 total subdistricts (Upazilas) in the Gaibandha and Jamalpur districts, and both districts are divided into 7 Upazilas equally. Therefore, a simple random sampling method was used to select four out of the seven Upazilas from each district. In the next stage, we selected twenty-two households from each Upazila, determining the sample size. Hence, sample sizes of 88 households (1 district × 4 Upazilas × 22 household heads = 88) were selected from one district. Likewise, a similar proportion of Upazilas and households were taken from another district. All study areas had 11 adopters and 11 non-adopters. Thus, finally, the sample number of two districts stood at 176. This study employed qualitative research techniques that involved the collection of data through semi structured personal interviews with farmers. This method (face-to-face, focus group discussion, and phone interviews) was effective for data collection, as it gave an opportunity for feedback between researchers and respondents. The interviewed respondents were engaged in agricultural work, with the majority of them being rice farmers and the rest producing vegetables. The questionnaire included household farmers’ demographic and socioeconomic condition, their adoption or non-adoption behavior, and their knowledge about and opinion regarding the service quality of soil testing and fertilizer recommendation facilities.
2.3. The Analytical Framework A great deal of the studies that investigated the adoption of a technology employed dichotomous choice data models (adopt or not adopt). Both logit and probit models can be used to assess the functional relationship between the probability of adoption and its determinants. Many studies used binary models to specifically analyze farmers’ adoption decisions on a single technology. This is considered the most suitable approach, as it provides more detailed information on the characteristics of farmers who would adopt a specific technology. For this specific study, we used the logit model to investigate the factors influencing the adoption of the soil testing and fertilizer recommendation facilities, as the adoption process itself is logistic in nature and is consistent with the literature on adoption. The theory that we considered, related to this study, is the threshold decision-making theory. Since the theory is related to farmers’ decisions of whether to adopt a technology or not, a reaction threshold that is dependent on a certain set of factors arises. As such, no adoption is observed when a certain value of stimulus falls below the threshold, while at the critical threshold value, a reaction is stimulated.
2.5. Data Analysis The Chi-square test was used to check the relationship between independent and dependent variables, at 95% confidence level (p < 0.05) using software for statistics and data science (STATA) version 14.0. The test is 2-tailed (non-directional), and in each case, the null hypothesis (Ho) states that there is no relationship between variables being tested, while the alternate hypothesis (Ha) states that there is a relationship. If the observed p was less than 0.05, the Ho was rejected and Ha accepted, and vice versa. Phi and Cramer’s V were measured using SPSS (Version 21) software to find out the strength of relationships, while Pearson’s test and likelihood ratio were used to compare the p-value to the rejection level when basic Chi-square assumptions were violated. Descriptive statistics and tables were used to present the distribution of results across variables. We conducted a logistic regression to show the factors most influencing the soil testing and fertilizer recommendation facilities in the survey areas. The variance-inflation factor (VIF) and tolerance were applied to test for multicollinearity among the independent variables using the “Collin” command in STATA software. Finally, the results of this analysis have been presented using frequency tables, cross-tabulations, and figures. STATA, SPSS, along with Microsoft Word and Excel (Version 2010) software were used for the data analyses.