Comparing the technical efficiency of farms benefiting from different agricultural interventions in Kenya's drylands

Abstract

Farmers in Kenya's drylands have difficulty accessing farm production resources and in consequence farm productivity is low. It is therefore important to find strategies for improving access to these scarce resources to help farmers use them efficiently. This paper analyses and compares the technical efficiency of five groups of small farms affected by five different agricultural interventions. The aim of the study was to identify intervention strategies that significantly improve farm efficiency. Data envelopment analysis was used to compute farm-level average technical efficiencies for each of the intervention groups. The results showed that average technical efficiency was highest for the farms that had participated in an irrigation intervention. The findings suggest that the strategies promoted by this intervention, such as access to irrigation, inputs and markets, have the most significant effect on farm efficiency.

Keywords:

• efficiency
• DEA (data envelopment analysis)
• agricultural intervention
• drylands
• irrigation
• Kenya

1. Introduction

Since the 1990s, productivity has been declining on Kenya's small farms (Ouma et al., 2002). Nyangito et al. (2004) link the problem mainly to the Structural Adjustment Programmes (SAPs) undertaken at the time. They argue that the suspension of the then government-subsidised input and market supply services through the SAPs created a vacuum in the resource supply system. This is because the private sector, which was expected to take over the responsibility for service provision, was inadequately equipped for the task. Consequently, the use of productivity enhancing resources such as fertiliser, quality seeds and farm implements declined among small farms. Nyoro et al. (2004) add that the problem of access to resources was aggravated by the collapse of the government supported agricultural lending institutions. The resulting lack of credit eroded farmers' input purchasing power. Funding for agricultural research and extension was also reduced, and this led to a decline in the level of technical knowledge passed to farmers. According to Nyangito et al. (2004), the resource access problem was further compounded by the then declining performance of the national economy, which had inflationary effects on farm input prices. This increased the cost of production and reduced the technical efficiency of small farms.

In the drylands, the impact of the SAPs on agricultural productivity aggravated the effects of drought. Nyariki et al. (2002) report more frequent droughts at that time, while Oxfam (2006) adds that farms in the region experienced drought-related crop failure in three of every four years. The resulting low agricultural productivity had severe consequences for the food security status of households in the region (Nyariki et al., 2002; Muyanga, 2004).

To bridge the resource gap created by the SAPs, agricultural interventions in the dryland region increased in the 1990s.The aim of these interventions is in line with the country's food security policy, which is to achieve self-sufficiency in food at both the national and household levels (GoK, 1986, 2001, 2008). The policy is pursued under challenging conditions: besides the problem of the resource supply vacuum created by the SAPs, arable land is scarce. About 80% of Kenya's land mass is classified as dryland, which is unsuitable for crops (Nyariki et al., 2002). The remaining cultivable land is highly populated, and crop production for home consumption competes strongly with export crops (Odhiambo et al., 2004). Consequently, the growing demand for food has to be supplied from improved land productivity. In this light, it is extremely important that small farms, which supply over 80% of the country's cereal demand, use the land and other scarce resources efficiently.

Agricultural interventions focus variously on enhancing technological innovation, expanding the natural resource base, improving living conditions, boosting market infrastructure, or developing institutional capacity, policy and institutional reforms, or may integrate several of these activities. Technological interventions generate innovative production technologies such as drought resistant crop varieties. Interventions for expanding the natural resource base do so by, for example, introducing irrigation systems. Interventions to improve general living conditions provide services such as technical advice to help households increase their income by increasing productivity. Interventions that aim to boost market infrastructure can raise farm productivity by, for example, building bridges to connect isolated villages to local markets. And interventions to develop institutional capacity, policy and institutional reforms indirectly increase productivity by, for example, training agricultural personnel, offering financial support for outreach activities, and enhancing market access to inputs and outputs. Abdulai & Huffman (1998) stress the role of markets in production-resource allocation decisions at farm-level.

These interventions can help improve a targeted group's productivity (Van Rooyen et al., 2002) and hence raise farm efficiency. This study evaluated the effect of different agricultural interventions on farm efficiency. Evaluation of agricultural interventions in Kenya has been limited to impact analysis of single interventions aimed at improving either resource access, farm yields or incomes (e.g. Nzomoi et al., 2007). However, studies that examine whether the type of intervention affects farm efficiency are lacking. This study analysed and compared the technical efficiency of 191 farms that had experienced five different types of agricultural intervention. The aim was to identify the most effective intervention type. We used the data envelopment analysis (DEA) technique to analyse the farms' technical efficiency, and statistical comparison to reveal any significant differences between the average efficiency scores of the five groups of farmers, and between these groups and a control group.

2. Research design

The study used ex post evaluation, comparing ‘with’ and ‘without’ intervention scenarios to identify intervention effects. The ‘with’ groups received the intervention while the ‘without’ group did not (Ravallion, 2001). Most ex post evaluations are limited to a comparison of two groups; in this study we compared five. We compared the efficiency of each intervention group first with the control and then with the other four groups, to determine which intervention had the most impact.

2.1 Research setting

The research was carried out in Makueni, one of Kenya's 36 dryland districts (see Figure 1). We chose this district because many agricultural interventions were implemented there and because of its high vulnerability to food insecurity. In 2002, 70% of the district's households were food insecure (Wanjama, 2002), and two years later 74% were reportedly living in absolute poverty (Danida, 2004). The poverty was attributed to high dependence on agriculture. According to Danida (2004), 75% of the households' income was derived from subsistence agriculture, supplemented by non-farm self-employment, remittances and casual employment. Agricultural production in Makueni District is a mix of crops and livestock (Tiffens et al., 1994; Mbogoh, 2000). Crop production is rain-fed and limited to six months of the year, with at most two crop seasons, apart from patches of irrigated vegetables and fruits along the main rivers (Nyariki et al., 2002). Livestock are mainly low-yielding traditional zebu cattle and some sheep, goats and donkeys (Mbogoh, 2000).

2.2 Data and sample selection

The data were collected in two phases. The first was a preliminary survey of the district between September and December 2005, which yielded secondary, explorative data on interventions in the district. This information was important for selecting the interventions for evaluation. The second phase, between July and September 2006, involved selecting the samples and collecting the empirical data for impact analysis. Although longitudinal data provide more accurate information (Malhotra, 1999), technical constraints obliged us to choose a cross-sectional survey. The data were mainly quantitative, collected with the aid of structured questionnaires completed by the 191 sampled households.

We used a two-step sampling procedure, selecting first the five interventions and then the respondent households. In the first step, five interventions were purposively selected from a list of about 21 that had been operating in the district since the 1990s. The following criteria were used to ensure that the interventions were representative:

•	placement – the intervention had to be located in an agri-ecological zone that was vulnerable to food insecurity, and where another intervention had not been selected,
•	relevance – the intervention had to have a food security element, such as food availability, food access, or food consumption and use, as one of its objectives (European Commission, 1999, 2000), and
•	evaluability – the intervention had to be evaluable in the sense that technical personnel were available to help identify the participants and facilitate the data collection.

For each of the three criteria we used contingency scores as used by Bojo & Reddy (2002). Each intervention in each of the five locations was ranked on a scale of 1 to 3 – low, moderate and high (see Table 1). The scores for each intervention were summed and selection was made on the basis of the highest ranking score. The number of selected interventions was limited to five because of resource constraints.

Table 1: Ranking criteria for selection of intervention case studies

	Scoring criteria
	Low (1)	Moderate (2)	High (3)
Placement	LM2	LM3	LM4/LM5/LM6
Relevance	Single strategy, e.g. availability	Double strategy: availability, access	More than two strategies: availability, access, consumption
Evaluability	Technical personnel neither coordinated nor participated in the intervention	Technical personnel only participated in the intervention	Technical personnel coordinated the intervention
Note: LM2 to LM6 represent agro-ecological zones Low Moisture potentials 2 to 6, with decreasing potential for crop production (GoK, 1986).

In the second step of the sampling process we selected the households that had participated in the interventions. The household was used as the unit of analysis because it is the level at which farm production and intervention targeting occurs (Deaton, 1997). The households were randomly selected from a list of participants in each of the selected interventions. The sample size was estimated using Fleiss's (1981) formula for sampling (Dell et al., 2002). In total 191 households were sampled: 134 from the five interventions, an average of 30 per intervention, and 57 as a control from the households that had not participated in any intervention.

2.3 Empirical analysis

Efficiency is defined as the global relationship between all outputs and inputs in a production process (Rodriguez Diaz et al., 2004), and can be expressed in technical, allocative or economic terms (Seiford & Thrall, 1990). The focus of this study is technical efficiency, based on the work of Farrel (1957). This refers to the farm's capacity to produce the maximum feasible output using a given bundle of inputs, or to use the minimum feasible amounts of inputs to produce a given level of outputs (Seiford & Thrall, 1990).These two measures of technical efficiency are known as the output-oriented and input-oriented model, respectively (Hartwich & Oppen, 2006). The former aims to optimise outputs and the latter to minimise inputs (Rodriguez Diaz et al., 2004).

Technical efficiency can be analysed using parametric or non-parametric methods (Hartwich & Oppen, 2006), the commonly applied techniques being the stochastic frontier production function for the former and DEA for the latter. The production function is explicit where the output of a farm is a function of a set of inputs, inefficiency and random error (Seiford & Thrall, 1990). It has, however, the disadvantage of imposing an explicit functional form and distribution assumption on the data. In contrast, DEA does not impose assumptions about the functional form, and is hence less prone to misspecification error. However, since DEA cannot take account of statistical noise, the efficiency estimates may be biased if the production process is largely characterised by stochastic elements. DEA is also disadvantageous in that when there is no relationship between the inputs and outputs, each farm will be viewed as unique and fully efficient, resulting in the loss of discriminating power (Thiam et al., 2001). Despite these drawbacks, however, the technique has been shown to produce analytical results which are highly correlated to those of the production function (Thiam et al., 2001; Alene & Zeller, 2005). A further advantage is that the technique is widely applied in efficiency analysis (e.g. Binici et al., 2006; Speelman et al., 2008).

DEA is based on multiple-input, multiple-output relations in which linear programming is used to construct a non-parametric piecewise surface (frontier) around the data (Charnes et al., 1995; Echevaria, 1998). It tries to maximise the relative technical efficiency score of each decision-making unit by minimising input (the input-oriented efficiency model) or maximising output (the output-oriented efficiency model). Pareto efficiency1 is attained when no input can be reduced without reducing the output (the input-oriented efficiency model) or when no output can be increased without increasing the input (the output-oriented efficiency model) (Hartwich & Oppen, 2006).

DEA analyses technical efficiency under the assumption of either constant returns to scale (CRS) or variable returns to scale (VRS). Technical efficiency can further be decomposed into pure technical efficiency (TE) and scale efficiency (SE). TE is estimated when assuming VRS. The assumption of CRS leads to the estimation of total technical efficiency. SE is then the ratio between the total (CRS) and pure (VRS) technical efficiency (CRS/VRS). Estimation of SE assumes calculation of TE under both the CRS and VRS. If there is a difference between the efficiency scores under CRS and VRS for a farm, the farm is said to be scale-inefficient. A unity value indicates it is scale-efficient (Rajcaniova, 2004). If a farm becomes VRS-inefficient, it can increase its efficiency by increasing the scale of operation until it reaches the level where it is only inefficient under CRS conditions. From then on it can increase technical efficiency only by changing the technology (Fare et al., 1994).

The technical efficiency scores for each unit are analysed on a scale between technically inefficient (close to zero) and technically efficient (close to one) (Coelli et al., 2002). The scores are relative as they are based on the frontier whose estimation refers to the best performing farms in the data set and hence reflects the best existing practice (Hartwich & Oppen, 2006). The frontier thus allows us to compare one unit with another in the analysis (Haji, 2006; Raju & Kumar, 2006), assuming that it is feasible for other farms to attain this level of efficiency (Hartwich & Oppen, 2006). It was this comparability, and the relative insensitivity of DEA to misspecification of the functional form, that led us to choose DEA over the production function for application in this study. We used DEA to estimate the technical efficiency (CRS, VRS and SE) of the 191 farms participating in the five interventions. Our analysis was based on pooling the data of the five interventions so we could determine the efficiencies of the farms on the basis of the same frontier, and thus make comparisons. The efficiency of individual farms could be attributed to the effectiveness of the intervention in which the farmer participated. Arguably, efficient farms are responsive to incentives provided by interventions, because production factors move to activities where they earn the highest returns. Since technical efficiency reflects perfection in the market system (Sadoulet & De Janvry, 1995), efficient farms in this study suggest that the intervention has been effective in addressing imperfections in the markets for production resources.

Since the aim of the study was to identify the interventions which lead to efficient use of scarce production resources, the analysis was based on the DEA input-oriented models (Hartwich & Oppen, 2006). The output variable used in the analysis was average farm productivity calculated as all crop and livestock produced, valued at current market prices. The input variables were the primary factors which influence farm productivity in Kenya, namely the size of the household as proxy for labour, cultivated area for land size and farm expenses for capital (Odhiambo et al., 2004). Farm expenses comprised the average variable costs of the farm such as planting materials and other inputs. The DEA models were run in LIMDEP version 8 program (Greene, 2002).

3. Results and discussion

3.1 Description of the selected interventions

The selected interventions, which used different strategies to raise farm productivity, were the Makueni Integrated Agricultural Project (MAP), the International Crops Research Institute for the Semi-Arid Tropics Project (ICRISATP), the Community Based Nutrition Programme Project (CBNP), the Kenya Rural Enterprise Programme Bank (K-REP), and the Kibwezi Irrigation Project (Irrigation). The MAP, an initiative of the Danish Technical Cooperation (Danida) with the collaboration of the Government of Kenya, aimed to raise community living standards by improving access to the supply of water and to agricultural extension services. The ICRISATP, undertaken within the research framework of ICRISAT, aimed to improve food security by developing and disseminating drought resistant crop varieties. The CBNP, a joint venture of Danida and the Government of Kenya, aimed to improve the nutritional status of poor households through improved access to farm production resources such as seeds, ploughs, oxen and skills. The K-REP, another joint venture of Danida and the Kenyan Government, aimed to build the financial resource base of rural communities through savings and credit services in the form of village banks. Lastly, the Irrigation intervention, an initiative of the Israeli technical cooperation, aimed to raise community living standards by improving skills for irrigated commercial horticultural production technology. It did this using a demonstration farm, where local communities could be trained to produce irrigated horticulture for local and overseas markets.

The MAP, ICRISATP and CBNP interventions focused on raising the agricultural productivity of rain-fed subsistence farms through improved resource access and use. Training in farming skills could lead specifically to improved allocation of resources and farm efficiency (Nyariki & Thirtle, 2000). The K-REP, on the other hand, had an indirect approach to improving farm productivity. It provided credit mainly for non-farm income-generating activities. The expected improvement in household income could trickle down to investment in farm resources and hence raise farm productivity (Nyariki & Thirtle, 2000). The Irrigation project aimed to increase irrigated crop production through training on drip irrigation at the demonstration farm, technical advice, provision of inputs and market organisation.

3.2 General characteristics of the households

Table 2 summarises the characteristics of the studied households. The figures depict a rather homogeneous population with relatively low variability in the observed characteristics. In terms of gender, about 88% of the households were male-headed, with no significant mean difference between the beneficiaries of any of the intervention and the control group. The average age was 46 years, and only the means for the MAP and CBNP intervention participants were significantly higher than those of the control.

Table 2: Statistical comparison of means of farm household characteristics between intervention and control groups

Project	MAP (N = 29)	ICRISATP (N = 25)	CBNP (N = 28)	K-REP (N = 29)	Irrigation (N = 23)	Control (N = 57)
Gender (female = 1)	0.14 (0.35)	0.20 (0.41)	0.37 (0.47)	0.27 (0.45)	0.13 (0.34)	0.22 (0.42)
Age (years)	53.3 (12.3)	47.5 (9.6)	53.5** (15.6)	52.1** (13.1)	40.1** (9.3)	43.0 (10.2)
Education level (schooling years)	9.6*** (3.4)	7.6 (3.6)	6.0 (4.8)	6.9 (4.1)	7.7 (2.7)	6.9 (2.3)
Land holdings (acres)	5.2** (4.5)	4.1 (2.5)	3.3 (1.5)	7.5*** (3.5)	2.0*** (1.6)	3.6 (2.2)
Household size (adult equivalents)	4.2 (1.3)	4.6 (1.8)	5.7*** (1.7)	5.5** (1.5)	3.9 (2.1)	4.3 (1.8)
Farm expenses (Kshs)	13 031** (13 444)	1 920 (1 173)	8 205 (10 018)	15 344** (12 060)	21 523** (15 133)	3 783 (5 096)
Farm productivity (Kshs/acre)	15 687** (26 785)	7 541 (8 733)	7 501 (331)	11 600** (9 154)	97 101** (76 073)	6 795 (3 884)
Note: Standard deviation in parentheses. N = sample size, KShs = Kenya shillings (exchange rate at 78 per US$ in 2006). **, *** = mean difference significant at the 95% and 99% confidence levels, respectively.

The average level of education was 7.4 years of schooling, equivalent to completing primary education. The mean education level of the MAP participants was significantly higher than that of the farmers in the control group, and somewhat higher than in the other interventions. This may have something to do with targeting. Although participation was open to anyone, active participation may have been limited to the more educated because of the cost of the intervention. More educated farmers are more likely to participate in interventions with monetary costs, perhaps because they understand the associated benefits better (Nzomoi et al., 2007). The opposite was observed for the CBNP, whose participants had the lowest average level of education. This was not surprising, since the intervention had targeted the poorest households. These are usually the least educated, with the most household members (Nyangito et al., 2004), and indeed the farmers in the CBNP intervention sample had the largest average household size (5.7), the overall average being 4.6. The observations generally match those of Nyariki et al. (2002) and reveal a poorly educated, middle-aged farming community. The younger and more educated population have most likely migrated to urban centres in search of non-farm employment opportunities, which offer higher and more stable incomes.

Similar variability was observed in the farm characteristics. Land endowment was generally low, but highest for the K-REP participants and lowest for the Irrigation participants. The difference could be explained in terms of aridity and water access. The two locations are the driest and most sparsely populated in the district. Without the possibility of irrigation, households plant large areas to increase production. In irrigable areas, farmers can farm smaller areas effectively because of the irrigation, producing higher yields than the rain-fed areas. The mean farm expenditure was variable across the interventions. The values for the farmers participating in the MAP, K-REP and the Irrigation interventions were significantly higher than those of the control group.

The rather homogeneous household characteristics suggested that any differences in technical efficiency were most likely due to differences in the endowment of production and management resources, rather than variations in household characteristics. Any differences in technical efficiency could then be attributable to variations in resource endowment (seeds, fertiliser, credit, irrigation water, skills, markets, etc.), which are intervention determined (Ravallion, 2001).

3.3 Technical efficiency of the farms

The distribution of the mean technical efficiency scores of farms within the interventions is shown in Figure 2. The scores are relative as they indicate the gap that separates each farm's behaviour from the best productive practices within the data matrix (Haji, 2006). The average efficiency level was quite low, and the scores reflect a large variance in efficiency levels between the farms in the sample, and a high level of relative inefficiency for a large number of them. About 70% of the farms had a CRS and VRS technical efficiency between 0 and 0.2, and only 3.2% and 6.2% of the farms were fully technically efficient under the CRS and VRS, respectively.

Figure 1: Location of Makueni (arrowed) among the dryland districts of Kenya

Figure 2: Distribution of CRS (constant returns to scale) and VRS (variable returns to scale) technical efficiency scores

Observations across the different intervention groups are summarised in Table 3. The mean CRS and VRS efficiency scores for the groups were also remarkably low (columns two and three). The means for the sample were 0.156 and 0.217 for CRS and VRS, respectively, and the mean for each intervention group was below the sample mean. The only difference was in the means for the farmers in the Irrigation intervention (0.365 and 0.471 for CRS and VRS, respectively).

Table 3: Statistical comparison of means of technical efficiencies of interventions and control group, and number of efficient farms

Intervention	CRS technical efficiency	VRS technical efficiency	Scale efficiency: CRS:VRS	Number of efficient farms
MAP (29)	0.126** (0.101)	0.193** (0.147)	0.800 (0.175)	0
ICRISATP (25)	0.147 (0.129)	0.232 (0.251)	0.794 (0.216)	1
CBNP (28)	0.089** (0.069)	0.115** (0.109)	0.839 (0.189)	2
K-REP (29)	0.063*** (0.058)	0.076*** (0.056)	0.763 (0.201)	1
Irrigation (23)	0.365*** (0.351)	0.471*** (0.354)	0.770 (0.204)	8
Control (57)	0.149 (0.138)	0.283 (0.316)	0.726 (0.255)	2
Total (191)	0.156 (0.203)	0.217 (0.269)	0.790 (0.208)	14
Note: Standard deviation in parentheses. **, *** = statistical significance at 95% and 99% confidence levels, respectively.

The results agreed with those of Narayanamoorthy (2001), who also observed higher technical efficiency scores for irrigated farms than for non-irrigated ones. However, the values for the Irrigation participants were lower than those from other reports (e.g. Iraizoz et al., 2003; Reig-Martinez & Picazo-Tadeo, 2004; Alabi & Aruna, 2006; Bravo-Ureta et al., 2007; Speelman et al., 2008). As an example, Bravo-Ureta et al. (2007) found an average technical efficiency value of 0.74 for irrigated small farms.

A comparison of means was carried out on the mean technical efficiency scores of the interventions to determine whether they were significantly different from the control. Significantly higher mean technical efficiency scores indicated that the intervention had a positive impact. The results indicated by the asterisks in Table 3 show that the means for the farmers in the MAP, CBNP and K-REP interventions were significantly lower than that of the control (p-value <0.05). This suggested that the interventions did not have an impact on technical efficiency. Conversely, the significantly higher mean scores for the Irrigation intervention participants suggested there was a positive impact. We found no statistical difference in technical efficiency between ICRISATP participants and the control group, who did not participate in any intervention, which suggests that this intervention had limited impact.

An examination of the scale efficiency showed that the average for the participants in each intervention group was less than 1, and 0.790 for all the farms (see column four in Table 3). The observation indicated the need to adjust the scale of operation towards the frontier (Tahir et al., 2009). The low average efficiency scores for participants in each intervention means that there were a few farms that were fully efficient, but at the same time a large group of very inefficient ones. Indeed, in terms of CRS, VRS and scale efficiency (SE), only 5, 9 and 11, respectively, of the 191 households analysed were found to be efficient. The highest proportion of these efficient farmers was from the Irrigation intervention: 100%, 56% and 73% for the CRS, VRS and SE, respectively. From this observation it appears that participation in this intervention improved farm efficiency. This improvement could be attributed to resource access by the participants in the intervention. The Irrigation intervention was distinguishable from the other interventions on the basis of the resources which were accessible to its participants, in particular irrigation and a market for produce. Adequate availability of these resources is said to improve farm productivity in Kenya (Nzomoi et al., 2007).

The study compared the VRS-efficient and -inefficient farms according to resource use or access and other characteristics in order to show the factors that influence technical efficiency. The results are shown in Table 4. The efficient farms were mainly from the Irrigation intervention and had a significantly larger mean irrigated area (0.8 acres), smaller rain-fed area (2.8 acres), better access to quality seed (0.9 or 90%) and access to markets for produce (0.8 or 80%) than the other groups. Furthermore, the mean crop yield of the efficient farms was 2.5 times that of the inefficient farms, even though farm expenditure was not significantly different. This shows that access to irrigation water enabled the irrigated farms to optimise the use of resources, in particular land, and keep producing all year round.

Table 4: Statistical comparison of means of household and farm variables for VRS efficient and not efficient farms

	Variable means
Variable description	Inefficient farms (VRS < 1)	Efficient farms (VRS =1)	t-statistic (Z)
Age (years)	46.9 (12.4)	41.2 (13.7)	–1.27
Education level (years)	7.7 (3.5)	6.9 (3.4)	–0.73
Household size (adult equivalents)	4.6 (1.9)	4.2 (2.6)	–0.57
Irrigated land (acres)	0.52 (1.3)	0.80 (0.70)	–2.34**
Rain-fed land (acres)	4.4 (3.0)	2.8 (2.3)	–1.97**
Farm expenditure (KShs)	10 513 (12 351)	11 031 (12 570)	–0.063
Crop yields (KShs)	79 650 (260 374)	205 527 (17 509)	–2.96***
Salaried income (KShs)	16 913 (40 240)	3 600 (11 354)	–0.34
Business income (KShs)	16 513 (28 258)	13 200 (27 714)	0.957
Access extension services (0,1)	0.54 (0.50)	0.60 (0.50)	–0.35
Access quality seeds (0,1)	0.44 (0.40)	0.90 (0.30)	–2.79***
Access credit services (0,1)	0.36 (0.48)	0.50 (0.52)	–0.86
Access produce markets (0,1)	0.41 (0.49)	0.80 (0.42)	–2.39**
Note: Standard deviation in parentheses. (0, 1) denotes access level where 0 = inadequate access and 1 = adequate access. **, *** = statistical significance at 95% and 99% confidence levels, respectively.

The inefficient farms were mainly from the other interventions, MAP, ICRISATP, CBNP and K-REP. These cultivated rain-fed crops on larger land areas. The results imply that the specific strategies used by the interventions did not raise farm productivity sufficiently to have an impact on farm efficiency. MAP's focus on improving farming skills, soil and water conservation was thus not effective. The outcomes of ICRISATP's drought-resistant crop varieties, CBNP's nutritional education and K-REP's credit service were similar. The failure of the K-REP credit strategy to raise farm efficiency matches findings by Taylor et al. (1986) and Idiong (2007). Even though access to credit is expected to improve the use of farm inputs and hence productivity, Taylor et al. (1986) argue that provision of credit alone is insufficient to address the technological constraints facing the smallholder in order to raise farm productivity. It can be understood that with the frequent rainfall failure, farm efficiency is difficult to attain in this region.

Household variables, namely the age and education level of the head and the size of the household could also influence farm productivity and hence efficiency in Kenya (Nzomoi et al., 2007). However, the means of these variables did not differ significantly between the efficient and inefficient farm categories. This implies that the observed differences in farm efficiency were attributable to Irrigation intervention strategies (irrigation, quality seeds and markets), rather than household characteristics.

4. Conclusions

This study investigated the technical efficiency of small farms participating in five different agricultural interventions, to identify the best strategies for future intervention. We used DEA (data envelopment analysis) to compute the farms' technical efficiency. The results showed that the efficiency levels varied considerably. Many had a very low efficiency score, with about 70% being within a range of 0 to 20% efficiency. Further analysis showed that only the participants in the Irrigation intervention had a mean efficiency significantly higher than that of the control group, who participated in no interventions. This Irrigation intervention group also had a higher mean efficiency than the groups participating in the other four interventions.

The difference was attributed to the fact that the participants in the Irrigation intervention had better access to farm production resources and markets. We observed during the survey for the research that the Irrigation participants received widespread support in terms of input supply, technical advice from contracting companies, irrigation equipment, and good markets for their produce. The conditions were conducive to adjusting their production towards optimal levels. It can be understood that access to irrigation facilitates optimal production as the risk of drought-induced crop failure characteristic of the drylands is minimised. The better (mainly export) markets for the irrigated produce guarantee farm revenues several times higher than the average for rain-fed produce. This enhances farm productivity and efficiency.

This combination and level of resource endowment was not observed in the other interventions, MAP, ICRISATP, CBNP and K-REP, which aimed at improving farm skills; adoption of drought resistant seed varieties; household nutrition and health; and credit, respectively. This observation supports the claim by Ruthenberg (1980) that small farms do not perform well unless all critical resources are adequately provided, and highlights the need to synchronise resource supply in agricultural interventions. With climate change threatening farm productivity in the drylands (FAO, 2008), interventions are needed that can counter the negative effects. This study has shown that technical efficiency of farms would be improved by interventions that combine provision of irrigation with other complementary farm input resources (such as good quality seeds) and linking farmers to good produce markets.

The low efficiency that we found even within the Irrigation intervention group of farms calls for agricultural intervention to extend beyond enhancing irrigation, markets and inputs and engage farmers in resource management. Nationally, agricultural policy should encourage a change from subsistence farming towards entrepreneurship; that is, towards ‘farm-firms’ (farms functioning as for-profit entities), which are most likely to produce efficiently.

The study showed that agricultural intervention outside irrigable areas did not improve farm efficiency. It was clear that the productivity of non-irrigated farms was very low. Intervention should therefore consider shifting the focus from household farms to other productive activities. As Table 4 shows, the inefficient farms had higher non-farm incomes (such as salaries, or income from businesses) than the efficient farms, which were focused on irrigation. Fostering service employment in the Kenyan drylands should thus be a key priority in the region's development policy.

Even though our results implied that the Irrigation intervention had a positive impact on the participants' efficiency, the possible effect of confounding factors (gender, age and education level) could not be fully discounted in this analysis. Further analysis with regression models should be carried out in order to test and perhaps reinforce the conclusion of this study. Researchers should also undertake a cost-benefit analysis of the interventions described in this study.

Acknowledgements

The researchers appreciate the funding of the research work by the African Economic Research Consortium, Nairobi.

Notes

¹This is an economic state where resources are allocated in the most efficient manner.