To provide a more accurate assessment of the household and environmental factors associated with household use of inorganic fertilizer, we undertake econometric analysis to explore determinants of fertilizer adoption and use intensity. Limited dependent variables models are often used to evaluate farmers’ decision-making process concerning adoption of agricultural technologies. Those models are based on the assumption that farmers are faced with a choice between two alternatives and the choice depends upon identifiable characteristics . In adopting new agricultural technologies, the decision maker is also assumed to maximise expected utility from using a new technology subject to some constraints . In many cases a Probit or Logit model is specified to explain whether or not farmers adopt a given technology without considering the intensity of use of the technology. The Probit or Logit models cannot handle the case of adoption choices that have a continuous value range. This is the typical case for fertilizer adoption decisions where some farmers apply positive levels of fertilizer while others have zero application . Intensity of use is a very important aspect of technology adoption because it is not only the choice to use but also how much to apply that is often more important. The Tobit model of Tobin can be used to handle such a situation. However, the Tobit model attributes the censoring to a standard corner solution thereby imposing the assumption that non-adoption is attributable to economic factors alone . A generalization of the Tobit model overcomes this restrictive assumption by accounting for the possibility that nonadoption is due to non-economic factors as well. Originally formulated by Cragg ,drainage collection pot the double-hurdle model assumes that households make two sequential decisions with regard to adopting and intensity of use of a technology. Each hurdle is conditioned by the household’s socio-economic characteristics. In the double-hurdle model, a different latent variable is used to model each decision process.

The first hurdle is a sample selection equation estimated with a Probit model.It is important to first define what is meant by fertilizer adoption. For Probit estimation, a household is regarded as an adopter of fertilizer if it was found to be using any inorganic fertilizer. The dependent variable in this model is a binary choice variable which is 1 if a household used inorganic fertilizer and 0 if otherwise. For the second hurdle , fertilizer adoption becomes continuous and the dependent variable is the amount of fertilizer applied per acre of cultivated land by a household. There is no firm economic theory that dictates the choice of which explanatory variables to include in the double-hurdle model to explain technology adoption behaviour of farmers. Nevertheless, adoption of agricultural technologies is influenced by a number of interrelated components within the decision environment in which farmers operate. For instance, Feder et al. identified lack of credit, limited access to information, aversion to risk, inadequate farm size, insufficient human capital, tenure arrangements, absence of adequate farm equipment, chaotic supply of complimentary inputs and inappropriate transportation infrastructure as key constraints to rapid adoption of innovations in less developed countries. However, not all factors are equally important in different areas and for farmers with different socio-economic situations. In this section, we discuss the appropriateness of different variables considered in our model. The household characteristics deemed to influence fertilizer adoption in this study include household heads characteristics , household size and dependency ratio. The conventional approach to adoption study considers age to be negatively related to adoption based on the assumption that with age farmers become more conservative and less amenable to change. On the other hand, it is also argued that with age farmers gain more experience and acquaintance with new technologies and hence are expected to have higher ability to use new technologies more efficiently. Education enhances the allocative ability of decision makers by enabling them to think critically and use information sources efficiently. However, since fertilizer is not a new technology, education is not expected to have strong effects on its adoption.

The effect of household size on fertilizer adoption can be ambiguous. It can hinder the adoption in areas where farmers are very poor and the financial resources are used for other family commitments with little left for purchase of farm inputs. On the other hand, it can also be an incentive for fertilizer adoption as more agricultural output is required to meet the family food consumption needs . Institutional and infrastructural factors considered important in fertilizer adoption in this study include access to credit, farm size, presence of a cash crop, distance to fertilizer market, distance to extension service provider and distance to motorable road. The size of landholding is expected to be positively correlated with fertilizer adoption, as farmers with bigger landholding size are assumed to have the ability to purchase improved technologies and the capacity to bear risk if the technology fails . However, the well-documented tendency for management intensity to decline with scale in tropical Africa suggests that land size will be negatively correlated with the intensity of fertilizer use. Lack of access to cash or credit does significantly limit the adoption of fertilizer but the choice of appropriate variable to measure access to credit remains problematic. On a discussion on the limitations, challenges and opportunities for improving technology adoption using micro-studies, Doss outlines the different measures often used but cautions the inherent problems of these methods, especially their endogeneity.

Doss suggests that whether a farmer had ever received cash credit is a better measure of credit access than whether there is a source of credit available to the farmer. This study measures credit access by looking at whether a household received or did not receive any credit during a cropping year. The presence of a major cash crop 1 in the household is included in the model to capture the influence of commodity based inputs delivery systems in fertilizer adoption. In Kenya, commodities such as tea, coffee and sugar cane have inputs credit schemes for farmers. Because inputs markets are widely distributed, farmers face travel costs when they buy inputs. Since the volumes of fertilizer purchases by smallholder farmers are not high and the location of fertilizer market can be inconvenient,round plastic pot the cost of travelling to purchase fertilizer is probably fixed over the quantities purchased. The distance to fertilizer market is thus expected to affect decision on whether or not to use fertilizer, but not the intensity of use. Exposure to information reduces subjective uncertainty and, therefore, increases likelihood of adoption of new technologies . Various approaches have been used to capture information including: determining whether or not the farmer was visited by an extension agent in a given time; whether or not the farmer attended demonstration tests for new technologies by extension agents; and the number of times the farmer has participated in on-farm tests. Due to absence of such data for this study, we use distance to extension service provider to capture the influence of information on adoption. To explore the impact of infrastructure, which influences market access for both inputs and outputs, on fertilizer use, we include the distance to motorable road as a variable in the model. To measure the influence of agro ecological factors on fertilizer adoption, we include dummies for agro ecological zones. The high potential maize zone is used as the base. The Coastal, Eastern and Western lowlands and Marginal rain shadow receive less rainfall and are prone to prolonged and frequent dry spells compared to the Central and Western highlands, Western transitional and High potential maize zone. Agro ecology variables pick up variation in rainfall, soil quality, and production potential. These variables may also pick up variation unrelated to agricultural potential, such as infrastructure and availability of markets for inputs and outputs. A summary description of the explanatory variables used in the model is presented in Table 1.Generally, the proportion of sampled households using fertilizer rose from 64% in 1997 to 76% in 2007. However, these proportions vary considerably across agro ecological zones. The High Potential Maize Zone, Western Highlands and Central Highlands had the highest proportion of the households applying fertilizer. On the other hand, the proportion of households using fertilizer has remained relatively lower in the drier regions of Coastal Lowlands , Western Lowlands , Marginal Rain Shadow and Eastern Lowlands . A notable increase in the proportion of households using fertilizer in Western Transitional was observed; from 58% in 1997 to 88% in 2007.Trends in fertilizer use by cultivated land size are presented in Table 3. Landholding size is considered one of the indicators of wealth in Kenya. Two observations are made on the trends. First, across all the panel years the proportion of households adopting fertilizer increased with increasing cultivated land size. This may indicate that households with larger landholdings have greater ability to acquire and use fertilizer. Second, the proportion of households using fertilizer increased between 1997 and 2007 across all categories of cultivated land sizes.A more detailed analysis of fertilizer use on selected crops across the panel period is presented in Table 4. The number of households producing maize has remained high and about the same over the panel period, pointing to the importance attached to maize by the smallholder farmers.

The proportion of these households using fertilizer on maize consistently increased during the panel period from 57% in 1997 to 71% in 2007. On the contrary, the intensity of fertilizer application on maize has fluctuated between 55kg and 60kg per acre over the panel period. It is important to note that the application rates reported here are far below those recommended per acre for maize by the Kenya Agricultural Research Institute ; 50 kg of DAP and 60 kg of CAN, resulting to a total of 110 kg. The proportion of households applying fertilizer on coffee declined between 1997 and 2007 by 16%. Similarly, fertilizer application rate on coffee plummeted by 20% over the same period. A closer look reveals that the application rate consistently declined from 364 kg/acre in 2000 to 147 kg/acre in 2007, an average decline of 148% in a span of seven years. The gloomy picture in fertilizer use patterns on coffee can be attributed to two main factors: alleged mismanagement of coffee cooperatives, which are the main channels through which members receive their fertilizer; and the poor international coffee prices. Mismanagement in the cooperatives has made some farmers abandon coffee production while other farmers have opted to directly access fertilizers from private traders. This has made them disadvantaged in that they no longer access input credit facilities offered by the cooperatives as was the custom during the days when the cooperative movements were active and efficiently managed.With respect to tea, the fertilizer application rate has declined from 385 kg/acre in 1997 to 371 kg/acre in 2007. This decline is, however, marginal. The proportion of tea growing households using fertilizer on tea has, on the other hand, increased from 84% in 1997 to 98% in 2007. The fertilizer distribution system in the tea sector is the reason behind the impressive performance in fertilizer adoption on tea. The Kenya Tea Development Agency supplies fertilizer on credit to smallholder tea farmers and then deducts the cost plus interest from their deliveries of tea, which is sold by KTDA on behalf of the farmers. Fertilizer adoption on sugarcane over the panel period has showed an impressive increase. Households using fertilizer has grown from 29% in 1997 to 69% in 2007. However, the application rate has fluctuated over the study period. Increased fertilizer adoption in smallholder sugarcane farming can be attributed to provision on credit of fertilizer and other inputs to small holder cane farmers by the cooperatives to which the farmers belong. On the other hand, the dwindling fertilizer application rate can be attributed to inadequate supply of fertilizer by the cooperatives relative to farmers’ demand, or it may be as a result of farmers’ diversion of fertilizer acquired from the cooperatives from use on sugarcane to use on other crops. Ariga, et al., observed that some of the fertilizer acquired for intended use on the cash crops such as coffee and sugarcane under cooperative schemes is appropriate for use on maize and most horticultural crops as well, and there is likely to be some diversion of fertilizer targeted for use on sugarcane and coffee to food crops.